Improving the estimation of solar radiation and evapotranspiration for crop models using the Hargreaves-Ssamani method at the European scale

In the context of climate change, the use of cropping system models to assess the productivity of agricultural systems under future scenarios is continuously growing. To provide the estimation of the photosynthetic activity, is important to adopt reliable data of solar radiation (Rs) and evapotranspiration (ET). Using daily meteorological data from ECAD, machine learning (ML) models and multiple linear regression (MLR) were trained to estimate the solar radiation coefficient (krs) of the Hargreaves-Samani (HS) equation at the European scale, using information on location, temperature, rainfall, and extraterrestrial solar radiation. The models were trained on 6.3M daily meteorological data and validated on 2.6M. The results show that ML methods outperformed MLR, and also the estimates obtained using standard HS coefficients, reducing the error of estimation for both Rs and ET. In validation, the RMSE of Rs with ML decreased by 13% (-0.5 MJ m-2 d-1), resulting in a 13.2% reduction in ET error (-0.22 mm d-1). These improvements could enhance the performance of cropping system models for assessing agricultural productivity in different climate scenarios.