This project explores methods to forecast daily log‑returns of U.S. agricultural futures (wheat, soybeans, corn, sugarcane, oranges) by integrating market and exogenous data. We constructed baseline models—including linear regression, ARIMA, vector autoregression and gradient‑boosting—using historical prices, volumes, contract‑cycle, and derived features. To capture agronomic influences, we incorporated geospatial weather metrics (temperature, precipitation and extremes from NOAA’s ACIS) and USDA crop‑yield reports, carefully aligning and lagging variables for growing‑season and storage effects. Time‑series diagnostics (e.g. sliding window PACF) guided feature design. Future work will investigate sentiment analysis of national and global news and lagged weather data to enhance signal extraction.