Valley Citrus & Agriculture
Yield Forecasting & Freeze Risk Analytics for Rio Grande Valley Growers
$890K
Yield Improvement
cooperative-wide
42%
Water Waste Reduction
irrigation
11 days
Freeze Warning
lead time
01The Challenge
A cooperative of five citrus growers in the Rio Grande Valley was losing crops and money to unpredictable freezes and inefficient water use. The 2021 Winter Storm Uri wiped out 60% of the harvest, and two years later, another late freeze caught growers off-guard again.
Three problems drove the engagement:
- No freeze early warning: Growers relied on NWS freeze warnings, which arrive 24–48 hours before the event — not enough time to deploy smudge pots, run wind machines, or apply micro-sprinkler frost protection across hundreds of acres.
- Water waste: Irrigation schedules varied 2x across growers, with some applying 42+ acre-inches per season when 24 was sufficient. No one knew what "optimal" looked like for their soil and microclimate.
- Yield blindness: The co-op tracked total harvest tonnage but couldn't attribute yield differences to weather, irrigation, variety, or grower practices. Good years were luck; bad years were weather — no one questioned the assumption.
Data Landscape
02Our Approach
We built three interlocking analytics: a yield prediction model that explains which factors drive harvest outcomes, a freeze risk system that extends warning time from 2 to 11 days, and an irrigation optimization framework that benchmarks water efficiency across growers.
- Random Forest + SHAP — yield prediction model with explainable feature importance showing which factors drive harvest outcomes
- Survival Analysis — freeze risk probability curves modeling the likelihood and severity of frost events across the winter season
- Irrigation Optimization — efficiency frontier analysis identifying growers who achieve high yields with less water
- Grower Benchmarking — multi-dimensional performance comparison across yield, water use, grade quality, and cost metrics
- Prophet + Pandas — seasonal yield forecasting pipeline and co-op planning dashboard
Yield Records
1,440+ observations
Weather Data
3,650 daily records
RF Yield Model
R² = 0.87
Freeze Risk
11-day advance warning
Co-op Dashboard
Grower planning tool
03Key Findings
Yield vs Minimum Temperature
Each dot is a crop observation during harvest/freeze season (Nov–Mar). Below 0°C, yields collapse — grapefruit and oranges are most vulnerable. The data clearly shows 3°C as the 'stress threshold' where yields begin declining even without a hard freeze.
Irrigation Efficiency Frontier
Water usage vs yield by grower. Ramirez Family achieves the highest yields per acre-inch of water — sitting on the efficiency frontier. Rio Citrus Co-op uses the most water per unit yield, indicating significant optimization opportunity.
Freeze Risk Probability Curves
Probability of freeze events by date (Day 1 = Nov 1). Three severity levels shown: light freeze (<0°C), hard freeze (<-3°C), and severe freeze (<-6°C). Peak risk window is mid-December through late January — the 11-day forecast model enables proactive frost protection deployment.
Grower Performance Comparison
Five-axis radar comparing growers on yield per acre, water efficiency, Grade A percentage, cost per ton (inverted — higher is better), and harvest timing optimization. Ramirez Family leads overall; Rio Citrus Co-op has the biggest improvement opportunity in water efficiency.
04Business Impact
Projected Annual Value
$890K yield improvement across the cooperative
The yield model identified minimum temperature, irrigation timing, and soil moisture as the three strongest predictors — together explaining 72% of yield variance. Growers who followed the model's irrigation recommendations saw a 15% average yield increase in the first season.
The freeze risk model correctly predicted the February 2024 cold snap 11 days in advance, giving growers time to deploy frost protection on 340 acres of high-value grapefruit. Estimated loss prevention: $420K in a single event.
Water efficiency benchmarking showed that adopting Ramirez Family's micro-drip practices could cut co-op water usage by 42% without yield loss — saving an estimated $180K in annual water costs and improving sustainability compliance.
05Technical Details
Yield Model (Random Forest)
- Features: min_temp, max_temp, precip, irrigation, soil_moisture, GDD, variety
- Target: yield_boxes_per_acre (regression)
- Performance: R² = 0.87, RMSE = 28 boxes/acre on holdout
- Explainability: SHAP values for per-prediction feature attribution
Freeze Risk (Survival Analysis)
- Method: Cox proportional hazards with time-varying weather covariates
- Horizon: 11-day rolling probability forecast
- Calibration: Brier score = 0.08 on 3-year holdout
- Alert thresholds: >30% hard freeze probability triggers grower notification
Irrigation Optimization
- Method: efficiency frontier analysis (DEA-inspired)
- Inputs: water_acre_inches, labor_hours, fertilizer_cost
- Output: yield_boxes, grade_a_percentage
- Benchmark: top-quartile grower practices as target efficiency
Facing similar challenges?
Let's discuss how data science can drive results for your business.