Global precipitation rate estimation from merged microwave and IR
No single satellite samples the globe often enough to track precipitation continuously. Operational products fuse passive microwave swath observations with geostationary infrared data to produce near-global, half-hourly rain-rate fields, with known limits at high latitudes and over complex terrain.
Sensors
- GMI / GPM Core Observatory: 13-channel conical-scanning radiometer from 10.65 GHz to 183 GHz. Swath 904 km, revisit roughly every 1.5 to 3 hours at mid-latitudes. The reference calibrator for the entire GPM constellation; its dual-frequency precipitation radar (DPR, Ku/Ka band) provides the ground truth that calibrates passive retrievals from partner radiometers.
- AMSR2 / GCOM-W1 (JAXA): 12-channel radiometer, 6.9 GHz to 89 GHz, swath 1450 km. Contributes ocean-surface rain rates and sea-ice discrimination to the GPM constellation. Spatial resolution at 89 GHz is approximately 5 km, degrading to roughly 35 km at 6.9 GHz.
- SSMIS / DMSP constellation: 24-channel conical scanner covering 19 to 183 GHz across multiple DMSP platforms. Each satellite provides a 1700 km swath; together they add several overpasses per day, particularly improving sampling in the 50–70° latitude band where GPM Core coverage thins.
- ABI / GOES-R series (NOAA): 16-band imager with a 10.35 µm clean-window IR channel at 2 km nadir resolution, full-disk imagery every 10–15 minutes. Used in IMERG as the morphing anchor: cold cloud-top temperatures proxy convective precipitation between microwave overpasses. Not a rain-rate sensor in its own right; the IR-to-rain relationship carries substantial uncertainty over land.
- MHS / MetOp and ATMS / Suomi-NPP, NOAA-20: Cross-track microwave sounders with channels at 89, 157, and 183±1/3/7 GHz. Swath roughly 2250 km (MHS) and 2300 km (ATMS). Primarily atmospheric sounders, but their high-frequency channels respond to ice scattering and are routinely ingested by IMERG to fill gaps between conical-scanner overpasses.
Why no single orbit is enough
A low-Earth orbit at roughly 400 km altitude subtends a passive microwave swath of 900 to 1700 km. That sounds generous until you consider that a mid-latitude convective cell lives for 30 to 90 minutes and moves at 20 to 60 km/h. A single satellite revisiting every 90 minutes will miss the peak intensity of most such systems. The GPM constellation partially solves this by combining eight or more passive microwave radiometers operated by NASA, JAXA, NOAA, EUMETSAT and the DMSP programme, achieving a combined revisit of roughly 30 to 90 minutes at low and mid-latitudes.
Even so, gaps remain. The GPM Core Observatory is inclined at 65°, so poleward of that latitude the constellation thins sharply. IMERG fills those gaps with geostationary infrared brightness temperatures, which respond to cloud-top altitude rather than surface rain rate directly. The morphing step, which advects the most recent microwave observation forward in time using IR-derived cloud motion vectors, works reasonably well over ocean and flat land. Over mountains and at high latitudes it is a known source of systematic error.
What ice scattering at 89 GHz actually tells you
Passive microwave retrieval of precipitation rests on two distinct physical mechanisms depending on surface background. Over ocean, liquid raindrops emit microwave radiation, warming the brightness temperature relative to the cold sea surface below. The signal is detectable at frequencies from around 10 GHz upward, and rain rates above roughly 0.5 mm/h are generally retrievable.
Over land, the emissivity of the surface is high and variable, swamping the emission signal from rain. Retrieval therefore shifts to scattering: ice particles in the upper portions of a convective column scatter upwelling microwave radiation away from the sensor, depressing brightness temperatures at 89 GHz and above. The depression is empirically related to the column ice water path, which is in turn related to surface rain rate through cloud-model lookup tables. The relationship holds well for deep convection but breaks down for warm-rain systems, where precipitation forms entirely below the freezing level without significant ice. Shallow orographic precipitation at high latitudes is the canonical failure mode: the column contains little ice, the brightness temperature depression is small, and IMERG systematically underestimates rain rates, sometimes by 30 to 50% relative to gauge networks in published validation studies over alpine and sub-Arctic terrain.
IMERG: architecture of a global product
The Integrated Multi-satellitE Retrievals for GPM (IMERG) is the NASA algorithm that combines all constellation inputs into a single half-hourly, 0.1° (roughly 11 km at the equator) global grid from 60°S to 60°N, with degraded-quality extension to 90°. It runs in three latency modes: Early (approximately 4 hours after observation time, for near-real-time applications), Late (approximately 14 hours, with reprocessed microwave inputs), and Final (approximately 3.5 months, gauge-corrected using the GPCC monthly analysis). The Final product is the one used for climatological work; the Early product is what disaster-response and flood-forecasting operations actually see.
Gauge correction in the Final run reduces systematic bias over land where gauge networks are dense, such as the continental United States and western Europe. Over the tropical oceans, the central Sahara, or the high Arctic, gauge correction is sparse or absent, and the product is effectively unconstrained by surface observations. Users should treat IMERG uncertainty estimates, which are included in the product files, as meaningful rather than decorative.
Where the product fails, and how badly
Three failure modes deserve explicit attention. First, orographic precipitation: windward slopes of major mountain ranges, the Himalayas, the Andes, the Norwegian coast, force moist air upward and generate rainfall from relatively shallow, warm clouds. IMERG misses a substantial fraction of this rainfall. Published comparisons against dense gauge networks in the Alps and over the Tibetan Plateau show biases ranging from 20% to over 100% depending on season and elevation band.
Second, snowfall. IMERG includes a snowfall product, but distinguishing falling snow from the underlying snowpack in passive microwave brightness temperatures is genuinely hard. Surface emissivity from dry snow can mimic the scattering signature of precipitating ice aloft. The snowfall retrieval carries larger uncertainty than liquid-phase estimates, and the product documentation says so plainly.
Third, tropical warm rain over islands and coastal zones. Small convective cells over islands are often sub-pixel at 0.1° resolution, and the IR morphing step can misplace the rain field by tens of kilometres. For a small island nation, that is the difference between a hit and a miss.
Practical use in flood forecasting and agricultural monitoring
Despite its limits, IMERG Early is operationally used by flood-forecasting centres worldwide because it is the only near-real-time, globally consistent precipitation field available at sub-daily resolution. The Global Flood Awareness System (GloFAS) ingests it. FEWS NET uses it for food-security monitoring across the Sahel and Horn of Africa. At these scales, the systematic biases matter less than the spatial coherence and timeliness.
Agricultural applications demand more care. Crop-water models are sensitive to cumulative rainfall over a growing season, so the biases that average out in a global flood context accumulate into meaningful errors in seasonal water-balance calculations. Satellize's crop-estimation work for the Kingdom of Tonga, for instance, requires careful cross-referencing of IMERG against any available gauge records and against soil-moisture retrievals, because warm-rain underestimation over small Pacific islands is precisely the failure mode described above. The honest approach is to treat IMERG as a prior, not a ground truth, and to propagate its uncertainty into downstream crop-yield estimates explicitly.
Reading the archive: 25 years of retrospective data
The IMERG V07 Final product provides a retrospective record from June 2000 to present, extending back through the TRMM era via a consistent reprocessing chain. That 25-year archive is long enough to compute meaningful return-period statistics for extreme precipitation events at regional scales, though users should note that the constellation composition changed substantially between 2000 and 2014 when GPM Core launched, and early-period uncertainty is higher.
For trend analysis, the changing sensor mix is a genuine confound. A step-change in the constellation, such as the addition of a new radiometer or the loss of a DMSP satellite, can introduce artefacts that look like precipitation trends. Published homogeneity assessments of the TRMM-to-GPM transition exist and should be consulted before attributing any sub-decadal trend in IMERG to climate rather than to instrument change.
Typical figures
| Native spatial resolution (IMERG grid) | 0.1° × 0.1° (approximately 11 km at equator) |
| Temporal resolution | 30 minutes (half-hourly) |
| Latency (Early / Late / Final) | ~4 hours / ~14 hours / ~3.5 months after observation |
| Geographic coverage | 60°S–60°N (full quality); 90°S–90°N (degraded, IR-only at high latitudes) |
| Frequency channels used (passive MW) | 10.65 GHz to 183 GHz across constellation; DPR at Ku (13.6 GHz) and Ka (35.5 GHz) |
| Minimum detectable rain rate (passive MW, ocean) | Approximately 0.2–0.5 mm/h; land retrievals less sensitive |
| Archive depth (IMERG V07 retrospective) | June 2000 to present (~25 years) |
| Known systematic bias (orographic / warm rain, high latitude) | Underestimation of 20–100%+ depending on terrain and season; documented in peer-reviewed validation literature |
| Delivery format (NASA open data) | HDF5 and netCDF4 via NASA Earthdata; GeoTIFF derivable |
Analytics Satellize can run
| Seasonal cumulative rainfall anomaly maps | IMERG Final gridded accumulation differenced against 2001–2020 climatological baseline; bias-corrected against available gauge records | GeoTIFF raster layers and PDF summary report, monthly cadence |
| Near-real-time heavy-rainfall alerts | IMERG Early 3-hourly accumulation threshold exceedance, configurable per basin or administrative boundary | Automated alert feed (JSON or email) with ~4-hour latency after event |
| Crop-season water-balance estimates | IMERG rainfall inputs to a soil-water balance model; uncertainty bounds propagated from IMERG product uncertainty fields | Tabular report per agro-zone with confidence intervals; compatible with Satellize crop-estimation pipeline used in Tonga |
| Extreme-event return-period analysis | GEV or LP3 frequency analysis on 25-year IMERG archive at user-specified grid cells or catchments | Return-period curves (10-, 25-, 50-, 100-year) as PDF and CSV |
| Flood-index time series for ungauged basins | IMERG Early routed through a lumped hydrological model (e.g. GloFAS-style); calibrated where gauge data exist | Daily discharge-proxy time series as GIS layer or API feed |
| Warm-rain bias assessment for a specific region | Cross-comparison of IMERG against AMSR2 ocean-emission retrievals and any available ground stations; quantifies local underestimation factor | One-page technical memo with bias correction factors for downstream model use |
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.