Abstract:
Many hydrological models for watershed management and planning require
rainfall as an input in a continuous format. This study analyzed four different
rainfall interpolation techniques in Nyando river basin, Kenya. Interpolation was
done for a period of 30 days using 19 rainfall stations. Two geostatistical
interpolation techniques (kriging and cokriging) were evaluated against inverse
distance weighted (IDW) and global polynomial interpolation (GPI). Of the four
spatial interpolators, kriging and cokriging produced results with the least root
mean square error (RMSE). A digital elevation model (DEM) was introduced into
the cokriging method and this improved the results considerably. The results
demonstrate that for low-resolution rain gauge networks, geostatistical
interpolation methods perform better than other techniques that ignore spatial
dependence patterns. The use of secondary information improved the prediction
results, as demonstrated by the inclusion of the DEM in this study.
