Aquanium Concept demonstrator

Where Central Asia should look for groundwater.

A knowledge-driven index of relative groundwater favorability for the Fergana Valley — built from open satellite data, and honest about its limits. It maps where conditions are more favorable. It does not locate water.

8 factors · ~51,300 km² at 200 m · open data via Google Earth Engine

Relative groundwater-favorability index for the Fergana Valley — a teal-to-gold raster where gold marks higher relative favorability. Knowledge-driven AHP overlay of 8 hydrogeological factors; not a prediction of water, yield, or where to drill.
Relative favorability — teal lower, gold higher. Not located water.
Scroll

The opportunity

Central Asia is drying — and almost no one has mapped its groundwater favorability.

Across arid Central Asia, water stress is rising and surface supplies are shrinking, yet systematic, region-wide assessments of groundwater favorability are still almost absent — the first peer-reviewed study of its kind in the region appeared only in 2025 (Akbakay, Kazakhstan). Where assessments do exist, they are largely manual: a specialist hand-digitizes geological sheets and weighs factors by hand, one region at a time. That manual-mapping bottleneck is exactly what an open-data pipeline can relieve.

The Fergana paradox

The Fergana Valley is a paradox. It is the best-documented basin in Uzbekistan — and there, the problem is often not too little water but too much shallow water: waterlogging, rising water tables, and salinization. So a favorability map here is a perfect honesty test: a physically favorable zone can still hold saline or shallow, problem water. Favorability is not the same as good water — and our map says so.

2025
first peer-reviewed groundwater-favorability study for Central Asia — the niche is effectively open
~51,300 km²
Fergana Valley pilot, assessed at 200 m resolution from open satellite data

Read this first

What Aquanium is — and what it is not.

It is

  • A knowledge-driven, relative groundwater-favorability index — a weighted overlay of open satellite factors.
  • A concept demonstrator of a working pipeline: open data → factors → weighted map → web UI.
  • An honesty-first tool with a built-in robustness layer that shows where the map is stable and where it depends on our weight choices.
  • A method and platform — the proven Morfo Labs stack (from minerals) applied to groundwater, with an explicit roadmap to a validated model.

It is not

  • Not a yield or flow prediction, and not a “drill here” recommendation.
  • Not a water-quality or potability map — a favorable zone can still be saline or brackish.
  • Not a trained / ML / AI model — the weights come from published literature, not from data. It is an expert hypothesis, not an empirical forecast.
  • Not complete: the dominant control on this aquifer — lithology (the alluvial-aquifer geology) — is not yet included and is flagged throughout.

The method

Open satellite data → 8 factors → an AHP-weighted overlay → favorability + robustness.

Every heavy computation runs server-side in Google Earth Engine. We pull open satellite layers, derive eight hydrogeological factors, normalize each one by its hydrogeological meaning, and combine them as a weighted sum using starting weights drawn from the arid-region literature. Then — and this is the honest core — we stress-test the weights with a Monte-Carlo run (±10%) and a Fuzzy-Gamma aggregation, so the map can show where it is sure and where it is only guessing about weights.

  1. 01

    Open satellite data

    Pull open layers server-side in Google Earth Engine — no heavy computation in the browser.

  2. 02

    8 hydrogeological factors

    Derive each factor and normalize it by its hydrogeological meaning (categorical layers stay nearest-neighbour).

  3. 03

    AHP-weighted overlay

    Combine the factors as a weighted sum using starting weights drawn from the arid-region literature.

  4. 04

    Favorability + robustness

    Stress-test the weights with a Monte-Carlo run (±10%) and Fuzzy-Gamma — so the map shows where it is sure and where it only guesses about weights.

The 8 factors and their starting weights

Σ = 1.00
TWI (topographic wetness) 0.190

higher is more favorable

Drainage density 0.166

lower is more favorable

Slope 0.154

gentler is more favorable

Rainfall (recharge proxy) 0.154

higher is more favorable

Distance to drainage 0.119

closer is more favorable (Fergana is a river-fed valley)

Land cover 0.095

categorical

NDWI (wetness index) 0.071

higher is more favorable · flagged: can be confounded by irrigation

Lineament density 0.050

preliminary terrain-roughness proxy, not true geological lineaments

These are literature-informed starting weights, not locally calibrated values. A hydrogeologist must review and adapt them. In the live app you can drag every weight yourself and watch the map and the area-by-class breakdown recompute.

Relative groundwater-favorability surface for the Fergana Valley — the composited output of the 8-factor AHP weighted overlay. Teal lower, gold higher. A method visualization, not located water, not a prediction.
TWI (topographic wetness) 0.190
Drainage density 0.166
Slope 0.154
Rainfall (recharge proxy) 0.154
Distance to drainage 0.119
Land cover 0.095
NDWI (wetness index) 0.071
Lineament density 0.050
01 Open the satellite data. Copernicus DEM, Sentinel-2, CHIRPS, ESA WorldCover and HydroSHEDS — pulled server-side in Google Earth Engine. 02 Stack the 8 factors as weighted layers. Each is normalized by its hydrogeological meaning; the printed weight is its AHP starting weight (Σ = 1.00). 03 Blend them into one relative-favorability surface — a weighted sum, teal lower to gold higher. 04 This is a visualization of the AHP weighted-overlay methodnot “AI scanning the ground”, not water found in real time, and not a drilling or yield recommendation.

The visualization above shows the AHP weighted-overlay method building step by step: open data → 8 weighted factors → a composited relative-favorability surface. It is not “AI scanning the ground” and not water being found in real time. It produces a relative favorability index — not a drilling recommendation and not a yield estimate.

Not a mockup

Interactive, on real data — explore the Fergana pilot yourself.

The live app runs the full pipeline on real open satellite data. Toggle the favorability map, the robustness layer, and all 8 factors; open the Analysis dashboard where every chart is computed from the actual rasters (area by class, factor weights, sensitivity, Monte-Carlo robustness, factor ranges); drag the weights in the Calibration panel; click anywhere to inspect the factor values at that point. Aqtrik, the in-app explainer, answers questions about the method and declines anything about “where to drill” or “how much water.”

Aquanium map view — relative favorability raster with the 8-factor layer panel and legend; on-screen note states it is not a water or drilling prediction.
  • Aquanium dashboard — area-by-class breakdown, the 8 starting factor weights, and ±10% Monte-Carlo robustness (9.5% stable-high; robustness is not accuracy).
    Analysis dashboard
  • Aquanium calibration view — adjust the AHP factor weights and recompute the relative favorability overlay.
    Calibration
  • Aquanium point inspector — per-factor contribution breakdown for a clicked location; explains relative favorability, not a yield or depth.
    Point inspector
  • Aquanium with the Aqtrik assistant — ask questions about the factors and method in plain language.
    Aqtrik assistant
  • Aquanium data provenance — open satellite sources (Copernicus DEM, Sentinel-2, CHIRPS, ESA WorldCover, HydroSHEDS) via Google Earth Engine.
    Data provenance
  • Aqtrik explaining the AHP factor weights as a table inside Aquanium.
    Factor-weights table

We show where we’re unsure

Robustness is not accuracy — and we never blur the difference.

A smooth colored heatmap can look more certain than it is. So Aquanium ships a robustness layer built from a Monte-Carlo run over the weights. It answers one specific question — how stable is the map to our choice of weights — and explicitly not how well it matches the reality underground. Confusing those two questions is the one thing we refuse to do.

Robustness ramp — stable → contested (never “hot” / never accuracy)
The biggest caveat stays in plain sight: the dominant control on this aquifer — lithology — is not yet in the model. Lineaments stand in only partially. A hydrogeologist and real geological data are required to close that gap.

The roadmap

From honest concept to a validated model — in the open.

  1. 1 · Concept demonstrator

    Now

    AHP-weighted overlay + Monte-Carlo + Fuzzy Gamma on the Fergana pilot, topo-climatic factors, full web UI. No blockers — this is what you can use today.

  2. 2 · Data + hydrogeologist

    Next

    Bring in a hydrogeologist (advisor is enough to start) and, under NDA, regional lithology and historical wells from Goscomgeology / GIDROINGEO.

  3. 3 · Labeled training

    Planned

    A hydrogeologist cleans and labels the well inventory; factors become features; a statistical baseline (Frequency Ratio / Weights-of-Evidence) comes first.

  4. 4 · Validated supervised model

    Planned

    Only with real labels: a supervised model with spatial cross-validation and an honest, leakage-aware accuracy assessment. This is where machine learning genuinely enters.

  5. 5 · Quality & monitoring

    Future

    Water-quality / freshness mapping (needs hydrochemistry) and storage-monitoring, plus reports for drillers and agencies.

The critical dependency for everything beyond stage 1 is human and physical: a hydrogeologist and real well data. Without them, stages 3–5 have no scientific legitimacy — and we won’t pretend otherwise.

Who is building this

Morfo Labs — a proven platform, applied to water.

Morfo Labs is a Tashkent-based team and an IT Park resident. Aquanium reuses roughly 70% of the infrastructure we already debugged on Mineralium (mineral-frontier mapping) — Earth Engine ingestion, export caching, and a FastAPI dark-theme web stack — and applies that proven pipeline to groundwater. The model part of Mineralium is deliberately left out here: with no labeled wells, there is nothing to train on yet, and we say so.

  • Tashkent
  • IT Park resident
  • Built on the Mineralium platform
  • 2026

Let’s build the next stage together

The map gets real when the data and the expertise arrive.

Aquanium’s concept stage is done. The next stage needs partners. We are looking for three things, through formal channels and under NDA.

  • 01

    Data

    Regional lithology, historical wells (levels, mineralization), and rainfall calibration — via Goscomgeology, GIDROINGEO, Uzgidromet, and satellite products from Uzbekcosmos.

  • 02

    Expertise

    A hydrogeologist — advisor or partner — to review the rank tables and weights and to anchor the work scientifically, alongside academic collaborators in the region.

  • 03

    A second pilot

    Fergana is the active pilot built on real data. Samarkand exists as an early second demo region; with a partner’s data it can become a genuine second pilot.

Fergana is the only region built on real, validated open-data output today. Other regions are demonstrations or planned — we don’t overclaim them.