Identifying precipitation uncertainty in crop modelling using Bayesian total error analysis

Abstract

Precipitation is an important source of soil water, which is critical to crop growth, and is therefore an important input when modelling crop growth. Although advances are continually being made in predicting and recording precipitation, input uncertainty of precipitation data is likely to influence the robustness of parameter estimate and thus the predictive accuracy in soil water and crop modelling. In this study, we use the Bayesian total error analysis (BATEA) method for the water-oriented crop model AquaCrop to identify the input uncertainty from multiple precipitation products respectively, including gauge-corrected grid dataset CPC, remote sensing based TRMM and reanalysis based ERA-Interim. This methodology uses latent variables to correct the input data errors. Adopting a single-multiplier method for precipitation correction, we simulate maize growth in both field and regional levels in China for a range of different possible climatic scenarios. Meanwhile, we use the average of multiple products for model driving in comparison. The results show that the BATEA method can consistently reduce uncertainty for crop growth prediction among different precipitation products. In regional simulation, the improvements for the three products are 1%, 7.3% and 2.8% on average in drought scenarios. These results imply the BATEA approach can be of great assistance for crop modeling studies and agricultural assessments under future changing climates.