Robustness of climate indices relevant for agriculture in Africa deduced from GCMs and RCMs against reanalysis and gridded observations

Abstract

Abstract This study assesses the ability of climate models to represent rainy season (RS) dependent climate indices relevant for agriculture and crop-specific agricultural indices in eleven African subregions. For this, we analyze model ensembles build from Regional Climate Models (RCMs) from CORDEX-CORE (RCM_hist) and their respective driving General Circulation Models (GCMs) from CMIP5 (GCM_hist). Those are compared with gridded reference data including reanalyses at high spatio-temporal resolution (≤ 0.25°, daily) over the climatological period 1981–2010. Furthermore, the ensemble of RCM-evaluation runs forced by ERA-Interim (RCM_eval) is considered. Beside precipitation indices like the precipitation sum or number of rainy days annually and during the RS, we examine three agricultural indices (crop water need (CWN), irrigation requirement, water availability), depending on the RS’ onset. The agricultural-relevant indices as simulated by climate models, including CORDEX-CORE, are assessed for the first time over several African subregions. All model ensembles simulate the general precipitation characteristics well. However, their performance strongly depends on the subregion. We show that the models can represent the RS in subregions with one RS adequately yet struggle in reproducing characteristics of two RSs. Precipitation indices based on the RS also show variable errors among the models and subregions. The representation of CWN is affected by the model family (GCM, RCM) and the forcing data (GCM, ERA-Interim). Nevertheless, the too coarse resolution of the GCMs hinders the representation of such specific indices as they are not able to consider land surface features and related processes of smaller scale. Additionally, the daily scale and the usage of complex variables (e.g., surface latent heat flux for CWN) and related preconditions (e.g., RS-onset and its spatial representation) add uncertainty to the index calculation. Mostly, the RCMs show a higher skill in representing the indices and add value to their forcing models.