Using climate variability for regional crop yield prediction with machine learning: a case study on sunflower, corn, and wheat in the Argentine pampas

Abstract

This study developed and evaluated yield forecasting models for sunflower, corn and wheat in the Argentine Pampas, using weather station variables and teleconnection indices as predictors. Three modeling approaches -multiple linear regression (MLR), multinomial regression (MLogR), and decision trees- were assessed, along with an ensemble model combining predictions from the primary models. Regionalization via K-means clustering identified coherent productive zones for each crop: three north-to-south clusters for sunflower, four clusters for corn, and two for wheat, reflecting different alignment with regional phenological stages. Results indicated that MLogR and MLR were the most effective models, with no clear predictors preference between weather station variables and teleconnection indices. The best predictions for sunflower’s central and southern clusters were attained with station-based models, highlighting the importance of winter and spring evapotranspiration and precipitation. Conversely, the northern cluster relied on teleconnection indices, including IOD and ENSO indicators, which influence rainfall and temperature patterns. The corn yield was effectively forecasted using teleconnection indices in the southern cluster and station-based models in the eastern and western clusters, emphasizing the role of soil moisture and temperature during critical growth phases. Lastly, wheat yield predictions varied by region, with teleconnection indices and station data capturing essential pre-sowing and flowering conditions. These findings underscore the importance of leveraging climate variability knowledge to develop regionalized models tailored to specific crop and environmental characteristics. MLogR and MLR, particularly with careful feature selection, offer robust tools for operational yield forecasting, enhancing agricultural decision-making in the Argentine Pampas.