Irrigation and yield forecasting agent for a US produce grower
Row-crop produce grower, Central Valley, California
- Challenge
- Irrigation scheduling across 3,100 acres was set on a fixed calendar rather than actual soil and weather conditions, so fields were often over- or under-watered. Yield forecasts for the packing house were built from a grower's experience and a spreadsheet, and were frequently off by enough to disrupt harvest labour planning.
- Solution
- We built an agent that pulls soil-moisture sensor data, weather forecasts and crop-stage information per field block and recommends an irrigation schedule, with a human agronomist approving changes before they reach the controllers. A separate forecasting model updates yield estimates weekly against the same data.
- Results
- Water usage down 18% with no yield loss across the season
- Yield forecast error narrowed from ±22% to ±7% by mid-season
- Irrigation schedule now recalculated daily instead of set once a month
