Methodology

For a point you select, WaterAgent checks which evidence exists in the datasets it has integrated and reports it with its source, date and resolution. It evaluates how much is known about a location. It does not evaluate whether water is present.

No single dataset describes groundwater at a site. Satellite gravimetry shows storage change over very large regions. Rainfall and terrain describe how water arrives and moves at the surface. Geological maps describe the rocks that may hold water. Only wells, monitoring points and field surveys observe the subsurface directly, and only where they exist.

A regional trend can say that water storage across a basin has been falling for years. It cannot say whether a well at one parcel will find water. WaterAgent keeps regional and local evidence separate and never turns a regional product into a site prediction.

Every evidence item belongs to one family. A family counts as available when at least one item of that family is returned for the location.

• RegionalRegional water-storage trend
• RegionalClimate and rainfall
• RegionalTerrain
• LocalGeology and aquifer
• LocalWells and monitoring points

The rules are applied in order and the first match wins.

1. Out of area — the point is outside the pilot boundary. No assessment is generated.
2. No evidence coverage — no evidence family is available.
3. Broad evidence coverage — both local families (geology and aquifer; wells and monitoring points) and at least one regional family are available.
4. Partial evidence coverage — anything else: some evidence exists but local evidence is missing or incomplete.

A well logged by the user counts toward the wells and monitoring family, labelled “User-reported, unverified”. The search radius and the consistency-check thresholds are provisional — pending hydrogeologist review.

Regional aquifer maps (about 5 km cells, 1:5,000,000) count toward the geology and aquifer family, labelled “Modeled (regional, ~5 km / 1:5M)”. When a family rests only on them, or only on user-reported wells, the status text states it.

These are coverage labels. WaterAgent does not compute a groundwater score, a probability of success, a depth or a yield, because no validated model exists to support one.

Next to the coverage label, each evidence family carries a note on how its evidence was obtained. The rule is fixed: a family takes the label of the most direct evidence available for it — measured before observed, observed before modeled.

Measured (in-situ)

Measured in the ground at a known point: a well logged by the user. Such wells are user-reported and unverified.

Observed (satellite)

Observed by satellite and processed into a map: elevation, slope derived from it, and land cover.

Modeled

Estimated by a model or a blended product: rainfall (CHIRPS), topsoil texture (SoilGrids) and total water storage (GRACE/GRACE-FO mascons).

Modeled (regional, ~5 km / 1:5M)

A class taken from a continental map: the BGS depth-to-groundwater and productivity grids (cells of about 5 km) and the Africa Groundwater Atlas hydrogeology map (1:5,000,000). It describes a region and cannot describe a single parcel.

Demo

Synthetic example value, loaded in demo mode only.

Missing

Nothing is loaded for this input.

These notes describe the kind of evidence. They are not weights and are not combined into a score.

The app runs in one of two modes and never combines them. Real mode loads only layers downloaded from their source by scripts, clipped to the test area and stored with their source, licence and download date, plus any wells logged by the user; a layer that has not been downloaded is reported as missing. Demo mode loads only a synthetic dataset so the workflow can be shown, and every screen carries a demo banner.

Review status: not yet reviewed

No hydrogeologist has reviewed the method, the coverage rules or the report wording yet. When a review is completed, the name and qualification of the reviewer and the review date will be listed here.