The Technology Behind AI Groundwater Detection

Updated July 2026 · ~8 min read

"AI finds water" explains nothing. Here is the actual data stack — which layers are read, what each one contributes, and where the uncertainty lives.

Layer 1 — Multi-spectral satellite imagery

Vegetation is a long-running moisture sensor. NDVI (near-infrared against red) measures plant vigour, and NDWI (near-infrared against green or shortwave infrared) responds to water content in leaves and soil. Persistent greenness through a dry season is a strong hint of accessible water below.

Layer 2 — Digital elevation and terrain

A global elevation model gives slope, curvature and depressions. Water infiltrates where terrain slows it down: valley floors, footslopes and closed depressions recharge, steep convex slopes shed.

Layer 3 — Drainage and catchment

Flow direction and accumulation build the stream network and show how large an area drains through a given point. A point receiving a big catchment has far more opportunity to recharge than a ridge top.

Layer 4 — Geology and lithology

Rock type sets the storage mechanism. Alluvium and sandstone hold water in pores and are forgiving to drill. Granite and basalt hold it only in fractures, so being fifty metres off can mean the difference between a good bore and a dry one. Limestone adds karst, which is productive and unpredictable.

Layer 5 — Lineaments and fractures

Lineaments are linear features traced from imagery that often correspond to faults and fracture zones. Their intersections are the classic high-yield targets in hard-rock terrain, and this is where remote sensing adds the most value over guesswork.

The model on top

The layers are normalised and weighted into a single groundwater probability score for the point, plus an estimated depth band, an indication of likely water quality and an optimised drill location with a survey direction. Weighting is geology-aware: fracture evidence matters more in basalt than in alluvium.

Where the uncertainty is — stated plainly

• Remote sensing infers from the surface; it cannot see a specific fracture at 90 metres.
• Depth bands are ranges, not promises.
• Recent abstraction by neighbouring bores is invisible to satellites.
• Hard rock is inherently less predictable than thick alluvium.

This is why the app recommends a resistivity survey before a high-value bore rather than claiming certainty.

Why it can be free

The heavy inputs — satellite archives, elevation models and public geology — are open data. Once the pipeline exists, an extra scan costs almost nothing, so there is no reason to charge for it.

JalDrishti is free worldwide: unlimited groundwater scans, reports, maps and a bore calculator, with voice guidance in 33 languages. You only ever pay if you choose an optional certified-geologist review.