ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
313 
of interpolation. These processes are performed by GIS. The 
system development also involves point estimation, updating 
and area-average estimation processes. Thus, the main 
processes involve: 
Figure 4.3 : Rainfall Estimation Processes 
4.4 Model Development 
As discussed in section 2, it involves preliminary 
statistical/graphical analysis and regression analysis in 
deriving the optimal rainfall estimation. SPSS is employed to 
perform regression analysis in deriving the optimal model. 
4.5 Graphical User Interface Development 
It is a GIS-based system whereby optimal rainfall estimation 
model is integrated with GIS to perform rainfall estimation. The 
system development includes point estimation, updating and 
area-average estimation processes as shown in Diagrams 2 
and 3 of DFD in Appendix E. It is implemented using the 
interface that is menu-driven application so that the system 
can be assessed easily by an inexperienced ARC/INFO user. 
The procedure is to customize ARC/INFO. This can be 
achieved by employing ARC Macro Language (AML). There 
are two types of AML files: command macros and menus. 
Macros are used to perform all ARC/INFO commands involved 
in the development of the system such as spatial analysis and 
computation. Whereas, AML menus are used to provide an 
easy-to-use interface that employs a mouse to select the 
desired menu choices. Thus, a user interface can be created 
by integrating AML programs and menus. That is, in a menu- 
driven application, the user is presented with a list of actions 
from which a choice can be selected. Once the choice is 
selected, it is implemented by macros. With AML, multiple 
menus and multiple menu types can be displayed on the 
screen simultaneously. AML menus can make use of 
interactive, visual features like buttons, slider bars, check 
boxes, and icons. In this way, GUI is developed to assess 
GIS-based rainfall estimation system estimate monthly rainfall. 
5. RESULTS AND ANALYSIS 
The optimal rainfall estimation model, constituting of the best 
multiple linear regression equation involving the best area of 
interpolation is assessed for its accuracy and performance. 
The performance of the optimal model and accuracy of rainfall 
estimates depends very much on interpolation of surrounding 
significance points as well as topographic parameters to be 
taken into account. The topographic parameters included as 
the independent variables in the best multiple linear regression 
equation are elevation and shortest distance of the stations 
from the coast. Besides, the best threshold value of distance 
of separation at 50km and optimal level of-line-sight at 700m 
defines the best area of interpolation. The assessment is 
carried out in terms of different considerations of these 
topographic parameters as well as spatial distribution of 
rainfall stations. It consists of analysis and validation that are 
measured by r-square values and RMSE values as well as 
percent errors in estimates respectively, as depicted in the 
figure below. 
Spatial Distribution of Rainfall 
Stations: 
-number of interpolated points 
-distribution of rainfall stations from 
the coast 
Best area 
of 
interpolatio 
n: 
-best 
threshold 
value of 
distance of 
separation 
-optimal 
level of line- 
of-sight 
AssessmeitT" 
Analysis 
R- 
square 
r 
MSE 
Percentage 
—error in 
Comparison with other 
alternate estimation methods 
Independen 
t Variables 
-elevation 
-shortest 
distance of 
the point 
from the 
coast 
Figure 5-1 Assessment of the optimal rainfall estimation 
model 
The optimal rainfall estimation model has the highest r-square 
value when: 
(i) the best area of interpolation is defined by both the 
two topographic parameters, best threshold value of 
distance of separation at 50km and optimal level of 
line-of-sight at 700m, instead of involving either one 
of them only, 
(ii) the two topographic parameters of elevation and 
shortest distance of the stations from the coast are 
involved as independent variables in the optimal 
model rather than either one of the parameters only. 
From the analysis, it is found that: 
(a) From the RMSE results, the best threshold 
value of distance of separation is 50km instead 
of 40km. This indicates that although it is 
important to reduce area of interpolation as 
correlation coefficient of rainfall between two 
stations increases with decreasing distance of 
separation, there is a minimum value beyond 
which the accuracy of the estimates is affected. 
(b) The involvement of more topographic 
parameters in the rainfall estimation produce 
better results than the models with just one 
parameter as the independent variable. In this 
study, the model with only elevation as 
independent variable did not perform well. 
Whereas, the model with shortest distance 
from coast as the independent variable fetched 
better results. Thus, it is important to use the 
spatial correlation of rainfall with more than one 
variable to reduce estimation variances when