Geostatistical methods for temperature estimation: A high-resolution Atlas of daily temperatures in Sicily

Abstract

Daily maximum and minimum temperature data are essential for different applications, such as climate change detection studies and the assessment of hydrological processes. These variables are generally obtained from meteorological stations, satellite products or reanalysis-based dataset. However, it is now well known that each of these products has limitations that compromise their use and/or reliability. As an example, reanalysis products generally provide datasets with very coarse spatial resolution, and their reliability may be affected by biases due to data availability during the assimilation processes. Even when statistical or dynamical downscaling techniques are used to improve their spatial resolution, computational resources and the high amount of observational data required limit these datasets to the last 20 years (Cornes et al. 2018). In contrast, ground stations offer a much more robust measurement but are often characterized by significant temporal discontinuities and limited spatial coverage. To address the lack of spatial and temporal coverage of ground-based temperature observations, several geostatistical techniques have been developed and used over time. For instance, Holdaway (1996) used the Ordinary Kriging (OK) method on monthly temperatures in Minnesota, USA, highlighting a temperature trend in some months. Wu and Li (2013) adopted the Residual Kriging (RK) method in the United States to interpolate monthly mean temperatures, demonstrating that the inclusion of elevation as an external drift improves the estimate. Di Piazza et al. (2015) compared different spatial interpolation methods in Sicily including OK and RK for estimating annual mean temperatures, finding that RK outperforms OK. Moving to the daily scale, Sekulić et al. (2019) developed a regression kriging model on daily average temperatures in Croatia at a spatial resolution of 1 km. They compare the outcomes with the MODIS land surface temperatures, founding reliable results. Here we propose a spatial interpolation methodology, based on geostatistical methods, for reconstructing a daily maximum and minimum temperature ATLAS for Sicily with a high spatial resolution (~2km) and for the period 1980-2024.