Validity of Regional Rainfall Spatial Distribution Methods in Mountainous Areas

Abstract

Estimation of spatial distribution of rainfall at different time scales is one of the basic steps in water resource studies. Problems in selection of proper methods for mapping rainfall can cause uncertainty in water budget studies and simulation of hydrologic models. Application of interpolation techniques is more problematic in mountainous areas where the number and spatial distribution of rain gauges is usually inadequate. There are numerous methods for estimation of spatial distribution of rainfall based on observed point data. Geostatistical methods are among interpolation techniques that are favored in many engineering studies since they consider spatial structure of data points. In this study, a number of methods were evaluated to estimate spatial distribution of annual rainfall in the southwest of Iran. These methods consist of thin plate smoothing splines (TPSS), weighted moving average, ordinary kriging, and cokriging. The cross-validation technique was applied in conjunction with analysis of cumulative distribution function and trends with elevation to evaluate the performance of different methods. While semivariogram analysis confirmed the existence of nonstationarity in the study area, division of the region based on cluster analysis allowed application of ordinary kriging. Moreover, although the TPSS method with an order of two was marginally the most accurate method based on the cross-validation results, the cokriging method showed better consistency with the topography of the area. As extrapolation is always required in mountainous regions, none of the methods produced acceptable results in extrapolation of rainfall point data to higher elevation areas.