How can phenology monitoring network data improve operational systems for regional yield prediction? A case study for winter wheat and grain maize in France with the WOFOST model. 

Abstract

Mechanistic, process-based crop models are a key component of operational systems for regional yield forecasting, such as the Mars Crop Yield Forecasting System (MCYFS) of the Joint Research Centre of the European Commission. Such systems usually rely on spatially explicit soil, weather and crop data to simulate crop growth, biomass accumulation and yield formation.  In order to simulate crop growth over large regions, the MCYFS uses strongly simplified crop and management information. Phenology parameterization is a typical example of such simplification, as fixed sowing dates and values for phenological parameters are used across years and sub regions, resulting in potentially large differences between simulated phenology and reports from the field, ultimately reducing the accuracy of other simulated variables, such as leaf area index, biomass and yield.  In this study, we use ground truth phenological data obtained during the past 10 years in France, with the crop model WOFOST (which is used in the MCYFS), to assess the effects of a more precise phenology parameterization on simulations and yield predictions. In addition, we built on findings from a recent study connecting phenological stages with Copernicus Sentinel-2 satellite time series to assess the operational potential of integrating remote sensing and crop modeling for the purpose of crop yield forecasting.  The crop model WOFOST is used to simulate the growth and development of the two crops. WOFOST works at a daily time step and calculates daily biomass gains on the basis of underlying processes, such as photosynthesis, respiration, and how these processes are influenced by environmental conditions, such as irradiation, temperature and soil water conditions. Daily biomass gains are partitioned among plant organs depending on thermal-time-determined phenological stages. Phenology simulation is based on a temperature sum approach. Key phenological parameters include base temperature, set at 0 °C for wheat and 4 °C for maize, the thermal time from emergence to anthesis (TSUM1), the thermal time from anthesis to physiological maturity (TSUM2) and, in the case of wheat, a vernalization factor (Fv).  Phenological data used in this study are derived from the “Céré’Obs” program (https://cereobs.franceagrimer.fr/cereobs-sp/), which aims to provide objective data on the status of major cereal crops in France. Data are provided at the level of administrative regions. In this study, winter wheat and grain maize are considered, from 2012 onwards.  Baseline simulations are first performed with the standard MCYFS setting of WOFOST. Then, updated simulations are performed, following a two-step approach: first, sowing dates are forced from Céré’Obs information for each year and region in France. Then, key phenological parameters TSUM1, TSUM2 and, in the case of wheat, Fv, are recalibrated so as to minimize differences between simulated phenology and ground observations. This parameterization update is performed both with Céré’Obs-collected information and remote sensing-derived information. Model’s outputs for both standard and updated simulations are finally compared against sub-national official yields.