Advances in technology and analytics to support data-driven agriculture has important implications for global food security and environmental sustainability. However, relatively few studies have investigated the potential to leverage the power of on-farm data for improved agronomy at scale using geospatial machine learning methods. Working in high-yielding rice systems of Uruguay, we developed a geospatial framework to identify yield-limiting factors across 55,000 ha annually of cropland over four seasons (2018–2021 harvest years), while also testing for tradeoffs in the environmental footprint related to nitrogen (N) fertilizer use. Our application of geographically-weighted random forest models showed that crop management decisions influenced rice yield more than variation in soil properties, highlighting the potential for improved agronomy to boost crop production by 1.4–1.8 Mg ha−1 across regions. Seeding date, variety, P rate, and K rate were the most important variables controlling yield, but with significant variation across fields. When these factors were optimized by farmers, the risk of environmental N losses or soil N mining did not increase, highlighting the potential for sustainable intensification by improving N use efficiency. These findings present a pathway for harnessing the benefits of increasingly available on-farm data to identify yield-limiting factors while minimizing negative environmental externalities at the field-level. To enable the development of such geospatial frameworks in other regions, new partnerships are required to engage stakeholders and promote data sharing and collaboration among farmers, researchers, and industry, helping guide regional extension programs and orient future investments in agricultural research.
Read full abstract