Current Limitations & Ongoing Refinements

Crop Risk is an early-stage product. This article is transparent about the current known limitations and what we are working on.

Current limitations

• Data lag - ERA5-Land data arrives with a 5 to 7 day lag, so risk scores reflect conditions from the past week rather than today. Rapidly developing events may not yet appear in the signal. We are evaluating near-real-time data sources to close this gap.
• Thresholds still being calibrated - Warning and Critical thresholds are grounded in agricultural science but have not all been empirically validated against yield data. Excessive Rainfall thresholds in particular are being recalibrated as they are likely too sensitive in the current version. Trigger frequency analyses are running continuously to identify and correct miscalibrated thresholds.
• Frost irreversibility not yet modelled - The current model treats frost like a recoverable stress event: if conditions warm up after a frost, the score can ease. In practice, a Critical frost alert during a sensitive phase - such as flowering or fruit set - causes permanent within-season yield damage that warmer subsequent weather cannot undo. Until a score floor is introduced for these events, be cautious about interpreting an improving trend after a critical frost spike.
• Equal hazard weighting - All four hazards currently contribute equally to the Season Risk score, and alert counts carry the same weight regardless of when in the season they occur. In reality, some hazards and some phases drive yield loss more than others. Empirically derived hazard weights and phase multipliers are on the roadmap.
• Irrigated regions - Weather-based soil moisture does not model irrigation. For irrigated crops, the drought signal reflects background climatic conditions rather than field-level water availability. This is flagged in the relevant crop views.
• Spatial resolution - The platform operates at 9 to 25 km resolution. In topographically complex areas, a single grid cell may average over diverse microclimates. Location-specific historical baselines partially compensate for this.
• Data coverage in some tropical regions - In parts of West and East Africa, sparse ground-station networks reduce absolute accuracy of precipitation and soil moisture. Relative anomalies (how conditions compare to local history) remain more robust than absolute values.

What is coming

• Near-real-time weather data to reduce the data lag
• Improved risk aggregation accounting for phase sensitivity and frost irreversibility
• Price and market intelligence - we are actively building the link between weather risk and commodity price impact, so that the platform can surface not just where production is at risk but what that means for prices. This is a top priority based on user feedback.
• Digest-style alerts - automated summaries delivered to your inbox when risk thresholds are crossed
• Expanded crop and region coverage

If you have feedback on specific crops, locations, or threshold behaviour, please reach out to the Nala team.