Performance of LINTUL-2 in simulating water-limited yields of maize using different sets of weather data in Ghana

The application of crop models in sub-Saharan Africa is often limited by the scarcity of reliable weather data. This study evaluated the performance of the LINTUL-2 maize model in simulating water-limited yields using three weather datasets: observed station data from the Ghana Meteorological Agency (GMet) and two gridded sources—ERA5-Land and NASA POWER. Model performance was assessed by quantifying and comparing the uncertainties associated with each dataset’s yield predictions. Calibration was conducted using 2020 field data on phenology and yield, yielding root mean square error (RMSE) of 0.82–1.44 t ha−1, mean error (Bias) of 0.34–1.2 t ha−1, and ratio of performance to interquartile distance (RPIQ) values ranging from 0.88 to 1.53 across the datasets, with GMet consistently producing the most accurate results. Observed 2020 yields ranged from 3.4 to 6 t ha−1, while simulated yields ranged from 4.4 to 6.9 t ha−1. Validation with 2021 data produced RMSE values of 1.2–1.8 t ha−1, Bias ranging from −0.6 to 0.9 t ha−1, and RPIQ scores of 0.8–1.2. Further evaluation using 60 legacy yield observations revealed that ERA5-Land showed better water-limited yield variability but still overestimated yields (Bias = 2.51 t ha−1), slightly outperforming NASA POWER (Bias = 2.85 t ha−1). Drastic underestimation at some locations were linked to poor rainfall distribution and low soil water-holding capacity, despite sufficient seasonal rainfall totals (> 400 mm), resulting in prolonged crop stress. Overall, the model tended to overestimate yields, likely due to field conditions not fully aligning with the model’s assumptions of water limitation and nutrient sufficiency. The findings suggest that while gridded weather datasets, when used with crop growth models, can provide reasonably accurate yield estimates in data-scarce regions, aligning field conditions with model assumptions–particularly regarding nutrient availability–is critical for improving simulation reliability.