How It Works
This article gives a plain-language overview of how Crop Risk produces its risk scores. For the full technical detail, see the dedicated articles on Hazard Alerts & Crop Phase Thresholds and Season Risk Scores & Aggregation.
The core logic in three steps
Step 1 - Monitor for hazard alerts. For each crop, region, and growth phase, the platform defines weather thresholds that represent the onset of stress. Each day, actual weather conditions at each monitored location are compared against these thresholds. When conditions cross a threshold, a hazard alert is generated - classified as Warning or Critical depending on severity. One alert can be generated per hazard per day per location.
Step 2 - Compare to history. A raw count of alerts does not tell you much on its own. Fifteen drought alerts in a season could be normal or exceptional depending on the crop and region. Alerts accumulate as a weighted total - Critical alerts count double (2 points) versus Warning alerts (1 point) - so a season with severe stress scores higher than one with marginal threshold crossings. To make the signal interpretable, this weighted total is compared against the same metric across the 30-year historical baseline, using ERA5-Land reanalysis data. This produces a percentile score: a score of 80 means the current season's weighted alert total exceeds 80% of all historical seasons to the same point in the year.
Step 3 - Aggregate into a risk score. Scores are combined in two stages. First, at the location level: each location's hazard scores are aggregated into a single Location Season Risk Score using a statistical method called a Gumbel copula, which is sensitive to co-occurring extremes. A season where drought and heat stress are both elevated simultaneously is treated as riskier than a season where only one stressor is elevated - because the combination itself is historically rare. Second, at the region level: each location's score is ranked against its own historical distribution, and these normalised ranks are averaged across all monitored locations to produce the overall Season Risk Score shown on the dashboard.
The result is a score that is interpretable ("this season is in the top 10% historically"), comparable across crops and regions, and updated daily as new weather data arrives.