Increasing the accuracy of geostatistical assessments involving extrapolation and zonal classification techniques: case study of Karun basin daily rainfall, IRAN

In this paper with analyzing various geostatistical methods in three different scenarios, the influence of zonal classification and rainfall extrapolation on increasing accuracy of the proposed methods for daily rainfall in Karun basin has been analyzed. In the first scenario, only various geostatistical methods including weighting moving average, Kriging, Co-Kriging and TPSS were analyzed. In the second scenario, the study area was divided into homogeneous regions based on cluster method and the accuracy of the selected models were analyzed within each region. In the third scenario, some hypothetical stations were considered in elevation points without real observations, and then, the accuracy of geostatistical method with considering extrapolation technique was analyzed. Analyzing the related Variogram well demonstrated a geometric anisotropy, while this problem was solved up when an all-round variogram was applied. In the first scenario, the mean absolute errors and the mean bias errors were varied in the range of 14.7-36 mm and -1.5-4.5 mm, respectively. In this scenario, the Co-kriging method had the lowest estimation error due to the positive effect of the considered auxiliary variable on increasing variogram "range of influence". Although, the interpolation of daily rainfall data, after zonal classification, had no positive impact on decreasing of estimation error, but the weighting moving average and TPSS methods with the powers of 3-5 were the Exceptions. In addition, the estimation accuracy increased up to 16%, and the estimation bias reduced up to 10%, when precipitation was extrapolated in elevations with no rain gauges. Under this scenario, the maximum threshold of error variance reduced by 45% (from 7.8 mm to 4.8 mm). Based on the results, integrating extrapolation into the Co-Kriging method with considering elevation as an auxiliary variable results the best estimation when assessing the spatial variation of daily rainfall.