Proccessing and geostatistical interpolation of data on annual precipitation for the meteostations of Western Ukraine

There is a significant demand for accurate and reliable spatially distributed precipitation data. Modeling and mapping of precipitation fields are best achieved by geostatistical interpolation procedures that also consider explanatory variables like the terrain morphometric parameters that influence the precipitation distribution. The purpose of this research was to create an accurate regional precipitation map by interpolating data on average annual precipitation sums gained through summarizing records of 50 meteorological stations located in Western Ukraine. Daily data have been downloaded from open GHCN database, then preprocessed and summarized in R to obtain average annual precipitation sums for each station. Auxiliary data on terrain morphometric parameters have been gathered by preprocessing of SRTM Version 4.1 DEM. Data were then interpolated in gstat R package using two methods: the ordinary kriging and the multiple regression model that uses a set of terrain morphometric parameter as explanatory variables. Both methods produced an estimated precipitation map accompanied by a map of estimation error quantified with RMSE. The estimation by leave-one-out cross validation revealed that the multiple regression method produced much better accuracy, accounting for more than 90% of initial variance, compared with 63% for an ordinary kriging method. Combining and synthesizing both of these interpolation methods is possible with regression-kriging (being considered the best linear unbiased prediction model for spatial data). However, in this case it is not justified, as one of the methods performed considerably better than the other.