Methodology

Optical imagery from ESA's Sentinel-2 mission covers South Africa at 10 m resolution with a revisit of roughly five days. For each cloud-free acquisition over a monitored dam the platform computes the Normalized Difference Water Index — NDWI = (Green − NIR) / (Green + NIR). Water absorbs near-infrared strongly while reflecting green, so open water returns high positive values; thresholding the NDWI raster classifies each pixel as water or non-water and the classified pixel count times pixel area gives surface extent in hectares.

Each reading is compared to a long-run baseline extent for that dam. Within 10% of baseline reports as Normal, 10–25% below as Below Average, and more than 25% below as Critical. Because dams are strongly seasonal, trend direction is assessed year-on-year against the same month rather than against the preceding months.

Flood mapping uses Sentinel-1 C-band radar, which images through cloud and at night — essential during an active storm when optical sensors are blinded. Smooth standing water produces very low radar backscatter, so a same-scene before/after comparison isolates newly inundated area, which is then intersected with settlement and cropland layers to estimate exposure.

Fire danger combines MODIS/VIIRS active-fire detections with vegetation dryness and recent rainfall to produce a provincial danger rating, while burned-area assessment uses post-fire spectral change (NBR difference) to quantify hectares affected. Detections are approximately 375 m–1 km resolution, so small fires can be missed.

Drought classification blends a rainfall deficit index against the long-term mean with vegetation-condition anomalies, expressed as Normal, Watch, Warning or Emergency per province and catchment.

Crop condition is tracked with NDVI = (NIR − Red) / (NIR + Red) from Sentinel-2, aggregated per farming region and compared to the same point in previous seasons. Persistent negative anomalies through the critical growth window translate into a Below Average or Poor yield outlook, which is what lenders and insurers use as an early risk signal.

Crop recommendations combine regional rainfall outlook, soil-moisture proxies and upstream dam status with each crop's water demand and growing window; they are advisory only and do not account for local soil chemistry, input costs or market access.

Rangeland carrying capacity converts a vegetation biomass proxy into estimated large stock units per hectare for communal grazing areas, compared with registered livestock numbers to flag overstocking.