TerraCast

In the UW–Databricks Hackathon, we built a Weather-to-Yield Signal Detection framework by integrating county-level annual crop yields with daily NOAA weather data on Databricks. We leveraged the full Databricks ecosystem—from large-scale data ingestion and transformation to model training and analysis—while focusing heavily on building a robust weather-driven prediction model. By engineering stress-based climate features and detecting anomalies where actual yields deviated from weather expectations, we identified regions where non-weather factors likely influenced production outcomes.