Deep learning improves brazilian soybean yield forecasts using weather data
An Input-Frugal Deep Learning Framework for Weather-Driven National Crop-Yield Forecasting: A Case Study of Brazilian Soybean
Machine Learning
Summary
Accurate forecasts of crop yields help farmers and markets plan better, but usually need expensive or complex data. The authors developed a simple deep learning method that only uses regular weather information plus two easy-to-get details about the crop year and growing environment. They tested this on 20 years of Brazilian soybean data and found it beats traditional methods in accuracy, especially later in the growing season. Their approach is easy to use in other regions and crops and works well with weather forecasts during the season.
What this means in practice
- •For agricultural monitoring teams: Use weather-only deep learning models to generate more accurate and timely soybean yield forecasts for national agricultural monitoring.
- •For crop insurance underwriters: Incorporate improved yield predictions based on routine weather inputs to better assess risk and set premiums in soybean insurance.
Tested on one dataset.
Authors
Fernando Dupin da Cunha Mello, Prashant Kumar, Erick G. Sperandio Nascimento
Abstract
Reliable, timely crop-yield forecasts are essential for market stability and risk management, yet many approaches rely on costly or hard-to-scale inputs. We present a frugal, transferable, and architecture-agnostic deep learning framework that uses routine weather as the only time-varying input plus two lightweight static context inputs (crop year and an agro-environmental label) to capture long-run change and regional heterogeneity, while supporting multiple sequence encoders under identical data requirements. Using a 20-season Brazilian soybean case study (2001/02-2020/21) with leave-one-year-out cross-validation, we benchmark MLP, CNN, LSTM, CNN-LSTM, a Transformer encoder and the Mamba state-space model against linear ridge regression and a five-year moving-average "farmer" baseline. All deep learning variants outperform ridge, and all sequential encoders surpass the non-sequential MLP. The Transformer achieves the best national accuracy (RMSE 149 kg ha^-1; rRMSE 5.3%; R^2 = 0.784), reducing error by 47.6% relative to the farmer baseline. In-season forecasts improve monotonically from early- to late-season issuance, reaching approximately 50% lower error than the baseline at the latest forecast point. Ablations indicate that the agro-environmental label and spatial instance expansion (multiple grid-node weather sequences per municipality-year) contribute positively without increasing input complexity. SHAP diagnostics suggest crop year explains most of the long-run trajectory, whereas within-season weather and agro-environmental context primarily drive interannual deviations, with moisture/cloud and thermal-demand variables dominating. Overall, the framework is straightforward to deploy across other crops and geographic regions and is naturally compatible with operational weather forecasts for routine monitoring.