ISPRS, Vol.34, Part 2W2, "Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
311 
Historically, most methods for estimating areal and gridded 
precipitation from point data have fallen into three major 
groups. They are graphical, topographical, and numerical 
methods. Graphical methods involve mapping of precipitation 
data, sometimes in combination with precipitation-elevation 
analyses, and include isohyet mapping [2;3] and Thiessen 
polygon estimation [4] . Topographical methods involve the 
correlation of pointprecipitation data with an array of 
topographic and synoptic parameters such as slope, exposure, 
elevation, location of barriers, and wind speed and direction. 
Over the past decade, the most commonly used precipitation 
distribution methods have been numerical. These are 
interpolation procedures in which a numerical function, 
developed or prescribed, is used to weight irregularly spaced 
point data to estimate a regularly spaced prediction grid. 
Various studies have been carried out using numerical 
interpolation method to analyze corelationship between rainfall 
and topography factors in rainfall estimation such as inverse - 
distance weighting interpolation, Shepard's weighted 
interpolation and optimum interpolation method. Among 
several methods for estimating rainfall using points data, can 
be concluded that geostatiscal methods (including Kriging and 
Cokriging) were superior to others. Kriging is a geostatistical 
approach that has gained acceptance as a tool for the 
interpolation of many types of data , including precipitation. A 
potential drawback of kriging is that it implicity relies on the 
data to directly represent the spatial variability of the actual 
precipitation field. If the data are not representative (as is often 
the case in complex terrain), the accuracy of the resulting 
interpolated field will be in question. In addition, more than one 
semivariogram (numerical function) may be needed to 
estimate precipitation at various time periods for which the 
processes producing the precipitation patterns differ. 
Recently, elevationally detrended kriging and cokriging with 
elevation as a covariate have been used to bring topographic 
influences into the calculations [2]. The resulting precipitation 
fields often show more topographically related spatial patterns 
in complex terrain than those from ordinary kriging. However 
the application is limited to areas characterized by a strong, 
overall precipitation-elevation relationship (i.e. regions 
dominated by one main orographic regime). 
2.2 Current Development of GIS Technology in Rainfall 
Estimation 
The demand for climatological precipitation fields on a regular 
grid is growing dramatically as ecological and hydrological 
models become increasingly linked to geographic information 
systems (GIS) that spatially represent and manipulate model 
output. GIS provide an array of interpolation techniques in 
which point estimates of rainfall are converted onto a 
rectangular surface. These techniques are based on 
integration of GIS tools with the interpolation procedures in 
which a numerical function, developed or prescribed, is used 
to weight irregularly spaced point data to estimate a regularly 
spaced prediction grid. Here, grid refers to a two - dimensional 
array of regularly spaced grid cells. A grid cell refers to a 
single pixel that has dimensions equivalent to the resolution of 
the grid. A value assigned to a pixel, such as rainfall estimate, 
is positioned at the cell center. However, it is not a point 
value; rather, it represents an average value over the entire 
cell. Examples of these techniques are: (1) the Thiessen 
polygon method of ARC/INFO system which is based on the 
Thiessen polygon estimation [4] which is a polynomial 
interpolation. Thiessen polygons are used to model or 
approximate the zones of influence around points in Thiessen 
interpolation method , (2) Kriging method of ARC/INFO . It 
interpolates a lattice from a set of variably - spaced points 
using Kriging which is an advanced geostatistical procedure 
that generates an estimated surface from a scattered set of 
points with z (rainfall) values. Recently GIS is also employed in 
the processing of Weather Surveillance Radar - 1988 Doppler 
(WSR-88D) radar reflectivity data in NEXRAD program. Here, 
GIS plays an important role in the management and 
processing of radar estimates of rainfall . WSR-88D radar 
rainfall estimates are an important new source of spatial 
rainfall rate for distributed hydrologic models. Since hydrologic 
models are GIS integrated, the native resolution of the WSR- 
88D radar data, which is in radial coordinates, must be 
resample into a georeferenced coordinate system in the GIS. 
Thus, a key aspect of GIS processing is the resolution of the 
rainfall estimates derived from radar. It is used both for data 
ingest and for hydrologic modeling. The ingest, processing, 
scale and resolution of WSR-88D radar data in hydrologic 
simulations can be managed using the raster GIS. 
3. RAINFALL ESTIMATION SYSTEM 
3.1 Conceptual Framework 
A conceptual framework is incorporated to address the 
analysis of corelationship of rainfall with an array of 
topography parameters to derive the best fitted multiple linear 
regression equation as the optimal model. It consists of four 
components as shown in Figure 3-1: 
Figure 3-1: The Conceptual Framework of GIS-based 
Rainfall Estimation System 
a. Database. 
The study of spatial correlationship of rainfall with topographic 
parameters involves spatial data and analysis. Thus, it is 
important to develop a database consisting of spatial and non- 
spatial data as required by the system development. 
b. Mathematical Model 
It consists of statistical and graphical analysis which studies 
spatial correlationship of individual topography parameter with 
rainfall, and regression analysis which investigates the effects 
of topography parameter in the presence of others to derive 
the best fitted multiple linear regression equation. The model 
development involves spatial and regression analysis 
attempting to find the optimal rainfall estimation model. The 
procedure is to interpolate from the surrounding points taking 
into account topography effects on rainfall. Its stages of 
development are listed as follows: