Learning main drivers of crop dynamics and production in Europe

An expanding world population combined with challenges brought by climate change pose totally new scenarios for managing agricultural fields and crop production. In the last decades, a variety of ground-based, modeled, and Earth observation (EO) data have been used to characterize crop dynamics and, ultimately, estimate yield. Typically, optical vegetation indices and, in particular, metrics like their maximum peak or integral during the growing season are exploited to estimated crop yield. Also, most studies are focused on large areas with homogeneous agricultural landscapes in which cultivation/production is centred in a unique main crop (e.g. the U.S. Corn Belt or the Indian Wheat Belt). 

In this study, we study the transportability of machine learning models for crop yield estimation across different regions and the relative relevance of agro-ecological drivers (input features). We use a previous methodology presented in [1] that synergistically combined optical and microwave vegetation data for crop yield prediction. We apply this methodology, which was trained in the homogeneous area of the US Corn Belt, to the highly heterogeneous agricultural landscapes across Europe. The fragmented and diverse European agro-ecosystems poses a greater challenge for the combination of multi-sensor data, and we need to establish first which is the set of variables providing the best skill for yield estimation of the main crops grown in Europe (corn, barley and wheat) under this new scenario. Subsequently, we study whether these variables are also able to capture potential disruptions on crop dynamics deriving from extreme events and their influence in final crop production. 

[1] Synergistic Integration of Optical and Microwave Satellite Data for Crop Yield Estimation. Anna Mateo-Sanchis, Maria Piles, Jordi Muñoz-Marí, Jose E. Adsuara, Adrián Pérez-Suay and Gustau Camps-Valls. Remote Sensing of Environment 234:111460, 2019.