Proceedings of 2013 IAHR World Congress

Spatial Rainfall Correlations and Their Implications for Raingage

Network and Rainfall Interpolation

Bellie Sivakumar1,2
1
Associate Professor and Australian Research Council (ARC) Future Fellow, School of Civil and
Environmental Engineering, The University of New South Wales, Sydney, NSW 2052, Australia. E-mail:
s.bellie@unsw.edu.au
2
Associate, Department of Land, Air and Water Resources, University of California, Davis, CA 95616,
USA. E-mail: sbellie@ucdavis.edu
Fitsum M. Woldemeskel
PhD Student, School of Civil and Environmental Engineering, The University of New South Wales,
Sydney, NSW 2052, Australia. E-mail: f.woldemeskel@unsw.edu.au
Vijay P. Singh
Professor, Department of Biological and Agricultural Engineering & Department of Civil and
Environmental Engineering, Texas A & M University, College Station, TX 77843, USA

ABSTRACT: Rainfall data at fine spatial resolutions are needed for various applications in the fields of
hydrology, water resources, and environment. However, high spatial resolution rainfall data are not
widely available, especially in developing regions. A common practice to obtain such data is through
interpolation, based on rainfall data available at coarser resolutions or nearby points. A fundamental
assumption in the development of many of the interpolation schemes is the existence of correlation in
rainfall across space, with nearby points considered to be more correlated than distant ones. Although this
assumption is reasonable, it is also important to note that rainfall correlation alone is not sufficient to
examine the spatial rainfall estimation problem, since other factors (e.g. scale, threshold) also play
important roles. In the present study, we propose a methodology that combines rainfall correlations,
spatial scale, and correlation thresholds for representing spatial rainfall correlations in a region. We apply
this method to high-quality monthly rainfall data at a resolution 0.25° x 0.25° latitude/longitude across
Australia. Important implications of the results are also discussed.

KEY WORDS: Rainfall, Spatial variability, Interpolation, Correlation, Scale, Threshold

1 INTRODUCTION
Rainfall data at fine spatial resolutions are needed for various studies associated with hydrologic,
water resources, and environmental systems. However, high spatial resolution rainfall data are not widely
available in many parts of the world, especially in developing countries, as collection of such data is
costly and time-consuming. In this situation, we normally resort to interpolation schemes for obtaining
rainfall at fine spatial resolutions based on rainfall measurements available at coarser resolutions and
nearby locations.
The last century has witnessed an enormous amount of research on raingage network design, rainfall
data estimation through interpolation, and estimation of uncertainty in rainfall (e.g. Thiessen, 1911;
Eagleson, 1967; Bras and Rodriguez-Iturbe, 1976; Bastin et al., 1984; Tabios and Salas, 1985; Lebel et al.,
1987; Hutchinson, 1995; Morrissey et al., 1995; McCullum and Krajewski, 1998; Bradley et al., 2002;
Garcia et al., 2008; Mishra and Coulibaly, 2009). Such research has also resulted in the development and
applications of various rainfall interpolation methods as well as comparison of their performances (e.g.
Simanton and Osborn, 1980; Creutin and Obled, 1982; Tabios and Salas, 1985; Michaud and Sorooshian,
1994; Ahrens, 2006; Borga and Vizzaccaro, 1997; Garcia et al., 2008; Kurtzman et al., 2009).
These studies have significantly advanced our understanding of rainfall variability across space and
our ability in estimating rainfall at fine spatial resolutions. The availability of more sophisticated
measurement technology and advanced scientific concepts also provides additional opportunities for
improving our understanding further. For instance: (1) more accurate rainfall estimation at some scales is
now possible through merging satellite/radar products and ground data (e.g. Huffman et al., 1995; Grimes
et al., 1999; Li and Shao, 2010); and (2) renewed and fresh insights into scientific theories (e.g.
complexity, scaling, entropy) offer new avenues for studying rainfall dynamics (e.g. Sivakumar, 2000;
Sivakumar et al., 2001, 2013; Özger et al., 2010; Gupta et al., 2011; Li et al., 2012).
Despite the advances, there remain some important issues in the existing approaches for rainfall
interpolation. For instance: (1) geographically nearby points are normally assumed to be more correlated
(and more connected) than distant points; (2) correlation between points often forms the basis for
assessing connectedness, while causation (together with scale and threshold) receives only limited
attention; and (3) smoothing of data often takes priority, when dealing with rainfall estimation at ungaged
points. In this study, we propose a new method that combines rainfall correlation, spatial scale, and
rainfall threshold, in an attempt to overcome some of the issues inherent in the existing approaches. To
this end, we analyze high-quality monthly rainfall data across Australia, the TRMM 3B43 dataset, which
is a combined product of rainfall observed through TRMM (Tropical Rainfall Measuring Mission) and
other satellites as well as gridded raingage data (e.g. Kummerow et al., 2010; Fleming et al., 2011).

2 METHODOLOGY
Figure 1 presents a schematic representation of the spatial rainfall correlation/estimation problem.
The figure shows that a given area (e.g. catchment) can be represented by a number of grids. Given that
rainfall data are available for some of these grids (i.e. gaged), the problem is to derive rainfall data for the
other grids (i.e. ungaged). To this end, study of correlations in rainfall among the grids is essential.
However, correlations alone are not sufficient to obtain realistic and reliable results. Here, we also
consider the spatial scale and threshold properties, together with correlations.
As shown in Figure 1, any grid i will have a certain number of “neighborhood” grids, denoted as j.
The number of neighborhood grids will vary depending upon both the “location” of the grid i and the
“spatial extent” (or size) of the neighborhood. If there is only one neighborhood grid, the rainfall
correlation between grids i and j (i.e. Ci,j) is simply given by:
Ci,j = Corr(Ri, Rj) (1)
where Ri and Rj are the rainfall values at grids i and j, respectively. Equation (1) can be extended or
generalized to any number of neighborhood grids j = 1, 2, 3, … N – 1, where N is the total number of
grids (including grid i) within the spatial extent of the neighborhood considered. The spatial extent and,
thus, the number of grids can be chosen in various ways, ranging from consideration of hydroclimatic,
topographic, catchment, and other properties to a purely arbitrary manner. Several methods exist in the
literature for neighborhood selection and spatial rainfall estimation, including nearest neighbors, inverse
distance weighting, kriging, and spline smoothing (e.g. Thiessen, 1911; Simanton and Osborn, 1980;
Creutin and Obled, 1982; Hutchinson, 1995; Jeffrey et al., 2011; Garcia et al., 2008; Kurtzman et al.,
2009). In this regard, “scale” is also an important factor in choosing the neighborhood size, as it provides
clues as to the “critical distance” (if exists) of correlation in one or more directions of grid i. Still another
factor that further improves the correlation analysis is the specification of the “extent of correlation”
between points, such as “threshold,” as a reliable measure of connection between the grids.
With this, we propose a “correlation-scale-threshold” method for studying spatial rainfall. The
method involves: (1) increasing the scale of the neighborhood n (i.e. box size) of grid i and evaluating its
effects on rainfall correlations between grid i and grids j; (2) calculating rainfall correlations Ci,j for all
grids j in the respective box sizes; and (3) determining the number of grids N′ having correlations
greater than a threshold (T) (where 0 ≤ T ≤ 1.0) and expressing it as a percentage of the total number of
grids j for the respective box sizes, as:

PCi,n = N′ /(N – 1) x 100 (2)

where PCi,n is the percentage number of grids for which correlation is greater than threshold T. The

2
procedure is repeated for each of the other grids.

Figure 1 Schematic representation of the spatial rainfall estimation problem. Grid i is base grid and grids j are
neighborhood grids in a given box size n (j = 1, 2, 3, … N – 1, where N is the total number of grids within a box size)

3 STUDY AREA AND DATA

In this study, spatial rainfall correlations across Australia are studied. Australia’s climate and, hence,
rainfall dynamics are significantly influenced by the hydroclimatic mechanisms that occur in the
surrounding oceans. The climate varies from tropical in the north to arid in the middle to temperate in the
south. Therefore, rainfall is highly variable, and also a large part of the country (especially the middle and
the west) is dry. About 80% of the country gets an annual rainfall of less than 600 mm, but the tropical
region of the far north receives an annual rainfall of over 4000 mm.
Rainfall in Australia is mainly monitored using the thousands of raingages installed in different parts
of the country. The regions near the coasts of east, southeast, and southwest have much denser gages,
while the number of raingages in the dry interior is very few. The launch of the Tropical Rainfall
Measuring Mission (TRMM) (Kummerow et al., 2000) has allowed rainfall measurements at finer spatial
(and temporal) resolutions across Australia. The TRMM rainfall products are available at different levels
of processing, and also at different versions that reflect the improvements in the estimation algorithm
since their launch (e.g. Kummerow et al., 2000; Chokngamwong and Chiu, 2008). Some efforts have also
been made to merge TRMM rainfall data with raingage observations (e.g. Li and Shao, 2010).
The rainfall data used in this study are from the TRMM 3B43 version. This version is a combined
product of rainfall observed through TRMM and other satellites as well as gridded raingage data
(Kummerow et al., 2000; Fleming et al., 2011). The rainfall data are available at a spatial resolution of
0.25° x 0.25° latitude/longitude and at the monthly scale. The data used here cover the period 1998–2007.
Although generally reliable nature, the TRMM 3B43 data set still has its own uncertainties due to various
factors, including errors in remotely-sensed data, ground-based data, and methodology in merging. We do
not address this issue in this study (see Tozer et al. (2012) for a recent study on the uncertainties in
gridded rainfall data in Australia).

3
4 RESULTS AND DISCUSSION
We perform the analysis for all the 0.25° x 0.25° latitude/longitude grids across Australia. Based on
results from a preliminary analysis, we consider only a limited number of box sizes (n) for detailed
analysis and presentation here: n = 1 (3 x 3 grids), 2 (5 x 5 grids), 4, 6, ..., 44 (89 x 89 grids). As for the
threshold (T), we consider four different rainfall correlation thresholds: 0.5, 0.6, 0.7, and 0.8.
Figure 2 presents some selected spatial rainfall correlations for the 0.25° x 0.25° grids across
Australia. The plots correspond to box sizes (n) of 1, 6, 12, 18, 24, 30, 36, and 42, and for a threshold
value (T) of 0.8, which is the most stringent case considered.

Figure 2 Spatial rainfall correlations for 0.25° x 0.25° grids across Australia for threshold (T) = 0.8. The results
correspond to eight different box sizes (n = 1, 6, 12, 18, 24, 30, 36, and 42) and indicate the percentage of grids
exceeding (PCi,n) the threshold value.

4
 The results reveal the following:
(1) when the box size is too small (e.g., n = 1), the spatial correlations are high all across Australia
(i.e. almost all grids have rainfall correlations exceeding 0.8);
(2) when the box size is too large (e.g., n = 42), the spatial correlations are generally low (i.e. only
0–20% of grids have rainfall correlations exceeding 0.8), except in the northern region (where about
20–40% of grids have correlations exceeding T = 0.8); and
(3) although spatial rainfall correlation decreases as the box size increases (and vice-versa) all across
Australia, very clear changes are observed at/across certain box sizes, with “pockets of regions” having
similar (or different) spatial correlations emerging.
Similar observations are also made for the other three threshold values considered, with appropriate
changes, in terms of percentage of grids exceeding a particular threshold, “pockets of regions,” and their
areal extent; see, for instance, Figure 3 for T = 0.5.

Figure 3 Spatial rainfall correlations for 0.25° x 0.25° grids across Australia for threshold (T) = 0.5. The results
correspond to eight different box sizes (n = 1, 6, 12, 18, 24, 30, 36, and 42) and indicate the percentage of grids
exceeding (PCi,n) the threshold value.

5
 The above spatial rainfall correlation results have important implications for a range of studies in
hydrology and water resources, as they are useful:
(1) to assess whether or not interpolation schemes are necessary or effective for a given region;
(2) to determine which interpolation scheme can be more effective for which region, and why;
(3) to identify “critical” points or areas at which installing raingages will significantly help in
rainfall estimation; and
(4) to “classify” regions in terms of spatial rainfall correlations, especially towards estimating
rainfall in sparsely-gaged or even ungaged areas.
For instance, rainfall dynamics in northern and northeastern Australia have far greater spatial
correlations when compared to that in southern and southeastern Australia. This may be an implication
that tropical climates generally have greater spatial rainfall correlations when compared to, for example,
temperate, oceanic, and continental climates, subject to other influencing factors. This seems to suggest
that rainfall interpolation schemes (if based on data available at 0.25° x 0.25° resolution or greater) will
not be particularly effective in southern and southeastern Australia when compared to that in northern and
northeastern regions. An important implication of this is that raingages need to be installed much closer in
the former areas when compared to that in the latter, in order to obtain more reliable rainfall estimates,
including using interpolation schemes.
In addition to climate type (e.g. tropical, temperate), several other factors drive rainfall dynamics
and, hence, dictate spatial rainfall correlations. These include topography (especially elevation), wind
(direction and velocity), and proximity to the coast, among others. Studying whether, how, and to what
extent these factors influence rainfall dynamics and spatial variability has been an important area of
research for many years now (see Garcia et al. (2008) and Kurtzman et al. (2009) for a sample of recent
studies). Although no attempt is made to study these issues here, the results reveal the potential
complications and inconsistencies in offering interpretations, in one way or another. For instance, while it
is known that coastal proximity plays a definite role in rainfall dynamics at a location, the differences
(sometimes significant) in spatial correlations obtained for the coastal regions (see Figures 2 and 3) reflect
the difficulties in providing general interpretations on its role and specific details about its extent.
Application of the above method to rainfall data at several major coastal cities across Australia (i.e. with
smaller areas and box sizes, as appropriate) also support this observation (results not presented here).

5 CONCLUSIONS
Adequate knowledge of spatial rainfall variability is crucial for undertaking a wide variety of
scientific and management studies associated with hydrology, water resources, and environmental
systems. However, our understanding of spatial rainfall variability continues to be limited, despite the
advances made thus far. This study has proposed a rainfall correlation–spatial scale–rainfall threshold
method for studying spatial rainfall variability, and also applied the method to high-quality monthly
rainfall data at a resolution of 0.25 degree x 0.25 degree (TRMM 3B43 version) across Australia. The
study addresses a very fundamental question on spatial rainfall correlations, i.e. to assess whether rainfall
interpolation based on existing data will be effective before any interpolation scheme is employed. The
results suggest the benefits of the methodology by providing useful information for identification of
“similar regions” or “clusters” of spatial rainfall correlation, especially to help with studies on estimation
of rainfall data in “ungaged” regions and also assessment of density of raingages.
Interpolation schemes for rainfall basically aim to establish “connections” between/among different
locations in a “network” of grids or raingages. In developing interpolation schemes, how to view such a
network and how to identify the dominant influencing factors are fundamental questions that continue to
be unresolved. Recent developments in the field of complex systems science have been found to be useful
in studing a wide range of networks in nature and society. Our future studies will focus on the
applications of such concepts to study spatial rainfall variability.

5 ACKNOWLEDGMENTS
This study is supported by the Australian Research Council (ARC). Bellie Sivakumar acknowledges the
support from the ARC through the Future Fellowship grant (FT110100328).

6
References
Bastin G., Lorent B., Duque C. and Gevers M., 1984. Optimal estimation of the average rainfall and optimal selection
of raingauge locations. Water Resources Research 20(4), 463–470.
Borga M. and Vizzaccaro A., 1997. On the interpolation of hydrologic variables: formal equivalence of
multiquadratic surface fitting and kriging. Journal of Hydrology 195, 160–171.
Bradley A.A., Peters-Lidard C., Nelson B.R., Smith J.A. and Young C.B., 2002. Raingage network design using
NEXRAD precipitation estimates. Journal of the American Water Resources Association 38, 1393–1407.
Bras R.F. and Rodriguez-Iturbe I., 1976. Network design for the estimation of areal mean rainfall events. Water
Resources Research 12, 1185–1195.
Chokngamwong R. and Chiu L.S., 2008. Thailand daily rainfall and comparison with TRMM products. Journal of
Hydrometeorology 9, 256–266.
Creutin J.D. and Obled C., 1982. Objective analysis and mapping techniques for rainfall fields: an objective
comparison. Water Resources Research 18(2), 251–256.
Eagleson P.S., 1967. Optimum density of rainfall networks. Water Resources Research 3(4), 1021–1033.
Fleming K., Awange J.L., Kuhn M. and Featherstone W.E., 2011. Evaluating the TRMM 3B43 monthly precipitation
product using gridded raingauge data over Australia. Australian Journal Meteorology and Oceanography 61(3),
171–184.
Garcia M., Peters-Lidard C.D. and Goodrich D.C., 2008. Spatial interpolation of precipitation in a dense gauge
network for monsoon storm events in the southwestern United States. Water Resources Research 44(5), W05S13.
Grimes D.I.F., Pardo–Iguzquiza E. and Bonifacio R., 1999. Optimal areal rainfall estimation using rain gauges and
satellite data. Journal of Hydrology 222(1-4), 93–108.
Gupta V.K., Mantilla R., Troutman B.M., Dawdy D. and Krajewski W.F., 2011. Generalizing a nonlinear geophysical
flood theory to medium-sized river networks. Geophysical Research Letters, 37(11), L11402.
Huffman G.J., Adler R.F., Rudolf B., Schneider U. and Keehn P.R., 1995. Global precipitation estimates based on a
technique for combining satellite-based estimates, rain gauge analysis, and NWP model precipitation information.
Journal of Climate 8(5), 1284–1295.
Hutchinson M.F., 1995. Interpolating mean rainfall using thin plate smoothing splines. International Journal
Geographic Information Systems 9(4), 385–403.
Jeffrey S.J., Carter J.O., Moodie K.B. and Beswick A.R., 2001. Using spatial interpolation to construct a
comprehensive archive of Australian climate data. Environmental Modelling & Software 16(4), 309–330.
Kummerow C., Simpson J., Thiele O., Barnes W., Chang A.T.C., Stocker E., Adler R.F., Hou A., Kakar R., Wentz F.,
Ashcroft F., Kozu T., Hong Y., Okamoto K., Iguchi T., Kuroiwa H., Im E., Haddad Z., Huffman G., Ferrier B.,
Olson W.S., Zipser E., Smith E.A., Wilheit T.T., North G., Krishnamurti T. and Nakamura K., 2000. The status of
Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. Journal of Applied Meteorology 39(12),
1965–1982.
Kurtzman D., Navon S. and Morin M., 2009. Improving interpolation of daily precipitation for hydrologic modelling:
spatial patterns of preferred interpolators. Hydrological Processes 23(23), 3281–3291.
Lebel T., Bastin G., Obled C., and Creutin J.D., 1987. On the accuracy of rainfall estimation: a case study. Water
Resources Research 23(11), 2123–2134.
Li M. and Shao Q., 2010. An improved statistical approach to merge satellite rainfall estimates and raingauge data.
Journal of Hydrology 385(1–4), 51–64.
Li C., Singh V.P. and Mishra A.K., 2012. Entropy theory-based criterion for hydrometric network evaluation and
design: Maximum information and minimum redundancy. Water Resources Research 48(5), W05521.
McCollum J.R. and Krajewski W.F., 1998. Uncertainty of monthly rainfall estimates from rain gauges in the Global
Precipitation Climatology Project. Water Resources Research 34(10), 2647–2654.
Michaud J.D. and Sorooshian S., 1994. Effect of rainfall-sampling errors on simulations of desert flash floods. Water
Resources Research 30(10), 2765–2775.
Mishra A.K. and Coulibaly P., 2010. Hydrometric network evaluation for Canadian watersheds. Journal of Hydrology
380(3–4), 420–437.
Morrissey M.L., Maliekal J.A., Greene J.S. and Wang J., 1995. The uncertainty of simple spatial averages using rain
gauge networks. Water Resources Research 31, 2011–2017.
Özger M., Mishra A.K. and Singh V.P., 2010. Scaling characteristics of precipitation data in conjunction with wavelet
analysis. Journal of Hydrology 395(3–4), 279–288.
Simanton J.R. and Osborn H.B., 1980. Reciprocal-distance estimate of point rainfall. Journal of Hydraulic
Engineering Division 106HY7.
Sivakumar B., 2000. Chaos theory in hydrology: important issues and interpretations. Journal of Hydrology 227(1–4),
1–20.
Sivakumar B., Sorooshian S., Gupta H.V. and Gao X., 2001. A chaotic approach to rainfall disaggregation. Water
Resources Research 37(1), 61–72.
Sivakumar B., Woldemeskel F.M. and Puente C.E., 2013. Nonlinear analysis of rainfall variability in Australia.
Stochastic Environmental Research and Risk Assessment, doi: 10.1007/s00477-013-0689-y.
Tabios III G.Q. and Salas J.D., 1985. A comparative analysis of techniques for spatial interpolation of precipitation.
Water Resources Bulletin 21(3), 365–380.
Thiessen A.H., 1911. Precipitation averages for large areas. Monthly Weather Review 39(7), 1082–1084.

7
Tozer C.R., Kiem A.S. and Vernon-Kidd D.C., 2012. On the uncertainties associated with using gridded rainfall data
as a proxy for observed. Hydrology and Earth System Sciences 16, 1481–1499.