The impact of high quality field data on crop model calibration

Process-based crop simulation models (CSMs) are valuable tools for assessing genotype by environment by management (GxExM) interactions and quantifying climate change impacts on crops. Ex-ante evaluations of adaptation options to drought stress require well-validated CSMs that are continuously improved and evaluated. This asks for high quality data from model-driven field experiments. We collected detailed data on weather, soil, and crop growth and development in one season of barley (cv. RGT Planet) field experiments at three locations in Denmark. The resultant dataset meets the highest standards for crop model improvement as defined by the modelling community. To evaluate the importance and impact of data quality on model calibration results, the CSM APSIM was calibrated for one location, first with a low, then with a medium, and finally with the high quality dataset generated in the field experiments. The low quality dataset represents a typical scenario of limited data availability for CSM calibration (e.g. limited soil description, few in-season phenology and biomass measurements). In a medium quality dataset usually better soil descriptions and phenology and biomass measurements at different crop stages are available, yet in lower temporal and spatial resolution than in a high quality dataset.

Phenology was predicted accurately with all datasets, but the highest accuracy was achieved using the high quality dataset (root mean square error RMSE low: 4.39, medium: 4.23, high: 1.56). LAI was overestimated with all quality datasets; however, the high quality calibration results were closest to the observations (RMSE low: 1.89, medium: 1.61, high: 1.09). Final grain yield was underestimated with the low and medium quality dataset but slightly overestimated with the high quality dataset, which facilitated the most accurate yield prediction (difference between modelled and observed yield: low: -6%, medium: -3.13 %, high: +1.38%).                                                                  Findings from this study support our basic hypothesis that calibrating a CSM with high quality data increases the prediction accuracy.    However, our results show that calibrating LAI and grain yield (complex traits) require more comprehensive datasets than calibrating phenology.

By generating such a high quality dataset, we contribute substantially to meeting the need for detailed and comprehensive datasets fit for model calibration and evaluation purposes, which are especially rare for northern Europe. We also found that APSIM possibly does not fully reproduce translocation processes, but this requires further field and modelling experiments.