Comparative Analysis of Spatial Interpolation Techniques for Mapping Annual Mean Rainfall Estimation within a Mountainous Region

The complex topography, poor gauge representativity and uneven density make it an uphill task to accurately map precipitation in mountainous regions. This challenge was confronted with the evaluation of four different mapping techniques: Inverse Distance Weighting (IDW), Ordinary Kriging (OK), Spline and Regression Kriging (RK). An evaluation of the resulting georasters using 1) cross-validation statistics, 2) a spatial cross-consistency test and 3) a water balance analysis reveals that the techniques ignoring the information on co-variables yield the largest prediction errors. Mean error and root-mean-square error values suggested that the most biased methods were IDW and spline, with a bias almost 2 to 5 times higher than ordinary kriging. The best model accounted for mean precipitation analysis is regression Kriging, with a mean error and root mean square error values of 1.38 mm and 72.36 mm respectively, which represents 42 % less bias and 16 % higher accuracy than OK results. Comparative performances show that the regression analysis made it possible to judiciously evaluate the variable patterns and get fairly accurate values at un-gauged locations where geographical information compensated the poor availability of local data.