Comparison of different geostatistical approaches to map climate variables: application to precipitation

The benefits of an integrated geographical information system (GIS) and a geostatistics approach to accurately model the spatial distribution pattern of precipitation are known. However, the determination of the most appropriate geostatistical algorithm for each case is usually neglected, i.e. it is important to select the best interpolation technique for each study area to obtain accurate results. In this work, the ordinary kriging (OK), simple kriging (SK) and universal kriging (universal kriging) methods are compared with three multivariate algorithms which take into account the altitude: collocated ordinary cokriging (OCK), simple kriging with varying local means (SKV) and regression-kriging (RK). The different techniques are applied to monthly and annual precipitation data measured at 136 meteorological stations in a region of southwestern Spain (Extremadura). After carrying out cross-validation, the smallest prediction errors are obtained for the three multivariate algorithms but, particularly, SKV and RK outperform collocated OCK, which needs a more demanding variogram analysis. These algorithms are easily implemented in a GIS, requiring the residual estimates and map algebra capability to generate the final maps. Results evidence the necessity of accounting for spatially dependent precipitation data and the collocated altitude, to accurately define monthly and annual precipitation maps. Copyright  2009 Royal Meteorological Society