Spatial Interpolation of Regional Rainfall Data of Thailand Using Different Methods

Abstract

Distributed areal rainfall is important data for hydrological modeling. Spatial interpolation is used to predict the values of a continuous variable at unmeasured points. This study compared the distribution and prediction error in the spatial interpolation maps of rainfall derived from different methods. Inverse Distance Weighted (IDW) using a power (p) equal to 2 and 3, and number of neighbor point (n) equal to 15, 30, and 50 points point and Kriging Ordinary (KrigOrd) with Circular, Exponential, Gaussian, Linear, and Spherical model were used and their predictions compared. Cohen’s kappa coefficient and mean absolute error were used to compare distribution and error in spatial interpolation, respectively. The results indicated that the distribution and parameters obtained from different spatial interpolation methods were in good agreement as indicated by k values (0.9202 to 0.9999) but results of IDW predicted a gradual change, while that from KrigOrd had localized features and abrupt change. The lowest spatial interpolation error from measurements in the north, northeast, central, east and south (east coast) parts was obtained for IDW with p of 2 with differing number of neighbors equal to 15, 15, 50, 50, and 15 points respectively, while that for the south (west coast) was obtained using the KrigOrd with a linear setup. Mean absolute error (MAE) for each method was 133.88 132.49 114.83 184.01 181.54, and 142.52 millimeters or 11.30, 10.25, 10.58, 11.76, 10.90, and 5.44 percent of the mean annual rainfall, respectively.