Sequential hybridization enhances the reliability of a statistical crop yield model – exemplified by wheat and sugar beet yields in the provinces of Poland

Abstract This study uses a data-driven modelling approach to integrate drought indicators and weather data to enhance crop yield simulations. By employing a sequential hybridization method, the research combines drought indices such as SPI, SPEI, or SMI with weather data to refine wheat and sugar beet yield predictions by the statistical model ABSOLUT at the voivodship level in Poland. The modelling approach systematically evaluates possible combinations of 15 input features to identify the most effective configurations for multiple linear regressions. The findings reveal that regression models incorporating both weather data and drought indicators (particularly SPEI and SMI) deliver superior performance compared to those relying solely on weather variables. This improvement is especially pronounced under conditions of variable moisture availability. For example, a model that includes SPEI and weather data more precisely estimates wheat yield, especially during extreme events like the 2018 drought. Additionally, using SMI as the only feature demonstrated that ABSOLUT performed better than when combined with weather data in the case of sugar beet. These results underscore the critical role of incorporating drought indicators to bolster the reliability of crop yield predictions, offering significant insights for agricultural planning in regions susceptible to climatic variability. The research also highlights the potential of hybrid modelling approaches, which combine the strengths of process-based and data-driven models to enhance predictive accuracy. Moreover, the study suggests that these models could be further refined by incorporating additional environmental factors for more robust agro-hydrological simulations.