Groundwater recharge zone identification using multi-source satellite data
Combining Sentinel-1 backscatter, Sentinel-2 land cover, TanDEM-X terrain and GRACE-FO gravity trends narrows the search for productive recharge zones. Satellite data constrains the hypothesis; borehole validation closes it.
Sensors
- GRACE-FO: Twin satellites measure inter-satellite range changes to millimetre precision, resolving terrestrial water storage anomalies at roughly 300 km spatial scale with monthly revisit. Useful for detecting multi-year recharge or depletion trends at aquifer scale, not for locating individual recharge corridors.
- Sentinel-1 SAR (C-band, 5.405 GHz): Interferometric Wide Swath mode delivers 10 m ground range resolution with 6-day revisit at mid-latitudes (12 days near the equator). Backscatter change after rainfall events reveals bare-soil moisture anomalies that proxy infiltration opportunity. Vegetation and surface roughness create ambiguity that must be disentangled.
- Sentinel-2 MSI: 13 spectral bands at 10-20 m resolution, 5-day revisit with both satellites. Used to classify land cover and derive permeability proxies: bare rock, alluvial sand, clay-rich soils and urban surfaces each carry different infiltration capacity assumptions. Cloud cover limits usable acquisitions in humid climates.
- TanDEM-X: Global DEM at 12 m posting (commercial) or 90 m (public release). Topographic wetness index and flow-accumulation grids derived from TanDEM-X identify convergent terrain where surface water concentrates. Vertical accuracy is typically better than 2 m in open terrain, degrading in dense forest or steep slopes.
What the surface actually reveals about what lies beneath
Recharge is a subsurface process. A satellite cannot see it directly. What satellites observe are the surface and near-surface conditions that make recharge more or less probable: topographic convergence, permeable soils, low evapotranspiration demand, and post-rainfall moisture retention. The inference from those proxies to actual aquifer connectivity is probabilistic, not deterministic.
This is not a weakness to apologise for. It is the honest framing that makes the analysis useful. A well-constructed multi-source satellite product reduces the area a hydrogeological survey must sample by ground methods, sometimes by an order of magnitude. That is the value proposition: cheaper, faster triage, not remote confirmation of subsurface geology.
Four data layers and what each one contributes
Sentinel-1 backscatter change is the most direct surface-moisture signal available at useful resolution. After a rainfall event, bare sandy soils show a measurable increase in C-band backscatter (typically 2-5 dB depending on soil texture and roughness) that decays over days as the surface dries. Persistent anomalies relative to a multi-year baseline, particularly in areas with low vegetation cover, flag locations where water is infiltrating rather than running off. The method breaks down under dense canopy, where the vegetation volume scattering dominates, and in areas of surface roughness change such as tilled agricultural fields.
Sentinel-2 land-cover classification assigns permeability classes to each pixel: coarse alluvial sediment, fractured limestone outcrop, clay pan, impervious urban surface and so on. These classes are not measured permeabilities; they are proxies drawn from spectral signatures and validated against published soil surveys. The NDVI and bare-soil indices derived from Sentinel-2 also constrain the Sentinel-1 interpretation by separating moisture-related backscatter change from vegetation-related change.
TanDEM-X provides the topographic skeleton. The topographic wetness index (TWI), computed as the natural log of upslope contributing area divided by local slope, identifies hollows and valley bottoms where water concentrates. High TWI cells that also show permeable land cover and post-rainfall Sentinel-1 anomalies are the strongest candidates. TWI is sensitive to DEM resolution and to the flow-routing algorithm chosen, so results from the 12 m and 90 m products can differ meaningfully in low-relief terrain.
GRACE-FO operates at a completely different scale. Monthly terrestrial water storage anomalies at the 300 km footprint cannot resolve individual recharge corridors, but they can confirm whether an aquifer system is gaining or losing water over seasonal to decadal periods. A basin where the three surface-layer indicators point to active recharge but GRACE-FO shows persistent storage decline warrants immediate scrutiny: either the recharge signal is smaller than the extraction signal, or the surface proxies are misleading.
Building the composite recharge-potential index
The standard analytical approach stacks the four layers into a weighted overlay or, in more rigorous implementations, a logistic regression or random-forest model trained against known recharge points from borehole records. Weights are not universal. In a fractured-basement aquifer in sub-Saharan Africa, lineament density derived from Sentinel-1 texture may outweigh TWI. In a sedimentary basin, soil permeability class may dominate. The model must be tuned to the hydrogeological context.
Published studies using similar multi-criteria approaches, including work documented in the MDPI Remote Sensing journal, typically report that 60-80 per cent of high-scoring pixels in the composite index correspond to locations with positive borehole yields when validated. That is useful but not conclusive. The remaining 20-40 per cent represent false positives driven by surface conditions that do not connect to a productive aquifer. Ground-truth sampling cannot be skipped.
The cloud problem and the archive solution
Humid tropical regions, where recharge rates are often highest, are also where cloud cover most severely limits Sentinel-2 optical acquisitions. In parts of West Africa or Southeast Asia, fewer than 20 clear-sky observations per year may be available in the wet season, precisely when soil moisture signals are most informative. The practical response is to build composites from multi-year archives rather than single-season data, accepting that the land-cover classification reflects a time-averaged state rather than a current snapshot.
Sentinel-1 SAR is cloud-transparent, which is why it carries disproportionate weight in the moisture-anomaly layer. The ESA archive extends back to 2014 for Sentinel-1A, giving a decade of backscatter time series for baseline construction. GRACE and its predecessor GRACE (2002-2017) together provide a 20-year gravity record, long enough to separate seasonal cycles from structural trends.
Validation is not optional: what borehole records must confirm
Satellite-derived recharge potential maps carry three irreducible ambiguities. First, surface permeability does not guarantee subsurface connectivity: a permeable sandy soil overlying an impermeable clay layer contributes nothing to a deep aquifer. Second, topographic convergence identifies where surface water collects, not where it infiltrates; a sealed hardpan can sit beneath an apparently ideal convergence zone. Third, GRACE-FO storage trends integrate all water compartments, including soil moisture and surface water, not just groundwater.
Borehole validation addresses all three. Yield tests at high-scoring index locations confirm or deny aquifer connectivity. Water-table hydrographs timed against rainfall events reveal the lag time between surface infiltration and water-table response, which is itself a recharge-rate indicator. Isotopic sampling (deuterium, oxygen-18) can fingerprint the altitude and season of recharge, providing independent corroboration of the spatial pattern the satellite analysis predicted.
National hydrogeological surveys that already hold borehole archives can use those records as training data to calibrate the composite index before any new drilling. Water utilities planning new wellfields can use the calibrated map to prioritise exploration targets. Neither application replaces the borehole; both reduce the number of boreholes needed.
From analysis to a decision-ready product
A recharge-zone mapping engagement typically delivers a GIS layer at 10-20 m resolution classifying each pixel into high, moderate, low and negligible recharge potential, accompanied by a per-pixel confidence score derived from data availability (cloud frequency, SAR acquisition gaps) and model performance against any available validation points. Uncertainty is not buried in a methods appendix; it is encoded in the layer itself so that planners can set their own risk threshold.
Satellize runs this multi-layer analysis on open constellations and can incorporate commercial TanDEM-X DEM licensing where the public 90 m product is insufficient. The workflow is the same one that underpins the spatial-analysis components of the Tonga crop-estimation programme: stack open data, validate against ground truth, deliver a decision layer rather than a data dump. A hydrogeological survey that wants to scope a basin before committing to a drilling campaign is the natural starting point.
Typical figures
| Spatial resolution (moisture anomaly layer) | 10 m (Sentinel-1 IW mode, ground range) |
| Spatial resolution (land-cover layer) | 10-20 m (Sentinel-2 MSI) |
| Spatial resolution (topographic layer) | 12 m (TanDEM-X commercial) or 90 m (public release) |
| Spatial resolution (water storage trend) | ~300 km (GRACE-FO mascon products) |
| Revisit cadence | Sentinel-1: 6 days (mid-latitude); Sentinel-2: 5 days; GRACE-FO: monthly |
| SAR archive depth | Sentinel-1A from April 2014; GRACE gravity record from 2002 (GRACE + GRACE-FO) |
| Minimum detectable backscatter anomaly | ~1-2 dB above noise floor; soil moisture changes below ~5 vol% may not register |
| Cloud penetration | Sentinel-1 SAR: full; Sentinel-2 optical: cloud-opaque, requires multi-year compositing in humid regions |
| Composite index output resolution | 10-20 m resampled grid; GRACE trend reported at aquifer-system scale only |
| Delivery formats | GeoTIFF (recharge-potential raster with confidence band), GeoPackage (vector polygons by class), PDF technical report |
Analytics Satellize can run
| Bare-soil moisture anomaly map | Multi-temporal Sentinel-1 backscatter change detection against rolling baseline; VV/VH polarisation ratio to reduce roughness ambiguity | GeoTIFF raster, per-event or seasonal composite |
| Land-cover permeability classification | Random-forest classifier on Sentinel-2 10-band composite; permeability class assigned from published soil-texture lookup | GeoTIFF land-cover layer with permeability score per class |
| Topographic wetness index grid | D-infinity flow routing on TanDEM-X DEM; TWI = ln(a / tan β) following Beven and Kirkby formulation | GeoTIFF TWI raster at DEM native resolution |
| Composite recharge-potential index | Weighted overlay or logistic-regression stack of moisture anomaly, permeability class and TWI; weights calibrated against available borehole yield records | GeoTIFF with four-class potential and per-pixel confidence score; GeoPackage polygon layer |
| GRACE-FO aquifer trend summary | Seasonal decomposition of JPL or CSR mascon time series; Mann-Kendall trend test on residuals after seasonal removal | Basin-scale trend report (mm/year equivalent water height) with confidence interval |
| Validation-ready target list | Ranked list of high-scoring composite-index pixels filtered by accessibility and existing borehole proximity; uncertainty flagged per site | CSV or GeoPackage point layer for field campaign planning |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.