Mapping average annual precipitation accounting for location-dependent variations

Geostatistical applications generally require the calculation of an experimental variogram and the fitting to it of a theoretical variogram model to describe and quantify spatial variation of the whole study area. Since in mountainous areas, it is not always reasonable to assume that a global variogram of the entire study area is suitable for modelling the spatial variation of precipitation, a solution could be using local parameters of the variogram. This study was aimed at determining whether the local geostatistics approach with varying variogram model parameters improves the modelling of average annual precipitation compared to the use of a global variogram with constant parameters for the entire study area. The study area was a region in southern Italy (Calabria) with a spatially variable Mediterranean climate. A dataset of 84 precipitation stations with long-term records (1951–2022) was used. A global variogram model of the precipitation data was calculated and fitted and used in the optimization of the local variograms parameters. Global and local variograms were used with block kriging with elevation as external drift to improve the estimation of precipitation. Using local parameters of variogram has allowed to improve the precipitation predictions as proven by all measures of accuracy (mean absolute error, root-mean-squared error of prediction, and mean relative error) and by the goodness-of-prediction of the estimates obtained by using the global and local parameters of variogram models. These results were also confirmed by R2 values obtained from two scatterplots of predicted precipitation values using global variogram (R2 = 0.58) and local parameters (R2 = 0.74) versus measured values. The two approaches produced a general similar spatial distribution of average annual precipitation, but with specific differences at local level in more orographically complex areas.