Hydro-resource assessment from satellite precipitation climatology
Multi-satellite merged precipitation products can estimate long-term runoff potential in basins where ground gauges are absent or unreliable, but orographic bias and coarse resolution demand careful uncertainty accounting before any investment decision.
Sensors
- GPM IMERG (GMI + DPR): The Global Precipitation Measurement mission's Integrated Multi-satellitE Retrievals product merges passive-microwave estimates from the GPM Microwave Imager with Ku/Ka-band Dual-frequency Precipitation Radar retrievals, then blends in IR-based estimates from geostationary satellites. Final-run IMERG delivers 0.1° (~11 km at the equator) gridded precipitation at 30-minute intervals, with a ~3.5-month latency for the calibrated Final product and near-real-time Early/Late runs available within hours. Archive runs from June 2000 (using TRMM-era intercalibration).
- TRMM TMPA (legacy): The TRMM Multi-satellite Precipitation Analysis operated from 1998 to 2019 at 0.25° resolution and 3-hourly intervals, covering 50°S–50°N. Though superseded by IMERG, TMPA's 20-year record remains useful for long-term climatology and trend detection in tropical and subtropical basins. Bias characteristics differ from IMERG, so merging the two records requires intercalibration.
- PERSIANN-CDR: The Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks Climate Data Record uses geostationary IR brightness temperatures, adjusted against GPCP monthly data. Resolution is 0.25° daily, coverage 60°S–60°N, and the archive extends from 1983, making it the longest consistent satellite-based precipitation record available. It carries higher random error than IMERG but its multi-decadal depth is valuable for exceedance-frequency analysis.
- ERA5 reanalysis precipitation: ECMWF's ERA5 reanalysis assimilates conventional observations and satellite radiances into a global atmospheric model, producing hourly precipitation estimates at roughly 31 km resolution from 1940 to present. ERA5 is not a pure satellite product, but it often outperforms satellite-only estimates in high-latitude and complex-terrain basins because the model's orographic representation adds physical constraint. Biases in ERA5 precipitation are well documented and region-specific.
What the satellites are actually measuring, and what they are not
None of these products measure rain at the surface directly. GPM's DPR measures the vertical profile of hydrometeors in the atmospheric column; the GMI retrieves column-integrated water vapour and precipitation-sized ice. Surface rainfall is inferred. The inference works well over ocean and flat terrain, where the microwave signal is relatively clean. Over land, especially vegetated land, the surface emissivity contaminates the passive-microwave retrieval, and the DPR's minimum detectable rain rate is approximately 0.2 mm/hr for the Ka-band channel and 0.5 mm/hr for the Ku-band, meaning light drizzle is systematically missed.
The IMERG merging algorithm fills the gaps between GPM overpasses, which recur every 1–3 days at any given point, by using morphed IR estimates from geostationary satellites. IR brightness temperature is a proxy for cloud-top height, not rain rate. The relationship between the two breaks down for warm-rain processes and for shallow orographic precipitation, which is precisely the mechanism that dominates in many hydro-resource basins. Buyers should treat the satellite product as a spatially continuous but physically approximate estimate, not as a dense gauge network.
The orographic problem is not a footnote
Orographic enhancement, the process by which moist air forced over a ridge deposits most of its precipitation on the windward slope within a few kilometres of the crest, is one of the most important processes for run-of-river hydro siting. It is also the process that satellite precipitation products handle worst. IMERG's 0.1° grid (~11 km) cannot resolve the precipitation gradient across a 2 km ridge. Published validation studies in the Himalayas, Andes and Ethiopian Highlands consistently find IMERG underestimates mean annual precipitation by 20–50% on steep windward slopes, with the error growing at higher elevations.
PERSIANN-CDR is coarser still at 0.25°, and ERA5, despite its physical orographic representation, inherits errors from the model's smoothed terrain. No single product is universally superior in complex terrain. The practical response is to run an ensemble: use IMERG, ERA5 and PERSIANN-CDR in parallel, quantify their spread as a first-order uncertainty bound, and then apply a bias-correction step anchored to whatever sparse gauge or river-flow data exist in or near the catchment. Even a handful of gauges, or a short-period flow record, can substantially reduce the ensemble spread through Bayesian updating.
From precipitation to runoff: the modelling chain and where uncertainty accumulates
Precipitation is an input, not the answer. Translating it to runoff potential requires a hydrological model that accounts for evapotranspiration, soil moisture storage, snowmelt (covered separately in the snowpack water-equivalent page in this library) and routing. Conceptual bucket models such as HBV or GR4J are commonly applied in data-sparse basins because they have few free parameters and can be calibrated on short flow records. Physically based models such as VIC or SWAT demand more input data and are not obviously more accurate when calibration data are scarce.
Uncertainty propagates multiplicatively through this chain. A 30% precipitation bias produces a runoff error that is larger in percentage terms because runoff is the residual after evapotranspiration is subtracted. In semi-arid catchments where annual precipitation is only modestly above potential evapotranspiration, a 20% precipitation underestimate can halve the estimated mean annual runoff. This is not a theoretical concern: it has caused material overestimates of hydro potential in feasibility studies that used uncorrected satellite precipitation without uncertainty bounds. Reporting a single-number mean annual energy yield without a P90/P50 exceedance analysis is not defensible practice.
What a 20-year record can and cannot tell you about future yield
IMERG's effective archive, using TRMM-era intercalibration, runs from mid-2000. PERSIANN-CDR reaches back to 1983. For hydro finance, a 20-year precipitation record is borderline: long enough to capture several ENSO cycles and to estimate P90 exceedance with moderate confidence, but short enough that a single anomalous wet or dry decade can skew the climatology. Inter-annual variability in tropical basins is often dominated by ENSO, and the 2015–16 El Niño was one of the strongest on record. If that event falls near the start or end of the analysis window, it distorts the frequency distribution.
ERA5's 1940-to-present archive is the most useful tool for extending the record, but ERA5 precipitation in data-sparse regions is constrained mainly by the model physics rather than observations, so it carries its own systematic biases. The honest position is that satellite-derived precipitation climatology can characterise the shape of the inter-annual distribution and flag high-variability regimes, but absolute yield estimates require bias correction and should always be presented with exceedance probabilities rather than point estimates.
What a satellite-driven assessment can deliver, and what it cannot replace
For early-stage screening of run-of-river or reservoir sites in ungauged basins, satellite precipitation climatology is genuinely useful. It can rank candidate catchments by mean annual precipitation, flag high inter-annual variability, identify seasonal flow regimes, and produce order-of-magnitude runoff estimates that justify or preclude the cost of a ground campaign. At this stage the uncertainty is acceptable because the decision being made is binary: worth investigating further, or not.
As a project advances toward bankable feasibility, the satellite record becomes a complement to, not a substitute for, ground measurement. A minimum of two to three years of in-situ flow gauging, combined with satellite-derived precipitation for spatial interpolation and gap-filling, is the standard that project finance lenders expect. Satellize can structure the satellite-side of that analysis, including ensemble bias correction, catchment-scale runoff modelling and exceedance-frequency reporting, as part of a broader data programme. The Tonga crop-estimation programme demonstrated a similar logic: satellite data establishes the spatial pattern; ground truth anchors the quantitative output.
The irreplaceable ground elements remain stream-flow gauges, local rain gauges on the windward slope, and sediment data. No satellite product currently resolves the 1–2 km precipitation gradients that determine whether a run-of-river scheme is viable or marginal.
Typical figures
| IMERG spatial resolution | 0.1° (~11 km at equator) |
| IMERG temporal resolution | 30 minutes (Final run); Early/Late near-real-time within hours |
| IMERG Final-run latency | ~3.5 months (calibrated against GPCC gauge analysis) |
| IMERG archive depth | June 2000 to present (TRMM-era intercalibrated back-extension) |
| PERSIANN-CDR archive depth | 1983 to present at 0.25° daily |
| ERA5 archive depth and resolution | 1940 to present; ~31 km hourly |
| GPM DPR minimum detectable rain rate | ~0.5 mm/hr (Ku-band); ~0.2 mm/hr (Ka-band) |
| Typical orographic bias (steep terrain) | IMERG underestimates 20–50% on windward slopes; range varies by region and season |
| Coverage | IMERG and PERSIANN-CDR: 60°S–60°N; ERA5: global |
| Deliverable formats | NetCDF, GeoTIFF time-series, CSV catchment-aggregated tables, GIS polygon layers with uncertainty bands |
Analytics Satellize can run
| Mean annual precipitation climatology with uncertainty ensemble | Multi-product ensemble (IMERG Final, ERA5, PERSIANN-CDR) with inter-product spread as uncertainty proxy; optional bias correction against available gauge records | GeoTIFF climatology maps per catchment with P10/P50/P90 precipitation layers and tabular summary |
| Inter-annual variability and ENSO sensitivity index | Correlation of annual precipitation anomalies with ONI (Oceanic Niño Index) over the available record; coefficient of variation computed per sub-catchment | PDF report with time-series plots, variability maps and ENSO-phase conditional distributions |
| Catchment-scale mean annual runoff estimate with exceedance probabilities | Conceptual hydrological model (HBV or GR4J class) driven by bias-corrected IMERG; Monte Carlo uncertainty propagation across precipitation and parameter uncertainty | P90/P75/P50 annual runoff tables per candidate catchment, GIS polygon layer, model parameter report |
| Seasonal flow-regime characterisation | Monthly precipitation climatology aggregated to catchment; flow-duration curve estimation via rainfall-runoff model; identification of low-flow months for firm-power calculation | Flow-duration curves (modelled), monthly regime charts, firm-power fraction estimate with stated uncertainty |
| Orographic bias-correction report | Comparison of IMERG and ERA5 precipitation against available gauge or river-flow data; PRISM-style or quantile-mapping correction applied where sufficient ground data exist | Correction factor maps, pre/post-correction precipitation time series, documented uncertainty reduction |
| Multi-site screening rank | Standardised scoring of candidate basins on mean annual precipitation, variability, seasonality and data quality; no hydrological modelling required at this stage | Ranked candidate-site table with data-confidence flags, suitable for go/no-go investment screening |
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.