Stepwise model parametrisation using satellite imagery and hemispherical photography: tuning AquaCrop sensitive parameters for improved winter wheat yield predictions in semi-arid regions

Crop models are complex with many parameters, which has limited their application. Here we present an approach which both removes the model complexity through reducing the parameter dimensionality through sensitivity analysis, and presents a subsequent efficient approach to model parameterisation using swarm optimisation. We do this for two key model outputs, crop canopy and yield, and for two types of observational data, hemispheric photographs and Landsat7 imagery. Importantly we compare the usefulness of these two sources of data in terms of accurate yield prediction. The results showed that the dominant model parameters that predict canopy cover were generally consistent across the fields, with the exception of those related water stress. Although mid-season canopy cover extracted from Landsat7 was underestimated, good agreement was found between the simulated and observed canopy cover for both sources of data. Subsequently, less accurate yield predictions were achieved with the Landsat7 compared to the hemispherical photography-based parametrizations. Despite the small differences in the canopy predictions, the implications for yield prediction were substantial with the parametrization based on hemispherical photography providing far more accurate estimates of yield. There are, however, additional resource implications associated with hemispherical photography. We evaluate these trade-offs, providing model parametrization sets and demonstrating the potential of satellite imagery to assist AquaCrop, particularly on large scales where ground measurements are challenging.