Open Access

Maximilian Zachow, Ivana Aleksovska, Riccardo Henin, Harald Kunstmann, Michele Meroni, Lorenzo Seguini, Elena Tarnavsky, Senthold Asseng

• The MARS Crop Yield Forecasting System combines process-based crop model outputs, satellite vegetation indicators and gridded meteorological data within an analyst-guided statistical modelling approach. In May, a critical stage of wheat development, MARS publishes its first forecast based on observed data. While 10-day weather forecasts are often considered at the analyst’s discretion, our study goes further by quantitatively incorporating four-week forecasts. To evaluate their added value, we introduce a new framework, MARS+Forecast, which extends MARS yield predictions using ECMWF four-week temperature and precipitation forecasts following the MARS publication date. This framework employs a data-driven yield model for 24 EU countries and is assessed for the harvest years 2007–2024. When assuming a perfect four-week weather forecast based on reanalysis data (MARS+Perfect), MARS forecasts would theoretically be improved in 16 countries, covering 60% of EU wheat area and 55% of wheatThe MARS Crop Yield Forecasting System combines process-based crop model outputs, satellite vegetation indicators and gridded meteorological data within an analyst-guided statistical modelling approach. In May, a critical stage of wheat development, MARS publishes its first forecast based on observed data. While 10-day weather forecasts are often considered at the analyst’s discretion, our study goes further by quantitatively incorporating four-week forecasts. To evaluate their added value, we introduce a new framework, MARS+Forecast, which extends MARS yield predictions using ECMWF four-week temperature and precipitation forecasts following the MARS publication date. This framework employs a data-driven yield model for 24 EU countries and is assessed for the harvest years 2007–2024. When assuming a perfect four-week weather forecast based on reanalysis data (MARS+Perfect), MARS forecasts would theoretically be improved in 16 countries, covering 60% of EU wheat area and 55% of wheat production. When adding an actual four-week forecast to MARS (MARS+Forecast), MARS forecasts were modestly improved in 8 countries, representing 39% of wheat area and 31% of wheat production. Extreme yield losses, such as in France in 2016 or Germany in 2018, are not captured by our model, because of the relatively short training datasets with a limited number of extreme years. MARS+Forecast extends the MARS system in May with a four-week weather forecast and demonstrates a novel approach with moderate improvement of the current operational MARS system. More training data, more accurate MARS predictions and improved four-week weather forecasts are needed for future improvements.…