Weather-corrected Scoring

Weather-corrected scoring is the technique by which EcoIntel separates the management signal in satellite ecological data from the weather noise.

The problem it solves is the Weather Halo: raw satellite greenness swings by roughly fifty points a year on any piece of land, independent of management. Without correction, a bad weather year looks like a degraded farm and a good weather year looks like effective management. Whole policy debates about whether regenerative methods “work” have been conducted on data that is mostly recording whether it rained.

The fix is not to abandon the satellite. It is to integrate the climate signal into the prediction model directly, parcel-specific, week by week.

EcoIntel does this using:

• Daily climate data from ERA5-Land (Muñoz-Sabater et al. 2021)
• Stress-week counts derived from zone-specific temperature and precipitation optima
• Vapour pressure deficit as a drought-stress proxy
• The ratio of actual to potential evapotranspiration as a water-availability proxy

These covariates are integrated directly into the score-prediction model, rather than residualised after the fact. The output is a climate-detrended trajectory that reflects management decisions, not rainfall.

The approach is in the lineage of RESTREND (Evans & Geerken 2004) and TSS-RESTREND (Burrell, Evans & Liu 2017). It is reproducible from public Copernicus and ECMWF data, and independently verifiable by any third party willing to redo the analysis from scratch.

On sites where raw NDVI swings fifty points a year, the climate-detrended trajectory typically moves by a handful of points around a clear management signal.