Uncertainty-based auto-calibration for crop yield – the EPIC+ procedure for a case study in Sub-Saharan Africa

Abstract

Process-based crop models are increasingly used to assess the effects of different agricultural management practices on crop yield. However, calibration of historic crop yield is a challenging and time-consuming task due to data limitation and lack of adaptive auto-calibration tools compatible with the model to be calibrated on different spatial and temporal scales. In this study we linked the general auto-calibration procedure SUFI-2 (Sequential Uncertainty Fitting Procedure) to the crop model EPIC (Environmental Policy Integrated Climate) to calibrate maize yield in Sub-Saharan African (SSA) countries. This resulted in the creation of a user-friendly software, EPIC+, for crop model calibration at spatial levels of grid to continent. EPIC+ greatly speeds up the calibration process with quantification of parameter ranges and prediction uncertainty. In the SSA application, we calibrated three sets of parameters referred to as Planting Date (PD), Operation (e.g., fertilizer application, planting density), and Model parameters (e.g., Harvest index, biomass-energy ratio, water stress harvest index, SCS curve number) in three steps to avoid parameter interaction and identifiability problems. In the first step, by adjusting PD parameters, the simulated yield results improved in Western and Central African countries. In the next step, Operation parameters were calibrated for individual countries resulting in a better model performance by more than 40% in many countries. In the third step, Model parameters were calibrated with significant improvements in all countries by an average of 50%. We also found that countries with less socio-political volatility benefited most from the calibration. For countries where agricultural production had trends, we suggest improving the calibration results by applying linear de-trending transformations, which we will explore in more detail in a subsequent study.