Using elevation to aid the geostatistical mapping of rainfall erosivity

Abstract This paper addresses the issue of incorporating a digital elevation model into the mapping of annual and monthly erosivity values in the Algarve region (Portugal). Besides linear regression of erosivity against elevation, three geostatistical algorithms are introduced: simple kriging with varying local means (SKlm), kriging with an external drift (KED) and colocated cokriging. Cross validation indicates that the straightforward linear regression, which ignores the information provided by neighboring climatic stations, yields the largest prediction errors in most situations. Smaller prediction errors are produced by SKlm and KED that both use elevation to inform on the local mean of erosivity; kriging with an external drift allows one to assess the relation between the two variables within each kriging search neighborhood instead of globally as for simple kriging with varying local means. The best results are generally obtained using cokriging that incorporates the secondary information directly into the computation of the erosivity estimate. The trade-off cost is the inference and modeling of three direct and cross semivariograms.