This is a screening hypothesis for the 4 × 4 km pilot area, not a measurement of underground water. The combined score draws on terrain shape and repeated Sentinel-2 optical observations. Its assumptions have not been calibrated against local wells or springs.
1. Terrain clue
Mapzen/AWS Terrarium elevation tiles provide a regional elevation surface at roughly 30 m scale. From it the map calculates ground slope and whether a cell sits below nearby terrain. Gentler, lower-lying places receive a higher terrain clue. Such places can sometimes collect surface water, but groundwater in karst and fractured rock need not follow the surface.
2. Satellite observations
Additional summer and autumn Copernicus Sentinel-2 Level-2A scenes were accessed through the public Element 84 Earth Search catalog. Summer scenes cover July–September 2023–2026; the usually wetter autumn comparison uses October–November 2023–2025. Sentinel-2's red and broad near-infrared bands used here resolve about 10 m, while its narrow near-infrared and shortwave infrared bands resolve about 20 m. The satellite's classification removes areas covered by clouds or shadows and excludes visible water, retaining vegetation and bare ground for this land analysis. The observations are placed on the map's roughly 30 m grid.
NDMI = (B8A − B11) / (B8A + B11). B8A is narrow near-infrared light; B11 is shortwave infrared. Their difference is sensitive mainly to vegetation water content and exposed surface properties. It does not sense buried groundwater directly.
NDVI = (B08 − B04) / (B08 + B04). B08 is near-infrared and B04 is red. NDVI indicates vegetation greenness. The script compares NDMI among broad groups of similar NDVI so dense forest does not automatically outrank sparse grass just because it is greener.
For each vegetation group, a relatively high summer NDMI threshold is derived from cells in the pilot area. Each year is summarized by its median available scene; the optical signal indicates how often a cell exceeds that threshold across observed years. A cell needs at least seven clear summer observations across at least three years to receive a combined score. The displayed summer NDVI is a median, not a measured drought anomaly. NDMI is a relative index, not a calibrated percentage of water in plants.
To compare seasons, the map pairs summer and autumn observations from the same calendar years, computes a median NDMI for each season and shows autumn minus summer for places with at least two paired years. This difference is a separate observation; it does not increase or decrease the combined score. October–November is usually wetter in the region, but the image dates were not certified as rainy at this parcel, and leaves, temperature, illumination and shadows can change NDMI. Seasonal index values should not be mistaken for underground water changes.
3. Other evidence and validation
Mapped streams and a spring come from OpenStreetMap and are shown as reference features; they do not enter the score. NASA POWER supplies modeled July–August rainfall context for the central point; rainfall also does not enter the score. A single mapped spring cannot validate the ranking, and there are no local well records, dry-well comparisons or geology inputs.
What the result can and cannot say
A higher score means these clues suggest underground water might be more likely at that spot than elsewhere in this pilot area. It is not a measured probability, a depth estimate or a predicted well yield. Optical shadows, forest cover, irrigation, seasonal weather and karst flow may mislead the ranking. Known flowing water in forest areas was missed by this map, so its scores should not be used to choose a drilling location.
For reproducibility, download the satellite processing notes, terrain metadata and coverage and spring check. See also the official Sentinel-2 band descriptions.