Groundwater depletion detection using satellite gravimetry
GRACE and GRACE-FO measure month-to-month shifts in Earth's gravity field to reveal changes in terrestrial water storage, including groundwater. Signal decomposition isolates the aquifer signal, but the ~300 km resolution floor means attribution requires ancillary data.
Sensors
- GRACE-FO K-Band Ranging (KBR): Measures the inter-satellite distance between twin spacecraft to within a few microns, translating gravity-field variations into terrestrial water storage anomalies at roughly 300 km spatial resolution and monthly temporal resolution. Launched May 2018, ongoing.
- GRACE (archive, 2002–2017): The original mission provides a 15-year baseline of monthly gravity-field solutions. Archive gaps exist (11 months between GRACE end and GRACE-FO first light), handled by interpolation or flagging in published mascon products.
- GLDAS (Global Land Data Assimilation System): Land-surface model output used to estimate and subtract soil-moisture and snow-water contributions from the total GRACE/GRACE-FO signal, leaving a residual attributed to groundwater. Output resolution is 0.25 degrees; model uncertainty propagates directly into the groundwater estimate.
- SMAP (Soil Moisture Active Passive): L-band radiometer providing 36 km soil-moisture retrievals every 2–3 days. Used as an observational constraint on the GLDAS soil-moisture subtraction, reducing one of the largest sources of error in the decomposition.
What the gravity field is actually measuring
GRACE and GRACE-FO do not image the ground. They measure the distance between two co-orbiting spacecraft, separated by roughly 220 kilometres, with a microwave ranging system accurate to a few microns. When the lead satellite passes over a mass concentration, it accelerates slightly, stretching the inter-satellite gap. That stretch, integrated over each month of orbital data, is inverted into a global gravity-field model.
Water is the dominant mobile mass on the land surface. Ice sheets, soil moisture, surface water, and groundwater all move seasonally and inter-annually. The GRACE signal conflates all of them. The spatial resolution of the resulting gravity anomaly maps is approximately 300 kilometres, a physical limit set by the satellite altitude and the mathematics of spherical harmonic inversion. No processing choice can push it below that floor.
Stripping out everything that is not groundwater
Isolating groundwater requires subtracting every other water-storage term from the total terrestrial water storage anomaly. The standard approach uses output from the GLDAS land-surface model to estimate soil-moisture and snow contributions, then subtracts surface-water storage changes derived from radar altimetry and lake-area products. What remains is attributed to groundwater storage change.
The method is honest about its weakness: GLDAS model error does not cancel. Published studies typically report groundwater anomaly uncertainties of 10 to 30 millimetres of equivalent water height per grid cell, depending on region and season. In areas with deep, confined aquifers and low soil-moisture variability, the signal-to-noise ratio is relatively good. In humid, vegetated regions with large seasonal soil-moisture swings, the groundwater residual can be buried in model uncertainty. SMAP observations, assimilated into the soil-moisture layer, tighten that uncertainty somewhat, but do not eliminate it.
Three depletion signals the public record has documented
The Central Valley of California is among the most studied GRACE targets. Published analyses covering the 2002 to 2016 period estimated groundwater loss rates on the order of several cubic kilometres per year during drought years, with the Sacramento and San Joaquin valleys showing distinct depletion patterns. The 300 km footprint means the two sub-basins are not fully separable from GRACE alone; well records and InSAR land-subsidence data are used alongside to attribute the signal.
The North China Plain, which supplies a large share of China's wheat and maize production, shows persistent depletion in GRACE and GRACE-FO records driven by irrigation from the shallow unconfined and deeper confined aquifer systems. Peer-reviewed studies have reported losses exceeding ten cubic kilometres per year during parts of the archive period, with the deep confined aquifer declining faster than the shallow layer.
The Arabian Peninsula aquifer system, a largely fossil aquifer with negligible modern recharge, shows one of the clearest long-term depletion trends in the GRACE record. Published work has linked the signal to agricultural expansion in Saudi Arabia and neighbouring states. Because the region has minimal soil-moisture variability and almost no surface-water bodies, the decomposition is more reliable here than in wetter climates, making it a useful validation case for the method.
What 300 kilometres cannot tell you, and what fills the gap
A single GRACE pixel covers an area comparable to France. It cannot distinguish one aquifer from another, cannot locate the well field driving depletion, and cannot separate a rapidly depleting confined aquifer from a slowly recovering unconfined one sitting above it. These are not software problems. They are geometry.
Ancillary data closes part of the gap. InSAR land-subsidence measurements from Sentinel-1 can resolve depletion-driven surface deformation at metre-scale spatial resolution, identifying which sub-basin is compacting. Groundwater-level well networks, where they exist and are shared, provide point-scale validation. Hydrological models constrained by GRACE totals and well observations can downscale the signal to basin level, though the downscaled product inherits model assumptions. The honest framing is that GRACE defines the budget; everything else argues about the allocation.
Latency is a practical constraint. The standard GRACE-FO mascon products from JPL and CSR are typically released one to three months after the observation month. This is not a near-real-time monitoring tool. It is a strategic accounting instrument, suited to annual or multi-year trend analysis rather than operational drought response.
Turning a gravity anomaly into a policy-relevant number
The raw product from GRACE-FO processing centres is a gridded time series of terrestrial water storage anomalies in centimetres of equivalent water height, referenced to a baseline mean. Converting that to groundwater volume requires multiplying by pixel area, applying the storage coefficient of the aquifer (which requires geological knowledge), and propagating uncertainty through each step. The resulting number, cubic kilometres of groundwater lost or gained per year, is the figure that appears in water-stress assessments and national water accounts.
Satellize runs GRACE-FO mascon time-series analysis as part of its hydrological analytics stack, combining JPL RL06 mascon products with GLDAS v2.1 decomposition and, where available, SMAP soil-moisture corrections. The output feeds into basin-level water-balance reports of the kind that inform irrigation policy and transboundary water negotiations. The approach is the same published method used in the peer-reviewed literature; the value is in operationalising it on a regular cadence and integrating it with other satellite layers a client already commissions.
Typical figures
| Spatial resolution | ~300 km (physical floor set by satellite altitude and spherical harmonic inversion; mascon products tile at 1° or 3° but do not recover sub-300 km detail) |
| Temporal resolution | Monthly gravity-field solutions; some near-daily accelerometer data available but not used for water-storage products |
| Product latency | 1 to 3 months post-observation for standard JPL and CSR mascon releases |
| Archive depth | April 2002 to present (GRACE archive 2002–2017; 11-month gap; GRACE-FO 2018–ongoing) |
| Measurement principle | K-band microwave ranging between twin spacecraft at ~220 km separation; inter-satellite distance resolved to a few microns |
| Signal quantity | Terrestrial water storage anomaly in cm equivalent water height, relative to 2004–2009 baseline mean |
| Groundwater uncertainty | Typically 10–30 mm equivalent water height per grid cell after soil-moisture and surface-water subtraction; higher in humid regions |
| Ancillary inputs required | GLDAS land-surface model (soil moisture, snow); radar altimetry or lake-area product (surface water); optionally SMAP for soil-moisture constraint |
| Coverage | Global land surface; polar regions above ~89° latitude excluded |
| Minimum detectable trend | Approximately 1 cm/year equivalent water height over multi-year periods in low-noise regions; larger in soil-moisture-dominated regions |
Analytics Satellize can run
| Groundwater storage anomaly time series | GRACE/GRACE-FO mascon decomposition (JPL RL06 or CSR RL06) with GLDAS v2.1 soil-moisture and snow subtraction | Monthly GIS layer (GeoTIFF or NetCDF) of groundwater anomaly in mm equivalent water height, with per-pixel uncertainty bounds |
| Long-term depletion trend map | Linear or seasonal-decomposition trend fitting on the groundwater anomaly time series across the full GRACE + GRACE-FO archive | Basin-level PDF report showing trend magnitude (km³/year), confidence interval, and comparison to published regional benchmarks |
| Annual water-balance closure report | Budget accounting combining GRACE-FO total storage change, GLDAS evapotranspiration, and precipitation from ERA5 or CHIRPS | Tabular water-balance summary per defined basin polygon, flagging years where the residual groundwater term exceeds a client-defined threshold |
| SMAP-constrained soil-moisture correction | Assimilation of SMAP L3 passive soil-moisture retrievals into the GLDAS subtraction step to reduce model-error contribution | Revised groundwater anomaly layer with reduced uncertainty estimate, delivered alongside the standard GLDAS-only version for comparison |
| Depletion hotspot alert | Threshold exceedance on rolling 12-month groundwater anomaly trend; triggered when trend exceeds a defined rate in cm/year | Automated alert (email or API push) with pixel location, trend magnitude, and link to the underlying time-series chart |
| InSAR-GRACE joint subsidence assessment | Co-registration of Sentinel-1 InSAR land-subsidence maps with GRACE-FO groundwater anomaly to spatially downscale the depletion signal | GIS layer showing subsidence rate (mm/year from InSAR) overlaid on GRACE groundwater trend, with a written interpretation of aquifer compaction risk |
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.