ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001 
314 
one of the variables is undersampled such as 
elevation. 
(c) The number of interpolated points within the 
best area of interpolation affects the accuracy 
of rainfall estimates at a point. It is discovered 
that too few interpolated points reduce the 
accuracy of the estimates. Thus, it is important 
to increase the number of interpolated points 
by obtaining more surrounding stations. 
(d) In this study, the best threshold value of 
distance of separation has more significance 
impact in rainfall estimation than the optimal 
level of line-of-sight. Besides, elevation does 
not contribute significantly to the optimal model. 
This indicates that the effect of barriers and 
elevation are not so significant as it should be, 
due to the locations of rainfall stations involved 
in this study which are mainly situated on the 
low terrain. Thus, the effects of these two 
topographic parameters are not studied 
intensively. In order to study their significant 
effects, more stations on the higher complex 
terrain are required. 
(e) In the distribution of rainfall relative to the 
distance of rainfall stations from the coast, 
coastal stations and inland stations varies 
greatly in rainfall with coastal stations receive 
much higher rainfall than inland stations. This 
may give rise to errors in estimates as both 
coastal and inland stations are involved during 
the process of interpolation. Thus, it is 
paramount to separate coastal stations and 
inland stations during interpolation or minimize 
the area of interpolation so as to reduce error 
induced by the large variation in rainfall in order 
to produce better estimates. However, as 
indicated in part c, number of interpolated 
points is also important as too few points may 
reduce accuracy of the estimates. In order to 
solve this problem, more points should be 
included in the study area in order to 'localize' 
area of interpolation without sacrificing number 
of interpolated points. 
Finally, the RMSE of the optimal model is compared with the 
RMSE of some of the alternate rainfall estimation methods. 
Cross-validation results for alternate estimation methods are 
listed together with RMSE value of the optimal model in order 
of RMSE performance in the table below: 
Estimation Methods 
RMSE values 
Neighborhood 
0.9 
Inverse-Distance 
0.81 
Inverse-distance 
0.81 
Inverse-distance 
0.74 
Kriging 
0.71 
okriging 
0.32 
Optimal Model 
0.253 
Table 5-1 RMSE Values for Alternate Estimation Methods 
From the table above, neighborhood average, inverse distance, 
inverse-distance squared, inverse-distance cubed and kriging 
methods are the interpolation methods using only rainfall 
measurements whereas Cokriging [8][9] used elevation and 
rainfall measurements. The least favorable estimation method 
was neighborhood average, with RMSE values greater than 
0.90. Among these alternate methods, Cokriging was the 
most favorable method when elevation factor was involved. In 
addition, estimation methods using elevation had more 
favorable RMSE results than interpolation methods using only 
rainfall measurements, indicating that the correlation of rainfall 
with elevation is more important for estimating rainfall than the 
spatial correlation of available rainfall measurements. In 
comparison, the optimal model has the best result indicating 
improved estimation performance with its RMSE value the 
lowest as compared with RMSE values of all the listed 
alternate estimation methods. Thus, rainfall estimation models 
using correlation of rainfall with more topographic parameters 
such as in the optimal model have more favorable RMSE 
results and thus better estimates. 
6. CONCLUSION 
This optimal model is developed by incorporating GIS. 
Without GIS, there are problems encountered in this technique 
in dealing especially in dealing with spatial data. The 
followings are achieved with the aid of GIS in this project. 
(a) In this rainfall estimation technique, it involves 
computation of spatial data and performance of spatial 
and surface analysis. This is achieved with the aid of GIS 
tools which has the full capabilities in dealing with 
geographical data and analysis. 
(b) Spatial correlationship of rainfall with more topography 
parameters can be studied and analyzed to derive the 
optimal rainfall estimation model to produce better 
estimates. As compared with other alternate rainfall 
estimation methods which take into account one 
topography parameter only, either distance of separation 
factor or elevation factor, in the analysis as review in 
literature review of Chapter two, RMSE results show that 
the optimal model performs better from RMSE results. 
Thus, it is paramount to study spatial correlation of 
rainfall with more topography parameters in rainfall 
estimation. 
(c) Graphical user interface is developed in GIS using the 
ARC Macro Langauge to assess the GIS-based rainfall 
estimation system for estimating point and areal-average 
rainfall from locational data randomly distributed over the 
study area. 
(d) Isohyetal maps are produced to indicate isolines of 
monthly rainfall estimates over the study area. 
In conclusion, GIS is urgently employed to solve the problems 
of spatial data and analysis in including the topography 
parameters in this optimal rainfall estimation model. That is, 
as rainfall varies spatially, GIS can be integrated with rainfall 
estimation technique to improve the accuracy of the estimates. 
7. FUTURE RESEARCH 
There are still other factors that give weightages to rainfall 
estimation. These factors include: 
• Physical variables. Examples of these variables are mean 
temperature in degree Celsius, mean number of sunshine 
hours per month and average relative humidity in 
percentage. These variables are significance in 
contributing to convective activities that cause convective 
type of rain in the study area such as showers and