Radiometric

 Correction  of  Remotely  Sensed  Data  

 
Douglas  A.  Stow  
Department  of  Geography  
San  Diego  State  University  
 
Introduction  and  Background  
Radiometric  correction  of  remotely  sensed  data  normally  involves  the  processing  of  
digital  images  to  improve  the  fidelity  of  the  brightness  value  magnitudes  (as  
opposed  to  geometric  correction  which  involves  improving  the  fidelity  of  relative  
spatial  or  absolute  locational  aspects  of  image  brightness  values).  The  main  purpose  
for  applying  radiometric  corrections  is  to  reduce  the  influence  of  errors  or  
inconsistencies  in  image  brightness  values  that  may  limit  one's  ability  to  interpret  
or  quantitatively  process  and  analyze  digital  remotely  sensed  images.  Throughout  
this  section,  radiometric  errors  and  inconsistencies  will  be  referred  to  as  "noise",  
which  could  be  considered  any  undesirable  spatial  or  temporal  variations  in  image  
brightness  not  associated  with  variations  in  the  imaged  surface.  
The  sources  of  radiometric  noise  and  therefore,  the  appropriate  types  of  
radiometric  corrections,  partially  depend  on  the  sensor  and  mode  of  imaging  used  
to  capture  the  digital  image  data.  Five  general  types  of  imaging  systems  and/or  
modes  are  utilized  for  generating  digital  remotely  sensed  data,  each  having  their  
own  characteristic  sources  of  radiometric  noise:  (1)  scanned  aerial  photography,  (2)  
optical  scanners,  3)  optical  linear  arrays,  (4)  optical  framing  arrays,  (5)  scanning  
microwave  radiometers,  and  (6)  side-­‐looking  radars.  (The  emphasis  of  this  section  
will  be  on  radiometric  correction  of  the  more  frequently  utilized  optical  imaging  
systems.)  Radiometric  noise  generated  by  remote  sensing  instruments  can  take  the  
form  of  random  brightness  deviations  from  electrical  sources  and  coherent  
radiation  interactions  or  more  systematic  variations  that  have  spatial  structure  or  
temporal  persistence.  
Sensor-­‐related  effects  are  not  the  only  sources  of  radiometric  noise.  Other  sources  
are  spatial  and/or  temporal  variations  in  illumination  quantity  and  quality,  
atmospheric  optical  properties,  terrain,  and  surface  properties.  Again,  these  variable  
environmental  factors  should  only  be  considered  "noise"  when  they  obscure  or  
reduce  image  brightness  signals  pertaining  to  surface  cover  types  and  conditions.  
There  are  five  primary  reasons  or  objectives  for  applying  radiometric  corrections  to  
digital  remotely  sensed  data;  four  of  which  pertain  to  achieving  consistency  in  
relative  image  brightness  and  one  involving  absolute  quantification  of  brightness  
values.  Relative  correspondence  of  image  brightness  magnitudes  may  be  desired  for  
pixels:  (1)  within  a  single  image  (e.g.,  orbit  segment  or  image  frame),  (2)  between  
images  (e.g.  adjacent,  overlapping  frames),  (3)  between  spectral  band  images,  and  
(4)  between  image  dates.  The  key  here  is  that  brightness  value  inconsistencies  
caused  by  the  sensor  and  environmental  noise  factors  listed  above  are  balanced  or  
"normalized"  across  and  between  image  coverage  and  spectral  bands.  The  other  
principal  objective  is  the  retrieval  of  surface  energy  properties  such  as  spectral  
reflectance,  albedo  or  surface  temperature,  which  requires  absolute  radiometric  
processing.  
 
Reflectance  Factor  Retrieval  
A  majority  of  remote  observations  of  earth  surface  forms  and  processes  are  based  
on  digital  images  captured  from  airborne  or  satellite  platforms  by  optical  imaging  
systems  operating  in  the  solar  reflective  portion  of  the  electromagnetic  spectrum.  
The  basis  for  extracting  information  on  earth  surface  objects,  types,  quantities  or  
patterns  is  the  variation  of  surface  reflectance,  usually  in  more  than  one  spectral  
band.  However,  other  factors  can  influence  the  amount  of  radiance  captured  and  
recorded  by  a  remote  sensing  instrument  besides  the  reflectance  properties  of  earth  
surface  materials.  These  factors  include:  
In  addition  to  artifacts  and  uncertainties  due  to  environmental  effects  on  radiation  
transfer  are  the  effects  of  the  sensor  and  it's  mode  of  sampling.  Even  when  the  same  
instrument  is  used,  the  earth-­‐platform-­‐sensor  geometry  relationship  is  different  for  
each  data  acquisition.  The  resultant  data  are  spatially  autocorrelated  because  of  
optical  diffraction  and  the  scanning  process.  Atmospheric  scattering  also  produces  
an  adjacency  effect  that  is  a  source  of  spatial  autocorrelation  in  the  image  data  
(Singh,  1988).  
The  other  type  of  requisite  image  transformation  is  radiometric  pre-­‐processing,  
meaning  that  remotely  sensed  digital  brightness  values  must  be  calibrated  and/or  
converted  so  as  to  improve  the  relative  spectral  and  temporal  fidelity  of  the  data.  
The  essential  first  step  is  to  radiometrically  calibrate  digital  brightness  values  by  
converting  them  into  spectral  radiances  (Robinove,  1982).  Such  calibration  is  based  
on  published  calibration  coefficients  derived  from  pre-­‐launch  laboratory  calibration  
and  post-­‐launch  empirical  corrections.  Radiometric  calibration  serves  to  normalize  
spectral  radiances  between  wavebands  of  the  same  image  and  between  images,  if  
the  calibration  coefficients  are  temporally  stable  or  updated.  
The  most  important  requirement  of  radiometric  correction  is  to  ensure  that  changes  
in  spectral  radiances  for  corresponding  pixels  of  a  multi-­‐temporal  image  sequence  
are  proportional  to  actual  changes  in  spectral  reflectance  of  the  surface  (Nelson,  
1985).  Differences  in  atmospheric  optical  properties  and  solar  illumination  between  
image  dates  influences  the  need  to  model  their  transient  effects.  More  approximate,  
empirically-­‐based  models  may  reduce  these  effects  through  scene  normalization,  
but  radiation  transfer  models  offer  a  more  precise,  deterministic  approach.  
Scene  normalization  models  are  based  on  empirically  matching  multi-­‐temporal  
images  based  on  scene  features,  e.g.,  features  with  stable  reflectances  such  as  rock  
outcrops  or  persistent  shadows,  (Chavez,  1988;  Laureau,  1991;  Pech  et  al.,  1986;  
Schott  et  al.,  1988).  A  particularly  promising  example  of  this  type  of  modeling  is  the  
"radiometric  registration"  approach  developed  by  Hall  et  al.,  (1991).  In  an  attempt  
to  normalize  atmospheric  and  illumination  differences  between  seasonal  Landsat  
TM  images  of  chaparral  and  pine-­‐oak  forest  in  southern  California,  Stow  et  al.  
(1993)  radiometrically  "registered"  an  image  data  from  a  April,  1987  scene  to  a  
scene  acquired  in  November,  1986  image  using  the  Hall  et  al.  (1991)  method  (Figure  
3).  TM  brightness  values  for  both  dates  were  converted  to  "Brightness"  (BR),  
"Greenness"  (GR)  and  "Wetness"  (WT)  based  on  the  "Tasseled-­‐cap"  transform  (Crist  
and  Cicone,  1984).  Dark  and  bright  control  sets  (pixel  samples)  were  selected  from  
BR-­‐GR  plots  and  used  to  develop  a  linear  fit  equation  for  normalizing  TM  spectral  
radiances  for  April,1987  relative  to  TM  spectral  radiances  for  November,  1986.  This  
enabled  seasonal  changes  in  BR,  GR  and  WT  to  be  derived  in  an  attempt  to  infer  
seasonal  changes  in  fire  fuels  amount  and  condition  (which  is  further  described  in  a  
later  section).  
Multi-­‐date  satellite  data  may  also  be  normalized  through  more  deterministic  
approaches  that  attempt  to  derive  reflectance  factors  of  land  surfaces  (Duggin,  
1980;  Moran  et  al.,  1990).  Reflectance  factors  are  approximations  of  surface  
reflectances,  (and  therefore,  should  be  temporally  stable),  that  are  based  on  an  
assumption  of  isotropically  reflecting  surfaces.  Such  approaches  are  based  on  
radiative  transfer  modeling  of  solar  illumination  and  atmospheric  effects  (Spanner  
et  al.,  1990).  In  the  typically  irregular  terrain  of  many  Mediterranean-­‐type  
landscapes,  the  differential  illumination  effects  from  variable  terrain  slope  facets  
should  also  be  modeled,  based  on  a  digital  elevation  model  (Holben  and  Justice,  
1979).  The  greatest  error  and  uncertainty  in  deriving  reflectance  factors  results  
from  the  atmospheric  modeling  component.  Particularly  problematic  is  the  lack  of  
concurrent  data  on  atmospheric  constituent  concentrations  and  their  horizontal  and  
vertical  distributions  (Caselles  and  Garcia,  1989).