Estimating surface air temperature from multiple gridded observations and reanalysis datasets over Ghana

Abstract

Global temperature datasets have become increasingly accessible in recent years, yet their performance in regions with sparse data remains a subject that requires thorough investigation, especially in areas where validation is essential. This study addresses this urgency, given the significant impact of the increase in global temperature, as highlighted in the IPCC's 6th Assessment Report, on regional and local levels. In this research, we evaluated air temperature datasets from Climatic Research Unit (CRU TS4.05), Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), Climate Prediction Centre (CPC), European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis (ERA-5), and TerraClimate (hereinafter referred to as Gridded Temperature Datasets - GTDs). Our assessment involved comparing and validating these datasets against reference ground observation data from 23 synoptic stations, spanning 40 years from 1981 to 2020, to ensure robustness in regions with sparse data. We conducted a spatio-temporal investigation to explore temperature patterns, and evaluated their capacity to reproduce inter-annual and intra-seasonal variability. To assess dataset performance, we employed a group of statistical metrics, including root-mean-square error (RMSE), mean absolute error (MAE), percentage bias (PBIAS), and correlation coefficient (R). The GTDs consistently captured the annual climatological cycle distribution over Ghana. The correlation coefficient values significantly varied across the 23 synoptic stations and three climatic zones. At the annual level, CPC exhibited R values ranging from 0.53 to 0.93, while CRU ranged from 0.48 to 0.88. The bias for CPC and CRU spanned from 2.40 to −0.35 °C and 2.07 to −0.91 °C, respectively. Overall, CRU, CPC, and TerraClimate demonstrated the most consistent near-surface temperature patterns over Ghana, outperforming other datasets. Based on our findings, we recommend these datasets for model evaluation and assessment in Ghana, given their strong performance in capturing temperature variations.