A pastoralist region spanning the Ethiopia-Somalia border area is chronically exposed to drought, and the NGO supporting livestock and food-security programs there needed an earlier, more objective signal of emerging vegetation stress than its existing generic-threshold approach could provide. Field teams were often the first to notice trouble, by which point livestock body condition had already begun to decline.
GeoVeris built a satellite-based monitoring workflow combining MODIS vegetation index data with rainfall anomaly records, computing a rolling Vegetation Condition Index and Standardized Precipitation Index for 18 administrative units. Rather than applying generic literature threshold values, we calibrated the stress thresholds against three seasons of the NGO's own field-reported pasture and livestock condition data, so the alerts reflected conditions on the ground rather than a one-size-fits-all standard.
The calibrated system flagged severe vegetation stress in five administrative units five to seven weeks ahead of the corresponding field-reported livestock decline — a meaningful head start for an organization whose response programs take time to mobilize. Monthly bulletins were delivered through three consecutive growing seasons, each combining the satellite indicators with a plain-language severity summary for program staff.
The monitoring workflow and calibrated threshold set have since been handed over to the NGO's own field-GIS staff for continued in-house operation.
Methodology
Built a MODIS NDVI (MOD13Q1, 250 m, 16-day composite) and CHIRPS rainfall anomaly time series (2001-2023 baseline) processed in Google Earth Engine to compute a rolling Vegetation Condition Index (VCI) and Standardized Precipitation Index (SPI-3) for 18 administrative units. Cross-validated VCI-flagged drought severity against three seasons of field-reported pasture and livestock body-condition scores supplied by partner field teams, then set threshold bands (VCI below 35 = moderate stress, below 20 = severe) calibrated to those field observations rather than generic literature defaults. Delivered outputs as a monthly bulletin plus a shapefile/GeoTIFF package for the NGO's own GIS staff.
Results & Outcomes
- Delivered monthly drought-severity bulletins across 18 administrative units through three consecutive growing seasons
- Flagged severe vegetation stress in 5 units roughly 5-7 weeks ahead of field-reported livestock condition decline
- Calibrated VCI stress thresholds against field data, improving alert precision over the NGO's prior generic-threshold approach
- Monitoring workflow and threshold set handed over for continued in-house operation by NGO field-GIS staff
Challenges
Persistent cloud and smoke contamination during the short rains season degraded a meaningful share of raw MODIS NDVI composites, risking false drought signals in the affected months. We applied a Savitzky-Golay temporal smoothing filter combined with a quality-flag-based compositing step to reconstruct a cleaner seasonal NDVI curve, validated against ground-reported conditions — an earlier unfiltered run had flagged a false severe-stress alert in 2 units during a heavily clouded month.