N014149CR

INSTITUTE of
HYDROLOGY
A N A SSESSM EN T O F TH E FSR RA I N FA LL- R U N O F F

M ETH O D O F DESI G N F LO O D ESTI M AT IO N
L'" •
INSTITUTEof HYDROLOGY
The Ins ti tute of Hydrolog y is a co mpo ne nt esta blis hme nt of the UK

Natural Environme nt Rese arch Co unc il, g ra nt-aided from Gove rnme nt
b y the Depa rtme nt of Ed uca tion and Scie nce . For over 20 years the
Ins titu te has b e en a t the fore front of rese arc h e xploration of hydrologica l
syste ms w ithin comp le te ca tchme nt areas and into the p hysical
p roc esse s b y which ra in or snow is transformed into fl ow in rive rs .
Applied stud ies, und ertake n both in the UK and ove rseas , e nsur es that
res e arc h activities are close ly re late d to p rac tical nee ds and tha t ne w ly
d eve lop ed me thod s and instrum e nts a re te ste d for a wid e range of
e nviro nme ntal cond itions.
The Institute , b ase d at Walling ford , e mp loys 140 staff, some 100 of whom
are g rad uates. Staff stru ctu re is multid isc ip linary involving p hysicists ,
geog rap he rs, g eo log ists, comp ute r sc ie ntis ts , mathema ticians , che mis ts ,
enviro nme ntal scientists , soil scie ntists a nd botan ists . Rese arch
de partments includ e ca tchme nt re searc h, re mote se ns ing,
instrume ntation, data p rocessing , ma the matica l mod e lling,
hyd rog eolog y, hyd rochemis try soil hyd ro logy evapo ration flux stud ie s,
vege tation-atmos phe ric inte ractions , flood and low-fl ow p red ictions ,
catc hrne nt respo ns e and e ng ine e ring hyd rology
The bud g et of the Ins titute comp rise s £4.5 million pe r year Abo ut 50
pe rc e nt re lates to re se arc h p rog rammes funded d ire ctly by the Natura l
Environment Rese arch Council. Exte nsive commiss ione d re se arch is
also carried out on be ha lf of gove rnme nt de partme nts (bo th UK and
ove rse as ), various inter nationa l ag e ncies, environme ntal org anisa tions
and p rivate se ctor clie nts The Institute is also re sp onsib le for
nationally archive d hyd ro log ical data and for pub lishing annually
HYDRO LOGICAL DATA: UNITED KINGDOM.
A N A S SE SS M EN T O F T H E FS R R A IN FA LL -R U N O FF M ET H O D O F
D ES IG N FLO O D E ST IM A T IO N
Jan uar y 29t h , 1990
0 a B oor man
M C A cr eman
J C Pac k man
I ns t i t ute of H y d r ol ogy
Wal l i ng f or d , Ox f or d shi r e
An assessment o f the FSR rain fa ll-runo ff method of design flood
est imation
•
• AN A SSESSMENT O F T HE FSR RA IN FA L L- R UNOFF MET HOD OF

FL OOD EST IMA T ION
0
411 D B Boorman , M C Acreman and J C Pack man

Institute o f Hyd ro logy , Wa llingfo rd , Ox fo rdshire .
• AB ST R AC T
411 The F lood Stud ies Repo rt , FSR , and its Supp lemen tary Reports prov ide
w ide ly used techn iques for d es ign flood est imation in UK ca tchments .
ID There has been cons ide rab le d ebate on the accu racy o f the va rious
methods , but few o f the object ions have been substantiated . Th is
repo rt desc ribes wo rk a imed at prov id ing autho ritative compa risons
be tween flood est ima tes derived from obse rved f lood da ta and
ID
- the o rig ina l FSR rainfa ll-runo ff method ,
the FSSR 16 ra in fa ll-runo ff method ,

411 the FSSR 16 me thod w ith observed da ta , and
• the FSR statistica l me thod (I equa tion p lus reg iona l g row th
curve) .
ID
The ana lys is was performed on a set o f 88 catchments wh ich had a t
411 least 15 years o f annua l max imum peak f low da ta (to gene rate the
obse rved flood frequency relationsh ips ), and d etailed ra in fa ll and
runo ff data d esc rib ing five o r mo re flood even ts (to p rov ide param-
eter est imates to rep lace those obta ined from catchment cha rac ter-
istics ) . Comparisons were made fo r a ll catchments and a ll retu rn
periods (2 , 5 , 10 , 25 and 50 years ) , fo r various subs ets o f
catchments , and fo r return pe riods be low a limit specified sepa rate ly
fo r each catchment . Resu lts show tha t estimates mad e us ing th e
40 statistica l me thod were unb iassed , wh ile the rainfa ll-runo ff me th-
ods , used w ithout cons idering hyd ro log ica l data recorded at the s ite ,
lb had a tend ency to overestimate . Th is b ias was reduced v irtua lly to
zero by inc lud ing observed da ta (particu larly percentage runo ff) .
411 The la rgest overest imates tend to be on catchm ents on re lative ly
pe rmeab le so ils . Res tr ict ing compa rison s fu rther to cons ider the
• retu rn per iods w ith in the spec ified lim its rend ered the estima tes
unb iassed .
ID
lb
An assessment o f the FSR rain fa ll-runo ff method of des ign flood
est imation
•
IV
An assessment o f the FSR rain fa ll-runoff method of design flood

estimation
41
41
41
41
41
41
41
• T A B LE O F C ONT ENT S
• 1.0 Int roduct ion
• 2 .0 Study catchmen ts
• 3 .0 Estima ting T-yea r flood peak s from observed annua l max ima 13
3 .1 Introduc tion . . 13
• 3 .2 Fitting a flood frequency d istr ibution . . . . 13
3 .3 Ob jective crite ria for se tting retu rn p eriod lim its 22
• 3 .3 .1 Standa rd erro rs 22
3 .3 .2 Goodness o f fit . 24
• 3 .3 .3 Comb ined standard errors and good ness o f fit 26
• 4 .0 Estimating T-year flood peaks by ind irect methods 29

4 .1 Introduction . . . . 29
• 4 .2 The rain fa ll- runoff me thod 29
4 .3 The sta tist ica l method 31
• 4 .4 Summa ry 31
• 5 .0 Resu lts 33
5 .1 In troduc tion . . . 33
• 5 .2 Performance sta tist ics . . . 33
5 .3 The standard set o f catchments and retu rn period s 34
• 5 .4 Compa r ison w ith resu lts from Lynn (1978 ) . 34
5 .5 Compa rison o f o rig ina l FSR and FSSR 16 m ethod s 36
41 5 .6 Use o f mode l pa rameters from flood even t data 36
5 .7 Resu lts from the standard set 37
• 5 .8 Resu lts from the fu ll set 39
5 .9 Exc lud ing catchment 39004 . . . . . 40
• 5 .10 Exc lud ing catchments underla in by WRA P type 1 soils 42
5 .11 Amount o f ava ilab le observed even t data 43
41 5 .12 Th e number o f annua l max ima ava ilab le 44
5 .13 Spa t ia l d istr ibution o f res idua ls 45
111 5 .14 Catchment s ize . 50
5 .15 Comb in ing subset se lection crite ria 51
• 5 .16 A ssess ing the entire flood frequ ency cu rve 55
41 6 .0 Examp le catchments 65
6 .1 In trod uction . . 65
41 6 .2 1900 1: A lmond at Cra ig ieha ll 66
6 .3 2900 1: Waithe Beck at Brigs ley 68
41 6 .4 390 12 : Hogsm ill at K ingston 71
41
• Tab le o f Con tents
41
41
41
vi
6 .5 4600 3 : Da rt at Austins Bridg e 73

6 .6 540 16 : Rod en a t Rod ing ton 76
6 .7 55008 : Wy e at Ce fn Brwyn 79
7 .0 Conc lus ions 83
8 .0 Recommenda tions 87
9 .0 Re ferences 89
Append ix A . Fu ll tab les of compa risons and resu lts 91
Append ix B . The FSR s imu lat ion exerc ise 103

B .1 Introduct ion . . . . 103
B .2 Reproduc tion o f flood frequency cu rves 103
8 .3 Cho ice o f a sing le set o f d esign inputs 104
Append ix C . Est ima tion o f som e extreme h istorica l UK f loods 107

C .1 No tat ion 108
C .2 Introd uc tion 109
C .3 S ign ifican t floods in Brita in 109
C .4 Ex trapo late o r interpo late? 114
C .5 Probab le max imum floods . . . . . 114
C .6 How can observed floods exceed PMF? 115
C .6 .1 Ra in fa ll input . 115
C .6 .2 Pe rcentage runo ff 117
C .6 .3 Un it hyd rograph . . . . 117
C .6 .4 Poo r est imation o f h isto rica l flood peak s 118
C .7 Conc lus ion 118
C .8 References 124
An assessment o f the FSR rain fa ll-runo ff method o f d esign flood

estimation
41
•
41
•
•
•
• L ist o f Illustrations
•
41
•
v iii
An assessment o f the FSR ra in fa ll-runo ff method o f des ign flood

estimation
0
0
0
0
0
LIST OF T A B LES
•
• Tab le 2 .1 Catchment cha racterist ics . 11
Tab le 3 .1 F lood quantiles . . . . . . . 20
• Tab le 5 .4 .1 Error statist ics : rain fa ll runo ff and sta tis ica l
method at 2 and 10 yea rs . . . . 35
• Tab le 5 .5 .1 E rror statist ics : FSR and FS SR 16 . . 36
Tab le 5 .6 .1 E rror statist ics : Use o f obs erved d ata 37
• Tab le 5 .7 .1 Error statistics : the standa rd catchment set . 38
Tab le 5 .8 .1 Erro r statist ics for a ll 88 catchments. . 39
• Tab le 5 .9 .1 E rro r statistics exc lud ing ca tchmen t 39004 . 42
Tab le 5 .10 .1 E rror statist ics exc lud ing SO IL type 1 catchm ents . 43
Tab le 5 .11 .1 Error stat istics fo r ca tchmen ts w ith >10 events . 44
• Tab le 5 .12 .1 Error statistics fo r catchm en ts w ith >25 an .max . 45
Tab le 5 .13 .1 Residua ls for 2-year flood us ing FSSR 16 . 48
• Tab le 5 .13 .2 Res idua ls for 10-yea r flood us ing FSSR 16 . 49
Tab le 5 .13 .3 E rror statist ics for g roups based on SAAR . 50
• Tab le 5 .14 .1 Error statistics for g roups based on AREA . 51
• Tab le 5 .16 .1 C lass ificat ion o f frequency cu rves (1). 58
Tab le 5 .16 .2 C lass ificat ion o f frequency cu rves (2). 60
Tab le 5 .16 .3 C lass ification o f frequency cu rves (3). . . . 62
• Tab le 6 .1 Estimates us ing urban me thod on ca tchment 390 12 . 71
Tab le A .1 Error statistics fo r standard set o f catchmen ts . 92
• Tab le A .2 E rro r statist ics fo r a ll 88 catchmen ts . . 93
Tab le A .3 Error statist ics exc lud ing ca tchment 39004 . 94
• Tab le A .4 E rror statistics exc lud ing SO IL typ e 1 ca tchments . 95
Tab le A .5 Error statist ics for ca tchmen ts w ith >10 events . 96
• Tab le A .6 Error statistics for catchments w ith >25 an .max . 97
Tab le A .7 Error statistics for g roups based on SAAR >800mm . 98
• Tab le A .8 Error statistics for g roups based on SAAR<800mm . 99
Tab le A .9 E rror statistics fo r g roups based on AREA< 100km 2 100
• Tab le A .10 E rro r statist ics for g roups based on AREA> 100km2 10 1
Tab le C .1 Peak d ischarge estimates fo r some documen ted
0 floods since 1795 . . 120
Tab le C .2 Catchment cha racterist ics fo r PMF estimat ion . 122
0 Tab le C .3 PHF est imates fo r 17 UK catchments . 123
L ist o f Tab les

x

estimation
•
•
xi
N O T A T IO N
AREA Topog raph ic d rainage area in km2

EDF Emp irica l D ist ribution Function
Non -exceedence probab ility
FSR F lood Stud ies Report
FSSR F lood Stud ies Supp lementa ry Repo rt
GEV G enera l Ex trem e Va lue
GEV shape pa rame ter
Mod ified Anderson -Da rling statist ic
ML Max imum L ike lihood
MSL Ma instream leng th in km
Numbe r o f catchments
Period o f reco rd in yea rs
O Observed non -exceedence probab ility
Exceed ence probab ility
R IF Probab le max imum flood
PWM P robab ility We igh ted Moments
M ean annua l flood
QT i Magn itude o f T year flood on catchment i
QT " Est ima ted va lue o f T year flood on catchment i
us ing method j
RMS Roo t mean squa re
RT 1.1 Relative erro r at retu rn per iod T on catchment i
using method j
Standard dev ia tion
SAAR Standa rd period annua l ave rage rainfa ll in mm
se Stand ard erro r
SOILn F raction o f ca tchment w ith so il (WRA P) type n
SPR Standa rd pe rcen tage runo ff
S l085 A ma inchanne l s lope measure in m/km
retu rn period
Tp Unit hyd rograph time to peak in hou rs
WRAP W inter Ra infa ll Accep tance Potent ia l (so i l classificat ion )
GEV locat ion parameter
URBAN F raction o f ca tchment cove r by urban area s
oC GEV sca le paramete r
Ch i-squared goodness o f fit statistic
No tation
41
x ii 41
41
Acrony ms fo r flood es timation method s
41
FSR/STATS(CC ) FSR statistica l method w ith ca tchment
cha racterist ics 41
FSR /RF-RO(CC ) FSR rainfa ll-runo ff method w ith catchment
cha racterist ics 41
FSSR 16(CC ) FSSR 16 ra infa ll-runo ff method w ith ca tchmen t
cha racteristics 41
FSSR 16(Tp& SPR ) FSSR 16 ra in fa ll-runo ff method w ith mod e l
pa ram eters Tp and SPR d erived from ob served data 41
FSSR 16(Tp ) FSSR 16 ra in fa ll-runo ff method w ith Tp
derived from observed d ata 41
FSSR 16(SPR ) FSSR 16 ra infa ll-runo ff method w ith SPR
d erived from observed d ata 41
DATA/S TATS Estima tes from fitting a d ist r ibution to observed
annua l max ima ; used as truth fo r assessment o f 41
othe r method s
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
An assessment o f the FSR rain fa ll-runo ff method o f. d esign flood 41
estimation
41
41
41
41
ID
ID
ID
41
ID
ID
1.0 INT RODUCT ION

ID
Estimat ing flood magn itudes of g iven probability or frequency o f
411 occurrence is an essential requirement for the design of drainage
systems , bridges, flood protection works and other river engineering
II schemes . If sufficiently long reco rds of river flow are ava ilable ,
the flood magnitude-frequency d istribution can be estimated d i-
II rectly . However , the majority of sites have litt le or no information
on previous flood flows , and the d istribution has to be estimated
ID indirectly
• The FSR 1 presents two indirect methods o f flood estimation , which

have been app lied to a large number of catchments throughout the UK ;
• firstly the statistica l method , in which observed flood peaks are
treated as random samples from some frequency distribution , and
• second ly the rainfa ll-runoff method , in which rainfa ll is treated
as the statistica l e lement and is converted to flow using a
II determin istic model of catchment response . In the ra infa ll-runo ff
method the estimated flood magnitude depends on several aspects of
• the input (eg . rainfall depth , duration , and pro file , and catchment
initia l condition), A simu lation exercise (FSR 1.6 .7, (437)) was
• undertaken to find a combination o f inputs that wou ld g ive a peak
flow o f the required return period . Append ix B describes this sim-
0 ulation exercise in some deta il.
• The statistical method estimates on ly peak flow , wh ich may suffice

for the design of culverts and bridges . However, the ra in fa ll-runo ff
• method synthesises the entire hydrograph and is therefore better
suited to the design o f flood storage or reservoir sp illways . The
411 rain fa ll-runoff method may a lso be preferred where the hydrolog ist
has a feel for the parameter va lues (eg . the percentage of rainfa ll
• that runs o ff as flood flow ). Adjusting such parameters to pred ict
the effect of catchment changes (eg . fie ld drainage , urbanisation )
• is intuitive ly easier than d irectly adjusting statist ical parame-
ters , such as mean annua l flood .
ID W ith both methods , the various mode l parameters were re lated v ia
mu ltip le reg ression equations to the physical characteristics o f the
•
catchments , enab ling flood quantiles to be estimated at sites without
ID
• Re ferences to the F lood Studies Report are made as FSR
vo lume .sectiOn(page) ; references to the F lood Stud ies Supple-
• mentary Reports (NERC ,1977-1987) are made as FSSRnumber .
Introduction
ID
flow (or rainfall-runo ff) data . These estimates might be improved
using observed data from (or loca l to) the site of Interest . The
methods have been upda ted a number of times ; of particu lar note here
is the revision o f the rainfa ll-runo ff mode l presented in FSSR 16 .
Considering the w ide app lication o f the methods , there have been few
studies comparing indirect flood est imates from the FSR methods with
values obtained d irect ly from flow data . One study, by Lynn (1978 ,
unpub lished ), compared the statistica l and rainfa ll-runoff method s
(amongst others) with such d irect estimates . He considered estimates
o f the mean annua l flood (for 82 catchments), and the 10 year flood
(for 39 catchments). However , he d id not cons ider the effects o f
using observed data to refine the ra infa ll-runoff method estimates .
Lynn found that the FSR statistical methcd underestimated the mean
annual flood by 6% overall (15% in catchments less than 100 km2),
wh ile the 10-year flood was underestimated by less than 1% overall.
The rain fa ll-runo ff mod el was more biassed , overestimating the mean
annual flood by 13% , and the 10-year flood by 56% . Moreover, it gave
a marked regiona l pattern o f errors , overpred icting in eastern En-
g land and underpred icting in south and south-west Eng land (a similar
pattern to that found in the simulation study , FSR 1.6 .7 .4(448)).
The somewhat d isappo inting performance o f the rainfa ll-runoff model

might be attributed to the design inputs derived from the simulation
study , or to model deficiencies d iscussed in the FSR 1.6 .5 .12(425),
FSSR3 and e lsewhere . Of these , the 5-fo ld classification o f so il
type (used to estimate the standa rd percentage runo ff) is perhaps
most cu lpab le , particu larly for c lasses 1 and 5 . For example ,
Boorman (1980) found that too low a percentage runoff is pred icted
from soil type 5 in north-west Eng land and that a soil type 6 might
be a more approp riate c lassification in some areas . Reynolds (198 1)
suggested that a similar prob lem existed in northern Scot land . A
recent study (Boorman et al, 1988) prov ides some support for this ;
small upland catchments in Scotland on soil type 5 were found to
y ield approximately 60% standard percentage runoff. The revised
parameter estimation equations presented by Boorman ( 1985 ) go some
way to rectifying the prob lems . These equations , summarised in
FSSR 16 , provide more runo ff from so il type 5 (SPR=53%) and reduce
SPR to 10% (from 15% ) on soil type 1. The requirement for even lower
SPR in some cha lk ca tchments (soil type 1) has been stated by Gurne ll
and M idgely (1987).
Errors in assessing response time or unit hydrograph shape may a lso

be to b lame for poor ra infa ll-runoff model performance . For examp le ,
Reed (1987) sugg ests that response time estimates for sma ll
catchments can be particular ly poor. FSSR16 and Boorman (1985 ) a lso
inc lude a more robust formulation o f the unit hyd rograph time-to-peak
equation , improv ing extrapo lation to sma ll catchments .
Archer (1980) examined the performance o f the statistical method in

north-east Eng land and found that sma ll-scale regiona l patterns in
An assessment of the FSR ra in fa ll-runoff method of design flood

estimation
41
• 3
41
mode l error could be identified . However , in a reply , Reran (1981)
• argued that the patterns might be apparent rather than real, and
could be due to samp le bias in the observed data caused by large
41 storms occuring over severa l catchments in an area . Archer and
Kelway (1987) used observed flood frequency data from 4 6 catchments
• in the Northumbrian Water Authority area to compare the FSR statis-
tical method w ith the FSR and FSSR16 rain fa ll-runoff methods . They
ID found that overall the statistical method overpredicted the mean
annual flood by 9 .5% , and the 30-yea r flood by 5 .5% . The ra infa ll
41 runo ff method underpred icted the mean annual flood by 4 .4% but
overpred icted the 30-year flood by 11.5% . A small-sca le regional
41 pattern of errors was found , similar to that for the statistical
method . The use o f observed rainfa ll-runoff data was not conside red .
41
Other users of the FSR methods have made informa l presentations at
41 meetings and conferences suggesting unhapp iness w ith certain aspects
of the methods but these are frequently anecdotal and cannot be
41 referenced or investigated .
• This report aims to
41 - provide definitive comparisons o f the various FSR techniques

aga inst observed data ,
40 test the effect of includ ing various amounts of observed data ,
41 identify classes o f catchment where the current ra in fa ll-runoff

method may be deficient (eg . size , urbanisat ion or soil type) .
41
The criteria adopted in selecting the catchments used in the report
41 are described in the next section . The derivation of floods o f re-
turn period 2 , 5 , 10 , 25 , 50 and 100 years is described in Chapter
• 3 , a long w ith consideration of the maximum return period to which
each catchment 's observed flood frequency can be taken . , A de-
ID scription of seven separate combinations o f method s and observed data
to be stud ied are g iven in Chapter 4 . Chapter 5 describes the re-
• su lts , using d ifferent mode l and catchment subsets, while Chapter 6
presents results for a subset of six test catchments. Conc lusions
41 and recommendations are given in Chapters 7 and H .
Appendix A contains fu ll tables o f resu lts presented in summary form

41 in the body of the report . Append ix 13 describes the FSR simulation
exercise to de fine the combination o f inputs requ ired for - the
41 rainfall-runoff model to produce floods o f des ignated return period .
Append ix C contains details of a supplementary study which compares
• estimates of the p robable maximum flood , PM? , using the FSR
rainfa ll-runo ff method , with maximum recorded historical floods in
• the UK .
41
41
41
41
Introduction
41
41
41
41
4
An assessment of the FSR ra infall-runoff method of design flood

estimation
2.0 STUDY CAT CHMENT S
Catchments selected for this study had to pass four stages o f qua lity
control. Each catchment must :
have a re liable estimate of the true flood frequency curve

aga inst which to compare the various methods ,
111
2 be suitable for application o f the FSR procedures ,
3. have numerous ana lysed individual flood events as a basis for

the assessment of the usefulness o f observed data , and
4 . have a stable and natura l hydro log ical reg ime .
The true flood frequency d istribut ion for any catchment is o f course
unknown . The quantiles of the d istribution may best be estimated
from observed sequences of recorded data . The precision of each
estimate depends primarily on the length of record , a lthough other
factors such as the accuracy of flow measurement are c learly in flu-
ential . FSR 1.2 .11.2 recommends that on ly floods up to return pe-
riods of 2N years , where N is the number of years of record , should
be estimated d irectly from at-site annua l maximum data .
ID Figure 1 (from Lees , 1987) shows the number of stations w ith a g iven
length of record held on the UK Surface Water .A rchive . The longest
record is 105 years on the R iver Thames at Tedd ington . Therefore ,
even at this site only floods up to a max imum return period of 210
years should be estimated d irectly from the flood data . The original
di m of this study was to estimate floods up to the 100 year return
period ; unfortunately there are on ly 12 stations in the UK whose
records exceed 50 years . To obtain a more comprehensive and repre-
40 sentative data set, the criterion was relaxed to include a ll
catchments w ith at least 15 annua l maximum flood events .
•
The FSR rainfa ll-runoff method assumes that the rainfa ll is evenly
411 d istributed over the catchment , therefore it is recommended for ap-
p lication on drainage areas up to 500 km2 . A ll but one o f the
catchments used in this study were smaller than this; the exception
being the Usk at Llandetty which has an area o f 544 km2.
In order to test the utility o f observed data when using the

rainfa ll-runoff method , the number of catchments emp loyed in the
study was further restricted to inc lude only those for which esti-
mates o f the unit-hydrograph mode l parameters had been derived for
Study catchments
ID
•
•
•
280
240
200
.061 120
z
80
40
0
0-10 11-20 21-30 31-40 41-50 51-60 61-70 71-80 81-90 91-100 100+
N um ber of years
F igure 1. Record leng ths for Sur face Water Archive sta tions
at least five separate flood events . As pa rt of the rev iew o f the

FSR rain fa ll-runo ff mod el pa rameter est ima tion equation s (Boo rman ,
1985) , rainfall and runo ff d ata were co llated fo r 210 catchmen ts ,
o f wh ich 128 had five or more events . N inety-one o f these a lso had
at least 15 yea rs o f annua l max ima .
Wh ile thes e data had a lready been checked when co llected , it was
cons id ered impo rtant to review their suitab ility for u se in this
study . Fo r examp le , if a catchment had good qua lity even t da ta for
flood s rema in ing in-bank , but the annua l max imum flood se ries con-
ta ined many ou t o f bank flows , it wou ld be unsuitab le for th is study .
At th is stage o f va lidation two stations were rejected from the

study . Th e flow reco rd for the R iver Is le at Ash fo rd M ill gaug ing
station show s that the h ighest five reco rded annua l max imum floods
were a ll o f s imilar magnitude (F igu re 2) . At th is s ite a large
vo lume o f flood wa ter is sto red on the flood-p la in imm ed iate ly up-
stream o f the station thereby a ttenua ting flood peaks . The Ouze l
at M illen was not inc luded since flood s a re g ross ly attenua ted by
the W illen Lak e as pa rt of the flood a llev iation schem e fo r H ilton
Keynes . F lows for the Tyne at Tarset a re now in fluenced by the
K ielder reservo ir , wh ich was comm iss ioned in 1980 ; hence on ly da ta
up to 1979 we re used . For the same reason data fo r the Brenig a t
Pont-y -Rhudd fa were restricted to the per iod 1922 -1974 .
An assessment o f the FSR rain fa ll-runo ff method o f d es ign flood

estimation
•
•
•
R e d u c e d v a r ia t e Y
•
F igu re 2 . F lood frequency ana lys ts for the Is le at A shford
•
• Study catchments
0
og
44, 7
$(4 0. 6
<0
Figure 3. Locations of catchments used in this study
An assessment o f the FSR ra infall-runoff method of design flood

estimation
• 9
•
Once unsuitable stations we re rejected the number ava ilab le was re-
411 duced to 88 ; the geographica l location of each is shown in
Figu re 3 . Unfortunately , for some areas o f the country , no
catchments were available for use . Therefore the sample set is not
entirely representative o f a ll types o f catchment upon wh ich the
• method s may be used in practice . In particular there are no
catchments north of the High land Boundary fault , in the Lake D istrict
or the Southern Uplands o f Scot land . There are , however , a number
of catchments from upland Wa les , urbanised south-east Eng land and
ID low- lying East Ang lia .
ID The range o f physical characteristics encompassed by this sample of
catchments is depicted in the histograms shown in Figure 4 and listed
411 in Tab le 2 .1 It can be seen that the sample contains a good range
of d rainage areas up to 400km2 , there being on ly 8 larger catchments ,
• and that about 75% of mainstream slopes (81085) are be low 10m/km .
For over half of the catchments , the proportion of the drainage area
411 urbanised is less than 0 .04 . A lthough not shown diagramatically ,
few of the catchments have more than 1% of their area d raining
through a lake . These d iag rams do not , however , disp lay the rang e
of combinations o f d ifferent characteristics, but there are certain
comb inations of characteristics which are more common than others .
Sma ll catchments tend to be steep, wet cåtchments tend to have
impermeab le soils , and urban ised catchments tend to be in areas of
low average rainfa ll and to have permeable so ils .
40
•
ID
ID
ID
111
ID
Study catchments
411
40
10
ID
F igu re 4 . H istog rams show ing d istribu tion o f catchment cha rac -
teristics

estimat ion
ID
11
No AREA NSL 5 1085 SAAR SO IL1 SO IL2 SO IL3 S0IL4 SO ILS URBAN
1900 1 369 .0 42 .0 5 .8 1 9 14 0 .000 0 .000 0 .000 0 .800 0 .200 0 .110

19002 43 .8 17 .9 5 .06 1024 0 .000 0 .000 0 .000 1.000 0 .000 0 .0 70
19005 229 .0 28 .2 6 .87 980 0 .000 0 .000 0 .000 0 .640 0 .360 0 .100
2000 1 30 7 .0 3 1.9 6 .08 736 0 .050 0 .000 0 .220 0 .720 0 .020 0 .020
23005 284 .9 36 .3 4 .85 1322 0 .000 0 .000 0 .000 0 .000 1.000 0 .000
24005 178 .5 3 1.7 6 .39 752 0 .000 0 .000 0 .000 0 .990 0 .0 10 0 .050
2400 7 44 .6 11.9 14 .89 797 0 .000 0 .000 0 .000 0 .930 0 .0 70 0 .000
2700 1 484 .3 84 .6 2 .54 975 0 .0 10 0 .000 0 .000 0 .690 0 .300 0 .020
27035 282 .3 3 1.7 4 .4 7 1134 0 .000 0 .000 0 .000 0 .530 0 .470 0 .020
28070 9 .1 4 .2 35 .95 985 0 .000 0 .100 0 .000 0 .000 0 .900 0 .000
2900 1 108 .3 20 .2 3 .33 729 0 .8 10 0 .000 0 .0 10 0 .190 0 .000 0 .000
29004 54 .7 12 .1 1 .98 630 0 .5 10 0 .080 0 .000 0 .420 0 .000 0 .000
3000 1 297 .9 46 .8 2 .13 625 0 .4 10 0 .040 0 .000 0 .550 0 .000 0 .020
30004 6 1 .6 15 .1 3 .26 697 0 .000 0 .000 1.000 0 .000 0 .000 0 .000
31005 4 17 .0 55 .4 1.44 643 0 .020 0 .000 0 .000 0 .980 0 .000 0 .0 10
330 14 272 .0 29 .9 2 .24 609 0 .550 0 .000 0 .450 0 .000 0 .000 0 .020
33029 98 .8 7 .0 1.63 637 0 .750 0 .250 0 .000 0 .000 0 .000 0 .000
33045 28 .3 7 .8 3 .26 627 0 .000 0 .000 1.000 0 .000 0 .000 0 .000
34003 164 .7 22 .4 2 .05 686 1.000 0 .000 0 .000 0 .000 0 .000 0 .000
34005 73 .2 22 .7 1 .73 647 0 .080 0 .000 0 .920 0 .000 0 .000 0 .000
35008 128 .9 14 .6 3 .4 1 606 0 .000 0 .100 0 .900 0 .000 0 .000 0 .020
3600 8 224 .5 37 .9 1.71 606 0 .000 0 .0 70 0 .930 0 .000 0 .000 0 .0 10
3700 1 303 .3 62 .6 1.22 6 10 0 .000 0 .000 0 .800 0 .200 0 .000 0 .100
3700 7 136 .3 26 .9 1.85 606 0 .000 0 .030 0 .900 0 .070 0 .000 0 .130
3800 7 2 1.4 5 .6 7 .47 6 11 0 .000 0 .300 0 .700 0 .000 0 .000 0 .290
39004 122 .0 2 .4 4 .36 764 0 .90 0 0 .100 0 .000 0 .000 0 .000 0 .390
39005 43 .6 7 .4 2 .28 640 0 .050 0 .700 0 .000 0 .250 0 .000 0 .8 10
3900 7 354 .8 32 .3 0 .98 710 0 .250 0 .140 0 .350 0 .260 0 .000 0 .330
390 12 69 .1 11 .8 3 .73 679 0 .250 0 .190 0 .0 10 0 .550 0 .000 0 .460
3902 2 164 .5 22 .1 1.62 751 0 .350 0 .000 0 .000 0 .650 0 .000 0 .020
39025 14 7 .6 23 .2 3 .20 798 0 .060 0 .090 0 .000 0 .850 0 .000 0 .0 10
39026 199 .4 27 .9 2 .10 700 0 .060 0 .000 0 .000 0 .940 0 .000 0 .020
3905 2 50 .2 11.0 3 .51 687 0 .000 0 .00 0 0 .200 0 .800 0 .000 0 .180
3905 3 89 .9 14 .6 2 .25 825 0 .000 0 .000 0 .000 1.000 0 .000 0 .090
40006 50 .3 13 .5 6 .20 733 0 .550 0 .060 0 .000 0 .380 0 .000 0 .030
40009 136 .2 19 .4 3 .24 808 0 .000 0 .000 0 .000 1.000 0 .000 0 .0 10
400 10 224 .3 30 .9 1.58 775 0 .140 0 .000 0 .110 0 .750 0 .000 0 .030
4 1005 180 .9 26 .7 2 .10 835 0 .000 0 .000 0 .000 1.000 0 .000 0 .040
4 1006 87 .8 16 .4 3 .99 837 0 .000 0 .00 0 0 .000 1.000 0 .000 0 .020
4 1007 403 .3 44 .8 1.50 755 0 .040 0 .000 0 .000 0 .960 0 .000 0 .0 10
4 10 15 58 .3 7 .3 4 .69 959 0 .840 0 .100 0 .000 0 .060 0 .000 0 .000
4 1028 24 .0 10 .0 4 .88 842 0 .000 0 .000 0 .000 1.000 0 .000 0 .0 10
45002 42 1.7 48 .1 5 .70 1420 0 .000 0 .8 70 0 .000 0 .020 0 .120 0 .000
45003 226 .1 26 .4 6 .15 996 0 .470 0 .020 0 .000 0 .5 10 0 .000 0 .000
Tab le 2 .1 Ca tChment cha racte rist ics .
Study catchmen ts
12
No AREA MSL 5 I085 SAAR SO IL 1 80 IL2 50 IL3 SO IL4 SO ILS UR BAN
45004 288 .5 33 .6 3 .58 1052 0 .500 0 .000 0 .150 0 .350 0 .000 0 .0 10

46003 247 .6 35 .2 16 .50 1696 0 .260 0 .240 0 .000 0 .000 0 .500 0 .000
46005 2 1.5 11 .8 22 .60 1987 0 .000 0 .000 0 .000 0 .000 1 .000 0 .000
4 700 7 54 .9 16 .6 17 .80 1477 0 .140 0 .500 0 .000 0 .000 0 .360 0 .000
48004 25 .3 10 .0 17 .48 1533 0 .000 0 .250 0 .000 0 .000 0 .750 0 .000
48005 19 .1 7 .2 13 .10 1121 0 .000 1.000 0 .000 0 .000 0 .000 0 .060
52005 202 .0 3 7.3 5 .60 993 0 .180 0 .4 70 0 .000 0 .350 0 .000 0 .060
52006 2 13 .1 16 .7 5 .50 907 0 .300 0 .000 0 .450 0 .250 0 .000 0 .050
520 10 135 .2 20 .4 4 .68 881 0 .070 0 .020 0 .630 0 .290 0 .000 0 .000
53005 14 7 .4 24 .6 3 .00 972 0 .600 0 .000 0 .100 0 .300 0 .000 0 .050
53007 26 1 .6 27 .7 2 .30 966 0 .270 0 .030 0 .350 0 .350 0 .000 0 .020
5300 9 72 .6 16 .1 8 .15 10 18 0 .570 0 .000 0 .430 0 .000 0 .000 0 .0 70
5400 4 262 .0 28 .8 1.92 69 1 0 .030 0 .000 0 .000 0 .970 0 .000 0 .250
54006 324 .0 35 .6 3 .0 7 70 1 0 .380 0 .140 0 .080 0 .400 0 .000 0 .140
540 11 184 .0 26 .9 4 .85 675 0 .420 0 .000 0 .0 10 0 .570 0 .000 0 .030
540 16 259 .0 40 .2 0 .92 713 0 .500 0 .030 0 .000 0 .470 0 .000 0 .000
540 19 34 7.0 56 .7 1.40 692 0 .300 0 .000 0 .700 0 .000 0 .000 0 .040
54022 8 .7 4 .7 63 .70 2235 0 .000 0 .000 0 .000 0 .000 1 .000 0 .000
55008 10 .6 5 .4 47 .44 240 1 0 .000 0 .000 0 .000 0 .000 1 .000 0 .000
550 12 244 .2 36 .0 7 .98 1643 0 .000 0 .630 0 .000 0 .000 0 .370 0 .000
5600 3 62 .1 20 .2 9 .02 1260 0 .000 0 .780 0 .000 0 .000 0 .220 0 .000
5600 4 543 .9 48 .7 4 .58 1488 0 .000 0 .600 0 .000 0 .000 0 .400 0 .020
56005 98 .1 25 .4 14 .23 1469 0 .000 0 .250 0 .300 0 .000 0 .450 0 .160
5600 6 183 .8 22 .4 8 .87 166 1 0 .000 0 .530 0 .000 0 .000 0 .470 0 .000
57004 106 .0 25 .8 7 .30 1759 0 .000 0 .000 0 .300 0 .000 0 .700 0 .040
5 7005 454 .8 42 .3 9 .23 1863 0 .000 0 .000 0 .400 0 .000 0 .600 0 .050
5800 1 158 .0 20 .1 10 .33 183 9 0 .000 0 .140 0 .430 0 .000 0 .430 0 .040
5800 2 190 .9 28 .3 13 .50 198 1 0 .000 0 .000 0 .100 0 .000 0 .900 0 .0 10
60002 297 .8 50 .0 4 .56 163 7 0 .000 0 .630 0 .000 0 .000 0 .370 0 .000
6 100 1 197 .6 27 .6 3 .24 1282 0 .000 0 .950 0 .000 0 .000 0 .050 0 .000
6 1003 3 1.3 9 .4 25 .4 7 1474 0 .000 1.000 0 .000 0 .000 0 .000 0 .000
6400 1 4 71 .3 37.5 5 .22 1836 0 .000 0 .500 0.000 0 .000 0 .500 0 .000
6500 1 68 .6 15 .2 33 .55 3030 0 .000 0 .020 0 .000 0 .000 0 .980 0 .000
660 11 344 .5 29 .0 17 .20 2 162 0 .000 0 .5 10 0 .000 0 .000 0 .490 0 .000
67003 20 .2 7 .1 13 .80 1308 0 .000 0 .000 0 .000 0 .000 1.000 0 .000
67008 227 .1 45 .8 4 .9 7 90 1 0 .150 0 .700 0 .000 0 .100 0 .050 0 .040
68006 150 .0 30 .9 10 .03 1053 0 .110 0 .080 0.000 0 .490 0 .320 0 .020
690 27 150 .0 4 1.4 5 .62 1179 0 .000 0 .000 0 .000 0 .4 10 0 .590 0 .220
71003 10 .4 5 .2 37 .80 1792 0 .000 0 .000 0 .000 0 .000 1.000 0 .000
71004 316 .0 37 .1 5 .02 12 11 0 .000 0 .000 0 .000 0 .710 0 .290 0 .090
72002 275 .0 34 .2 7 .74 125 1 0 .000 0 .070 0 .000 0 .460 0 .470 0 .0 10
72002 495 .0 53 .4 3 .69 1497 0 .000 0 .000 0 .340 0 .000 0 .670 0 .000
84008 5 1 .3 18 .9 13 .45 1187 0 .000 0 .000 0 .140 0 .750 0 .120 0 .260
840 12 227 .2 6 1.2 6 .62 1264 0 .000 0 .000 0 .320 0 .500 0 .180 0 .270
Tab le 2 .1 (Con tinued ) Catchment cha racteristics .
An assessment o f the FSR rain fa ll-runo ff method o f des ign flood

est imation
•
• 13
•
•
•
•
•
3 .0 EST IMAT ING T - YEAR FLOOD PEA KS FROM OBSERV ED ANNUA L
• MAX IMA
3.1 INT RODUCT ION

41 As described in the last section , the catchments chosen for ana lysis
each had at least 15 years of observed annua l maximum flood data .
41 Th is section describes how floods of various return periods were
estimated from these data , becoming the truth agains t which other
41 methods of estimation wou ld be compared . It had been intended to
make such comparisons over a wide range of return pe riods , namely
41 2 , 5 , 10 , 25 , 50 and 100 years. However, inspection o f the observed
data for many catchments suggested that although va lues for the
41 longer return periods m ight be adequate as best estimates fo r engi-
neering design at that s ite, they were not reliab le enough to be used
10 as a basis for comparing other estimation methods. Thus , for each
catchment , an upper lim it on T was chosen by visual inspection o f
41 the p lotted annual maximum data and fitted frequency curve . These
IIeyeba ll" limits o f trustworthiness have been applied for most o f
41 the comparisons in Chapter 5 . More objective ways o f defin ing the
limit were a lso investigated , including those based on goodness-of-
41 fit and estimation errors . A lthough these techn iques were not gen-
erally successfu l, they d id improve consistency i n the eyeball
ii classification by focussing attention on those catchments where the
techniques showed greatest d iscrepancy .
41 3.2 F IT T ING A FLOOD FRE UENCY DIST RIBUT IO
The FSR presents a number o f procedures for estimating T-year floods

from annua l maximum data ; the choice o f method depends on the re-
• quired return period , T , and the number o f years of data , N (see FSR
1.2 .11.2(243)). Thus , with N between 10 and 24 , floods of return
period up to 2N shou ld be estimated using the EV 1 (Gumbe l) d istrib-
ution (fitted by the method o f maximum likelihood); with N of 25
or more , the GEV distribution shou ld be used (aga in fitted by maximum
likelihood ) . In each case , floods of return pericd beyond 2N shou ld
be found by scaling the observed mean annual flood , q, by the ap-
propriate regiona l growth factor . FSSRs 13 and 14 contain mod ifi-
cations to this advice that affect the b lending of flood frequency
curves derived from data with reg iona l and nat ional g rowth curves .
41 Firm guidance on how to assess the qua lity o f derived flood frequency
curves is not given ; users are encouraged to use several methods ,
41 including examination of the p lotted data .
41
Estimating T-year flood peaks from observed annual maxima
5
•
14
•
•
The method s o f estimating T-year floods adopted in this report depart
from these recommendations in two ways . ID
Firstly , Hosk ing et a l (1984) have suggested that more stable esti-
mates of flood frequency are obtained when distributions are fitted
by the method o f probab ility we ighted moments (PWM). For the GEV ID
d istribution , the probability , F , of an annual maximum value , x ,
being less than any va lue , X , is g iven by : 411
F(x<X ) = exp( - [ 1 - k( u ) /0,( )1/k) (3 .2 .1)

ID
This quantity F is usua lly called the non-exceedence probability .
Hosk ing et a l give the fo llow ing parameter estima tion equations .
411
k = 2 .9554c2 + 7 .859c
DC = (261-b0)k/{(1-2-k)r (l+k)) (3 .2 .2)
u = b0+ oc[r (1+k)-11/k ID

where c = (2b1-b0)/(3b2-b0)-1n2/1n3 ID
bo= xo/N 40
1:
31= 21pixo/N 110
be 2: p 12x 1/N
p1=(i-0 .35)/N
and i = ascend ing rank order of the annual maxima ID

The likely errors in estimat ing T-year floods us ing this method with ID
15 years o f data are broad ly similar to or better than those of using
maximum like lihood w ith 25 years o f data (Hosking et a l, 1984 , Table
6 , page 15). Since the catchments used in th is study had at least
10
15 years of data , it was considered reasonable to estimate T-year
floods using the GEV d istribution fitted by PWM throughout .
Second ly , the FSR recommendation to use regiona l growth factors in

ID
preference to the fitted d istribution for return periods beyond 2N
has not been fo llowed . Th is is because the regiona l approach is
ID
intended to improve on uncertain estimates obtained from extrapo-
lation o f the at-site record , rather than to de fine the true value
ID
of the quantile .
411
For each catchment the GEV d istribution was fitted by the PWM method
and d isp layed w ith the annual maximum data on a graph of d ischarge 40
ID
An assessment o f the FSR rain fall-runo ff method of design flood ID
estimation
ID
ID
ID
41
• 15
41
versus return period . The non-exceedence probability o f each flood
41 was calculated using the Gringorten formula :
41 F = (i - 0 .44) / (N + 0 .12) (3 .2 .3)
41 Inspection of these frequency plots showed a number o f catchments
where the observed data gave an ind istinct trend or where the d is-
tribution seemed to provide a poor fit. In such cases , the accuracy
of 'the higher quantile estimates was considered to be poor and so a
41 limit on return period was sought , below which the flood estimates
cou ld be trusted . Visual assessments o f the p lots were sensitive
41 to :
41 .. departure of observed data from the fitted curve at high return
periods (remembering though that the data points do not have
41 constant variance and the return periods and magn itudes of the
largest floods are poorly defined)
41
2 . d iscont inuities in the observed data (suggesting changes in flow
41 mechanism and/or compounded frequency d istr ibutions)
• 3 . groups o f data points at a certa in d ischarge (suggesting overbank

flow is limiting d ischarge at some point ups tream)
41 4. downward curvature of they fitted GEV distribution imp lying an
41 upper bound ; this could result from overbank storage , which might
fill eventually , and a llow the flood frequency cu rve to resume
41 upward curvature at higher return periods .
Reliance on visual definition using flood frequency p lots can be

• criticised on the g rounds o f subjectivity. However , as discussed
later in this section , more objective criteria based on the standard
41 error o f estimate gave limits wh ich were often intu itive ly unac-
ceptable .
•
Figure 5 gives twelve examp les of the frequency p lots and fitted GEV
41 distributions (a further six examp les are g iven in Chapter 6 of this
report) . Plots (a) to (c) show genera lly good fits where the eyeba ll
41 limit was set at 25 years . P lot (d ) shows a reasonable overall fit
but contains large loca l departures (not uncommon w ith longer re-
ID cords); the 10-year limit seems appropriate particular ly when it was
found that fitting the GEV by maximum likelihood increased the
41 50-year estimate by 23% . P lots (e) and (f) show slightly in ferior
fit and were limited to 10 years (the downward curvature in (e)
41 looked to be strong ly influenced by the smallest flood ). Plots (g )
to (1) show poo r fits , and were all limited to 5 years . P lots (g)
41 and (h) indicate breaks in trend at about 5 years. P lots (i) and
(j) exhibit downward curvatu re and the effect o f the p lotting pos1.-
11 tion of the largest flood . Plots (k) and (1) exhibit g roups of
floods at similar discharges .
41
Table 3.1 gives , for each catchment , the length of record , the eye-
ball limit, and the 2- to 100-year estimates . O f the 88 catChments ,
41
41
41
41
41
16
18 were g iven eyeba ll limits of 25 years , 5 3 we re given 10 years ,

14 were given 5 years , and three were g iven 2 years . In every case,
the eyeba ll limit was considerably less than the FSR recommendation
of 2N , and was often less than N/ 2 (see Figu re 6) . The effectiveness
of these eyeball limits in filtering out poor qua lity'quantile es-
timates is assessed in Chapter 5 of this report.
The remainder of th is section describes investiga tions aimed at de-

fining the eyeba ll limits more objective ly , or at least ensuring that
the lim its were chosen consistently . It may be om itted by the more
casua l reader .

estimation
I. /
Ma n n • 4 1. 11 V 111124X I 0 14627 11 Y
Figure 5 . Example flood frequency curves
Estimating T-year flood peaks from observed annua l maxima

( h) 400 10
* RETURN P52 I 0
7 7 7 3 Y e ar s
• 2 5 10 25 50 100
- 1 0 I 2 3 4 5
Re m = 44 0 / 4 7E
( 1) 330 14
la i il i e PLP 100
c 7 i i 1 Yen
2 5 30 25 50 100
r _ r.
0 2 3 4 5
of ()P.CU3 v an A1E 7
0 ) 540 19
RETU RN F U SCO
5 10 21
5 50
i I L Y e a"
2 3 4
RED UCED VARI ATE M UCCI) 4 4.4 14 1E
An assessment of the FSR rain fall-runo ff method of design flood

estimation
19
10 0
e d114
50
-(7) -AA
l e'
a)
25 4 • • • 0 • • •
X•112
--s e* 1A14
• • S • • S S • • • • • • •
• •
5 • • • •
2
12 -5 25 50
R e c o rd le n g th N (y e a rs )
F igure 6 . Eyeba ll lim it p lotted aga inst leng th of reco rd
Est imating T -yea r flood peaks from o bserved annua l maxima

•
Tab le 3 .1 Numbe r o f annua l max ima , eyeba ll lim it and GEV-P101 flood
quant iles (m3s -1 ) for 88 catchm ents .
An assessmen t o f the FSR rain fa ll-runo ff method o f d es ign flood

estimation
ID
11
41
411
Tab le 3 .1 (Continued )
Estimat ing T -yea r flood peaks from obse rved annua l max ima
22
3 .3 OBJ ECT I E CR IT ER IA FOR SETT NG RET U N PER IOD L IMIT S
There is some chance , however sma ll, that a particular samp le of

annual maxima cou ld have come from any o f a range o f d istributions ,
however un like the population the samp le m ight be . Thus , in evalu-
ating the suitability of a particular d istribution and parameter set ,
the fo llowing questions may be asked .
How like ly is it that the samp le comes from this distribution?
What is the like ly error in T-year flood estimate if this d is-

tribution is adopted?
If these questions can be answered by quo ting the value o f a derived

statistic , the same measure may be useful in he lp ing to define the
max imum trustworthy T .
3 .3. 1 St andar d er r or s
It might seem that the above questions are answered by considering

the standard error o f estimation , both of the parameters o f the
d istribution and the associated quantile estimates . The max imum
trustworthy T cou ld then be chosen as the return period at which the
standard error reached some critical va lue . Un fortunately , esti-
mates o f standard errors are not obtained when fitting a d istribut ion
by probability weighted moments (PWM). However , fitting by max imum
lik lihood (ML) g ives a variance-covariance matrix of parameters from
wh ich standard errors may be derived . The GEV d istribution was
therefore fitted aga in to each catchment , this time by ML , and the
resu lting standa rd errors were used to assess the goodness of fit
of both the ML and the PWM fitted quantiles.
Un fortunately , these standard error lim its were qu ite d ifferent from
the eyeball limits . The standard errors seemed to relate more to
the parametric form o f the d istribution than to any perceived lack
of fit on the frequency plot . Th is was particu larly true when the
derived va lue for the GEV k parameter was positive (correspond ing
to downward curvature of the frequency p lot w ith an upper limit on
flood magnitude) . In such cases the predicted errors might even
reduce w ith rising T as the upper limit was approached . Flood fre-
quency curves for two of the catchments seen in Figure 5 on page 17
are shown again in Figure 7; this figure shows both the P101 and MI.
curves and one standa rd error either side of the latter curve . The
two catchments have s imilar record lengths , but catchment 4 1028 was
g iven an eyeball limit of 25 years , wh ile catchment 72002 was g iven
a limit o f ten years . In contrast , the standard error derived for
the 10 year flood on catchment 4 1028 was 12% , while for catchment
72002 the standard error of the 100 year estimate was on ly 2 .5t
(wh ich was considerab ly less than the d ifference between the PWM and
ML curves).
An assessment o f the FSR rainfall-runo ff method of design flood

estimation
23
28
4 00 4
MY
- 2 - 1 0 1 2 3 4 5
M Y= VARI ATE
Figure 7. Max imum likelihood standard errors
Such prob lems associated w ith the estimat ion o f standa rd errors were
recognised in the FSR 1.2 .5 .6(170), where a sing le 'practica l'
standard error formula was proposed :
se(Q(T)) = C / (3.3 .1)
where C cou ld be taken as 0 .35 - 0 .8 ln(-1n(1-1/T)). This depended

only on the sample size and return period , and was therefore inde-
pendent o f d istribution form and parameter values . However , standard
errors which result from using th is approach were still poorly cor-
related w ith the eyeball limits .
Est imating T-year flood peaks from observed annua l maxima

24
In an attempt to overcome these prob lems , standard errors were es-

timated by another method known as the jackknife (Miller 1974). The
distribution is fitted N-times , omitting each data po int in turn ,
thus g iving N different estimates for the parameter and -quantile
va lues . The means (m) and standard deviat ions (S) of these N va lues
may be used to correct bias and to pred ict standard errors (se) in
the values derived from the fu ll data set . Thus
QT = N QTN - ( N-1 ) mT (3 .3 .2)
se = ( N-1 ) ST / N (3 .3 .3)
where QT is the bias free est imate of the quanti le , QTN is the es-
timate based on a ll N years data , and mT and S . are the mean and
standard dev iation o f the the N jackknifed estimates of QT each based
on N- 1 years o f data .
This method wou ld appear less sensitive to the p resence (or other-
wise) of outliers , but since the method samples (in effect) from
within the available da ta , if those data are unrepresen tative o f the
true flood d istribution , then the jackknife estimates w ill a lso be
unrepresentative . Note also that any out lier w ill appear in a ll but
one of the N sub-samp les .
Jackknife standard error estimates for the PWM fitting method were
derived for each catchment and return period . Figure 8 shows the
results for the same two catchments as shown in Figure 7 on page 23
Taken over a ll catchments the results seemed intu itive ly more real-
istic than either the ML values or those from Equation 3 .3 .1. How-
ever , they were still poor ly correlated with the eyeba ll limits ,
g iving higher limits to catchments w ith a downwa rd curvature .
Despite these reservations, the jackknife error estimates were used

to find the return period at which the standard error first exceeded
X% o f the corresponding flood estimate (with X = 10 , 12 .5 , 15) .
Of these , the 12 .5% error seemed best correlated w ith the eyeball
limits (though fo r some catchments even the 2-year flood fa iled the
criteria , wh ile for others the 100-year flood easily passed). For
the example catchments seen in Figure 5 on page 17, jackknife limits
correspond ing to p lots (a) to (f) were a ll 25 years or more , (100
years surprising ly for both (d ) and (e)), 2 yea rs for (g) and (h),
and 10 years for (j) to (1). The limit correspond ing to plot (i) was
100 years .
3 . 3.2 Goodness of f it
Since standard errors of estima te d id not seem as sens itive as the

eye to dev iations from the fitted curve , an independent measure o f
goodness of fit was sought . In the F lood Stud ies Report, two good -
ness of fit indices were used CX2 , Kolmogorov-Smirnov) but we re found
insufficiently powerful for the sma ll samples typica lly availab le .

estimation
25
V a
24 0 ( • ) 4 10 2 8
21 C —
rr.
co I
RI TURN ?I n c()
I I I I 1 Y ea U
5 10 25 50 100
0 I 3 3 • 5
REDU CED WA I ATE Y
( b) 7 20 0 2
:13
!)
Pan s
00 0 —
20 . 0 —
RC I URN Pu m a
Y ears
3 5 10 25 50 i 00
a I 1 3 4 5
W OO= V4A / ATC Y
Figure 8. Jackknife standard errors
The Ko lmogo rov -Sm irnov statistic is one of a number o f so-ca lled
empirical d istribut ion function (EDF) statistics which compare , at
each data value (x ), the observed non-exceedence probability
(0=rank/number) w ith the expected value (F) derived from the chosen
d istribution . A more powerful EDF statistic is the Anderson-Darling ,
representing an integ ral we ighted square error between 0 and F .
Ahmad et a l'(1988) have recently studied a modified fo rm (M) of th is
statistic with the weighting funct ion (1-F(x )) biassed towards er-
rors at high return period .
(0(x)-F(x ))2
N N 10 dF(x ) (3.3 .4)
(1-F(x))

26
Integrating for constant 0(x) between xi and xi,i, and summing over
the xi g ives the calculation formu la :
M = N/2 - 4 : F1 - ( 1/N )E 1(21-1)1n (1-Fi)i (3 .3 .5)

1= 1 1= 1
In a series o f simu lation experiments , they derived samples from a

known GEV parent d istribution , fitted a GEV to the samp le (to give
F va lues), and derived the corresponding M va lues . In this way they
bu ilt up a probability d istribution for M , and then derived an
equa tion for exceedance probability :
p (M ) = sin2(h(M)) (3.3 .6)
where h(M ) = -0 .9394 + 0 .9939M - 0 .054 11/M312 + 0 .3476/M

- 0 .7785M/N1/2 + 0 .057 15/(MN1/2)
Us ing this equat ion , it is possib le to estimate the probab ility o f

a samp le w ith a g iven M value coming from the fitted d istribution .
Thus for each catchment , the 11 value o f the fitted GEV was found and
the correspond ing probability derived . In genera l, these probabil-
ities seemed well correlated with the trustworthiness of the T-year
flood estimates , and indeed the three catchments g iven the lowest
eyeba ll limit (27035,29004 ,4 1007) had the lowest probabilities (less
than 0 .5% ). However , the correlation was not felt to be strong
enough to provide a suitab le objective method o f setting maximum
trustworthy T . In particular , long records were pena lised where ,
a lthough the d istribut ion departed from the data , intu itive ly rea-
sonab le T -year estimates cou ld still be obtained . For the examp le
catchments in Figure 5 on page 17 , plots (b) and (d ) gave probabil-
ities of 14% and 2% , while p lots (a) and (c) gave probabilities o f
95% and 98% . As expected , p lots (e) and (f) gave lower probabilities
o f 35% and 15% , but unexpected ly similar values (44% and 23% ) were
g iven by p lots (g ) and (1). P lots (h ) to (k) a ll gave 1% or less .
3.3.3 Comb ined st andar d er ror s and goodness of f it
From the ana lysis described in Sections 3 .3 .1 and 3.3 .2 it seems that
ne ither standard erro rs nor goodness o f fit statistics a lone can
quantify the intuitive confidence a hydro logist has in flood fre-
quency estimates based on visual inspection of flood frequency p lots .
Standard errors , in genera l, reduce w ith the length o f record , but
seem to be too closely associated with the form of the d istribution
rather than any apparent lack o f fit. Good ness o f fit genera lly gets
worse w ith longer records where , a lthough the overa ll fit may be
adequate , small local departures are heav ily penalised .
An assessment of the FSR ra infa ll-runo ff method of design flood

estimation
•
• 27
•
• KEY
Ey e b a ll l i mi t s
• - 2 5 y e ar s
• 0 - 10 y e a r s
A - 5 years
• v - 2 years
.
ID 0s‘ . •
., 0
-9 \
• o
e
2 5 y e a r s*
9
• s
o o •
.8
0 0
• ., •
II ., 0
\ •
• >, . 7 oo 0 N.
0
.-.
=
• a \
al
n •6
• 2
o. o 10 y e ar s \ o
• a,
S .5 o o o
v. •
co
• a A o
i *%.
c
• o •4 \
co
., 0 §
if,
• .o
c ..
< .3 `•.- o
• • o
o 0 0
'1.% o
• .2 5 years \
0
AA
A 0. , 0 0
0 A
0 A
. .0 . 08
• •1 0
.,
A 0 .,
A
ID —
— — — —A -A — — — _ — —
0
• 2 5 10 25 50 100
• 12.5% error limit (years)
F igure 9. Comparison of assessment criteria

•
•
•
• Estimating T-year flood peaks from observed annual maxima
•
•
•
28
Since standard errors and goodness of fit are comp lementary , it was
felt that a combination might be usefu l in de fining the maximum
trustworthy T . Figure 9 shows shows each catchment p lotted on a graph
of Mod ified Anderson-Darling probab ility (goodness o f fit) against
the jackknife (standard error) lim it , w ith different symbo ls used
to show the chosen eyeball limits . Overa ll, this figu re seems to
confirm that the eyeball limits comb ined both ideas . Furthermore ,
a lthough the arguments are somewhat circular, the figure sugg ests
that the eyeba ll limits have been app lied in a reasonab ly consistent
manner . Catchments wh ich d id not fit the trend were re-examined ,
but the eyeball limits were not rede fined . The eyeball limits seemed
consistent enough to use in the next stage of the comparisons .
A number of ideas for further work have been identified in the in

the course of the investigat ions described in this chapter and are
d iscussed in Chapter 8 .

estimation
ID
• 29
ID
41
•
•
41
4. 0 EST IMAT ING T - YEAR FLOOD PEA KS BY I NDIRECT MET HODS
ID
4. 1 INT RODUCT ION
41
The main objective o f this study was to estab lish how we ll the FSR
• rainfa ll-runoff method of flood estimation works . This was achieved
by comparing va lues calculated from flood data w ith estimates ob-
ID tained using FSR methods . Chapter 3 described how "true" va lues were
obtained as a bas is for comparison . Th is section describes the
ID various estimated va lues .
4 .2 T HE RA INFA LL - RUNOFF METHOD

41
The FSR rainfall-runoff method .can be app lied at s ites w ith no
II hyd rological data by using estimates o f mode l parameters based on
catchment characteristics derived from maps . The details of this
111 method are g iven in the FSR and are not repeated here : FSSR 16 re-
vised the parameter estimation equations , and therefore slightly
10 d ifferent estimated flood peaks are obtained . Two sets of estimates
correspond ing to the FSR and FSSR 16 mod el parameter estimation
0 equations were calcu lated .
• In both the FSR and FSSR16 it is recommended that, where possible ,

the model parameter values obtained from the regression equations
II are replaced with , or revised using , values from observed data . By
ana lysing flood events , va lues of percentage runoff and unit
4110 hyd rog raph time-to-peak can be derived . Two types of data can be
d istinguished . Firstly , data collected at the site o f interest,
411 which may be ca lled observed data . Since flood estimates are usually
requ ired at ungauged sites , this is rare ly available . However , on
II large schemes there may be time to install equipment and collect data
from at least a few storm events . The second type of data is usually
ID referred to as local data , meaning that it has been collected at a
station loca l to the site o f interest. If the gauged catchment is
• sufficient ly similar to that for which the estimates are required ,
in formation can be transferred between the sites . In this study only
• the utility of observed data was examined , and then on ly w ith the
FSSE I6 version since this is the current ly recommended rainfa ll-
• runoff method .
II Percentage runoff is not transferred d irect ly from ana lys is to the

design method , but is adjusted to a standard percentage runoff tic-
• cording to the catchment wetness and the rain fa ll depth . An ad-
40
Estimating T-year flood peaks by ind irect methods
•
411
41
30
justment is a lso made to remove the effect of urbanisation . Thus

from an observed value of percentage runo ff, PR , rura l percentage
runoff is calculated using the equation :
" rural = ( PR - 21.0 URBAN ) / ( 1.0 - 0 .3 UR BAN ) (4 .1)
Standard percentage runoff (SPR ) is calculated by subtracting two

dynamic terms , OPRC" , which is the dynam ic contribution to per-
centage runoff based on catchment wetness index (CWI), and DPRrain,
the dynamic term based on ra infa ll depth (P). Thus :
SPR = PR
--rural- " Rcwi- DPRr.in (4 .2 )
where
DPRCWI = 0 .25 ( CW I - 125 ) (4 .3 )
in which
CWI = 125 + API R ID (4 .4 )
API is a 5-day antecedent precip itation index defined in FSR 1.6 .4 .4 ,

R ID is the estimated soil moisture de ficit , and
= 0 .45 ( P - 40)" for P > 40mm (4 .5)
= 0 for P 4 40mm
It is recommended that SPR va lues shou ld be calcu lated from at least

five events . If they agree reasonab ly , then their average should be
used in design flood calcula tions .
The observed event da ta can a lso be used to derive unit hydrographs.

Th is can be done in two ways . The triangular FSR un it hyd rograph
can be replaced w ith an ord inate-by-ordinate representation , or the
triangular form can be retained but w ith a time to peak derived from
observed da ta . On ly the latter form of mod ifica tion was used in this
study ; again average va lues o f time to peak were calcu lated from at
least five events .
Revised T-year flood peak estimates were calculated using the FSSR 16
method for four cases :
'. w ith model parameters estimated from catchment characteristics

("no data"),
2 . w ith SPR from observed data ,
3 . Tp from observed data , and
4 . w ith both observed SPR and Tp .

estimation
• 31
40 Some o f the catchments ava ilab le contained larg e urban areas . Fo r

such catchments it is recomm ended that the bas ic method o f flood
• est imation is mod ified ; the rev ised techn ique is described in FSSR5 .
The ma in d ifference be tween the FSR method and FSSR5 was the d if-
ferent way in wh ich the inc reased runo ff from u rban area s was han -
d led . Howeve r, FSSR 16 uses the same urban correct ion as FSSR5 . The
• methods fo r rura l and urban flood estimation now d iffer on ly in the
retu rn period o f the ra in fa ll even t requ ired to estima te the flood
• o f spec ifed return period , and the shape o f the rain fa ll p ro file .
Th is study was conce rned on ly w ith the method fo r rura l catchments ;
estimates ca lcu lated for catchments w ith large urban fractions use
the ru ra l method . It was hop ed tha t these estima tes wou ld ind icate
• the su itab ility o f the method on urban catchments .
• 4. 3 T HE STAT IST ICA L MET HOD

40 A lthough the primary aim o f the study was to assess the per fo rmance
o f the rain fa ll-runo ff method , estima tes were a lso ca lcu lated us ing
• th e FSR statistica l techn ique based on catchmen t characteristics .
Aga in the ru ra l catchment method w as app lied to a ll catc hmen ts even
whe re an urban adjustment was app ropriate .
• 4. 4 SUMMA RY
• "True" flood quantiles were ca lcu la ted from observed annua l max ima
• in Ch apter 3 . Six ind irect me thod s o f estimating flood magn itudes
have been desribed in this chapte r :
40 '. Using the FSR statist ica l method based on catchment cha racter-
istics : FSR/STATS(CC ) .
•
2. Using the FSR rain fa ll-runo ff method based on catchment cha rac-
• terist ics : FSR/RF-RO(CC )
3. Us ing the ra in fa ll-runo ff method based on catchment cha racter-

ist ics w ith FSSR 16 mod ifications : FSSR 16(CC )
4. As 3 but w ith observed da ta va lues o f Tp and SPR : FSSR 16(Tp&SPR )
S. As 3 but w ith observed data va lues o f Tp : FSSR 16(Tp )
411 6. As 3 bu t w ith observed data va lues o f SPR : FSSR16(SPR )
40
40
40
Estimat ing T-yea r flood peaks by ind irect method s
40
32
An assessment o f the FSR ra in fall-runo ff method of design flood

estimation
33
5 .0 RESULT S
5. 1 INT RODUCT ION
For each of the 88 catchments , floods of five return pe riods (2 , 5 ,

10 , 25 and 50 years) were estimated using the six different methods
described in Chapter 4 . These estimates were then compared w ith the
respective flood quantiles derived from observed flow data as de-
scribed in Chapter 3 . Resu lts are presented in two ways .
The accuracy o f flood quantile estimates are compared across a ll

catchments.
F lood frequency curves for each catchment are exam ined in terms
o f s lope and index flood .
In order to assess the individua l influence of particu lar catchments

and ranges o f catchment type (indexed by the physica l character-
istics), analyses were performed on various subsets o f the 88
catchments . Some of these are described in detail in the text whilst
the resu lts from a ll subsets analysed are given in Appendix A .
5.2 PERFORMANCE STA I ST ICS
For each catchment a relative error was calcu lated for each of the
30 estimates (five return periods and six methods) using :
RT" = (QT" - QT1) / QT1 (5.2 .1)
where QT" is the T year return period flood quantile estimated on

catchment i using method j and QT1 is the same flood quantile from
the observed flow da ta at the gaug ing station . A pos it ive value
indicates that the method is pred icting a peak flow greater than that
observed , ie . overestimation , whereas a negative value indicates
underestimation .
A residua l statistic , defined in log space , was used by Lynn (1978)

in his comparison of several flood frequency estimation methods .
RT '„ = log ( QT1/ (Tr i ) ( 5 .2 .2)
Rather confusingly underestimation resu lts in negative value using

Equation 5 .2 .1 but a pos itive value using Equation 5 .2 .2 .
Resu lts
34
To exam ine how we ll the mode ls performed over a range o f catchments ,

these residua ls were used to calculate two summary statistics .
meanj = 1/n)ERTij (5 .2 .3)
where the meanj is the average value of the residual for return pe-
riod T calcu lated using method j over a ll catchments 1 to n . This
equation g ives the mean residua l, o r bias , desc ribing how well the
method is do ing , on average , over the range of catchments inc luded
in the calcu lation . The statistic ind icates the expected accuracy
o f an estimate on a catchment chosen at random from the samp le .
RMSj = ,/ l/nL ( RT112 ) ( 5 .2 .4)
This equation provides the root mean square residua l, RMS , ind icating
the variability o f the estimates about zero rather than about the
mean residua l. This root mean square residual should only be used
where the mean residua l is close to zero .
5 .3 T H E ST A N D A R D SE T O F C A T C H M EN T S A N D R ET U R N PE R IO D S
The fu ll data set contained flood magnitudes from 88 catchments , 5

returns periods and 6 estimation methods . However , as a lready noted ,
some o f these estimates come from urban ised catchments for which the
methods are considered inappropriate , or are for return periods be-
yond our eyeba ll limit . A data set was identified that comprised
those 74 catchments w ith less than 107; o f the d rainage area under
urban development and with the quantiles restricted to those w ithin
the eyeball limits described in Chapter 3 . This "standard data set"
is used as the benchmark for many comparisons in the fo llowing
sections . It is noteworthy that th is set of 74 catchments contains
quantiles up to the 25 year return period only , ie . no 50 year flood
estimates were felt to be su fficiently reliable .
5 .4 COM P SO N IT H R ESU LT S FR O M LYN N 1978
The residua ls calculated for the 74 catchments , using the

logarithm-based residua l defined in Equation 5 .2 .2 for the 2 and 10
year return period floods were compared w ith the correspond ing re-
sults reported by Lynn (1978). These two sets of statistics are
g iven in the upper section of Table 5 .4 .1. They d isp lay a similar
pattern o f resu lts ie . the statistical method is out-performing the
ra infa ll-runo ff method in terms of both mean and RMS residua ls . The
statistical method is a lmost unbiassed whereas the ra infall-runoff
method overestimates , on average , both the 2 and 10 year floods .
The RMS statistics are larger, in all cases for the Lynn data set .
This may be partly due to the data sets comprising d ifferent
catchments , but is more like ly to result from exc lusion of observed
An assessment of the FSR rainfall-runoff method of design flood

estimation
35
quan tiles wh ich were fe lt to be o f poo r qua lity in this study

(Boo rman , Ac reman and Packman , abb rev iated to BAP in Tab le 5 .4 .1).
Show n in the lower ha lf o f Tab le 5 .4 .1 are the mean and RMS stat is-
tics based on the relative error (Equation 5.2 .1). Th ese show the
same basic pattern o f the ra in fa ll-runo ff mode l overes tima tion and
the better pe rfo rmance o f the statistica l method in bo th bias and
var iab ility . It shou ld come as no surp rise tha t the sta tist ica l
method performs better at low return pe riods s ince it ha s been ca l-
ibrated d irectly aga inst the mean annua l flood .
The loga rithm-based residua ls have on ly been used to compa re resu lts
w ith those reported by Lynn . In the rema inde r o f the chapte r com-
pa risons are based on the .re lative e rror as g iven by Equa t ion 5 .2 .1.
Howeve r , App end ix A contains a ll four stat istics for a ll subsets o f
catchments exam ined .
RT '"
mean RMS
BAP Lynn BAP Lynn BAP Lynn
2 1 FSR /STATS (CC ) 74 43 0 .0 1 0 .05 0 .16 0 .2 1

2 FSR /RF-RO(CC) 74 43 -0 .05 -0 .06 0 .21 0 .32
10 1 FSR/STATS(CC ) 57 38 0 .02 0 .00 0 .15 0 .16

2 FSR/RF-RO(CC) 57 38 -0 .06 -0 .19 0 .19 0 .35
RT
Method mean RMS
2 1 FSR/STATS(CC ) 74 0 .06 0 .43

2 FSR /RF-RO(CC ) 74 0 .27 0 .89
10 1 FSR/STATS(CC) 57 0 .02 0 .37

2 FSR/RF-RO (CC ) 57 0 .28 0 .72
Tab le 5 .4 .1 Statistics for retu rn pe riod s 2 and 10 yea rs

fo r the FSR statistica l and rain fa ll-runo ff
methods fo r the standa rd catchment set used in
this study (BAP) and those derived by Lynn (1978) .
Resu lts
•
36 •
•
5 .5 COMPA RISON OF ORIG INA L FSR A ND FSSR16 MET HODS
•
As described in Chapter 1, Boo rman (1985 ) found that the rev isions
to the FSR rain fa ll-runo ff pa rameter estimation equa tions (FSSR I6 ) •
in gene ra l left the flood estimates on ly s light ly chang ed from those
ob ta ined us ing the o rg ina l equation s . Tab le 5 .5 .1 shows a compa rison •
o f resu lts us ing the FSSR 16 and the o rig inal FSR equa tions . It can
be seen tha t overa ll the FSSR 16 me thod pe rforms s ligh tly be tter than •
the o rig ina l FSR me thod in terms o f bo th mean and RMS relative er-
rors , w ith both me thod s ove rest imating by , on average , 22-4 1% . The •
FSSR 16 me thod is the cu rrent recommend at ion and hence the FSR me thod
is no t considered further in th is chap te r , a lthough comp rehens ive •
resu lts a re g iven in Append ix A .
•
•
RT"
Method n mean RMS •
2 2 FSR/RF-RO(CC ) 74 0 .27 0 .89 •
3 FSSR I6(CC ) 74 0 .22 0 .73
•
5 2 FSR/RF-RO(CC ) 71 0 .37 0 .92
3 FSSR 16(CC ) 71 0 .34 0 .77 •
10 2 FSR/RF-RO(CC ) 57 0 .28 0 .72 •
3 FSSR 16(CC ) 57 0 .28 0 .64
•
25 2 FSR/RF -RO (CC ) 15 0 .39 0 .96
3 FSSR 16(CC ) 15 0 .41 0 .83 •
•
Tab le 5 .5 .1 Statistics for return per iod s 2 , 5 , 10 and 25 y ears
for the FSR/RF-RO(CC ) and FSSR16(CC ) method s .
•
•
5 .6 USE OF MODEL PA RAMET ERS FROM FLOOD EVENT DA T A •
The FSR strong ly recomm ends that va lues fo r the ra in fa ll-runo ff mode l
pa rameters derived from flood events obse rved on the catchment shou ld
•
be used in preference to thos e va lues g iven by the ca tchment char-
acterist ic based equat ions . F lood estimates were derived using the
•
FSSR 16 mod e l w ith pa ram eters ob tained from observed SPR and Tp data .
The res idua ls were then comp ared w ith those from using the no-data
•
equa tions in the same method . The resu lts are g iven in Tab le 5 .6 .1.
Fo r a ll retu rn per iods the b ias is redu ced by u sing ob served data ;
•
s ligh t ly when observed Tp is used , mo re so when observed SPR is used .
Using bo th observed Tp and SPR mak es a substant ia l imp rovemen t , fo r
•
exam p le , reduc ing the average overest imat ion o f the 25 yea r flood
from 4 1% to 11% . Us ing bo th ob served Tp and SPR a lso reduces the
•
•
An assessment o f the FSR rain fa ll-runo ff method o f des ign flood •
estimation
•
•
•
•
• 37
RMS residua l. Howeve r , it can be seen that th is dec rease resu lts
411 predominantly from using SPR , since using observed Tp a lone increases
the variab ility o f the estimates o f a ll bu t the 10 year return pe riod
•
flood quan tiles .
411 RT1J
Method n mean RMS
•
2 3 FSSR 16(CC ) 74 0 .22 0 .73
II 4 FSSR 16(Tp& SPR) 74 -0.01 0 .35
5 FSSR 16(Tp ) 74 0 .16 0 .83
411 6 FSSR 16(SPR ) 74 0 .07 0 .39
411 5 3 FSSR 16(CC ) 71 0 .34 0 .77

4 FSSR 16(Tp& SPR ) 71 0 .07 0 .32
ID 5 FSSR 16(Tp ) 71 0 .26 0 .83
6 FSSR 16(SPR ) 71 0 .16 0 .38
II
10 3 FSSR 16(CC ) 57 0 .28 0 .64
II 4 FSSR 16(Tp&SPR ) 57 0 .06 0 .27
5 FSSR 16(Tp ) 57 0 .17 0 .6 1
II 6 FSSR 16(SPR ) 57 0 .16 0 .37
II 25 3 FSSR 16(CC) 15 0 .4 1 0 .83

4 FSSR 16(Tp&SPR ) 15 0 .11 0 .4 1
ID 5 FSSR 16(Tp ) 15 0 .4 1 1.04
6 FSSR 16(SPR ) 15 0 .16 0 .40
II
41
Tab le 5 .6 .1 Statist ics for return periods 2 , 5 , 10 and 25
years comparing the utility o f observed da ta .
Since observed Tp and SPR are usua lly ava ilab le together and because
us ing bo th g ives the best estimates , fu rthe r resu lts ob tained using
111 method s FSSR16(Tp) and FSSR 16(SPR) a re not cons id ered further in the
body o f th is repo rt but are contained in Append ix A . The rema inder
o f th is chap ter cons id ers the three method s :
1. FSR/STATS (CC ) ,
3 . FSSR 16 (CC ) and
4 . FSSR 16 (Tp&SPR )
1111
5.7 R ESU LT S FR O M T H E ST A N D A R D SET
Tab le 5 .7 .1 show s the statist ics fo r retu rn period s 2 , 5 , 10 and 25

5 yea rs fo r the standard catchment set , for the three method s chosen
fo r deta iled ana lys is . The most strik ing resu lt is that the majority
ID o f the va lues in the first co lumn are pos itive indicat ing that the
40
Resu lts
ID
111
411
40
38
me thod s are , on averag e , ove r-estimating the flood peak s . On ly fo r

on e me thod (FSSRTp&SPR ) and one return p eriod (2-yea rs) is there ,
on averag e , underestimation . The figu re o f -0 .0 1 ind icates that on
average this method is underes timating the 2-year return .pe riod flood
peak by 1% . Both th is and the sta tist ical method show a sma ll
ov era ll b ias ; on ly fo r the 25 yea r quan tiles does it reach 10% . The
RMS statist ics a re sma llest for the stat istica l method for the 25
yea r quan tile , bu t the FSSR 16 method w ith observed da ta show s sma ller
variab ility in the 2 , 5 and 10 year cases .
RT
M ethod n mean RMS
2 1 FSR /STATS (CC ) 74 0 .06 0 .43

3 FSSR 16 (CC ) 74 0 .22 0 .73
4 FSSR 16 (Tp&SPR ) 74 -0 .0 1 0 .35
5 1 FSR/STATS (CC ) 71 0 .02 0 .36

3 FSSR 16(CC ) 71 0 .34 0 .77
4 FSSR 16(Tp& SPR ) 71 0 .0 7 0 .32
10 1 FSR/STAT S(CC ) 57 0 .02 0 .37

3 FSSR 16 (CC ) 57 0 .28 0 .64
4 FSSR 16(Tp&SPR ) 57 0 .06 0 .27
25 1 FSR /STATS (CC ) 15 0 .09 0 .32

3 FSSR 16 (CC ) 15 0 .4 1 0 .83
4 FSSR 11
5(Tp&SIT ) 15 0 .11 0 .4 1
Tab le 5 .7 .1 Statist ics for return pe riod s 2 , 5 , 10 and 25 yea rs

comparing the performance o f three mod e ls on
standa rd set o f catchments .
It is necessa ry to remember the characteristics o f the two methods

wh en cons idering these resu lts . FSR/STAT S(CC) uses a reg ression o f
the mean annua l flood on s ix catchment charac terist ics and a re-
g iona lly based mu ltip lier ; it shou ld therefore be expec ted tha t very
good estimates o f the 2-year f lood a re obtained . To estimate mo re
extreme floods w ith th is method a fam ily of regiona l g row th curves
is used and the qua lity of the es tima tes w ill dec rease . On the other
hand the rain fa ll-runo ff method uses catchment cha rac terist ics in a
less d irect fash ion and contains no reg iona lisa tion . Est imates from
FSSR 16(CC ) a re there fo re un like ly to b e as good as those us ing
FSR /STATS(CC ) at low return pe riod s . What m igh t not hav e been an -
ticipated is that the use o f observed data in FSSR(Tp& SPR ) improves
resu lts on ly to the same leve l as ob tained w ith FSR /STATS (CC ) . It
An assessment o f the FSR rain fa ll-runo ff method o f d esign flood

est ima tion
39
is , un fortuna te ly , no t possib le to estab lish the rela tive perform-

ance o f the me thods a t higher return periods .
5 .8 RESULT S FROM T HE FUL L SET
Tab le 5 .8 .1 shows the statist ics fo r return periods 2 , 5 , 10 , 25 and

50 years us ing each o f the three methods on a ll 88 catchments , thus
relax ing the eyeba ll lim its and inc lud ing the mo re heav ily u rban ised
catchmen ts . The pe rfo rmanc e o f the rain fall-runo ff me thod is con -
s iderab ly worse than on the standa rd set , even w ith observed da ta .
These resu lts wou ld seem to justify the need to have a specific
method fo r urban catchments and to restrict the return pe riod s to a
reasonab le lim it .
Cu rious ly the statistica l method d isp lays a sma ller bias on this se t
o f catchm ents than on the standard set , a lthough the RMS is s ligh t ly
wo rse .
Tab le 5 .8 .1 Statistics for return periods 2 , 5 , 10, 25 and SO yea rs

comparing the performance o f three mode ls on a ll 88
catchments . Statist ics for the stand ard set a re show n
in brackets .
Resu lts
40
5.9 EX CLUD ING CAT CHMENT 39004
It was noted that from one catchment 39004 , the Wand le at Beddington ,
the flood frequency curves generated by the rain fa ll-runoff methods
provided very poor estimates of those derived from the observed data
(Figure 10). Even when using observed flood event data , estimates
of the percentage runoff were far too high . In contrast the
FSR/STATS(CC) method performs we ll on this catchment . The R iver
Rand le is under lain predominantly by cha lk but has an urban fraction
of 0 .39 , characteristics that together present particu lar prob lems
for flood frequency estimation . In Chapter 6 details are presented
of flood estimation problems on another catchment with a high pro-
portion of WRAP type 1 soils , the Waithe Beck at Brigs ley ; the same
prob lem occurs on the Wandle and it wou ld be inappropriate to de lve
too deep ly into causes for poor estimation on an ind ividual catchment
at this point . It is , however, worth noting that while observed SPR
data does improve the estimates , using observed time to peak makes
estimates worse . This is because the derived un it hyd rographs have
a very different shape to the triangu lar unit hydrog raph used in
making the flood estimates.
To test th is catchment 's influence on the overa ll resu lts the sta-
tistics in Tab le 5 .8 .1 were recalculated after exc lud ing this
catchment . The resu lts are shown in Tab le 5.9 .1. A comparison o f
this table with Tab le 5.8 .1 shows that, for the rainfa ll-runoff
method s , the degree of improvement is marked . For each return pe-
riod the b ias is reduced by around 10% and the RMS is also signif-
icantly sma ller , whereas the results for the statistica l method are
virtua lly unchanged . This demonstrates the cons iderable effects
that a sing le poorly modelled catchment can have on the overa ll re-
sults . The comparison of Table 5 .9 .1 w ith the standa rd set g iven
in Table 5 .7 .1 perhaps gives a more rea listic impression of the ef-
fects o f inc luding the urban catchments and removing the quantile
limit constraints . To aid this comparison the statistics for the
standard set are g iven in b rackets on Tab le 5 .9 .1. For the 5 and
10 year floods the bias and variability of both rainfa ll-runoff
methods has increased by including the urban catchments and re laxing
the return period limits . Resu lts for the 2 and 25 yea r floods are
about the same . It can be concluded that the overall performance
of the model is not be ing undu ly in fluenced by the inclus ion o f urban
catchments or poorly estimated observed quantiles .
An assessment o f the FSR rainfall-runo ff method o f design flood

est imation
41
0 i 2 3 4 5
R e d uc e d v a r ia t e Y
F igu re 10 . F lood frequency curves for the Wand le at Bedd ing ton
(39004 )
Resu lts
42
Tab le 5 .9 .1 Stat istics for retu rn per iods 2 , 5 , 10 , 25 and 50

yea rs comparing the perfo rman ce o f th ree mode ls on a ll
88 catchmen ts except 39 004 the Wand le at Bedd ing ton .
Statistics for the standa rd set are shown in brackets .
5 . 10 EXC LUD ING CA T CHMENT S UNDERLA IN BY RAP T Y PE 1

SOILS
Th e Wand le catchmen t is unde rla in by so ils with a h igh W inter Ra in

Accep tance Potentia l . It was specu lated that the ra in fa ll-runo ff
mode ls were per form ing re lative ly bad ly on other ca tchments w ith h igh
p ropo rt ions o f WRAP type 1 so ils . Tab le 5 .10 .1 shows the resu lts
from a set o f catchmen ts wh ich have less than 20% WRAP type 1 so il .
Th ese figures can be compared w ith those from the standard set wh ich
are g iven in brack ets . The FSR/ST ATS resu lts are abou t th e same ,
o r sligh tly wo rse , sugg esting that it is perform ing relat ive ly we ll
on the catchments und erlain by WRA P type 1 so il. An improvement in
the resu lts is ev id ent for FSSR 16(CC ) , and , with FSSR 16 (Tp&S PR ), the
stat istics a re s ligh t ly better . Th is imp lies that the rain fa ll-
runo ff method is ind eed work ing poorly on these catchmen ts wh en no
ob served data a re ava ilable and that ob served Tp and SPR prov ide
va luab le in format ion . It is no tewo rthy tha t on th is set o f
ca tchm ents the ra in fa ll-runo ff mod e l w ith observed data shows
An assessment o f the FSR ra in fa ll-runo ff method o f d esign flood

es tima tion
43
sma ller va riab ility than the statis tica l method and the b ias is on ly
g reater for the 5 year retu rn period flood s . Fu rthermo re , overes-
tima tion is , for th is set , no g reater than 12X fo r any return period ,
and is v irtua lly unb iassed a t the 2 and 10 ye a r quant iles when ob -
served data are emp loyed .
Tab le 5 .10 .1 Statistics for retu rn period s 2 , 5 , 10 , 25 and 50

yea rs compa ring the performance o f three mode ls
on standa rd set exc lud ing ca tchmen ts w ith mo re than
20% WRA P type 1 soils . Stat ist ics for the stand ard
set are shown in brack ets .
5 . 11 MOU NT OF A ILA B LE OB SER ED E EN T DA A
A fu rthe r set o f statist ics w ere derived to investigate whether im-

p roved quantile est imates wou ld resu lt from increasing the amount
o f observed event da ta . Th is wa s ach ieved by reducing the standa rd
set o f catchm ents to inc lude on ly those w ith at least 10 ava ilab le
even ts from wh ich SPR and Tp had been d erived . Resu lts fo r this
subs et o f catchmen ts are g iven In Tab le 5 .11.1 . C learly the no -da ta
methods shou ld g ive sim ilar resu lts if the two subsets con tain the
same range o f catchments . Th is appears to be the case for the
ra in fa ll-runo ff method s ince the resu lts are no t sign ifican tly d if-
ferent from those fo r the standard set , g iven in brack ets . Con -
ve rse ly , the statistica l method pe rfo rms s ligh t ly wors e a ll round ,
wh ich is p resumab ly a cu riosity o f the m ix of ca tchments in the two
sets . The mo st important sta tist ics a re those fo r the observed data
method . On ly fo r the 25 yea r quan tiles is va riab ility and bias
Results
44
sma ller . Th is resu lt is ba sed on on ly 13 catchments but sugg ests

tha t the va lue o f increasing the number o f events availab le is on ly
impo rtan t fo r h igh er return per iods . Comp lementa ry to the find ings
in Section 5 .6 , the full vers ion o f Tab le 5 .11.1 in Append ix A (Tab le
A .5 ) shows that the type o f ob se rved da ta a vailab le may be impo rtant .
In a ll cases the va riab ility o f es timates is grea ter us ing obse rved
T p than when using no observed da ta at a ll . On the o ther hand , es-
tima tes us ing on ly obse rved SPR have s imi lar RM S va lues to the use
o f bo th types o f obse rved da ta . Th is sugg ests tha t there are prob-
ab ly a few catchmen ts fo r wh ich the no -data estima tes o f Tp a re much
bette r tha t tho se d erived from the availab le events . On ave rage the
perfo rmance o f the un it hydrog raph mode l is imp roved by inc reas ing
the numbers o f events w ith es tima tes o f pe rcentag e runo ff.
Tab le 5 .11.1 Stat ist ics fo r return period s 2 , 5 , 10 , 25 and 50

yea rs comparing the pe rformance o f three mode ls
on standard set exc lud ing ca tchmen ts w ith less than
10 events . Statistics fo r the standard set a re shown
in brackets .
5 . 12 HE NUMBER OF ANN UA L MAX IMA AVA I LA B E
In add ition to eyeba ll lim its , the amoun t of da ta avai lab le fo r de-
riv ing obse rved flood frequ ency es timates may g ive an ind ica tion o f
the ir accu racy . To test th is , statistics were d erived for a further
subset o f the standa rd set o f catchments with at least 25 annua l
maxima . Th e resu lts a re g iv en in Tab le 5 .12 .1. For F SSR 16 (Tp&SPR )
An assessment o f the FSR ra in fa ll-runo ff method o f d es ign flood

estimation
ID
• 45
40
the RMS stat ist ic is sma ller for the 2 and 5 yea r floods , thoug h no t
sign ificant ly so . Resu lts fo r the 10 and 25 yea r floods show a de-
c line in per formance bu t are based on on ly 11 and 6 ca tchm ents re-
10 spec tive ly . Surprising ly the bias is greater fo r FSR/ST ATS (CC ) and
FSSR 16(Tp6SPR ). W ith ob serv ed data there is little d ifference .
ID Overa ll , these resu lts suggests that the errors a re not d ue to poo r ly
defined observed flood frequency cu rves caused by too few annua l
ID max ima .
Tab le 5 .12 .1 Statistics for retu rn period s 2 , 5 , 10 , 25 and 50

110 yea rs compa ring the pe rforman ce o f three mod e ls
on standard set exc lud ing ca tchmen ts with at least
ID 25 years annua l max imum flood s . Stat istics
for the standa rd set a re show n in b rackets .
411
5 . 13 S T IA L D IST R IBUT ION OF RES ID UA LS
ID
F igu re 11 shows the spa tia l d istr ibu tion o f residu a ls from the FSSR 16
411 observed da ta method for the standa rd set at the 2 year flood . These
residua l va lues a re g iven in Tab le 5 .13 .1, and fo r the 10 yea r flood
in Tab le 5 .13 .2 . The mod e l underestimates in sou th-western pa rts
of Eng land and Wa les , and there is a tendency for overest imation in
south -eas t Eng land . Th e find ings fo r south-west Eng land reproduce
those repo rted by Lynn ( 1978) wh ich a lso co incid e w ith the residua ls
mapp ed in the FSR 1 .6 .7 .4(44 8), reproduc ed as F igure 26 on page
105 . However , bo th Lynn and the FSR found underes timation in
south-east Eng land . The mixture o f ove r- and u nderest imation de-
Resu lts
40
46
<fie;
Qs°
o
o Key
under e stimation
0
• ov er es t imat io n
•
o° o w ithin 10%
0
0 _ •
m o•
• 0 0
• •
0 _
0 •
0
•
o •
• le
•
oo-
0
-o
Figure 11. 2-year flood residuals using FSSR 16(Tp&SPR)
An assessment o f the FSR rain fa ll-runo ff method of des ign flood

estimation
41
• 47
•
picted in Figure 11 for the rest o f the UK does not suggest reg iona l
patterns . However, it appears that underestimation dominates near
to the west coast, w ith , genera lly , overestimat ion elsewhere .
The d ivision line between these two reg ions fo llows , very rough ly ,
• the 800mm average annual rain fall isohyet . Tab le 5.13 .3 shows that
for catchments wetter than 800mm the average underestimation of the
• 2 year flood by the FSSR16 method us ing observed data is , on average ,
6% . For catchments drier than 800mm the mean overestimation is 10% .
II The remainder o f the table shows that the relat ive overest imat ion
in the d rier east of the UK is true a lso of the 5 and 10 year flood
411 quantiles . Variability appears to be abou t the same in both reg ions .
411 It wou ld be foo lhardy to read to much into these resu lts . Annua l
average ra infa ll is just prov id ing a convenient way of sp litting the
II catchments . The reason for the observed pattern of residua ls is
like ly to be a combination o f factors that w ill inc lude soil type ,
• topography and possibly design storm specification .
II
48
ID
411
II
Ca tchmen t Res idua l Catchment Residua l 411
19002 0 .09 46003 -0 .45
19005 -0 .08 46005 -0 .45 10
2000 1 0 .29 4 7007 -0.14
23005 -0 .46 48004 0 .04 111
24005 -0 .08 48005 -0 .43
24007 -0 .06 52005 -0 .23 II
2700 1 -0 .02 52006 -0 .19
27035 0 .45 520 10 -0 .30 ID
280 70 0 .36 53005 -0 .31
2900 1 0 .69 53007 -0 .28 ID
29004 -0 .03 53009 -0.27
3000 1 0 .05 540 11 -0 .02 411
30004 -0 .19 540 16 0 .12
3 1005 0 .22 540 19 -0 .3 7 411
330 14 -0 .48 540 22 -0 .17
33029 -0 .25 55008 -0 .22 ID
33045 0 .05 550 12 -0 .34
34003 0 .00 5600 3 0 .17 411
34005 0 .12 56004 -0 .20 I
35008 0 .30 56006 -0 .26 0 1
36008 0 .09 57004 0 .00
3700 1 -0 .05 57005 0 .04 40
39022 0 .32 5800 1 -0 .4 1
39025 -0 .04 58002 -0 .36 ID
39026 0 .02 60002 0 .10
39053 0 .16 6 100 1 -0 .17 ID
40006 0 .28 6 1003 -0.17
40009 0 .29 6400 1 -0 .04 ID
400 10 1.39 6500 1 -0 .20
4 1005 0 .03 660 11 -0.12 40
4 1006 -0 .29 67003 0 .33
4 100 7 0 .30 67008 0 .37 II
4 10 15 1.67 68006 0 .05
4 1028 0 .06 7100 3 -0 .18 II
45002 -0 .03 71004 -0 .19
45003 -0 .19 7200 2 0 .03 411
45004 -0 .06 7700 2 -0 .30
II
Tab le 5 .13 .1 Res idua ls for the 2 year return period show ing
the performance o f the FSSR 16(Tp&SPR ) on the ID
standard set . Fo r examp le -0 .02 ind icates
und erest imat ion by 2% .
ID
An assessment o f the FSR ra in fa ll-runo ff method o f d es ign flood •

est imation
•
•
•
•
41
• 49
41
41
Catchmen t Res idua l Catchmen t Residua l
41 19002 0 .10 48004 -0 .09
19005 -0 .06 48005 -0 .40
41 2000 1 0 .11 52005 -0 .18
23005 -0 .29 52006 -0 .30
41 24005 0 .05 520 10 -0 .25
24007 0 .0 1 53005 -0 .20
41 2700 1 0 .04 53007 -0 .14
2900 1 0 .89 53009 -0 .17
41 30001 0 .15 540 11 0 .04
30004 -0 .04 540 16 0 .34
• 3 1005 0 .11 54022 -0 .06
33029 0 .2 1 550 12 -0 .14
• 33045 0 .08 56 003 0 .18
34003 0 .10 56004 -0 .2 1
• 34005 0 .08 56 006 -0 .19
35008 0 .32 57004 -0 .04
• 36008 0 .02 5 7005 0 .08
3700 1 0 .08 5800 1 -0 .24
• 39022 0 .63 58002 -0 .29
39025 0 .17 60002 0 .34
• 39053 0 .57 6 1003 0 .08
40009 0 .50 6400 1 0 .35
• 4 1006 -0 .18 6500 1 -0 .08
4 1028 0 .27 66 0 11 0 .13
• 45003 -0 .21 67003 0 .50
45004 -0 .07 68 006 0 .13
• 46005 -0 .32 7 1003 -0 .11
47007 0 .38 72002 0 .53
• 77002 -0 .2 1
•
Tab le 5 .13 .2 Residua ls fo r the 10 year return pe riod show ing
• the pe rformance o f the FSSR 16(Tp& SPR) on the
standa rd set . For examp le -0 .12 indicates
• underestimat ion by 12% .
•
•
•
41
41
ID
411
Resu lts
111
•
•
•
50
Tab le 5 .13 .3 Sta tistics fo r retu rn period s 2 , 5 , and 10

yea rs comparing the perfo rmance o f three mode ls
on standard set d iv id ing the catchments into two
g roups on the bas is o f a th resho ld va lue o f SAAR
o f 800 mm .
5 . 14 CA T CHMENT SIZE
Fu rther invest igation s were undertaken to examine wheth er the mode l

perfo rmed better on la rge o r sma ll catchm ents . The standa rd set o f
74 catchments was d iv ided into two g roup s with AREA g reater o r less
than 100 km2 . The resu lting statistics fo r the two g roups are g iven
in Tab le 5 .14 .1. Th e statist ica l and no -d ata ra in fa ll-runo ff methods
perform much better on the la rger catchm ents in terms o f bo th b ias
and va riab ility for a ll but the 25 year floods , wh ich are based on
g roups o f on ly 5 and 10 catchmen ts . In contrast , the ra in fa ll-runo ff
method using obse rved da ta d isp lays cons istent pe rfo rmance for both
g roups o f catchments . In can be conc luded from these resu lts that
ob served da ta are mo re bene ficia l on sma ller ca tchments . Th is may
resu lt from the prob lem o f accurate ly abstract ing physica l charac-
terist ics , su ch as so il type , on sma ll catchments , wh erea s erro rs
tend to averag e-out on larger catchments . Ano ther po ssib le exp la-
na tion is that on sma ll bas ins the respo nse is dom ina ted by the land
phase o f catchment respons e , wh ich is d ifficu lt to mode l, whereas
on la rger bas ins response is dom ina ted by the channe l pha se , wh ich
is re lative ly easy to mode l.

es tima tion
• 51
•
40
• Method n meanRT11 RMS
< > < 100 > 100 < 100 > 100
40
2 1 FSRISTATS (CC ) 25 49 0 .24 -0 .03 0 .55 0 .34
• 3 FSSR 16(CC ) 25 49 0 .42 0 .12 1.00 0 .54
4 FSSR 16(Tp& SPR ) 25 49
• 0 .0 1 -0 .02 0 .40 0 .32
5 1 FSR/STATS (CC ) 24 47 0 .22 -0 .09 0 .4 7 0 .29

• 3 FSSR 16(CC ) 24 47 0 .57 0 .21 1.03 0 .59
4 FSSR 16(Tp&SPR ) 24 47 0 .11 0 .05 0 .39 0 .28
41
10 1 FSR/STATS (CC ) 20 37 0 .25 -0 .10 0 .48 0 .28
• 3 FSSR 16(CC ) 20 37 0 .48 0 .17 0 .73 0 .59
4 FSSR 16(Tp& SPR ) 20 37 0 .05 0 .06 0 .25 0 .28
ID
25 1 FSR/STATS (CC ) 5 10 0 .23 0 .02 0 .33 0 .32
• 3 FSSR 16(CC ) 5 10 0 .46 0 .39 0 .46 0 .96
4 FSSR 16(Tp& SPR ) 5 10 0 .28 0 .03 0 .40 0 .4 1
•
•
Tab le 5 .14 .1 Stat istics fo r return periods 2 , 5 , 10 and 25
years compa ring the pe rfo rmance o f three mode ls
on standa rd set d iv id ing the catchments into two
g roups on the bas is o f a thresho ld value o f area
of 100 km2.
• 5 . 15 COMBINING SUBSET SELECT ION CRIT ER IA
40 There are a large number o f possib le comb ina tions o f restrictions

wh ich can be app lied to produce subsets o f catchments fo r compa rison
o f the various techn iques . In ord er to d isp lay the relative per-
formance o f the three primary method s on a numbe r o f sub sets , a "bush
d iagram" was produced separate ly fo r each o f fou r statistics , name ly
.the 2 yea r mean and RMS and the 10 year mean and RM S . These are show n
• in Figu re 12 to Figure 15 . The d iag rams conta in a number o f boxes
each o f wh ich represents a comb inat ion o f restrict ions . A rrows
• lead ing to each box ind icate the add itiona l restrict ion imposed
compa red w ith the box from wh ich the arrow o rig inates . Henc e the
• box a t the extreme le ft contains the resu lts for the fu ll (88
catchment) data set , wh ereas the box on the fa r righ t con ta ins the
same statistics for a reduced set o f catchments which satisfy all
the restrictions imposed at once . In genera l , the numerica l va lues
• d ecrease from le ft to righ t as the qua lity o f the ob serv ed flood
frequency curves inc reases and as the catchments on wh ich the mode l
411 performs poorly a re om itted . Figu re 12 shows that , in a ll subsets
o f the da ta , the basic ra in fa ll-runo ff mode l is , on average , ov er
• est imating the 2 year retu rn period floods by a round 20% . In con -
•
Resu lts
ID
411
40
•
52
•
•
trast , when observed da ta are employed , this method d isplays little
bias or slight ly underestimates . This model performs pred ictab ly II
across the range o f subsets . The bias reducing from 9% for all 88
catchments virtua lly to zero in many boxes on the right-hand side . II
Anoma lies occur w ith the restriction on the number o f annua l maxima
availab le . As indicated in Chapter 2 , this is not a good ind icator II
of the accuracy o f observed quantiles . An exp lanation is that the
subset of long record catchments just happens to inc lude those on II
wh ich the model performs re latively poorly . It may a lso be argued
that this restriction reduces the number of catchments to a level ID
where the resu lts may be insign ificant .
For the 10 year floods (Figure 14), both methods overestimate. This
positive bias is sma ll when observed data are used (less than 10% )
but around 20-30% for the no-data case .
It is important to note that the number o f catchments decreases from

ID
left to right . In the most restricted set on ly seven catchments
remain , thus the sample is not very representative o f the UK as a
111
who le .
The RMS statistic for FSSR16(CC ) decreases from over 70% on the left
to just over 30% on the right for both the 2 and 10 year floods ; the 111
correspond ing figures for FSSR16(Tp&SPR) are 22 and 11% respec-
tively . In nearly a ll but the fu ll 88 catchment set F5SR16(Tp&SPR) ID
out-performs FSR/STAT5 (CC).
411
40
40
411
40
40
An assessment o f the FSR rainfa ll-runoff method of des ign flood ID

estimation
ID
40
ID
ID
53
02 57
0.35 7
0.0 13 90 0.0 52 0
02 3 2
I.
0.0 80 Oot
02 0 02 2 2
0.0 40 - 0 .0 08 d. 0. 14 7 02 2 1
„ Ot
0.2 0 0 2 7• - 0.0 2 7
IO
- 0.0 0 5 sCrt 0.0 5 2 0.00 1
r. 0. 172
0 .0 82 9? 0.0 15 4
a
4.•
0 . 17 2 •6 4/
se
0.0 3 1 0 2 18
0. 17 7 il 12.2 15
- 0 .00 5 41 . 0. 14 7 0 00 6
66 0 175
cum 00 0.00.1.
t o* 0. 11 7
0.04 7 0.00 0 02 22 re
•
19551/ 812 9 7 11 0.0 11 02 8 4 0.2 2 1
6 6 6 6 66 CC C) 0.3 44 0.0 14 0. 143 00 - 0 .02 7
t i 02 33 ID
2 8 5 1118 ( T p a lI PR ) 0.0 12
$.4. • , • • 41w. ol s I I ° Do 0.0 7 1 e l 0.00 6
°V 0.0 18 0 .23 1
0.2 54 4 4. 0.0 53 0 0 .0 2 1 0. 143
0 .00 3 0.4 40 02 33
6/ 0 .02 3 02 87 p4 - 0.00 8
r. 01 6 7
0.0 3 1 99 0.00 0
0 2 77
0.0 18 sr - 0.00 3
0.2 54 •1
0.00 3
F igu re 12. Bush d iagram show ing 2-year KEAN
e mn
0. 734
0 . 4 4 0 94' 02 13 0
0 , 72 4 ,4 04.
0 . 4 6 2 34 t au, ,r0
0 . 72 7 .4 0 2 30
0 .4 8 2 02 52 0 0 .4 0 1 0. 0 .3 10
„ Ot 0 .4 11 9' 0. 2 14
0.7 2 7
I .
02 5 2 90d, 0 .• 4 5 0 .3 13
" 0. 5 3 $ 4,
0 .4 4 5 e 0 .2 9 5
44
0 .5 3 5 $1
0 .4 4 1 `p 0 .2 9 5 4 02 27
0.5 7 1 11 2 228
S4 0 . 229
0.2 8 5 4 0 .4 8 1
44 0 . 4 19 I I
9d. 0 .4 8 5 093 0 .3 13 14,

0 .4 2 .11
0.4 4 8 0 .3 0 3 0 2 38
0.73 1 02 10
' SR I 0 .• 3 1
8 8 8 88 ( C C ) 0 .34 4 0 . 3 4 11 0.378 .30 0.2 18
I . 0 .3 5 8
146 . . 61W4 ,4 6 12
052
m
4 0 .3 4 8 94 .2 21 .9
11
I:19 0 .4 3 1 0 . 3 78
0 . 3 8 4 90 0 .37 8
I S
7 3 51 4 4
0 .3 0 2 118 0 .38 6
0 .3 0 1 0 .4 6 5 09 0 .2 28
24 0 .4 26
94 0 .4 4 I 9 4 0 .3 0 3
0 .5 7 1
0 . • 3 1 03 3 . 0 .2 8 5 •
0 . 73 1
0,338
F igu re 13. Bush d iagram show ing 2-year RMS
Resu lts
54
0 30 3
Im o. 0 0 089
02 8 7
0. 0 3 3 SO 0. 0 00
0 .3 3 5 ss
0 .0 5 5
sis
F igu re 15 . Bush d iag ram show ing 10-year RMS
An assessment o f the FSR ra infa ll-runo ff method o f d es ign flood

es tima tion
• 55
5 . 16 A SSESSI NG T H E ENT I RE FLOOD FRE UENC Y C UR E

•
The resu lts presented so far have been in terms of the average erro r
ID across all catchments in a set for ind ividual quantiles . To exam ine
the performance o f the FSSR 16 mode l (FSSR 16(Tp&SPR)) on each
ID catchment individua lly an index o f the who le flood .frequency curve
was needed , ind icating , for examp le , whether the estimated curves
ID were too steep , too shallow , or about the right s lope but a lways
over- or underestimated . To achieve th is objective the slope of the
ID flood frequency curve was assumed to be adequate ly described by the
slope o f two segments of the curve
ID
• the s lope o f the curve between the 2 and 10 year floods
the s lope of the curve between the 10 and 50 year floods .

ID
In other words , how much larger is the 10 year than the 2 year return
period flood , or the 50 year than the 10 year? The magn itude o f the
10 year flood can be used to ind icate the typical size of flood s on
ID the catchment and therefore fix the general location o f the curve
on the flood magnitude sca le . In o rder to arrange the 88 curves into
a sma ll number o f groups a 25-way c lassification system was devised .
The s lope of each segment was c lass ified according to whether the
40 value was
1. a marked underestimate o f that observed < -20%
2. a slight underestimate -10% to -20%
3. about the same + 10% to -10%

ID
4. a slight overestimate 10% to 20%
5. a marked overestimate. > 20%

40 Each o f the 10 year flood ind ices was classified accord ing to a
similar 5-way scheme . Thus each catchment was assigned to one o f
25 c lasses , each of wh ich has a unique description eg . c lass two
consists of those catchments where the s lope o f the curve is s lightly
underestimated and the index is marked ly underestimated .
Tab le 5 .16 .1 shows the results o f th is 25-way class ification as

indexed by the 10 year return period flood and the ratio o f the 10
and 50 year floods . Table 5 .16 .1 shows that the errors in both s lope
and index are not symmetrica lly d istributed ie . there a re relatively
few catchments where the index o r slope is slightly over- or under
estimated , with the majority being either about right or very wrong .
There are 26 (out of 78) catchments where the estimated flood fre-
quency curve , between the 10 and 50 year floods , is too sha llow and
30 where th is s lope is too steep . For 12 of these oversteep curves
the index 10 year flood is a lso marked ly overestimated .
Tables 5 .16 .2 and 5 .16 .3 show similar c lassifications for the 2 and
10 year return period floods , the former using the 10 year flood as
Resu lts
ID
ID
ID
56
the index , the latter using the 2 year flood . Table 5 .16 .2 shows
that for 22 of the 88 catchments the s lope is overestimated a long
w ith the 10 year flood . It is also noteworthy that in over ha lf o f
the catchments (47) the ratio o f 10 to 2 year flood has been over-
estimated by more than 20% and three-quarters by more than 10% .. One
possib le exp lanation for these resu lts is that the relationship be-
tween rain fa ll and flow return periods derived in the FSR simu lation
exercise (see Append ix B) is too steep .
The numbers in the right-hand boxes are reversed between Tab les
5 .16 .2 and 5 .16 .3 . This suggests that , for many catchments , the
estimated flood frequency curves cross the observed , between the 2
and 10 year flood quantiles , thus the 2 year flood is underestimated
and the 10 year flood is overestimated . The majority o f catchments
in Tab le 5 .16 .3 are still in the right-hand boxes ind icating that
the s lope of the curve between the 2 and 10 year floods is marked ly
overestimated on as many catchments as the slope between the 10 and
50 yea r floods .
An assessment o f the FSR rainfa ll-runo ff method of design flood

estimation
N.
in
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
58
% ERROR RAT IO OF 10 AND 50 YEAR FLOODS
< -20% <- 10% - 10% - + 10% >+ 10% >+20%

4I
Too flat About right Too steep
% E880 8 10 * * ** * * ** * * * ** * * * * * * * * * * * * * * * * ** * * * * * * * * * * * * * * * * ** * *
le] *2 *3 *4 *5 *
<-20% * 3 * * 9 * * 4 *
*
Too sma ll * * * * *
**************************** ***********************
*6 *7 *8 *8 * 10 * 40
-20% - -10% * 3 * * * * 1 *
* * * * * *
************** ******* ********************* *********
* 11 * 12 * 13 * 14 * 15 * 0
41
* 9 * 2 * 7 * 1 * 2 *
About righ t * * * * * *
********* ********************************* *********
* 16 * 17 * 18 * 19 * 20 *
10% - 20% * 1 * 1 * 4 * 3 * 1 *
*
* * * * * 4I
******************************************* **** ****
*21 * 22 * 23 *24 * 25 *
>+20% * 2 * 1 * 11 * 4 * 12 *
Too big * * * * * *
********************************** ********** **** ***
ID
Number o f catchm ents appea rs in centre o f box .

Box numbe r is in top le ft o f box .
An assessment o f the FSR ra in fa ll-runo ff method of d esign flood

estimation
40
59
Box Ca tchment numbers
1 52006 55008 56004

2 45003 540 19 56006
3 4 1006 46003 48005 52005 520 10 53005 53009 5800 2 77002
4 69027
5 23005 330 14 46005 5800 1
6 39026 40006 71004
7 56005
8 7 1003
9 53007
10 550 12
11 34005 36008 4 1005 4 1007 4500 4 48004 54006 57004 5 7005
12 27001 54022
13 24005 2400 7 29004 33045 3700 1 540 11 6 1003
14 6500 1
15 30004 3800 7
16 3 1005
17 68006
18 19002 2000 1 37007 45002
19 3000 1 34003 660 11
20 840 12
21 28070 400 10
22 56003
23 19005 35008 39022 39025 39052 40009 4 10 15 4 1028 540 16 60002
67008
24 39004 54004 6700 3 84008
25 1900 1 27035 2900 1 3302 9 3900 5 3900 7 390 12 390 53 4 7007 6 1001
6400 1 72002
Tab le 5 .16 .1 C lass ifica tion o f flood frequency curve estima tes by
method 5 (FSSR 16(Tp& SPR )) accord ing to the
accuracy o f estimation of the 10 year flood and ratio
o f 10 to 50 yea r flood s as compared with obse rved da ta .
Resu lts
60
X ERROR RAT IO OF 2 AND 10 YEAR FLOODS
< -20% < -10% - 10% - + 10% >+ 10% >+20%
Too flat Abo ut right Too steep
% ERRO R 10 **************************** ****** ************ ** ***

*1 *2 *3 *4 *5 *
< -20% * * 0 * 3 * 5 * 11 *
Too sma ll * * * * * *
************************** ** ************* **********
*6 *7 *8 ore * 10 *
-20% - -10X * 2 * 0 * 0 * 2 * 3 *
* * * * * *
*********** ***** ********** ********* ************** **
* 11 * 12 * 13 * 14 * 15 *
- 10% - + 10% * 3 * 2 * 5 * 4 * 7 *
A bout right * * * * * *
*************************** ************************
* 16 * 17 * 18 * 19 * 20 *
10% - 20% * 1 * 1 * 1 * 3 * 4 *
* * * * * *
********* ************************ ******************
*21 *22 *23 *24 * 25 *
>+20% * 2 * 0 * 5 * 1 * 22 *
Too b ig * * * * * *
*************************** ********* *********** ****
Numbe r o f catchments appea rs in cen tre o f box .

Box numb er is in top le ft o f box .

es tima tion
411
• 61
ID
ID
ID
Box Ca tchment numbers
1 52006
3 45003 540 19 56004
ID 4 48005 52005 520 10 55008 56006
5 23005 330 14 4 1006 46003 46005 53005 \ 53009 5800 1 58002 69027
ID 77002
6 39026 40006
9 71003 71004
10 53007 550 12 .56005
11 4 1007 48004 54006
12 36008 4 1005
ID 13 33045 34005 45004 57004 57005
14 24007 2700 1 29004 540 11
411 15 24005 30004 3700 1 38007 54022 6 1003 6500 1
16 2000 1
17 3 1005
18 19002
ID 19 3000 1 37007 68006
20 3400 3 45002 660 11 840 12
411 21 280 70 400 10
23 19005 35008 4 10 15 56003 84008
24 1900 1
25 27035 2900 1 33029 39004 39005 3900 7 390 12 39022 390 25 39052
ID 39053 40009 4 1028 47007 54004 540 16 60002 6 100 1 6400 1 67003
67008 72002
40
ID
ID Tab le 5 .16 .2 C lass ification o f flood frequency curve estimates by the

method 5 (FSSR 16(Tp&SPR )) accord ing to the
accu racy o f estimat ion o f the 10 year flood and ratio
o f 2 to 10 year flood s as compared w ith ob se rved data .
40
Resu lts
ID
41
62
ERROR RATIO OF 2 AND 10 YEAR FLOODS
<-20% < -10% - 10% • + 10% >+ 10% >+20%
Too flat About right Too steep
% ERRO R 2 *************************** ******* *************** **

*1 *2 *3 *4 *5 *
< -20% * * 0 * 3 * * 21 *
Too sma ll * * * * * *
************************ *** ****************** ******
*6 *7 *8 *9 * 10 *
-20% - -10% * 0 * 0 * 0 * 0 * 2 *
* * * * * *
********** ***************** ****************** **** **
* 11 * 12 * 13 * 14 * 15 *
- 10% - + 10% * 2 * 2 * 5 * 7 * 12 *
About righ t * * * * * *
**** **************** ****** ***************** **** ****
* 16 * 17 * 18 * 19 * 20 *
10% - 20% * 0 * 0 * 2 * 0 * 1 *
* * * * * *
**************** **** ********************** **** *****
* 21 *22 * 23 *24 *25 *
>+20% * 6 * 1 * 4 * 1 * 11 *
Too b ig * * * * * *
**** ******************** ****************** ***** ****
Number o f catchmen ts appea rs in centre o f box .

Box number is in top left o f box .
An assessment o f the FSR ra in fa ll-runo ff method of des ign flood

est ima tion
41
• 63
41
41
41
41
•
Tab le 5 .16 .3 C lass ification o f flood frequency curve estimates by the
• method 5 (FSSR 16(Tp&SPR )) accord ing to the
accuracy o f estimation o f the 2 year flood and ratio
41 o f 2 to 10 yea r flood s as compa red w ith ob served data .
41
41
411
Resu lts
41
41
41
41
64

estimation
ID
• 65
ID
•
ID
41
41
6 .0 EX AMPLE CAT CHMENT S
41
6. 1 INT RODUCT ION
41
In Chapter 5 results were given for the who le data set and for var-
41 ious subsets based on physical characteristics o f the basins or the
quantity and qua lity of hydro logical data . In this section data and
411 results are presented for six example catchments . These catchments
are :
ID
19001 A lmond at Craig iehall
29001 Weithe Beck at Brigs /ey
39012 Hogsm ill at Kingston
41 46003 Dart at Austins Bridge
54016 Roden at Rodington
ID 55008 Wye at Cefn Brwyn
41 Three o f these (19001, 39012 and 54016) were tria l catchments in the
FSR and therefore not used in developing the FSR regression equations
ID or in the simu lation exercise (Append ix B). These three catchments
and 29001 were not used in the rev iew o f the FSR ra in fa ll-runo ff
411 mode l parameter estimation equations on which FSSR16 is based . The
extra two catchments are inc luded to give a better distribution and
range o f catchment types . Va lues o f catchment characteristics can
be found in Table 2 .1.
ID
For each of the six catchments , two g raphs are g iven comprising (a)
the annual maximum data and the fitted curve , and (b) th is curve
p lotted w ith the various estimated flood frequency curves . Beware
o f scale changes within each pair o f figures .
ID Percentage runoff has a lready been ident ified as a most important

variable to estimate we ll if accurate flood estimates are to be made .
411 With th is in mind , for some catchments , graphs are presented which
illustrate how percentage runoff varies w ith ra infall for observed
events , and how estimated percentage runoff increases using the
FSSR16 model. A lso shown in this figure is a curve that shows what
percentage runo ff wou ld be required to est imate perfectly the fitted
curve . The difference between this line and the FSSR 16 model shou ld
40 not be viewed as the error in the PR mod el. E rrors are obvious ly
present at various stages of the estimation procedure. This perfect
fit line g ives a representation that assumes a ll such errors are in
the PR term . If this line is quite d ifferent from the observed data
411 it points to there being errors in some other part of the procedure .
Example catchments
111
ID
66
6 .2 19001: A LMOND AT CRA IG IEHA LL
Period of record , N : 29 years

Eyeball limit on return period : 10 years
Return period for jackkn ife standard error of 12 .5% : > 100 years
Mod ified Anderson-Dar ling statistic : 0 .237
F irstly , conside r the fit to the annual maximum data (Figure 16a).
The fitted GEV has a positive k (0 .14 ) and therefore curves down-
wards . The largest two peaks are we ll below the fitted line and are
large ly resposib le for its downwa rd curvature. Looking at the data
points it might be thought that there is a kink at a return period
o f about 5 years (160 m3/s). Such features in the p lotted data can
be caused by rea l catchment or hydraulic effects , or may be tota lly
spurious . In a study involving so many catchments the data could
not be investigated in detail. The authors wou ld feel unhappy using
the GEV curve to extrapo late beyond 10 years and wou ld prefer to use
an EV 1 d istribution . Be low 10 years these two curves are very simi-
lar, hence the eyeball limit o f 10 years . This is greatly less than
the 60 years o f the 2N ru le . The modified Anderson-Darling sta-
tistic of 0 .237 reflects the variation o f the annual maxima about
the fitted curve above 4 years. At first sight it seems surprising
that the jackknife limit is greater than 100 years . Th is is because
o f the positive k ; as return period increases the curve flattens and
the standard dev iation decreases .
From the estimated flood frequency curves (Figu re 16b ) , it can be

seen that the no data rainfa ll-runoff estimates are a ll too large
and the cu rves too steep (hence the catchment appears at the bottom
and to the right in the box d iagrams in Tables 5 .16 .1, .2 & .3) .
On ly four observed events were availab le but they gave model param-
eter va lues very similar to those obtained from regressions . These
resu lted in s lightly larger estimated peaks . The statistical method
underestimates the observed data .
An assessment o f the FSR rainfa ll-runo ff method o f design flood

estimation
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 1 11
C o m p a ris o n o f f re q u e nc y c u rv e s
F S S R I O( T p A S P R )
(b) of
F s s Ft I el ( C C )
E y • b • II l i m i t i /
FSR/ STATS
I // /
, / / /
i .9/ / j•
/ _.. . OATA/ STATS
°
i /
R e t u r n p e r io d T years
5 10 25 50 10 0
0 1 2 3 4 5
R e d u c e d v a r ia t e Y R e d u c e d v a r ia t e Y
68
6 .3 29001: A IT HE BECK A T BRIGSLEY
Period o f record , N : 26 years

Eyeball limit on return period : 25 years
Return period for jackknife standard error o f 12 .5% : 8 .2 years
Mod ified Anderson-Dar ling statistic • 0 .982
Figure I7a shows that 25 of the 26 annua l maximum flows plot on a

near perfect straight line ; because the 26th p lots we ll above this
line the fitted QEV curves upwards (negative k ). The modified
Anderson-Darling stat istic is 0 .982 , closer to a perfect fit o f unity
than might have been thought by inspection . A lso surprising is that
the return period corresponding to the jackknife limit is 8 .2 years .
The eyeba ll limit was 25 years , indicating that the data comprise
one o f the most consistent sets in the study . Note the scale of peak
flow in the figure ; the mean annual flood is about 2 m3/s and the
25 year flood 4 .5 m3/s . The catchment has an area of 108 km2 but has
80% WRAP type 1 soils .
FSSR16(CC) estimates are much too big but are g reatly improved by
using parameter estimates based on observed data . However, as the
peaks are so small an estimated mean annua l flood of 4 m3/s repres-
ents a large error in percentage terms . The estimated curves are
steeper than the observed one so the catchment is in the bottom
right-hand corner o f the box d iagrams (Tab les 5 .16 .1, .2 & .3)
It is worth look ing slightly deeper at the percentage runoff va lues

obtained from the events . F igure 18 shows all observed percentage
runoffs were between 1.0% and 3 .3% , yet when they were converted to
SPR (as indicated by the arrows) a mean value o f over 10% was ob-
tained . The lower line representing the percentage runo ff requ ired
for a perfect fit passes through the observed percentage runo ff
va lues , but the upper line , used in calculating the des ign flood
peaks , is much higher . However, the abso lute errors in estimating
percentage runo ff are small; at the 5-year level 4% is estimated when
2% is requ ired for a perfect fit.

estimation
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
C A T C H M ENT 2900 1
W a it h e B e c k a t B r ig s le y C o m p a r is o n o f f r e q u e n c y c u r v e s
9 40
(a) (b)
35
30
to 0)
i s S R 1 e (C C )
6 E y. b t Il ln+11
E 25
m I ..,
O oe
... .09
a) 2 0 6,
ta /
._ ..." FM Ie (Tpa sPR )
O
n . 0"
fl
O 15 7 ° ••••"
u, 3 O tsR/STATs
0 / ..-'-- .-
.7 ...-
10 --' I •- •
/ ..- ••••
•••• DA T A / ST A TS
• i
Re turn p e r io d T ye ar s 5 • e t u r n p e r io d T y e a r s
2 5 10 2 5 50 10 0 2 5 10 25 50 10 0
0 0
-2 -1 0 1 2 3 4 5 0 i 2 3 4 5
R e d u c e d v a r ia t e Y R e d uc e d v a r ia t e Y
‘ 10
2900 1
20
K EY
S P R • l lm a t • d f r o m
W RAP min)
P R re qu ir e d to r
16
Pe r fe c t c o ns truc tio n of
flood fre quency curve
- - - PR fro m F S S R 16
12 w $t h S PR fr o m lo c a l d a t e
L in e s d a s n e d a b o v e e y e b a ll lim It
8 P e r c e n t a g e r u no f f ( P R )
S PR from a na ly sis
of e v en t da ta
c Or r e e p o n d in g s t a nd a r d
I p e r c e n t a g e r u n ta l t ( S P R )
4
. ••
i tt
0
0 20 40 60 80 10 0
R a in f a ll ( m m )
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
•
• 71
•
I 6 .4 39012: HOGSMILL AT K INGST ON
Period o f reco rd , N 27 years

• Eyeba ll lim it on retu rn period 10 years
Re turn pe riod fo r jackkn ife standard erro r of 12 .5% 10 .9 years
• Mod ified Anderson-Da rling statistic 0 .188
41 Th e p lotted annua l max ima (F igu re 19a ) show some deviations from the
fitted GEV d istribu tion , hence the low va lue o f the mod ified
• And erson -Darling statistic . The k va lue is ve ry c lo se to zero so
the fitted d istribu tion is very c lose to be ing an EV 1 . The eyeba ll
• and jackkn ife lim its a re bo th 10 years .
41 Th e FSSR 16(CC ) est imates are too large (see Tab le 6 .1) . Us ing ob -
se rved da ta va lues o f SPR and Tp g ives the co rrect va lue for the
. 2-year flood , bu t the estimated flood frequency curve is too steep .
The statist ica l method g ives est imates that are too sma ll .
41
However , bo th sets o f estimates use the rura l method a lthough the
41 catchment is 46% urban ised and the urban co rrections to the method s
should be app lied . Tab le 6 .1 g ives the va lues us ing the urban
• methods .
41
41 Return Data Ra infa ll-runo ff method s Statistica l
Period method
41 Ru ra l Urban U rban Rura l U rban
Tp+SPR
41
41
41 Tab le 6 .1 Compa rison o f flood est imates us ing the rura l method
and the u rban corrections .
41
The ra in fa ll-runo ff estimates have hard ly chang ed as there is on ly
41 a s ligh t d ifference in pro file shape and use o f ra in fa ll retu rn
pe riod . The statist ica l method est imates have increased to be ve ry
41 c lose to the data . Tha t the statist ica l method wo rked poo rly be fo re
is not surp ris ing as the method made no a llowance for the
41 u rban isat ion (un like the ra in fa ll-runo ff method ). Th e adjustment
made in the statist ica l method is based on the mode l used in the
41 rain fa ll-runoff method .
41
Examp le ca tchmen ts
41
41
41
41
Dr > -4
2 - 00 AD
2 12
14 -
CT 0
0
X re
7
(0 . al
1•••
C A T C H M EN T 3 9 0 12 a, n
Cie0
1-1
,..,
co0
Fl o g s m il l a t K in g s t o n Co m p a r is o n o f f re q u e nc y c u rv e s a 0.
ca a
0
48 70 0 =
(a) (a)
r•- 1
42 co c°
60 0 13
_ . 13
cb M
Ca g :.
e l' S
.-.. 3 6 40 50 o
O l il
U
U
op 11• • ol
W
E - FS SR I I ( Tp & SPA )
E 30
o ' 40
u .— 1—
(2 ...
— / .. . - S
u 24 o I ra .° li m i t .. • " .
o o • rt
30 I - -- DAT A/ 3 TM " 7-
to r ,-- - -- - - es
r i8 • • 17+
u u I, - - -
- - --- -
VI el . .. . Cr
• .•-• I ......----- - ,..,
• 0 re
a •• a 20
12 ..... ..... I . F II R I S T A T 3 . rt
•' / 0
•. .. - . . ./ 3
10 I . .../ • - Pe
6 Re turn p e r io d T y ears __.- -- • - - R e t u r n p e r io d T y e a r s i-1
i—
2 5 10 25 50 10 0 2- -- 5. 10 25 50 10 0 Coo
5.
0 0 r t•
- 2 -I 0 I 2 3 4 5 0 1 2 3 4 5 a
o
Re duc e d v a r ia t e `I Reduc ed v a r ia t o 1' z
0.
n
0
,I
S
fe
1-1
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
41
• 73
•
6.5 46003 : DART AT AUST INS BR IDGE
•
Period of record , N : 27 years
II Eyeball limit on return period : 5 years
411 Return period for jackknife standard error of 12 .5% : 10 .9 years
Modified Anderson-Darling statistic • 0 .221
•
As can be seen in Figure 20a there is one very large flood in the
annual maximum series ; it is almost twice the size of the next big-
gest . W ith such an extreme flood in the series„the fitted GEV o f
411 course curves upwards ; without this flood it would have curved the
other way . There is obvious ly some concern about the p lotting po-
411 sition o f this large flood . The authors made their eyeball limit
5-years , less than the 10 years jackkn ife limit. The mod ified
Anderson-Darling statistic is relatively small.
Figure 206 shows that all the estimates are low ; the statistical
method is the best . Using model parameters based on event data has
made the estimates worse , main ly by reducing the SPR va lue from &bout
3 6% to 30% . The rev ised va lue has come from 23 events d istributed
•
throughout the year . Figure 21 shows that the event ra infalls were
• generally between 20 and 60mm (one event had 122mm). This figure
also shows that percentage runo ffs varied between 17% and 42% , giving
• an average SPR o f 30% . The large rainfa ll event has a lmost exact ly
this average SPR . The line p lotted on the diagram to represent the
PR needed to estimate perfect ly the observed flood frequency curve
is much h igher than any o f the event data which sugg ests strong ly
that an error in PR estimation is not the cause of the poor flood
peak estimates .
That estimates should be so poor even though over 20 events are

available is worrying and is the subject of further research . Apart
from looking again at percentage runoff estimation , there are severa l
other concerns that should be considered .
10 The event flow peaks were between 60 and 200 m3/s (ie . a ll less
• than the mean annual flood ). Events w ith larger peak flows may
give larger percentage runoffs .
The triangular unit hydrograph may not be suitab le for the

catchment . A peakier unit hydrograph wou ld give some increase
in des ign flood peaks.
•
The rainfa ll input is too low or the profile shape too d iffuse .
Rainfa ll depths may have been overestimated for the observed

• events .
• It is most likely that a combination of factors is respons ib le .
Example catchments
41
ID
ID
411
74
The catchment appears in the top line o f Tab les -5 .16 .1, .2 & .3 ,
which shows the magn itudes are underestimated , but in the centre and
right hand columns ind icating the slope o f the estimated flood fre-
quency curve is about right or too steep .
CD
a
P
0
z =
10 7;
.t .
o 71 7
1—
ø(
U 0
( s o a un i a ) a 6 l e y a s K )
F igure 20 . Da rt at Austins Br idg e (flood frequency curves )
An assessment o f the FSR ra in fa ll-runo ff method of d esign flood

estima tion
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
46003
75
KEY
.. 0
60 S P R • s t im • t e d f r o m ...- - - — — P R r e q u ir e d t o r
.. -• p • r f e c t c o n • tr u c t Io n o f
. .- W R A P ma p ..---" ......... ., • " ..
flood fre que nc y c ur ve
0
c
c
._ PR fro m F S S R 1 0
45
a) w it h S P R fr o m lo c a l d a t a
o)
ai
Linn d a sh e d a bo v e e y e b • 11 !kn it
4a)a
,(.-2 30 t it I fr i -- - -------- p e rc e nt a ge runo ff ( PR )
0 t I
O.
, y1 i Cor re SPOnd Ing s t a nda rd
p e rc e nt a g e r un o ff ( S P R )
I
15
I
S P R f r o m • n • ly s is
o f e n n t cl • t •
0
0 30 60 90 12 0 150
R a in f a ll ( m m)
76
6 .6 54016 : RODEN A T ROD INGT ON
Period o f record , N . 23 years

Eyeba ll limit on return period :.10 years
Return period for jackknife standard error of 12 .5% : 15 .8 years
Mod ified Anderson-Darling statistic : 0 .496
The p lot o f annual max imum data has a strong trend but has two
,
'waves ksee Figure 22a); the scatter about the fitted line is typ-
ical of a curve w ith a modified Anderson-Darling statistic o f 0 .5 .
The eyeball limit was 10 years , s lightly less than the 16 years
correspond ing to the 12 .5% jackknife standard error .
A ll the methods g ive peaks larger than the observed data (see
Figure 22b) . The performance o f the rainfall-runo ff method is im-
proved by using observed data , but is sti ll not as good as the sta-
tistica l method wh ich agrees well w ith the observed data at low
return periods . The event data are from seven events , but they are
all small (the largest is just over 11 m3/s). F igure 23 shows that
three events have SPR values greater than PR , arising as they do
from events with large SMDs . As on catchment 29001, for the design
case , using mean PR rather than mean SPR would g ive a better repre-
sentation o f the observed flood frequency curve .
The catchment fa lls in boxes 23 , 25 & 20 of Tab les 5 .16 .1, .2 & .3
indicating that the estimates and the slope of the estimated flood
frequency curve are too large .
An assessment of the FSR rainfall-runoff method of design flood

estimation
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
C A T C H M EN T 5 4 0 16
R o d e n a t R o d in g t o n C o m p a r is o n o f f r e q u e n c y c u r v e s
b F S S R 1 0 (C C )
F 3 3 11 1 0 (T p & S P R )
E y • b ibli lim it / / e F O R /S T A T S
1 / • •• /
/ e. •
/
/ • D A T A /S T * 1'S
/
/ / •
/ , ' 1 / 4/
/ / / 7 1>
R e t u r n p e r io d T years
./ / , .Z . I
2 5 10 25 50 10 0
0 1 2 3 4 5
R e d u c e d v a r ia t e Y R e d uc e d v a r ia t e Y
54 0 16
50
K EY
40 P R re q ui re d t o r
S PR e st ima te d from
Pe rf e c t c o ns tr uc tion of
W R AP ma p
flo od fr e oue nc y c ur ve
— PR fr om PS S R 16
30
w it h S PR fro m lo c a l d a t e
L ine s da sh e d a b o ve e y e ba ll limit
.1= • • .. = • •
20 i p e rc e n ta ge r unoff ( P R)
c orre sponding st a nd a rd
pe rc e nt a ge ru no ff ( S P R )
SP R from a na ly si s
10
of event dote
0 20 40 60 80 10 0
R a in fa ll ( m m )
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
79
The GEV d istribution fitted to the annua l maxima has a large negative
k (-0.31) and therefore curves steep ly upwards as shown in
F igure 24a . The figure also shows that the annual maxima are d is-
tinctly stepped and that there are two very large floods . The mod i-
fied Anderson-Darling statistic is extremely low . The authors felt
happy using the fitted curve only be low 5 years despite 34 years o f
data . The return period correspond ing to a jackknife standard de-
v iation o f 12 .5% is slightly higher .
The FSSR 16 no-data rainfall-runo ff estimate agrees w ith the data at

2 years but is too b ig at 5 years . Incorporating obse rved data de-
creases the estimates to below the observed values . The statistica l
method over-estimates l or low return periods but is then too flat
so it intersects the observed curve at 10 years , as shown in
Figure 24b . Figure 25 shows the event data represent a good range
of rain fa ll depths and that variations in percentage runoff are
roughly as pred icted by the estimation equation .
The catchment appears in the top line o f Tab les 5.16 .1, .2 & .3 ,
which shows underestimation o f magnitudes , but moves from the left
to the right side of the tab le showing that the estimated flood
frequency curve is too steep at low return periods and not steep
enough for high return periods .
Example catchments
(to CO
41 > 0
o
•-,,
rp 64 60 DATA/ STATS
r• (a) (b)
m
IT1 /
CA 56 54
PO
11
ø 48 48 /
I- . F S S R I OI C C )
7 /
Co U U / " ./ . F OR / 3 7 A T *
1—• c 40 tz 4 2
,— E E
Pa o o F S S R I O( T o & S P R )
c / ./ /
c 32 36
o cz ca 7
1-h E y • b o l l li mi t 7
pi , :El /
;1
7
a 0 2 4 0 30 I . ..- - - -
o co
,CD
r• a _•• 6
a. 16 ... . 24
o 6
-,, .d; 5; : r °
8 R e t u r n p e r io d T y e a r s 18 R e t u r n p e r io d T y e a r s
/ I
R e d u Ce d v a r i a t e Y R e d u c e d v a r ia t e Y
0
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
• • • • • • • • • • • • • • 411" • • • • • • • • • • • • • • • • • • •
55008
10 0
KE Y
80 — — PR re quir e d for
pe rfe c t co ns truction of
SP R es timat ed from flood freque ncy curve
...
—.. WR AP map
o
c — — PR Ir on FS SR I 6
2 60 wit h SP R fr om loc a l d a t a
a) /
o)
- - - - - - ----- - - 1 l ines da she d a bove e y e ba ll limit
.iCi-O
c i
CD
C.) 40 \ Perc entage runoff (PR)
05
1
a 1 co rre sponding sta nd a rd
1 (torc onta go runoff ( S PR )
SP R fro m • na ly sla
of e ve nt da ta
20
0 30 60 90 120 150
R a in f a l l ( m m )
CO
82
An assessment o f the FSR ra in fall-runo ff method of des ign flood

estimation
• 83
•
41
41
41
7.0 CONC LUSIONS
The objective of th is study was to make a de finitive assessment o f

flood estimates (up to return periods of 100 years) on rura l
catchments using the F lood Stud ies Report rainfall-runoff method . '
Many prob lems , both expected and unexpected , were encountered , but
the study has provided a quantitative insight into how the FSR
rainfa ll-runoff method performs and has ind icated some of its po-
tentia l weaknesses .
The first problem encountered was identify ing a set of catchments

41 for which flood quantiles could be calcu lated re liably , and for which
flood event data were ava ilable to permit refinement of the
41 rainfa ll-runoff mode l parameters . This latter consideration imposed
the greater constraint A S only 128 catchments had the requ ired 5
41 events available . The orig inal idea of trying to examine flood mag-
nitudes up to the 100-year return period had to be mod ified as no
41 catchment for which event data were available had su fficient annual
maxima data to prov ide reliable estimates of the 100-year flood .
41 Eighty-eight catchments were selected for study by lowering the re-
qu irement for annua l maxima data to just 15 years . Accord ing to the
• 'twice-the-period -o f-record" ru le-of-thumb it was ant icipated that
all catchments could have been used for a comparison o f estimates
• up to 25 years , which wou ld still have been a considerab le im-
provement on resu lts pub lished from other stud ies .
In co llating the true quantiles against wh ich the estimates were to

41. be compared , p lots were produced show ing the goodness of fit of the
observed annual maxima to the fitted d istributions . They showed very
41 c learly the problem o f trying to identify the true d istributions from
limited amounts of data . Some data sets p lotted consistently and gave
confidence in using quantiles from the d istribut ion, but other data
sets were very inconsistent and gave little con fidence , even at re-
ID turn periods less than the period of record . A limiting return period
was set by inspection (and termed an eyeball limit) to mark the
max imum return period at wh ich the fitted d istribution WA S considered
re liab le . In a ll cases th is WA S less than tw ice the period of record
(2N ), and usually less than N .
This comparision w ith the 2N ru le is unfair as the rule g ives a guide

to the point at which a deve loped flood frequency curve shou ld depart
from the at-site line to the regiona l curve . The ru le is concerned
w ith how to get the best estimate of the T-year flood from observed
41 annual max ima , rather than how to be sure that the estimate is ac-
e
Conc lusions
41
84
curate enough to be used as a benchmark for testing ind irect methods

of estimation . However , the compar ison does serve as a reminder o f
how carefu l one must be in us ing at-site annual maximum data .
Considerab le effort was expended in try ing to replace the eyeball

limit w ith a statistically derived one that cou ld be app lied ob-
jectively and , therefore , by anyone unfamiliar w ith the prob lems o f
fitting d istributions to observed data . Some success was achieved
by combining the mod ified Anderson-Darling statistic w ith a standard
error from a jackknife analys is , but this is still far from being a
standard technique that can be app lied generally .
When catchments with urban areas greater than 10% were exc luded , and
the eyeball limits were app lied , just 74 catchments were left on
which to assess estimates of the 2 year flood . The number of
catchments reduced as return period increased , so that on ly 15
catchments were availab le for compar ison at the max imum return period
of 25 years . These catchments are not we ll d istributed geograph-
ically ; there are none in Northern Ireland , none in Scotland outside
the central low lands , and none in the Lake D istrict or northern
Pennines . Comparisons using subsets o f the data (eg . small or wet
catchments ) were based on very few catchments.
Using this standard data set , the no-data rainfa ll-runoff method ,
w ith parameters estimated by the FSSR16 regression equations, tended
to over-estimate flood magnitudes . However, estimates were greatly
improved when model parameters were derived from observed data ; bias
was then zero at the 2 year return period , increasing w ith return
period to 11% at 25 years (the correspond ing figures in the no-data
case were 22% and 4 1%). It is reassuring that w ith parameters from
observed data the method is seen to work fairly we ll. Va lues of
standard percentage runoff were seen as particularly va luab le in
improving estimates . The statistical method , w ithout observed data ,
performed as we ll as the rainfa ll-runoff method w ith data .
In the various subsets of catchments that were examined the per-

formance o f the ra in fa ll-runoff method varied much as expected .
Estimates were better at higher return periods if more events were
availab le , they were worse on catchments w ith main ly permeable soils ,
and , in the no-data case , better on larger catchments. The variation
in the perfo rmance o f the statistical method was more random between
these subsets . A p lot of the residua ls on a map of Brita in showed
genera l over-estimation in the south-east o f England and under-
estimat ion in south-west Eng land and Wa les ; in other reg ions resi-
dua ls were mixed . Divid ing the catchments into those w ith more or
less than 800mm average annua l ra infall showed that the estimates
tended to be better on wet catchments than on dry ones . The various
derived statistics have been presented in three d ifferent ways in
summary form in the body o f the report , and comprehensively in an
appendix .
An assessment of the FSR ra infall-runoff method of des ign flood

estimation
41
• 85
•
This study has provided statistics describing the performance of the
• rainfa ll-runoff method of flood estimat ion . It is seen to work
reasonab ly well in most cases where model parameter va lues can be
41 derived from observed data . If such data are available , the method
performs about as well as the statistica l method without observed
41 data , but has severa l advantages : it can prov id e a comp lete design
hydrograph , it is based on a mode l of catchment response , and it
41 performs in a pred ictable manner. However , several areas for further
work remain and are described in the next chapter .
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
Conclusions
41
41
41
41
41
86 41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
An assessment of the FSR rain fall-runo ff method of design flood 41
estimation
41
41
41
41
41
• 87
41
8 .0 RECOMMEND AT ION S
The lack of catchments availab le for this study c learly demonstrates

41 the continuing need to gather both event data and annua l maximum
data . Such data underpins the des ign o f land dra inage schemes which
• accounted for 4: 67M o f Water Authority cap ital expenditure in 1986/7
(the last year for which figures are currently availab le , Water Au-
41 thorities Association , 1987) . Authorities measuring data must be
made aware of the va lue of long records in estimating flood statis-
41 tics ; there is a national need for such data . There is a particu lar
need for data from remote catchments in Scot land and Northern
41 Ire land .
Work shou ld continue on trying to identify measures o f accuracy as-

sociated w ith fitting d istributions to annual maxima . Perhaps an
. expert systems approach is needed to encode the princip les used by
the authors in fixing their eyeball limits .
41
The rainfa ll-runoff method w ith observed data is unbiassed overa ll,
41 but there are particu lar catchments on which the flood frequency
curve is poorly estimated . Since errors are not elim inated when
41 observed mode l parameters are used , the method of defin ing the des ign
inputs, or the way in which model parameters are assumed to vary on
• individual catchments , must be at fault . It is not acceptab le to
state that the method is unb iassed on average when errors on par-
111 ticular catchments can be large . A great dea l o f further work is
required to improve the ac cu ra cy o f estimation on all catchments .
41
The value of using loca l data (ie . data from a gauged catchment
• 'hydrologica lly similar to the site o f interest) rather than data from
the site o f interest itself, should be considered , as it is th is type
• o f data which is normally available . The use of observed unit
hyd rographs instead of the triangular FSR one shou ld be investigated ;
41 if this proves beneficial , the use o f a more rea listic unit
hydrograph in the no-data situation should be cons idered . The type
• and qua lity of local data that g ive the g reatest improvements shou ld
a lso be investigated since larger storms and floods may be found to
41 be much more useful than smaller events .
• Wh ile improving the qua lity o f estimates obtained using the no-data
equations is a valid objective , no flood estimate for a site in the
6 UK shou ld be made without some reference to loca l data . Estimating
the variation in percentage runoff is of most importance . It is hoped
• that the ex isting project to rep lace the 5-class Winter Ra infa ll
•
ID Recommendations
41
88 41
41
Acceptance Potentia l map with a more detailed map w ith a greater
number of classes will produce sign ificant improvement in flood es- 41
timation . An important objective for research is the further de-
ve lopment of method s o f calibrating percentage runoff by reference 41
to easily extracted measures of catchment response , such as base flow
index . 41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
41
An assessment o f the FSR rainfa ll-runo ff method of design flood 41
estimation
41
41
41
41
ID
• 89
•
41
111
•
41
9 .0 REFERENCES
•
Ahmed , M .I ., Sinc la ir , C .D . & Spurr , B .D . (1988) 'Assessm ent o f
• F lood Frequ ency Mod e ls U s ing Emp irica l
D istribution Funct ion Sta tistics ',
411 Water Resou rces Research , Vol 24 , No 8 ,
p 13 12-1328 .
ID
Archer , D .R . (1980 ) 'A catchment app roach to flood estimation ',
ID J Inst Wat Eng rs Sci , 3 , 275-289 .
Archer , D .R . & Ke lway , P .S . (1987) 'A comp uter system for flood
estimation and its use in eva lua ting the
F lood Stud ies ra in fa ll runoff method ',
Proc Inst C iv Eng rs , Pt 2 , 9178 , 83 ,
ID 60 1-6 12 .
Beran , M .A . (198 1) Commun ication on 'A ca tchmen t approa ch to

flood est ima tion ', J Ins t Wat Eng rs Sc i,
•
529-535 .
ID Boorman , D .B . ( 1980 ) 'A rev iew of the Flood Stud ies Report
ra in fa ll-runo ff mode l p rediction equa tions
111 for No rth-Wes t Eng land ', Report to M in istry
o f Ag ricu lture F isheries and Food , No rth W est
411 Weather Rada r Proj ect .
Boorman , D .B . (1985 ) 'A rev iew of the Flood Stud ies Repo rt
ID rain fa ll-runo ff mod e l pa rameter est ima tion
equat ions ', Report No 94 ,
Institute o f Hyd ro logy , W a lling ford .
•
• Boorman , D .B ., Acreman , N .C . & C lay ton , N .C . (1988 ) 'A study o f

percen tage runo ffs on Sco ttish catchments ',
• Repo rt to No rth o f Scotland Hyd ro -electric
Board .
411
Gurne ll, A .M . & M idg ley , P . ( 1987) 'Re fin ing the estimat ion o f
41 percentage runo ff in catchments w ith
extreme hyd rogeo log ical condition s ',
411 National Hyd ro log ica l Sym posium , Hu ll,
14-16th Sep tember 1987 , British Hyd ro log ica l
• Society .
•
Re ferences
•
40
•
90
Hosk ing , J .R .M . Wa llis , J .R . & Wood , E .F . (1984 ) 'Est imation o f the

genera lised extreme value d istr ibution by
the me thod o f probab ility we igh ted moments ',
Report No 89 , Inst itute o f Hyd ro logy ,
Wa lling ford .
Lees , M .D . (1987) 'In land wa ter survey ing in the United K ingdom -
a sho rt h isto ry ', Institute o f Hyd ro logy .
Wa lling ford .
Lynn , P .P . (1978) 'A comparison o f severa l methods of F lood

Estimation ', App lied Hyd ro logy In fo rma l Note
23 , Inst itute o f Hydro logy , Wa lling ford .
M iller , R .G . (1974 ) 'The jackkn ife - a rev iew ',

Biome trika , 6 1, 1- 15 .
Natu ra l Env ironment Resea rch Counc il (1975 ) 'F lood Stud ies
Report ', NERC , London .
Natura l Env ironment Research Counc il (1977-87) 'F lood Stud ies
Supp lementa ry Repo rts (1-18) ', Inst itute o f
Hyd ro logy , Wa lling ford .
Reed , D .W . (1987) 'Engaged on the ungauged - ref lection s on the

app lication o f the FSR ra in fa ll-runo ff
method ', Nat iona l Hyd ro log ica l
Sympos ium , Hu ll, 14-16th September 198 7 ,
British Hyd ro log ica l Soc iety .
Reyn o lds , G . (198 1) D iscussion on 'Recent advances in flood

estimation techn iq ues based on
rain fa ll-runo ff' in "FSR-F ive Y ears On " ,
Proceed ings o f a con feren ce o rgan ised
by the Instn C iv Engnrs , Manchester ,
22-24 Ju ly 1980 , Thomas Te lford Ltd ,
London .
Sutc liffe , J .V . (19 78) 'Methods o f flood estimat ion :

a gu ide to the FSR ', Repo rt No . 49 ,
Inst itute o f Hyd ro logy , Wa lling ford .
Water Authorities Assoc iation (1987) 'Water Facts ', Wa ter

Autho rities Assoc iation , London .
An assessment o f the FSR ra in fa ll-runo ff method o f d es ign flood

estima tion
91
A PPEND IX A . FU L L T A B LES OF COMPA R ISONS A ND RESULT S
Append ix A . Fu ll tab les o f compa risons and resu lts

41
92
41
41
RTij RT 'ij
Me thod n mean RMS mean RMS 41
2 1 FSR /STATS(CC ) 74 0 .06 0 .43 0 .0 1 0 .16 41
2 FSR /RF -M CC ) 74 0 .27 0 .89 -0 .05 0 .2 1
3 FSSR 16(CC ) 74 0 .22 0 .73 -0 .04 0 .19 41
4 FSSR 16(Tp& SPR ) 74 -0 .0 1 0 .35 0 .03 0 .14
5 FSSR 16(Tp) 74 0 .16 0 .83 -0 .01 0 .19 41
6 FSSR 16(SPR ) 74 0 .0 7 0 .39 -0 .00 0 .14
41
5 1 FSR /STATS(CC ) 71 0 .02 0 .36 0 .02 0 .14
2 FSR /RF-RO(CC ) 71 0 .37 0 .92 -0 .09 0 .21 41
3 FSSR 16(CC ) 71 0 .34 0 .77 -0 .09 0 .20
4 FSSR 16(Tp& SPR ) 71 0 .0 7 0 .32 -0 .0 1 0 .12 41
5 FSSR 16(Tp) 71 0 .26 0 .83 -0 .05 0 .19
6 FSSR 16(SPR ) 71 0 .16 0 .38 -0 .05 0 .13 41
10 1 FSR /STATS(CC ) 57 0 .02 0 .37 0 .02 0 .15 41
2 FSR /RF-RO(CC ) 57 0 .28 0 .72 -0 .06 0 .19
3 FSSR 16(CC ) 57 0 .28 0 .64 -0 .07 0 .18 41
4 FSSR 16(Tp& SPR ) 57 0 .06 0 .27 -0 .0 1 0 .11
5 FSSR 16(Tp ) 57 0 .17 0 .6 1 -0 .04 0 .17 41
6 FSSR 16(SPR ) 57 0 .16 0 .37 -0 .05 0 .13
41
25 1 FSR /STAT S(CC ) 15 0 .09 0 .32 -0 .02 0 .14
2 FSR/RF -RO(CC ) 15 0 .39 0 .96 -0 .09 0 .22 41
3 FSSR 16(CC ) 15 0 .4 1 0 .83 -0 .11 0 .22
4 FSSR 16(Tp& SPR ) 15 0 .11 0 .4 1 -0 .02 0 .15 41
5 FSSR 16(Tp ) 15 0 .4 1 1 .04 -0 .08 0 .24
6 FSSR 16(SPR ) 15 0 .16 0 .40 -0 .05 0 .14 41
Tab le A .1 Fou r statistics fo r return per iods 2 , 5 , 10 and 50 years 41
compa ring the pe rformance o f s ix mod e ls on standa rd set
o f catchmen ts . 41
41
41
41
41
41
41
41
41
41
An assessmen t o f the FSR ra in fa ll-runo ff method o f des ign flood 41
es tima tion
41
41
41
41
41
• 93
41
RT RT 'll
41 Me thod mean RMS mean RMS .
• 2 1 FSR/STATS (CC ) 88 0 .0 1 0 .43 0 .03 0 .18

2 FSR/RF-RO(CC ) 88 0 .38 1 .23 -0 .07 0 .23
41 3 FSSR 16(CC ) 88 0 .34 1 .13 -0 .06 0 .22
4 FSSR 16(Tp&SPR ) 88 0 .10 0 .89 0 .0 1 0 .16
41 5 FSSR I6(Tp ) 88 0 .31 1 .49 -0 .03 0 .22
6 FSSR 16(SPR ) 88 0 .15 0 .69 -0 .02 0 .17
41
5 1 FSR /STATS(CC ) 88 -0 .02 0 .39 0 .04 0 .17
41 2 FSR/RF-RO(CC ) 88 0 .5 1 1 .44 -0 .11 0 .24
3 FSSR 16(CC ) 88 0 .48 1 .30 -0 .11 0 .23
41 4 FSSR 16(Tp&SPR ) 88 0 .2 1 1 .06 -0 .04 0 .16
5 FSSR 16(Tp ) 88 0 .45 1 .75 -0 .08 0 .23
41 6 FSSR 16(SPR) 88 0 .27 0 .78 -0 .0 7 0 .17
41 10 1 FSR/STATS(CC ) 88 -0 .02 0 .38 0 .04 0 .18

2 FSR/RF-RO(CC ) 88 0 .48 1 .42 -0 .10 0 .24
41 3 FSSR I6(CC ) 88 0 .4 7 1 .28 -0 .11 0 .23
4 FSSR 16(Tp&SPR ) 88 0 .20 1 .05 -0 .03 0 .16
41 5 FSSR 16(Tp ) 88 0 .44 1 .73 -0 .07 0 .23
6 FSSR 16(SPR ) 88 0 .26 0 .77 -0 .07 0 .17
41
25 1 FSR/STATS(CC ) 88 0 .0 1 0 .40 0 .03 0 .17
41 2 FSR/RF-RO (CC ) 88 0 .45 1 .38 -0 .09 0 .24
3 FSSR 16(CC ) 88 0 .46 1 .31 -0 .10 0 .23
41 4 FSSR 16(Tp& SPR ) 88 0 .20 1 .04 -0 .03 0 .18
5 FSSR 16(Tp ) 88 0 .42 1 .68 -0 .06 0 .24
41 6 FSSR 16(SPR ) 88 0 .26 0 .83 -0 .06 0 .17
41 50 1 FSR/STATS(CC ) 88 0 .05 0 .46 0 .02 0 .18

2 FSR/RF-RO(CC ) 88 0 .46 1 .4 1 -0 .09 0 .25
41 3 FSSR 16(CC ) 88 0 .48 1 .36 -0 .10 0 .24
4 FSSR 16(Tp& SPR ) 88 0 .23 1 .14 -0 .03 0 .19
41 s FSSR 16(Tp ) 88 0 .45 1 .73 -0 .07 0 .25
6 FSSR 16(SPR ) 88 0 .29 0 .85 -0 .07 0 .19
41
Tab le A .2 Fou r stat ist ics for return pe riods 2 , 5 , 10 , 25 and 50
41 yea rs compa ring the pe rformance o f s ix mode ls on a ll 88
catchments .
41
41
41
41
41
41
Append ix A . Fu ll tables o f comparisons and resu lts
41
41
41
41
Tab le A .3 Fou r statistics for return pe r iods 2 , 5 , 10 , 25 and 50
years compa ring the perfo rmance o f s ix mod e ls on a ll
catchments except 39004 the Wa nd le a t Bedd ington .

estimation
• Tab le A .4 Four stat ist ics for retu rn pe riods 2 , 5 , 10 , 25 and 50
years comparing the per formance o f s ix mod e ls on the
• standard set w ith less than 20% SO IL l.
•
411
ID
•
ID
•
ID
41
Append ix A . Fu ll tab les o f comparison s and resu lts
411
Tab le A .5 Fou r stat istics for retu rn pe riods 2 , 5 , 10 , 25 and 50
yea rs comparing the perfo rmance o f six models on the
standard set with at least 10 obse rved events .

estimation
97
RT 1.1 RT 'IJ
Method n mean RM S mean M S
2 1 FSR /STATS(CC ) 16 0 .16 0 .30 -0 .05 0 .11

2 FSR/RF-RO(CC ) 16 0 .44 0 .99 -0 .11 0 .22
3 FSSR 16(CC ) 16 0 .35 0 .71 -0 .10 0 .18
4 FSSR 16(Tp&SPR) 16 -0 .03 0 .31 0 .04 0 .15
5 FSSR 16(Tp ) 16 0 .3 1 0 .92 -0 .07 0 .20
6 FSSR 16(SPR) 16 0 .03 0 .32 0 .0 1 0 .14
5 1 FSR/STATS(CC ) 16 0 .11 0 .25 -0 .04 0 .09

2 FSR /RF-RO(CC ) 16 0 .55 1.06 -0 .14 0 .23
3 FSSR 16(CC ) 16 0 .47 0 .80 -0 .14 0 .20
4 FSSR 16(Tp&SPR ) 16 0 .05 0 .27 -0 .0 1 0 .11
5 FSSR 16(Tp ) 16 0 .4 1 0 .94 -0 .11 0 .20
6 FSSR 16(SPR) 16 0 .14 0 .35 -0 .04 0 .13
10 I FSR/STATS(CC ) 11 0 .12 0 .27 -0 .04 0 .10

2 FSR/RF-RO (CC ) 11 0 .57 1.13 -0 .15 0 .23
3 FSSR 16(CC ) 11 0 .52 0 .89 -0 .15 0 .2 1
4 FSSR 16(Tp&SPR ) 11 0 .12 0 .33 -0 .03 0 .11
5 FSSR 16(Tp ) 11 0 .5 1 1.15 -0 .13 0 .23
6 FSSR 16(SPR) 11 0 .21 0 .42 -0 .06 0 .14
25 1 FSR /STATS(CC ) 6 0 .29 0 .36 -0 .11 0 .13

2 FSR /RF-RO(CC ) 6 0 .77 1.4 1 -0 .18 0 .28
3 FSSR 16(CC ) 6 0 .71 1.14 -0 .19 0 .26
4 FSSR 16(Tp&SPR) 6 0 .20 0 .50 -0 .05 0 .16
5 FSSR 16(Tp ) 6 0 .81 1.54 -0 .18 0 .29
6 FSSR 16(SPR) 6 0 .23 0 .49 -0 .0 7 0 .16
Tab le A .6 Four statist ics fo r retu rn periods 2 , 5 , 10 , 25 and 50

years compar ing the perfo rmance of s ix mod e ls on
standard set of catchmen ts wh ich have at least 25
yea rs o f annua l max imum data .
Append ix A . Fu ll tab les o f compa risons and resu lts

•
98 •
•
Method n
RT"
mean RMS
RT 'ij
mean RMS •
2 1 FSR/STATS(CC ) 50 0 .05 0 .44 0 .0 1 0 .16 •
2 FSR /RF-RO(CC ) 50 0 .06 0 .79 0 .02 0 .18
3 FSSR 16 (CC ) 50 0 .07 0 .64 0 .0 1 0 .17 •
4 FSSR 16(Tp&SPR ) 50 -0 .06 0 .34 0 .05 0 .14
5 FSSR 16(Tp ) 50 0 .06 0 .82 0 .03 0 .18 •
6 FSSR 16 (SPR ) 50 -0 .03 0 .32 0 .03 0 .13
•
5 1 FSR /STATS (CC ) 48 0 .0 1 0 .39 0 .02 0 .15
2 FSR/RF -RO (CC ) 48 0 .16 0 .78 -0 .02 0 .18 •
3 FSSR 16(CC ) 48 0 .19 0 .66 -0 .04 0 .17
4 FSSR 16(Tp&SPR ) 48 0 .04 0 .33 0 .00 0 .12 •
5 FSSR 16 (Tp ) 48 0 .17 0 .8 1 -0 .02 0 .17
6 FSSR 16 (SPR ) 48 0 .08 0 .32 -0 .02 0 .12 •
10 1 FSR /STATS (CC ) 39 -0 .00 0 .40 0 .03 0 .15 •
2 FSR /RF-RO(CC ) 39 0 .02 0 .32 0 .0 1 0 .14
3 FSSR 16(CC ) 39 0 .08 0 .35 -0 .0 1 0 .13 •
4 FSSR 16(Tp&SPR ) 39 -0 .0 1 0 .26 0 .02 0 .11
5 FSSR 16(Tp ) 39 0 .0 1 0 .29 0 .0 1 0 .13 **
6 FSSR 16 (SPR ) 39 0 .05 0 .29 -0 .0 1 0 .11
•
25 1 FSR /STATS(CC ) 11 -0 .0 1 0 .29 0 .03 0 .13
2 FSR/RF-RO(CC ) 11 0 .03 0 .25 -0 .00 0 .11 •
3 FSSR 16(CC ) 11 0 .12 0 .32 -0 .03 0 .13
4 FSSR 16 (Tp&SPR ) 11 0 .03 0 .34 0 .0 1 0 .14 •
5 FSSR 16 (Tp ) 11 0 .02 0 .30 0 .0 1 0 .14
6 FSSR 16 (SPR ) 11 0 .13 0 .39 -0 .32 0 .13 •
Tab le A .7 Fou r statistics for return pe riods 2 , 5 , 10 , 25 and 50 •
years comparing the perfo rman ce of six mode ls on
stand ard set o f catchments wh ich have SAAR g reater •
than'800 mm .
•
•
•
•
•
•
•
•
•
An assessmen t o f the FSR ra in fa ll-runo ff method o f des ign flood •
es t imation
•
•
•
•
• Tab le A .8 Fou r sta tistics fo r return pe riods 2 , 5, 10 , 25 and 50
years comparing *the pe rfo rmance o f s ix mode ls on
• standa rd set of catchments wh ich have SAAR less
than 800 mm .
41
41
•
41
41
41
41
41
41
41
Append ix A . Full tab les o f comparisons and resu lts
41
41
41
100
Tab le A .9 Four statist ics for return per iods 2 , 5 , 10 , 25 and 50

yea rs compa ring the performance of s ix mode ls on
standa rd se t of ca tchments wh ich have AREA less
than 100 km2 .

estimation
•
• Tab le A .10 Fou r stat istics for retu rn period s 2 , 5, 10 and 25

yea rs compa ring the pe rforman ce o f s ix mod e ls on
0 standard set o f ca tchments wh ich have AREA g reater
than 100 km2 .
0
•
•
•
S.
•
•
•
Append ix A . Fu ll tab les o f compa rison s and resu lts
•
102

estimation
•
• 103
A PPEND IX B . T H E FSR SIMULAT ION EX ERC ISE
B.1 I NT ROD UCT ION

•
The F lood Studies Repo rt rain fa ll-runoff model prov ides a method o f
• deriving a flood hydrograph from a sing le rainfa ll storm event . It
is , of course , possib le that different combinations of storm char-
acteristics and catchment state produce flood peaks o f the same
magn itude , and it is to be expected that the magnitude of the derived
111 flood peaks will be more sensitive to some of these variab les than
to others . For example , perhaps rainfall depth a ffects flood peaks
more than the rainfall profile . A large number of computer simu-
lations was performed to examine the way in which the return period
411 o f the peak flow was affected by these variables so that a sing le
set of design inputs cou ld be spec ified to generate a T-year flood
• peak . In fact the analysis had two stages . Firstly it had to be
proven that the technique of using a set of design inputs and an
event based mode l wou ld work (ie. that it could reproduce observed
flood frequency curves). Once this was established , the second stage
was to fo rmulate a way o f selecting a sing le set o f inputs that wou ld
g ive the flood peak of requ ired return period . The fo llowing two
sections review the two stages o f the simu lation exercise .
B.2 REPRODUCT ION O F FLOOD FRE UENC Y C U R ES
The four variables that are required for design flood estimation
•
using the rainfall-runoff method are :
1. ra infall storm duration ,
2. ra infall profile shape ,
3. rainfall storm depth (or return period ), and
4. catchment wetness .
Each o f these variables has a corresponding probab ility d istribution

• and these can be combined to yield an overall probability d istrib-
ution of peak flow (statistically they are the marginal d istibutions
• of a joint probability surface ). As the marg ina l d istributions and
their interdependence were known , numerical integration cou ld be
used to obtain the jo int probability of particular combinations of
•
inputs (for uncorrelated marg ina l d istribut ions , A and B ,
• p(AnB)=p(A) .p(B)) . The correspond ing flow peak was derived from the
•
Appendix B . The FSR s imulation exercise
•
•
104
ra infa ll-runo ff model. The probability o f obtain ing a flood magni-

tude in the interval qi to q2 was then found by summing a ll the joint
probabilities for derived peaks in that interval. The flood frequency
curve was built up by performing this summation over successive in-
terva ls and thereby covering the requ ired range o f flood peaks .
While this process was exhaustive in that a ll possible combinations

of the four variab les were considered , it was greatly simplified by
defining just six to twe lve sub-divisions to represent the entire
range of each of the four variables.
Such simu lations were carried out on 98 catchments for wh ich

rainfa ll-runoff model parameters , and a suitable length of annual
max imum flows , were available . Seventeen catchments were later re-
jected because their response was too flashy for successfu l simu-
lation based on hourly rain fall. Genera l comparisons were made
between the flood frequency curve derived from annual maxima and the
the one resu lting from the simulation exercise. However , subsequent
ana lysis was restricted primarily to comparing observed and simu-
lated va lues o f the mean and 10 year floods .
In genera l, catchments with large floods were underestimated , but

ind ividua l departures were worse on sma ll and med ium flood
catchments . Figure 26 shows the pattern of residua ls from a re-
gression of the observed mean annual flood (BESMAF ) on the simulated
mean annua l flood (SIMMAF), the latter having been adjusted by
raising it to the power of 0 .98
BESMAF SIMMAF" g (B .1)
The simulations tend to underestimate in the south and south-west ,

and to overestimate floods in East Ang lia and on the east coast .
The pattern resembles that of residua ls from the regression of BESMAF
on catchment characteristics, suggesting that the mean annual flood
wou ld be similar ly over- or underestimated in the same areas by both
the statistica l method and fu ll numerical integration of probab ili-
ties and simu lation of peaks using the rain fall-runoff method .
The conc lusion was that "the probability d istributions of floods from
real catchments can be adequately pred icted by the simulation tech-
n ique" , FSR 1.6 .7 .4(444) .
B .3 CHO ICE OF A SI NGLE SET OF DESIGN INPUT S
The second stage of the analysis invo lved selecting a single choice
of variab les for each flood return period . This was achieved by
choosing suitable fixed va lues of the three less important variables
and then optimising the remain ing variab le such that the mode l re-
produced the requ ired flood magnitude . Since storm pro file was found
to be the least important variab le it was fixed as the 75% w inter
pro file , since th is profile gave resu lts c losest to the average o f
An assessment of the FSR rainfa ll-runo ff method o f des ign flood

estima tion
105
Small positive Residua l
Large positive Residual

Small negative Residu al
Large n e g a t i v e Residual
Th e s i ze d i v i s io n i s t a k e n
i h e s t a n d a rd e r r o r v a l u e
••
•
-4-
F igure 26 . Res idua ls from estimat ing the mean annua l flood
Append ix B . The FSR simulation exercise

106
a ll profiles . It was found that flood magnitude was less sensitive

to storm duration than to e ither o f the remain ing variables (ie .
antecedent wetness and storm depth). Thus duration , D , was fixed
by the equation
D = ( 1.0 + SAAR / 1000 ) Tp (B .2)
This equation was intended to estimate the du ration g iving the

largest flood magnitude but , as curves o f flood magnitude aga inst
storm duration are very flat , the choice of D is not critical.
Antecedent wetness (CWI) and storm depth were found to be equa lly
important in influenc ing flood peaks . When CWI was fixed , the re-
lationship between flood return period and rain fa ll return period
was similar between catchments . Conversely , fix ing rainfa ll depth
(by return period ) led to inconsistent CW I values for d ifferent
catchments . Therefore CW I was fixed and rainfall return period was
chosen by opt imisation . For each catchment the return period of
rainfa ll requ ired to produce floods of a range of return periods W a s
eva luated . Curves depicting the relaL onships between the resu lting
return periods are g iven in FSR Figure 1.6 .54(456) for seven
catchments . An average curve was recommended for select ing the ap-
propriate storm return period to give the peak discharge o f requ ired
return period when combined w ith the other variab les .
Two points in particular may be made about the second stage o f the
ana lysis . Firstly , in se lecting the sing le choice of va riab les , a
match was sought w ith the simulated flood frequency curves , rather
than those derived from observed data . Thus the reg ional dev iations
present in the simulations (see Figure 26) were built into the sing le
choice of variab les . Second ly , it is not c lear how many catchments
were used , and how much variability was present , when defining the
relative return periods o f des ign rainfa ll and peak flow . FSR Figure
1.6 .54(456 ) shows considerable scatter in the relationship for seven
catchments where the rain fall return period varies from (i) 5 to 10
years for the 5 year flood , (ii) 12 to 27 years for the 10 year flood
and (iii) 60 to 128 years for the 50 year flood . The corresponding
recommendations of the FSR are 8 , 17 and 8 1 years respective ly .
An assessment o f the FSR rainfall-runoff method of design flood

estimation
10 7
A PPEND IX C . EST IMAT ION OF SOME EX T REME HIST OR IC A L UK

FLOODS
Append ix C . Es tima tion o f s o m e extreme h istorica l UK floods

108
C . 1 NOT AT ION
A cross-sectional area o f river channel

AREA catchment d rainage area (km2)
ARF area l reduction factor
CW I catchment wetness index
DPR..1 dynam ic term o f percentage runo ff mode l
contro lled by catchment wetness
DPR,„i n dynam ic term of percentage runo ff mode l
contro lled by storm rain fa ll depth
MSL mainstream length (km)
n roughness coefficient in Manning equation
R IF norma l max imum flood (m3/s)
PMF probab le maximum flood (m3/s)
P IP probab le maximum precip itation (mm)
PR percentage runoff
R hydrau lic rad ius o f river channe l
S water surface s lope (m/km )
5 1085 slope of ma instream between 10% and 85%
d is tan ce between outlet and source (m/km)
SAAR average annual rainfa ll for the standard
period 194 1-1970 (mm)
SOIL1 proportion o f d rainage area underlain by
soil o f WRAP class 1
SPR standa rd percentage runo ff
Tp mean time-to-peak o f un it hyd rog raph (hr)
Tp minimum time-to-peak o f unit hyd rograph (hr)
URBAN propo rtion of catchment area urbanised
WRAP w inter rain fa ll acceptance potential
An assessment o f the FSR rainfa ll-runoff method of design flood

estimation
41
• 109
ID
C .2 INT RODUCT ION
•
Dam failures are amongst the most catastrophic calamities ; a tota l
41 o f a lmost 350 people lost the ir lives in just three d isasters in
Britain (Bilberry in 1852 , Da le Dyke 1864 and Da lgarrog in 1925) .
41 G runer (1963 ) reported that a qua rter o f dam failures documented
between 1799 and 1944 resulted from insufficient spillway capacity .
41 The hydro logist therefore has an important role in developing design
flood estimat ion techniques wh ich accurately estimate the largest
41 flood likely to be encountered , thus minimising the risk o f catas-
trophe , whilst avoid ing costly over-des ign .
40
It is fifteen years since the Flood Studies Report method (NERC ,
• 1975) for estimating a maximum flood hydrograph was first app lied
in the UK . As a design tool it rep laced some approximate ru les-o f-
thumb w ith recommendations based on the first rigorous study of na-
tiona l rainfall and river flow data , though it too has been
41 criticised for its simplistic assumptions . The question to be asked
in 1989 is whether we can now do any better. A number o f Flood
• Studies Supp lementary Reports (NERC , 1977-1985) have been pub lished
refining the FSR procedure and many other papers have been w ritten
41 on this or close ly related subjects ; there should be , therefore , some
new insights . On the other hand , as we are d iscussing max imum floods
41 we do not expect much in the way o f new data to prove o r d isprove
the accuracy of our estimates . Indeed , if, in that space o f time ,
41 we had had a major flood somewhere which equalled or exceeded our
estimate , we wou ld be concerned to say the least .
40
In normal estimation techniques it is intended that the best estimate
41 is (rough ly ) equa lly like ly to be under or over the true value . W ith
maximum floods not only are we deprived (by de finition ) of the true
40 va lue but even if we had some values that were c lose to being true
we wou ld not a llow ourse lves to underestimate any of them ; we p lace
40 all .the error of estimate on one side . Th is is not like a factor
of safety , app lied as a multip lier to the fina l figure to reflect
40 the lumped uncertainties ; in the FSR procedure the approach is to
max imise , or make the worst reasonab le assumptions about , each com-
e ponent o f the procedure as we go . In doing so , our estimates on many
catchments may be greatly - perhaps an order of magnitude - in excess
41 o f any experienced flood . Th is paper compares some estimates o f
recorded flood s with FSR max imum flood estimates made for the same
40 catchments . It a lso examines the FSR procedure and suggests which
aspects are most open to review .
411
C .3 SIGN I F ICA NT FLOODS IN BRIT A IN
40 Records of h istorica l flood events are ava ilab le from many sources
inc lud ing water authority archives , newspapers and journa ls , some
41 o f which include photographic evidence and eyew itness accounts . For
some other floods , the peak water levels are recorded as flood marks
41 on bridges , walls and houses or as specia lly sited stones . A know-
110
Append ix C . Estimation of some ex treme historical UK floods
40
41
40
41
110
ledge of the maximum recorded water leve l is only the first step in
hydro log ical ana lysis; an estimate o f the peak discharge is required
if we are to estimate runoff potentia l and transfer our findings to
other catchments .
Tab le C .1 contains a list of some major floods over the past 200
years w ith est imates of peak discharge . A lso given is the method
o f estimation . The symbo l G denotes a flow gaug ing station with an
existing relationship between stage and d ischarge . This relation-
ship w ill a lmost certa in ly have been extrapo lated well beyond any
flows used for its calibration . Uncerta inty is compounded by the
probab ility that the flow will have overtopped the measuring struc-
ture or river banks , and that the river bed may have been scoured
during the flood , chang ing the stage-d ischarge relationship . Occa-
siona lly a gaug ing station is built at , or near , a site where a
historica l flood was recorded , such as on the Dee at Woodend (no 2
in Tab le C .1) where the peak d ischarge o f the 1829 flood was esti-
mated (NERC , 1975) by extrapo lating the present stage-d ischarge re-
lationship to the peak level g iven in the account of the flood s by
Lauder (1830).
The symbo l SA in Tab le C .1 ind icates that the d ischarge , Q , was

calculated by the s lope-area method eg . us ing the Manning equation
Q = A R213 s 1/ 2 / n (C.1)
where A is the cross-sectiona l a rea , R is the hydraulic rad ius , n ,

the channe l roughness and S , the water sur face s lope . The Manning
equation was deve loped to describe flow in an infinite channel w ith
constant cross-section , energy g rad ient and roughness ; conditions
rare ly encountered in natura l channe ls . Use of th is retrospective
d ischarge estimation technique re lies on post-flood surveys to pro-
vide an accurate picture of the hydrau lic properties o f the channe l
at the peak of the flood . Critica l to the calculation is the nu-
merical va lue assigned to n . Text books provide suitab le values for
regu lar surfaces such as concrete , but suitable values for mixed
surfaces including cobbles and grass containing fallen trees and
supermarket trolleys are more d ifficu lt to determine . Desp ite these
d rawbacks , the technique has been w ide ly used in the UK (see for
examp le Acreman , 1983b for no 57 in Table C .1; Hyd raulics Research ,
1968 for nos 36-39 ; and Wh iter , 1982 for nos 62-65).
Hydraulic equations can a lso be used to compute a minimum estimate

o f the peak flow by calculating the critical velocities needed to
entrain materia l, such as large bou lders , which were known to have
been transported during the flood (see for examp le Metca lfe , 1979 ;
Tab le C .1 no 9 1). Making certain assumptions about the hyd raulic
conditions , the d ifference in peak flows at two gauging stations (eg .
Shenchie and Forres on the Findhorn , no 42 ) can g ive an estimate of
the inflow from the intervening catchment (in th is case the D ivie
and Dorback tributaries).
An assessment of the FSR rain fall-runo ff method o f des ign flood

est imation
• 111
•
Techniques may be combined where the channe l geometry and hyd rau lic
ID cond itions are comp licated . Sargent (1982 ; no 12) ca lculated the
peak flow for 1948 at Haddington us ing a back-water approach , to
4, mode l the e ffects of weirs , combined w ith slope-area estimates where
the water leve l was contro lled by channel friction . A variety of
411 other methods have been used ; Dobb ie and Wo lf (1953; 19-23) built a
scale mode l from paraffin wax to estimate the peak d ischarges of
411 severa l streams around Lynmouth affected by the floods o f 1952 ;
Acreman (1986 ; nos 3 & 4 ) showed how a rating curve for the site o f
ID a h istoric peak level could be constructed us ing flow data from
e lsewhere on the same river ; and erosion damage was used by Baxter
(1949 ; no 14) to estimate the depth of water pass ing over a spillway .
In each case the authors point out the uncertainties involved and
wou ld usually adm it to errors o f estimate of at least 20% and o ften
much more .
Figure 27 shows the location of the sites mentioned in Tab le C .1.

ID It can be seen that there are concentrations o f flood events recorded
in the h igh lands of Scotland , the Southern Up lands and south-west
Eng land . This pattern reflects , to some extent , the d istribution
of flood producing mechanisms . F loods resu lt from a combination of
411 intense rainfa ll falling on a responsive catchment , Thus the largest
floods wou ld be expected on steep , impermeable catchments , on small
ID catchments in thunderstorm prone areas and on large catchments where
long duration rainfa lls are intense . However , the pattern of floods
411 in F igure 27 is a lso partly due to the a geographical bias in
available estimates of peak flows . The Hampstead storm of 14th Au-
• gust 1975 (Keers & Wescott , 1976), resu lted in severe flooding o f
parts of north-west London but precise estimates' of peak d ischarge
ID are not available for the worst affected areas (Binnie and Partners
(1976) estimated the peak runoff from an area of 0 .5 km2 to be between
1111 about 2 and 7 m3/s, lack of data prec luded a more exact figure). A
further example for wh ich no flow data were recorded is the great
Till flood of 1841 (Cross, 1967) wh ich resulted from rain fa lling
on frozen ground . Under normal cond itions on this type of catchment ,
411 underla in by cha lk , most of the rain fa ll wou ld percolate into ground
411 with. only a low proportion producing stream-flow .
Despite some shortcomings these historica l extreme events can be used

to provide an ind ication o f the like ly max imum size of floods in the
UK and their d istribution . Figure 28 is a graph which shows each
ID estimate from Tab le C .1 with its reference number attached . It a lso
shows , w ith a + symbo l, the maxima from each gauged catchment held
41 on the Surface Water Archive (Institute of Hydro logy , 1988) .
Figure 28 on page 113 a lso shows two lines . A-A defines the origina l
ID Normal Max imum F lood , NMF (Institution o f Civil Engineers , 1933) and
ID B-B is the suggested spillway design flood curve for upland reser-
voirs which accompanied a later rev iew (A llard et al, 1960) . This
41 latter line appears not to have forma lly superseded the earlier
practice o f taking some multip le ('at least tw ice') of the NMF as
ID the spillway design (or 'catastrophic ') flood .
Append ix C . Estimation of some extreme h istorical UK floods

40
40
ID
411
112
• 26
5 70
49
• 42 5
10 0 •
6 40
•
48
•se
20 34 18
44
14 .
On
5
60
0 65
• 17 31 O M
41"
so
•
45 41
•
81
•
•
• 52
90
•
2 74 1 47
da
55
89
7
19 - 73
70
2-40 66
29 •
46 53 0 ,C>
F igure 27 . Location o f sites in Tab le C .I
An assessment o f the PSR ra in fa ll-runo ff method o f des ign flood

estimation
1 13
" • _ 1 '
7:7 1- H - H — 7 1 111 1 •
I I I ij . : ! I I I i I '
- 1- 1 - II I •
1 I
I I 1 ! 111-r I
II Vriti
• , • ,1 LT —
d I 0•p irl.d
0
• . .._ r i- /
o. I •
7
I
FFFT- _ -
. p ••
41, • ..•••
'
! 4
Pd
z -t- :
. I • I.
i l i .
i 1 11 1 ! i 1
'
! 1, i 1 i 1 ! i ! : 1 !
j IQ ! : ! i • 11 , • :
•• ' . i 1 .
. •
, •.
-4
o 0 0
O 0
- .-
O
-
( z . tin i 1 . s O m) e a n y a c ip o u p e d s
F i g u r e 28 . Ob s e r v ed ma g n i t ud e s a g a i n s t c a t c h me n t a r e a
Ap p e nd i x C . Es t i ma t i o n o f s o me ex t r e me h i s t o r i c a l UK f l o od s
•
114 •
•
C .4 EX T R A PO LA T E O R IN T ER PO LA T E?
41
The presentation of max ima in Figure 28 a llows the eng ineer to in-
terpolate a va lue for his catchment . It is interpolation in the 41
sense that recorded maxima are themselves being used d irect ly to
estimate similar maxima which might be expected at other and ungauged II
sites . It is implic it in such a use of envelope curves that the very
highest floods , expressed in this case as runoff per unit area are II
the worst that our climate is capab le o f producing . We suggest later
that , for larger catchments , this is a dubious proposition . ID
Interpolation can be supplemented by engineering judgment if cond i- II
tions o f the design catchment are sign ificantly different from those
o f the observed sites . C learly , a method o f estimation based solely II
on catchment area , though attractive ly s imple, is rather restric-
tive . However , this was not a problem when British design usage was 40
dom inated by up land reservoir construction . The post war deve lopment
of low land reservoirs and contro l o f large r catchments , helped to II
h igh light the need for new guidance where the a llowances cou ld be
made exp licit. Such guidance is provided in the F lood Stud ies Re- II
port.
Instead of draw ing an enve lope around recorded flow maxima , the FSR
method o f max imum flood estimation in effect d raws the envelope
around recorded rainfall maxima2 . Rainfa ll is then converted to flow
using a simp le linear ra infall-runoff mod el. Mod el parameters were
re lated to those physical characteristics of the catchment wh ich
quantify the up land v . lowland factor . The key d ifference between
this technique and the one it has rep laced is that the envelope
ID
method ga ins nothing from lesser flows recorded on the same
catchment ; the method relies on interpo lation between observed
maxima at a ll ava ilable sites . The rainfa ll-runoff methodo logy , on
the other hand , uses data from a w ider range of events, many of which
411
are sma ller - o ften much smaller - than those featuring in Tab le C .1;
this method can be considered an extrapolation from recorded data
40
on the same catchment .
C .5 PR O B A B LE M A X IM UM FLO O D S 411
The FSR method trans forms maximum rainfa ll estimates into flow
hyd rographs to produce the Probab le Max imum Flood , PMF . This is
achieved by assuming the worst possib le cond itions regard ing 40
antecedent wetness , design storm construction and speed of catchment
response . One obvious test o f the procedure is that its estimates 41
' The term rainfall is used loosely here to include snowme lt. There
is no doubt that snowmelt is a sign ificant factor in many large
floods , especially on large catchments , but , in this country , ID
snowme lt a lone can not generate flows in the PMF range .
ID
An assessment o f the FSR rainfa ll-runoff method o f des ign flood
estimation
11
411
ID
115
shou ld a lways be greater than any recorded maxima . These observed

max ima , on the other hand , might be exceeded as time passes . At the
time o f pub lication o f the FSR it was noted (Lowing , 1975) that the
peak flow from the Red-a-ven event (Worth , 1930 no 8) had been es-
timated to be s lightly higher than that g iven by the FSR procedure .
To test whether this was a unique occurrence , Pif s were ca lcu lated
(using the Institute o f Hydro logy 's m icro-FSR computer package , de-
ve loped by Boorman , 1988) for a se lection of the catchments listed
in Tab le C .1. Tab le C .2 provides the necessary catchment and climate
characteristics required by micro-FSR . Boorman (1985) prov ides es-
timates o f model parameters required for PMF estimation for a large
number of catchments throughout the UK includ ing the Tyne at East
Linton (used for estimate 12). Boorman et al (1988) give va lues for
severa l other catchments in Scotland . Tab le C .3 g ives the resu lts
from the PMF procedure. These are shown in Figure 29 p lotted aga inst
the maximum recorded floods . On the larger catchments (ie . those
w ith larger abso lute flood peaks), R IF is around two and ha lf times
the historica l maximum . In add ition to Red-a-ven , it can be seen
that estimated flows at five sites (Stobshiel, no 14 ; C laughton , 39 ;
D iv ie/Dorback , 42 ; Ca ldwell, 51; Chu lm le igh , 62) exceed ed the esti-
mated PMF . To consider whether these exceedences pose a serious
threat to the credib ility of the P IP estimation procedure , we need
to examine the estimates in a little more detail.
C .6 HO CAN OBSER ED FLOODS EXCEED PMF?
There are a number of possib le reasons why observed flood peaks might
exceed R IF . The PMF mode l may be de ficient in rainfa ll input , per-
centage runo ff or unit hydrograph , or the model parameters may be
inappropriate . A lternatively , the recorded flood peak may be in
error .
C .6 . 1 Rainfall input
Unfortunate ly there are no short duration rain fall data for the
largest event , the Ca ldwe ll Burn flood , a lthough a pro fessiona l me-
teoro logist from Eskda lemuir Observatory , who was caught in the
storm , estimated that the intensity p robab ly equa lled the 90 mm/hr
which had been recorded in 1953 (Metca lfe , 1979). This is still far
less than the 166 mm/hr probab le max imum precip itation , PMP , used
to estimate the P g . There has been no defin itive study comparing
R IP and recent olpserved storms . However , the largest da ily rain fa ll
since the FSR data were co llected , 238 .4mm at Sloy Ma in Ad it in
January 1974 (Reynolds , 1982), is considerab ly less than the
300-350mm PMP for that site. Of the six historical floods wh ich have
exceeded R IF , five are on sma ll catchments (< 10 km2), for which the
critical storm duration is much shorter than 24 hours. The Hampstead
storm of 14th August 1975 , during wh ich 169 mm rainfa ll fell in two
and a ha lf hours (Keers & Wescott , 1976), is the closest to PMP re-
Appendix C . Estimation of some extreme h istorical UK floods

II
116
II
ID
II
0
ID
II
II
II
50 0 0
0
/ II
II
10 0 0
I n
42
0
13
• II
50 0 II
• 65
0 9
19
4,
1
41
• 67
II
10 0
0
so • S1
II
•
1• • 39
II
0 62 II
0
66
•
10
5 10 50 10 0 SOO 10 0 0 50 0 0 II
O m. , - H i s t of I CA I 110 94 p4a1r I f/O s - I I
II
ID
II
F igure 29 . Observed magnitudes p lo tted agains t estima ted PMF
II
II
An assessment o f the FSR ra in fa ll-runo ff method o f des ign flood II
estimation
0
II
II
0
117
cently recorded ; the maximum two hour ra in fa ll for this area is 190
mm . Tab le C .3 shows the peak runoff expressed in mm/hr over each
catchment . Even during these six events runoff intens ity was far
less than the estimated PMP . This sugges ts that any deficiency in
the PMF mode l is un likely to arise from the ra in fall input .
C .6 .2 Pe rc e nt ag e r u no f f
Henderson (1986) estimated that percentage runo ff, PR , exceeded 90%

in all parts (up to 107 km2) of the Water o f Leith catchment in
Mid lothian during the flood of November 1984 , wh ich was not influ-
enced by snowmelt or frozen ground . However, it is not always clear
that the method s used for the calculations are consistent with those
specified in the FSR . A recent report by We lsh and Burns (1987)
illustrates the sensitivity o f PR estimates to the selected duration ,
showing that for sma ll upland catchments in the south o f Scotland
PR can vary from around 40% for the first 5 hou rs of the storm to
nearly 100% when ca lculated for the first 24 hours o f the storm .
Using the recommended procedure , Boorman et al (1988) con firmed that
in sma ll catchments underlain by impermeab le so ils (WRAP class 5) ,
PR could be h igher than g iven in the FSR . However, the PMP inten-
sities are , in most cases , g reater than twice the runo ff rate re-
corded du ring the h istoric floods . Therefore even a PR as low as
50% wou ld supply a sufficiently high runo ff rate , suggesting that
the percentage runo ff part of the model is not inadequa te .
C .6 .3 Unit hy d ro g r ap h
The FSR procedure uses a unit hydrograph to transform rainfa ll to

runoff. The linear mode / has one parameter, its time to peak , Tp .
For IM F estimat ion , Tp is reduced by one-third to simulate the worst
possible cond itions (this is the average ratio of min imum to mean
Tp in the UK ). This reduction may be inadequate to mod e l the very
rapid runoff experienced during extreme events . A lternative ly , use
of a linear mod el may not be appropriate . Neve rtheless it is d if-
ficult to visua lise what sort of model cou ld reproduce the flood of
189 m3/s estimated for the Ca ldwe ll Burn flood (no 51) .
During some large floods the peak d ischarge resu lts from a surge of
water caused by the release of temporary b lockages upstream . There
is evidence that this occurred during the Lynmou th flood when a 15m
high railway embankment co llapsed on the R iver Heddon above
Parracoombe (west of the Lyn); three peop le d ied (Delderfield , 1978).
Such effects are not a llowed for in the PMF model, but are usually
short lived , and wou ld not be important for spillway des ign as the
peak wou ld be attenuated when routed through the reservoir .
There are situations where the flood wave form and the channel ge-
ometry may combine to produce an unusual effect . Rather than at-
tenuating as it moves downstream , the rising limb is steepened and
the peak enhanced . Th is phenomenon may have occurred on the Findhorn
Appendix C . Estimation o f some extreme h istoriCal UK floods

118
in 1970 and wou ld explain the large inferred flood peak for the
intermed iate catchment o f the Divie/Dorback (no 42 ). If this was
the case it wou ld be w rong to take the d ifference In peaks as an
estimate of the inflow from the extra contributing catchment area .
Wh ilst the FSR rainfa ll-runoff procedure may be de ficient for in-
tense rainfa ll events on small catchments , it may lead to overesti-
mation o f PMF on large catchments . When a large floodp lain is ID
invo lved , the flood wave trave l time may be increased during large
floods as the water spills into overbank storage and flow resistance ID
increases .
C .6.4 Poor est enat h, n of histor ical f lood peaks

111
The method of flood peak derivation for the Red-a-ven flood is not
clear from the artic le by Worth (1930). The floods at Chu lmleigh , ID
Forest of Bow land and Berryscaur were estimated by the slope/area
method . As described above, even when the technique is applied by
experienced hydraulic eng ineers , the accuracy of estimation can be
poor , and the true peak may we ll have been c loser to PMF . Apart from 411
the doubts about in ferring the peak d ischarge from the intervening
area between two gaug ing stations , the Div ie/Dorback flood d ischarge
estimate relies heav ily on the stage-d ischarge relationsh ip at
Forres gauging station . Previous to the flood , the highest current
meter measurement used for the rating equation was around 2 .1m . The
flood peak stage was 4 .71 m , thus the peak discharge was based on 411
a large extrapo lation . Futhermore , the control at the station is a
grave l bar which su ffered considerable scour and redeposition during
the flood . Therefore , despite being recorded at a forma l gaug ing
station , the peak d ischarge may be a poor estimate .
There is a danger o f d ismissing a ll six o f the est imated peaks which 411
exceeded PMF . It is possible that some were unde restimates . It is
interesting that , w ith the exception of the Div ie/Dorback - which 40
is not a true catchment - the PMF exceedances all relate to
catchments under 10 km2 . Indeed , on ly one of the events from such ID
a small catchment (no 66) d id not exceed PMF and this is underlain
by cha lk . Perhaps we wou ld expect PMF to be app roached more fre- 111
quently on sma ll catchments . The chance of maximum rainfa ll of small
area l extent coincid ing w ith a small catchment is much g reater than 411
a larger storm sitting squarely over a larger catchment . Even the
Lynmouth storm was not centred over the Lyn catchment.
C .7 CONC LUSION
ID
Extreme flood estimation is prone to uncertainty whether the estimate
10
re lates to an actua l event or to a hypothetical d esign storm . The
two design approaches - interpo lating with an enve lope curve or ex-
trapo lating w ith a model - are seen to have strengths and weaknesses ,
but the former still has a strong intuitive attraction wh ich he lps
ID
ID
An assessment of the FSR rainfa ll-runoff method o f des ign flood
estimation
ID
119
to aid interpretation o f the latter. Recorded floods have high-

lighted the potentia l extreme response from small (<1.0 km2)
catchments .
Appendix C . Estimation o f some extreme historica l UK floods

120
Date Water course/s ite Peak Area Estimat ion Reference

flow method
(mom -l) (km 2)
1 Feb 1795 Trent , Nottingham 14 16 .0 7490 .0 VA G NERC (19 75)
2 Aug 1829 Dee , Woodend (Aberdeensh ire) 1900 .0 1370 .0 VA G NERC (1975 )
3 F indhorn , Shenachie (Moraysh ire) 1050 .0 4 17.0 VA Acreman (1986)
4 Du lnain , Balnaan Br (Morayshire) 500 .0 272 .0 VA Acreman (1986)
5 Spey , Boat o 'Brig 1665 .0 2850 .0 VA Werritty & Areman (1985)
6 Jan 1849 Ness (Invernesssh ire) 1700 .2 1792 .3 Nairne (1895)
7 Nov 1894 Thames , T edd ington 1059 .0 9948 .0 TH G SWA
8 Aug 19 17 Red -a -ven , Dartmoo r 110 .4 4 .0 Worth (1931)
9 May 1920 Lud , Louth (Linco lnsh ire) 138 .0 55 .1 C rosthwa ite (1921)
10 Dec 1936 Moriston , Invermoriston 557 .5 391.0 VA G Mc lean (1945)
11 Nov 1947 G len Cann ich , Inverness 433 .3 128 .1 SA Wo lf (1952 )
12 Aug 1948 Tyne , Haddington (East Lothian ) 255 .0 264 .0 SA BW Sargent (1982)
13 Aug 1948 Ga la Water , Ga lashie ls 200 .0 207 .0 VA G NERC (19 75)
14 Aug 1948 Tyne , Stobshiel (East Lothian ) 40 .8 4 .1 TH Baxter (1949 )
15 Jan 1949 Tweed , Peebles 10 79 .0 694 .0 SA G NERC (19 75)
16 Sep 1950 Ken , Ear lston Dam (Ga lloway ) 708 .0 372 .3 TH Chapman 6, Buchanan (1966 )
17 Sep 1950 Po lharrow , Cars fad Dam (Ga lloway) 254 .9 59 .5 TH Chapman & Buchanan (1966 )
18 Nov 195 1 T ay , Caputh 148 1.0 3211.0 VA G SVA
19 Aug 1952 West Lynn , Lynmouth 252 .8 22 .8 PM Dobb ie & Wo lf (1953 )
20 East Lynn , Lynmouth 436 .1 76 .0 PM
21 Hoaroak Water , Lynmouth 148 .7 8 .1 PM
22 Hoaroak Water , Lynmouth 286 .0 17 .0 PM
23 Badgeworthy Water, Lynmouth 97 .7 25 .3 PH
24 Sep 1953 A llt Uaine (Dumbartonshire) 11.3 3 .1 TH NERC (1975)
25 Dec 1954 Lune , Lancaster 116 1.0 10 11.7 VA G Chapman & Buchanan (1966 )
26 Jan 1955 T irry , Rh ian Bridge 110 .9 64 .2 VA NERC (1975)
27 Aug 1957 Foston Brook (Derbyshire) 200 .4 27 .4 SA Barnes & Potter (1958 )
28 Sne lston Brook (De rbyshire) 39 .1 3 .6 SA

29 Sep 1960 A lphin Brook , Exeter 59 .5 7 .2 SA Brierley (1965)
30 Oct 1960 W ithycombe Brook , Exmouth 99 .0 36 .0 SA Harrison (1961)
31 Jan 1962 N ith , Fr iars Carse 12 74 .0 799 .0 VA G SWA
32 Feb 1962 A llt Lar ig nan Lunn (Argy ll) 18 .1 6 .8 Chapman & Buchanan (1966 )
33 Feb 1962 Loch Awe (Argy ll) 10 76.2 79 7.0 Chapman & Buchanan (1966 )
34 Feb 1962 Lyon (Perthsh ire) 324 .3 16 1.5 VA G Chapman & Buchanan (1966 )
35 Feb 1962 Beauty , Erchless (/nvernesssh ire) 608 .8 84 1.8 VA G Morgan (1966)
36 Aug 1967 H indbu rn , Bou land Forest 637.0 83 .4 SA Hydraulics Rsearch (1968 )
37 Dunsop W ater , Bow land Fo rest 271.8 28 .7 SA

38 Blacko Water , Bou land Forest 52 .3 6 .9 SA
39 C laughton Beck , Bow land Forest 66 .5 2 .2 SA
An assessment o f the FSR rainfa ll-runoff method of des ign flood

estimation
12 1
SWA Surface W ater A rch ive , Institute of Hydro logy

SA velocity from water surface s lope ,
• channe l roughness and cross -sectiona l area
W river leve l from back -water effect o f w eir
critica l ve loc ity requ ired to ent rain
• transported materia l
VA velocity/area from current meter ing
PM physical mode l
• TH thoorectical calibration o f hydrau lic structure
at forma l gauging station
11. Tab le C .1 Peak discharge est imates for som e documented flood s s ince 1795
•
Append ix C . Estimat ion o f some ex treme h istorica l UK floods
•
T ab le C .2 Catchm ent characteristics for R IF est imation
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •
41
• 123
•
41
41
Site Date of Area Estim ated peak PMP PR R IF
• historical runo ff max imum
flood intens ity
41
(km2) (m 3s- 1 ) (mm hr - 1)(mm hr -l) % (m 3s-1)
• 3 Shenachie 4 -AUG-1829 4 17 .0 1050 .0 9 .06 76 .0 79 .4 2400 .
• 4 Ba lnaan Br 4 -AUG -1829 272 .0 500 .0 6 .62 82 .3 73 .4 1400 .

8 Red -a-ven 17-AUG-1917 4 .0 110 .4 99 .36 264 .8 79 .1 78 .1
41
9 Louth 29-MAY -1920 55 .1 138 .0 9 .02 114 .6 34 .3 317 .
41 12 Hadd ington 12-AUG-1948 30 7 .5 255 .8 2 .99 82 .6 66 .3 1150 .
• 13 Ga lashiels 12 -AUG -1948 207 .0 200 .0 3 .48 89 .2 6 1.4 782 .
14 Stobshie l 12-AUG -1948 4 .1 40 .8 35 .8 164 .1 59 .7 36 .7
•
15 Peeb les 7-JAN -1949 694 .0 1079 .0 5 .60 78 .4 64 .8 2790 .
41
19 Lynmouth 15-AUG -1952 23 .5 252 .1 38 .6 167 .0 68 .5 258 .
• 39 C laughton 8-AUG -1967 2 .3 66 .6 104 .2 264 .0 70 .5 33 .3
• 42 Do rback 16 -AUG -1970 365 .0 1939 .3 19 .12 79 .4 66 .5 1400 .

49 Oyk e l 5-OCT -1978 330 .7 847 .5 9 .23 5 1.9 85 .2 22 10 .
41
51 Ca ldwe ll B 13-JUN -1980 5 .7 189 .0 119 .4 166 .0 62 .2 49 .6
41 57 A rdessie 25-SEP -1981 13 .5 65 .0 17 .3 172 .0 77 .0 157 .
• 62 Chu lm leigh 12-JUL -1982 1.7 68 .0 144 .0 294 .0 59 .4 25 .2
65 Herm itage 26-JUL -1983 35 .9 165 .0 16 .5 14 1 .6 79 .1 362 .
•
66 Lyons Gate 20-MAY -1986 0 .83 7.0 31.5 286 .0 56 .7 10 .8
41
• Tab le C .3 R IF estim ates for 17 UR catchm ents togethe r w ith
• the peak flow for the largest recorded floods
41
41
•
411
40
•
•
• Append ix C . Est imat ion o f some extreme h istorica l UK flood s
41
41
411
41
124 41
•
C .8 REFERENCES
41
Acreman , M .C . (1983a) A hydro log ica l ana lys is o f the flood o f 26th
Ju ly 1983 on the He rm itage Wa ter , Roxbu rghsh ire Repo rt to MP 40
fo r Roxbu rghsh ire .
Acreman , M .C . (19836 ) The sign ificance o f the flood o f Septembe r 198 1

on the A rdess ie Burn , Wester Ross , Sco tt Geog Mag 99 , 150- 160 . 41
Ac reman , M .G . (1986 ) Est imating flood statistics from bas in 41
cha rac teristics in Scotland , PhD Thes is , Un iv St And rews .
0
A llard , W ., G lasspoo le , J . & Wo lf, P .0 (1960 ) F loods in the British
Is les , Proc Instit C iv Eng rs , 15 , 119-144 .
Barnes , F .A . & Potter , H .R . ( 1958) A flash flood in Derbysh ire ,

40
East M id lands G eogr , 10 , 3- 15 .
41
Baxter , G . (1949 ) Ra in fa ll and flood in south-east Scotland 12th August ,
1948 , J Inst Wat Eng rs Sc i , 3 , 26 1-268 .
41
B inn ie and Partners (1976) F lood ing at 225 and 22 7 West Heath Road
London , NW3 on 14th Augus t 1975 , Repo rt to Messrs C linton .
41
Boo rman , D .B (1985 ) A review o f the F lood Stud ies Repo rt
rain fa ll-runo ff pa rameter est ima tion equations , Institute
41
o f Hyd ro logy Repo rt No . 94 .
41
Boo rman , D .B (1988 ) M icro-FSR . A m icrocomputer-based package
to ass ist des ign flood estimation using F lood Stud ies Repo rt
method s , Institute o f Hyd ro logy , Wa lling ford .
41
Boo rman , D .B ., Acreman , M .C . & C layton , M .C . (1988 ) A study
o f pe rcen tage runo ffs on Scottish catchments , Repo rt to the
41
No rth o f Scot land Hyd ro-E lectric Board .
40
Brier ley , J . (1965 ) F lood ing in the Exe va lley , 41
Proc Instn C iv Eng rs , 6753, 15 1-588 .
Chapma n , E .J .K . & Buchanan , R .W . (1966 ) Frequency o f floods o f 'norma l

max imum ' intens ity in up land areas o f G reat Br itain , in 40
"R ive r F lood Hydro logy" , Ins tn C iv Eng rs , Lond on .
41
C ross , D .E . (1967) The G reat T ill flood o f 1841, Weather , 22 ,
430-433 .
C ros thwa ite , P .M . (192 1) The Lou th flood o f 1920 , T rans Wa t 41

Eng rs , 26 , 204 -217 .
De lder field , E .R . (1978) The Lynmou th flood d isas ter ,

40
An assessment o f the FSR ra in fa ll-runo ff method o f des ign flood 40
estimation
41
41
41
125
ERD Pub lications Ltd , Exmouth .
Dobb ie , C .H . & Wo lf, P .O . (1953) The Lynmouth f lood o f Augus t 1952 ,

Proc Instn C iv Eng rs , Pt III, 2 .
Frost , J .R . (1988 ) The Tyw i at Ty-Caste ll . Th e Octobe r 198 7

flood , Interna l Repo rt W e lsh Wa ter .
G runer , E . (1963 ) Dam d isasters , P roc Ins tn C iv Engrs , 24 , 4 7-60 .
Harrison , A .M .J . (196 1) The 1960 Exmouth floods , The Surveyo r

4th Feb , 127- 132 .
Henderson , R .J . (1986 ) Pre lim inary repo rt on investigations

into river flood ing in Ed inbu rgh , (i) Water o f Le ith catchment
Wa ter Research Centre (Eng r) , Sw indon .
Horrocks , R . (1986 ) Repo rt on the Know lstone flood s 11 Augu st

1986 , Interna l Report South West Wate r .
Hyd raulic Research S tation (1968) Fo rest o f How land and

Pend le flood s o f August 1967 , Hyd rau lics R esearch Station
Report EX 382 , Wa lling ford .
Ian How ick Assoc iates (1986 ) Lyons Ga te : Repo rt on flood ing and
proposa l for a llev iation wo rks , Report to West Do rset D istrict
Council .
Institute o f Hydro logy (1988 ) Hydrometric reg iste r and

statistics 198 1-5 Ins titute o f Hydro logy , Wa lling ford .
Institution o f C iv il Eng ineers (1933) F lood s in relat ion to

reservoir practice , Instn C iv Engrs , London .
Institution o f C iv il Eng ineers ( 1978) F lood s and reservo ir

sa fety : an eng inee ring gu ide , Instn C iv Eng rs , Lond on .
Kavanagh , M .D . (1976 ) Po lperro flood allev iat ion scheme ,

In terna l Repo rt South West Wa ter.
Keers , J .F . & Wescott , P . (1976) The Hamp stead storm - 14 August 19 75

Weather , 31, 1, 1-10 .
Laud er , T .D ., Sir (1830 ) An account o f the Grea t F lood o f 1829 in

the province o f Mo ray and the adjoin ing d istricts , Adam B lack ,
Ed inburgh .
Low ing , H .J . (1975) Pred ict ion o f the runo ff hyd rograph from a d esign
sto rm , in "F lood Stud ies Con ference" , Instn C iv Eng rs , London .
McEwen , L .J . (198 1) An assessmen t o f the geomo rph ic impact o f the
Appendix C . Est imation o f some extreme h istorica l UK floods

126
flash flood occu rring on 4 th August , 19 78 on the A llt Mor , G lenmo re

Inve rnesssh ire , Unde rg raduate d issertation , G eog Dep t , Univ
St And rews .
McC lean , W .N . (1945) River F low Records , Private Pub lication .
Metca lfe , M . (1979) Th e F lood at Berryscau r , June 13th 1979

Underg radua te d isse rtation , Geog Dept , Univ St And rews .
Mo rgan , H .D . (1966 ) Est imation o f des ign floods in Scot land and Wa les ,
in "R ive r F lood Hyd ro logy" , Instn C iv Eng rs , London .
Na irne , D . (1895) Memorab le floods in the H igh land s du ring the

n ineteen th century w ith some accounts o f the G reat F rost o f 1895 ,
No rthe rn Counties Pr inting and Pub lish ing , Inv erness .
Na tu ra l Env ironment Resea rch Coun cil (1975) F lood Stud ies
Repo rt NERC , London .
Natura l Env ironment Research Counc il (1977- 1985) F lood Stud ies
Supp lemen tary Repo rts 1-17 , Institute o f Hyd ro logy .
Pirt , J . & Ka rle , M . (1987) Summ ary o f hyd rologica l events in the T rent
basin 22-25th Augus t 1987 , Interna l Repo rt Seve rn T rent Wa ter .
Reyn o lds , G . (1982) Heaviest da ily rain fa ll in Sc ot land ,

Met Mag , 111, 158 .
Sargent , R . (1982 ) Pred iction o f exten t and frequ ency o f flood ing at
Hadd ington , Repo rt to Loth ian Reg iona l Counc i l, Wate r Research
Centre , Sco ttish O ffice .
Tyhurst , M .F . (198 1) The yea r the Stou r tu rned sou r - a review of the
review o f the 1979 Do rset floods , C iv Engr Techn ,
7 , 3 , 5-17 .
Werrit ty , A . & Acreman , M .C . (1985) The flood haza rd in Scotland , in

"Cima tic Hazard s in Scot land" , Harrison , S .J . 1
Geobooks , No rw ich .
We lsh , W .T . & Burns , J .C . (1987) The Loch Dee Project , runoff and
sur face water qua lity in an area subject to a cid prec ipitation
and a ffo restation in SW Scot land , Trans Roy Soc
Ed in , Ea rth Sc i , 78 , 249-260 .
Wh iter , N .E . ( 1982) Report on the Chu lm le igh F lood s 12 Ju ly

1982 , Interna l Repo rt South West Wa ter .
Wo lf, P .O . (1952) Fo recast and records o f flood s in G len Cann ich in

194 7 , J Instn Wet Eng rs , 6 , 298-324 .
An asse ssment o f the FSR ra in fa ll-runo ff method o f des ign flood

estimat ion
• 12 7
•
40 Wo rth , R .G .H . (1930) Thirty -sixth repo rt (third ser ies ) o f the
Comm ittee on the C lima te o f Devon , T rans Devonsh ire Assoc
fo r the Advancement o f Sc i , L it and A rts , 62 , 98- 100 .
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Append ix C . Estima tion o f some ex treme h isto rica l UK floods

•
•
•
•
128

estimation
•
• I Xi AT 11 E0 1 A I A SSO C IA TI O N
I N I1 I II O M F A P AI S P LA V 1 I i l l s . ',
• UN ] t fl W A T I X II I su l R C P ,
SY SI I M S HE S L A M H
•
The d emand for lo ng-te rm sc ie ntifi c cap a b ilities conce rning the
resources of the land a nd its fres hwate rs is rising sha rp ly as the
powe r of man to chang e his e nvironme n t is growing, and with
it the sc ale of his imp a ct. Co mp re hens ive res earc h facilities
(labo ra tories, fi e ld stud ie s, comp ute r mo d e lling, mstrume ntation,
re mote se nsing) are need e d to pro v id e solutions to th e
challenging p ro blems of the modern w o rld in its conc e rn for
approp riate and sy mpa th etic ma nage me n t of the frag ile syste ms of
the land s surface .
The Terre strial and Freshw at er Scie n c es Direc tora te of the

Na tura l Environment Rese arch Coun c il b rings toge the r an
e xcep tionally wid e rang e of ap p rop riate d isc ip lines (che mistry
b iolog y, e ng ine e ring, ph ysics, geology, g e og rap hy, ma the ma tics
and compu te r sc ience s) comp rising on e of the wor ld s larg est
bod ies of es ta blis hed e nvironmental e x p ertise . A staff of 550,
larg e ly g rad ua te and p rofessio nal, from fo ur Institutes a t e le ven
labo ra tories a nd fi eld s tations and two U n ive rsity units p ro vide
the sp e cialise d knowled ge and e xperie n c e to me e t nationa l and
inte rnationa l needs in three m ajor are as
Land Us e and Natural Re s o urces
Environme ntal Quality and Pollution
Ecology and Conse rva tion

N014149CR

Uploaded by

Copyright:

Available Formats

N014149CR

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

N014149CR

Uploaded by

Copyright:

Available Formats

INSTITUTE of

A N A SSESSM EN T O F TH E FSR RA I N FA LL- R U N O F F

The Ins ti tute of Hydrolog y is a co mpo ne nt esta blis hme nt of the UK

Jan uar y 29t h , 1990

• AN A SSESSMENT O F T HE FSR RA IN FA L L- R UNOFF MET HOD OF

411 D B Boorman , M C Acreman and J C Pack man

the FSSR 16 ra in fa ll-runo ff method ,

An assessment o f the FSR rain fa ll-runoff method of design flood

• 1.0 Int roduct ion

• 4 .0 Estimating T-year flood peaks by ind irect methods 29

6 .5 4600 3 : Da rt at Austins Bridg e 73

7 .0 Conc lus ions 83

Append ix A . Fu ll tab les of compa risons and resu lts 91

Append ix B . The FSR s imu lat ion exerc ise 103

Append ix C . Est ima tion o f som e extreme h istorica l UK f loods 107

An assessment o f the FSR rain fa ll-runo ff method o f d esign flood

An assessment o f the FSR ra in fa ll-runo ff method o f des ign flood

L ist o f Tab les

An assessment o f the FSR rain fa ll-runoff method of design flood

AREA Topog raph ic d rainage area in km2

1.0 INT RODUCT ION

• The FSR 1 presents two indirect methods o f flood estimation , which

• The statistical method estimates on ly peak flow , wh ich may suffice

The somewhat d isappo inting performance o f the rainfa ll-runoff model

Errors in assessing response time or unit hydrograph shape may a lso

Archer (1980) examined the performance o f the statistical method in

An assessment of the FSR ra in fa ll-runoff method of design flood

• This report aims to

41 - provide definitive comparisons o f the various FSR techniques

41 identify classes o f catchment where the current ra in fa ll-runoff

Appendix A contains fu ll tables o f resu lts presented in summary form

An assessment of the FSR ra infall-runoff method of design flood

have a re liable estimate of the true flood frequency curve

3. have numerous ana lysed individual flood events as a basis for

4 . have a stable and natura l hydro log ical reg ime .

In order to test the utility o f observed data when using the

at least five separate flood events . As pa rt of the rev iew o f the

At th is stage o f va lidation two stations were rejected from the

An assessment o f the FSR rain fa ll-runo ff method o f d es ign flood

Figure 3. Locations of catchments used in this study

An assessment o f the FSR ra infall-runoff method of design flood

An assessment o f the FSR rain fa ll-runo ff method o f d es ign flood

1900 1 369 .0 42 .0 5 .8 1 9 14 0 .000 0 .000 0 .000 0 .800 0 .200 0 .110

Tab le 2 .1 Ca tChment cha racte rist ics .

No AREA MSL 5 I085 SAAR SO IL 1 80 IL2 50 IL3 SO IL4 SO ILS UR BAN

45004 288 .5 33 .6 3 .58 1052 0 .500 0 .000 0 .150 0 .350 0 .000 0 .0 10

Tab le 2 .1 (Con tinued ) Catchment cha racteristics .

An assessment o f the FSR rain fa ll-runo ff method o f des ign flood

3.1 INT RODUCT ION

41 3.2 F IT T ING A FLOOD FRE UENCY DIST RIBUT IO

The FSR presents a number o f procedures for estimating T-year floods

F(x<X ) = exp( - [ 1 - k( u ) /0,( )1/k) (3 .2 .1)

DC = (261-b0)k/{(1-2-k)r (l+k)) (3 .2 .2)

u = b0+ oc[r (1+k)-11/k ID

and i = ascend ing rank order of the annual maxima ID

Second ly , the FSR recommendation to use regiona l growth factors in

• 3 . groups o f data points at a certa in d ischarge (suggesting overbank

Reliance on visual definition using flood frequency p lots can be

18 were g iven eyeba ll limits of 25 years , 5 3 we re given 10 years ,

The remainder of th is section describes investiga tions aimed at de-