Groundwater depletion rate estimation for irrigated-agriculture credit-risk assessment
GRACE and GRACE-FO measure basin-scale groundwater loss that no surface sensor can see. Combined with SMAP soil-moisture and Sentinel-2 irrigated-area mapping, the result is a long-run yield-risk signal for lenders exposed to aquifer-dependent farming.
Sensors
- GRACE-FO gravimetry (JPL/GFZ): Measures terrestrial water storage anomalies (TWSA) by tracking inter-satellite range changes at roughly 1-micron precision. Spatial footprint approximately 300 km; monthly temporal resolution. Groundwater storage change is isolated as a residual after subtracting soil moisture, surface water and snow-water-equivalent estimates from ancillary models.
- SMAP L-band radiometer (NASA): 1.41 GHz passive radiometer retrieving surface soil-moisture (0–5 cm layer) at 36 km resolution, with a 2–3 day revisit. The enhanced product uses Sentinel-1 disaggregation to reach approximately 9 km. Used to constrain the soil-moisture term in the GRACE water-balance equation.
- Sentinel-2 MSI (ESA Copernicus): 13-band multispectral imager at 10–20 m resolution with a 5-day revisit at the equator. Used to map irrigated-area extent, crop type and greenness (NDVI, NDWI) within the GRACE footprint, providing the surface-agriculture context that explains why groundwater is being drawn down.
- MODIS Terra/Aqua (NASA): 250 m to 1 km resolution, daily revisit. Long archive from 2000 onward makes it the backbone of multi-decadal irrigated-area and vegetation-anomaly time series, allowing lenders to see trend direction over 20-plus years rather than just recent seasons.
What the gravity signal actually measures
GRACE and its successor GRACE-FO do not image the ground. They measure tiny fluctuations in Earth's gravitational field by tracking the distance between two co-orbiting spacecraft to sub-micron accuracy. When water accumulates or disappears beneath a basin, the mass change shifts the gravity field, and the satellites register it. The terrestrial water storage anomaly (TWSA) that comes out of processing is the sum of all water in the column: surface water, soil moisture, snowpack and groundwater.
Isolating groundwater requires subtracting the other terms. Soil moisture comes from SMAP or land-surface models such as GLDAS. Surface-water extent comes from optical or SAR observations. Snow-water equivalent comes from passive-microwave or model output. What remains is the groundwater storage anomaly, with an uncertainty that published studies typically place at 10–20 mm of equivalent water height per month at basin scale. That is not a precise well-level reading. It is a basin-integrated mass signal, and that distinction matters enormously for how a lender should use it.
Where the method is credible and where it breaks down
The GRACE record, stitched together with GRACE-FO from 2002 onward (with a roughly 11-month gap in 2017–18), is long enough to separate seasonal recharge cycles from secular depletion trends. Basins such as the North China Plain, the Central Valley of California, the High Plains Ogallala, the Indus-Ganges system and the Arabian Peninsula have all shown statistically significant multi-year decline signals in peer-reviewed literature. These are precisely the basins that underpin large concentrations of irrigated-agriculture lending.
The hard limits are real and should not be papered over. The 300 km footprint means GRACE cannot distinguish which part of a basin is depleting fastest, nor can it resolve individual aquifer units that may behave very differently from basin averages. Signal leakage from adjacent basins is a known processing artefact. And the monthly cadence means GRACE captures seasonal storage change only in aggregate; it cannot track the rapid drawdown that follows a single dry spell. For sub-basin credit decisions, GRACE provides the macro trend but must be paired with well-level data, geological surveys or higher-resolution proxies such as InSAR land-subsidence mapping.
Translating a gravity trend into a credit-risk indicator
A lender with a loan book concentrated in, say, the Indus basin needs to know not just that groundwater is declining but at what rate, how that rate compares to recharge, and what the trajectory implies for irrigated-area productivity over the loan term. The satellite water-balance approach delivers a depletion rate in km³ per year or mm per month of equivalent water height. Pairing that with Sentinel-2 irrigated-area maps gives a per-hectare depletion intensity figure. Pairing it further with MODIS NDVI time series shows whether declining water availability is already suppressing crop greenness or whether farmers are compensating by drilling deeper.
The credit-risk translation involves two further steps that satellite data alone cannot provide: an estimate of remaining economically recoverable storage (which requires hydrogeological data), and a price-of-water or cost-of-pumping model that converts depth-to-water into farm-level operating cost. Satellite analytics supply the demand-side signal and the trend direction. They do not replace aquifer characterisation. Lenders who treat GRACE-derived depletion rates as a standalone underwriting tool are misreading the method; those who use it to flag basins for deeper due diligence are using it correctly.
Irrigated area as the exposure multiplier
Groundwater depletion risk is proportional to irrigated area. A basin depleting at 10 km³ per year matters more to a lender if 80% of that basin is under centre-pivot irrigation than if it is 20%. Sentinel-2 at 10 m resolution, with its red-edge and shortwave-infrared bands, supports irrigated-area classification at field scale. Multi-temporal composites distinguish irrigated from rainfed agriculture by the persistence of greenness through dry seasons, when rainfed fields go dormant.
MODIS extends this classification back to 2000, providing a 20-plus year record of irrigated-area expansion or contraction. In some basins, irrigated area has grown substantially even as groundwater declined, which compresses the remaining aquifer lifetime and concentrates credit risk in the most recently expanded areas. That dynamic is visible in the satellite record and is not visible in borrower-reported data.
Practical delivery for portfolio-level screening
For a bank or insurer running a portfolio screen, the most useful output is a basin-level groundwater trend score updated quarterly, flagging basins where the GRACE-derived depletion rate exceeds estimated natural recharge and where irrigated area is expanding. This is a monitoring product, not a point-in-time assessment. The value accumulates over successive quarters as the trend either continues, reverses after a wet season, or accelerates.
Satellize runs this class of water-balance analytics on open constellations, combining GRACE-FO monthly mascon solutions with SMAP soil-moisture fields and Sentinel-2 irrigated-area mapping. The Tonga crop-estimation programme established the operational pipeline for integrating multi-sensor agricultural signals; the groundwater application extends that pipeline into the subsurface water-balance domain. Outputs are delivered as basin-level time-series datasets with uncertainty ranges stated explicitly, not suppressed.
What a lender should ask before commissioning the analysis
Three questions determine whether this method will actually inform a credit decision. First, is the loan book concentrated in basins large enough for GRACE to resolve? Basins smaller than roughly 200,000 km² are at the edge of GRACE's reliable detection range. Second, does the institution have access to, or willingness to commission, complementary hydrogeological data on remaining storage? Without it, the depletion rate is a trend without a terminus. Third, over what time horizon is the credit exposure? GRACE trends are most meaningful for loans with five-year-plus terms; for shorter tenors, seasonal variability dominates the signal.
If the answers are yes, yes and long-term, satellite gravimetry offers something that no ground-based data network can match at comparable cost: a consistent, politically neutral, basin-wide view of where the water went. Wells get capped, reported data get smoothed, and aquifer maps go out of date. Gravity does not lie.
Typical figures
| GRACE-FO spatial footprint | Approximately 300 km (mascon solutions; some products at ~200 km with regularisation) |
| GRACE-FO temporal resolution | Monthly mascon solutions; near-real-time products with ~60-day latency |
| TWSA detection sensitivity | Approximately 10–20 mm equivalent water height per month at basin scale (published uncertainty range) |
| SMAP soil-moisture resolution | 36 km native; ~9 km enhanced (Sentinel-1 disaggregated); 2–3 day revisit |
| Sentinel-2 irrigated-area mapping resolution | 10 m (visible/NIR bands); 20 m (red-edge, SWIR); 5-day revisit at equator |
| MODIS vegetation time-series archive | 2000 to present; 250 m to 1 km; daily revisit |
| GRACE/GRACE-FO combined archive depth | April 2002 to present (11-month gap 2017–18 bridged by statistical interpolation) |
| Minimum basin size for reliable GRACE signal | Approximately 200,000 km² (smaller basins subject to signal leakage and reduced reliability) |
| Delivery format | Basin-level time-series CSV or GeoJSON with uncertainty bounds; optional quarterly PDF summary |
Analytics Satellize can run
| Groundwater storage anomaly time series | GRACE-FO mascon solution minus SMAP soil-moisture and GLDAS surface-water terms (water-balance residual method) | Monthly basin-level groundwater storage change in km³ and mm equivalent water height, with stated uncertainty range, delivered as time-series dataset |
| Multi-decadal depletion trend and rate estimate | Linear or piecewise trend fitting on GRACE/GRACE-FO combined record (2002–present), with seasonal decomposition to separate cyclic recharge from secular decline | Trend report per basin: depletion rate in km³/year, statistical significance, comparison to published recharge estimates |
| Irrigated-area extent and expansion mapping | Multi-temporal Sentinel-2 NDVI compositing to classify irrigated versus rainfed agriculture by dry-season greenness persistence; MODIS time series for decadal trend | Annual irrigated-area GeoJSON layer at field scale, with 20-year expansion time series at basin scale |
| Depletion intensity per irrigated hectare | GRACE-derived groundwater loss divided by Sentinel-2 irrigated-area estimate within the same basin footprint | Normalised depletion-intensity score (m³/ha/year) enabling cross-basin comparison for portfolio screening |
| Basin groundwater risk score for portfolio screening | Composite of depletion trend, depletion intensity, irrigated-area growth rate and NDVI anomaly; scored against peer basins in the same climate zone | Quarterly basin risk-tier update (flagged/watch/stable) for each basin in client's exposure universe, delivered as structured data feed |
| Crop greenness anomaly correlated with water-storage decline | MODIS MOD13 NDVI anomaly relative to 20-year baseline, overlaid on GRACE depletion signal to test whether surface productivity is already responding to subsurface stress | Correlation analysis report and annual NDVI anomaly raster per basin |
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.