How CropHeat AI calculates risk
FortyGuard Environmental Intelligence
Real hyperlocal heatmap data: snapshot temperature (tcm), exceedance hours, and persistence hours, plus point-level environmental parameters (apparent temperature, wet-bulb, humidity).
Heat Exposure
How many hours a field's tiles spent above the crop's growth-stage reference threshold, from FortyGuard's exceedance analytic.
Persistence
The longest CONTINUOUS run of hours above threshold, from FortyGuard's persistence analytic — continuous exposure without a recovery window is weighted more heavily than the same total hours spread across multiple short episodes.
Crop Sensitivity
A static, sourced rating (low/moderate/high/critical) per crop, drawn from peer-reviewed agronomy research — see citations below.
Growth Stage
The crop's current phenological stage (e.g. flowering, silking). Every cited study identifies growth-stage TIMING as the dominant factor in whether a given temperature actually causes damage, so this carries the highest weight in the engine.
Risk Aggregation
A transparent, weighted sum of the six components above (weights documented in config/risk_config.yaml as engineering defaults, not fitted ML coefficients — no labeled crop-damage dataset exists to fit against).
AI Advisory
An LLM is given the ALREADY-COMPUTED risk score and component breakdown as structured input, and asked only to explain it in plain language and suggest actions. It never recalculates or overrides the score.
Honesty notes
- This is an explainable hybrid risk engine, not a trained ML model. No labeled crop-heat-damage dataset exists that's accessible for this hackathon — see Section 4 of the original spec.
- The 0–24 / 25–49 / 50–74 / 75–100 risk bands are CropHeat engineering defaults for communicating relative severity, not a published agronomic standard.
- env_params is only fetched for the top-ranked fields by preliminary heatmap-derived risk, to conserve the hackathon's fixed FortyGuard credit budget — not because the data isn't wanted for every field.
Crop heat-stress citations
Wheat
Girousse et al., Field Crops Research 316 (2024) 109489
Maize
Djalovic et al., The Plant Genome (2024)
Rice
Jagadish et al., J. Exp. Bot. 58(7):1627 (2007)
Cotton
Oosterhuis & Snider, J. Cotton Research (2023)