Satellite rainfall bias assessment for crop growth simulation – A case study of maize growth in Kenya

In this research, the performance of satellite rainfall estimates (SREs) for crop growth simulation was investigated. Rainfall products selected were CHIRPS 2.0, CMORPH 1.0, MSWEP 2.2, and RFE 2.0. In-situ rainfall from 20 stations within the Lake Victoria basin in Kenya served as reference. Rainfall products were evaluated for onset days, rainfall depths, dry spells, and rainfall occurrence for four crop growth stages. Assessment was on a daily time step for the period 2012–2018 and on a point-to-pixel basis. Results showed that SREs exhibit large variation in timing of rainfall arrival. SREs exhibited largest interannual and spatial spreads in representing dry spell length during the flowering stage with CMORPH and CHIRPS showing best and weakest results, respectively. Bias of SREs in representing dry spells was smaller during early growth stages. Detecting rainfall occurrence by the SREs weakened as the growing season progressed. MSWEP, followed by RFE2, produced the best results in detecting rainfall events, while falsely detected rainfall was frequent in CHIRPS, particularly in later growth stages. SREs generally performed better during a wet than a dry growing season. SREs indicated less bias in rainfall depths during the early stages of crop growth but deteriorated at later stages. MSWEP and CMORPH exhibited the least and highest interannual spread in relative bias, respectively. In associating biases to severe and extreme water stress, based on crop water requirement satisfaction index, effects were more prevailing in the ripening than flowering stages. Findings of this study suggest that SREs can serve as input to crop growth modelling, but validation of SREs with rain gauge observed counterparts is essential.