Estimation of Missing Rainfall Data in Northeast Region of Thailand Using Spatial Interpolation Methods

Ground-based rainfall observations are the primary sources of precipitation data used in most developing countries. However, those observations are frequently damaged or incomplete, thus missing data is always a problem. This comparison study examines a number of spatial interpolation methods used to estimate missing monthly rainfall data in the northeast region of Thailand. The comparison was grouped into global and local methods. In global methods, trend surface analysis was compared to back-propagation neural network. The results showed that back-propagation neural network is more capable of tolerating to richnoised data. However, such neural network must be carefully used because it could provide unreliable results at the boundary area. In local methods, common used kriging methods were compared and it was found that the characteristics of the datasets have significant effects on the estimation performance. This study recommends using the kurtosis value of observations’ histogram and nugget to sill ratio of fitted semivariogram models as a guideline to select between ordinary kriging and universal kriging methods. Since the study area is a large plateau, in which there is low correlation between rainfall and altitude, ordinary cokriging method cannot make use of the altitude as a supplementary feature to improve the estimation performance.