Abstract
Recent climate change motivates the creation of high‐resolution and high‐quality reference datasets of essential environmental variables as the necessary base for adaptation planning. To obtain these maps, a widely used procedure is to interpolate in situ observations, for which several different methods were developed in the past decades. In this study, we calculate gridded daily maps of precipitation and temperature at the regional scale in Abruzzo (central Italy), comparing different interpolation methods: universal kriging, radial basis function and gradient boosting forest. We validated the results against an independent set of stations from the same network used to produce the gridded dataset (1994–2013). The interpolated values were also compared with those obtained from two widely used global datasets (CHELSA and WorldClim). Universal kriging achieved the best performance, with a daily root mean square error of ~0.45 mm/day for precipitation and ~1.2°C for temperature. Seasonality affects the bias values, being larger in winter for precipitation and in summer for temperature, as well as in isolated stations in mountainous areas. Our gridded dataset (ADAMO, ~0.01° and daily resolution) shows decreased bias with respect to the global databases. There was a considerable discrepancy for precipitation (RMSE ≥ 60 mm/month for CHELSA and WorldClim , while ADAMO was 35 mm/month), and a too strong altitude effect for temperature, especially in WorldClim. Temperature increased its RMSE in summer compared to winter error, more pronounced in global datasets, while precipitation had an increase in winter with respect to summer, more evident in the ADAMO dataset. These results show that in regions with large topographic variability, the implementation of datasets with diffuse local observations is expected to be more accurate than global datasets. This is particularly relevant for fields such as climatology, ecology and biology that require climate information with high accuracy.