Abstract

The accuracy of hydrological assessments in mountain regions is often hindered by the low density of gauges coupled with complex spatial variations in climate. Increasingly, spatial datasets (i.e. satellite and other products) and new computational tools are merged with ground observations to address this problem. This paper presents a comparison of approaches of different complexities to spatially interpolate monthly precipitation and daily temperature time series in the upper Aconcagua catchment in central Chile. A generalised linear mixed model (GLMM) whose parameters are estimated through approximate Bayesian inference is compared with simpler alternatives: inverse distance weighting (IDW), lapse rates (LRs), and two methods that analyse the residuals between observations and WorldClim (WC) data or Climate Hazards Group Infrared Precipitation with Station data (CHIRPS). The assessment is based on a leave-one-out cross validation (LOOCV), with the root-mean-squared error (RMSE) being the primary performance criterion for both climate variables, while the probability of detection (POD) and false-alarm ratio (FAR) are also used for precipitation. Results show that for spatial interpolation of temperature and precipitation, the approaches based on the WorldClim or CHIRPS residuals may be recommended as being more accurate, easy to apply and relatively robust to tested reductions in the number of estimation gauges. The GLMM has comparable performance when all gauges were included and is better for estimating occurrence of precipitation but is more sensitive to the reduction in the number of gauges used for estimation, which is a constraint in sparsely monitored catchments.