XGBoost Precipitation Forecasting
Hackathon-winning ML solution for agricultural precipitation prediction
XGBoost Precipitation Forecasting (2023)
Winner: Ag Aid Hackathon, Water Challenge, Oregon State University
Developed a machine learning solution to forecast cool-season precipitation in the Sacramento Valley to help farmers prepare crops effectively.
Key Results
- 75% accuracy on temporal data prediction
- 84% accuracy on spatial data prediction
- Enabled actionable agricultural planning insights
Technical Approach
- Leveraged 73 years of precipitation data, climate indices, and 20 years of spatial data
- Features included soil moisture, ice extent, and sea temperature data
- Applied Dynamic Time Warping for temporal pattern matching
- Used spatial aggregation and correlation analysis for feature engineering
- Built XGBoost regression model combined with classifier
Challenges Overcome
- Coastal range rain shadow effects
- Pacific current influences on regional weather
- Inherently chaotic nature of precipitation patterns
The solution provides farmers with reliable forecasts to make informed decisions about crop preparation and water resource management.