Cloud liquid water path retrieval from passive microwave radiometry

Passive microwave radiometers on AMSR2, GMI and SSMIS measure cloud liquid water path to roughly 10–20 g/m² over ocean, giving climate modellers and precipitation forecasters a variable optical sensors cannot reach. The method fails almost completely over land.

Sensors

AMSR2 / GCOM-W1 (JAXA): Six frequency pairs from 6.9 to 89 GHz; the 18.7 and 36.5 GHz channels are the primary pair for cloud liquid water path retrieval over ocean. Swath 1,450 km, daily global coverage at mid-latitudes. Spatial resolution at 18.7 GHz is approximately 22 × 14 km.
GMI / GPM Core Observatory (NASA/JAXA): 13-channel conical scanner from 10 to 183 GHz. The 18.7 and 36.5 GHz channels retrieve cloud liquid water path; the higher frequencies extend sensitivity toward precipitation-sized drops. Swath 885 km; non-sun-synchronous orbit gives variable local overpass times, reducing diurnal sampling bias. Resolution at 18.7 GHz approximately 19 × 11 km.
SSM/I and SSMIS / DMSP series (NOAA/DoD): The SSM/I archive begins 1987, making it the backbone of long-term cloud liquid water path climatologies. SSMIS extends the record with additional channels. Seven channels from 19.35 to 85.5 GHz; cloud liquid water path retrieved primarily from the 19 and 37 GHz window channels. Swath 1,400 km.
MWR / Sentinel-3 (ESA/Copernicus): Two-channel microwave radiometer at 23.8 and 36.5 GHz, carried primarily to correct the radar altimeter for wet tropospheric path delay. It produces cloud liquid water path as a by-product over ocean. Footprint approximately 20 km; coverage is a narrow nadir strip rather than a wide swath, so it is not suited to area mapping but is useful for validation along the ground track.

What a wet cloud looks like at 18 and 37 GHz

Liquid water droplets in clouds absorb and emit microwave radiation in proportion to the total mass of water in the column, a quantity called cloud liquid water path (LWP), measured in grams per square metre. Over a cold, specularly smooth ocean surface, the background brightness temperature is low, typically 100–150 K at 18 GHz, giving liquid cloud a warm, high-contrast signature against it. That contrast is the physical basis of the retrieval.

The two-frequency approach exploits the fact that liquid water and water vapour have different spectral dependencies across the microwave window channels. At 18.7 GHz, both liquid water and vapour contribute to brightness temperature, but at 36.5 GHz liquid water dominates more strongly relative to vapour. Combining the two channels in a linear or log-linear regression, calibrated against radiative transfer model simulations, separates the two contributions. Published algorithms such as the Wentz (1997) formulation and its successors, widely used in the Remote Sensing Systems product suite, achieve LWP retrievals with an uncertainty of roughly 10–25 g/m² under typical conditions.

The 10–20 g/m² floor and what sits below it

The detection floor for LWP over ocean is generally cited at 10–20 g/m² in the peer-reviewed literature. Below that threshold, the signal is buried in noise from sea-surface emissivity variability, foam, wind-roughening effects and instrument noise equivalent delta-T, which for AMSR2 is approximately 0.3–0.5 K at 18.7 GHz. Thin cirrus, optically thin stratus and newly forming convective anvils all fall below this floor. The method is therefore not a cloud-detection tool; it quantifies clouds that are already optically thick enough to matter for radiation budgets and precipitation onset.

At the upper end, heavy precipitation contaminates the retrieval. Rain drops are large enough to scatter at 37 GHz, and the scattering signal drives brightness temperatures down rather than up, mimicking low LWP. Most operational algorithms screen rainy pixels using the 85–89 GHz channels, where scattering is stronger and more easily identified, then flag or discard them. The result is that LWP products have a gap precisely where convective clouds are heaviest, which is inconvenient for precipitation studies but unavoidable with current passive sensors.

Why land is nearly useless for this retrieval

Over ocean, surface emissivity is low and relatively predictable from wind speed and sea-surface temperature. Over land, emissivity at 18–37 GHz ranges from roughly 0.6 over sandy desert to above 0.95 over dense vegetation, varies with soil moisture, changes after rainfall, and is spatially heterogeneous at scales far smaller than the sensor footprint. The surface emission swamps the cloud signal. A 50 g/m² cloud produces a brightness temperature increment of perhaps 5–8 K over ocean; over a mixed land surface, the emissivity uncertainty alone can produce errors of 20–40 K.

There is no satisfactory workaround with current window-channel radiometers. Attempts to use ancillary land-surface emissivity maps reduce the error somewhat, but residual uncertainty still exceeds the cloud signal for all but the thickest cloud layers. The practical consequence is that passive microwave LWP retrievals are ocean products. Continental cloud properties must be inferred from other methods, including ground-based microwave radiometers or active lidar and radar, none of which are covered here.

Climate model evaluation: why the long SSM/I record matters

The SSM/I archive, continuous since July 1987, provides nearly four decades of ocean LWP observations. Climate models have historically overestimated LWP in marine stratocumulus regions, particularly off the west coasts of continents, and the satellite record has been central to identifying and quantifying that bias. The AMSR2 and GMI records, beginning in 2012 and 2014 respectively, extend and refine the climatology with improved calibration and additional channels.

Seasonal and interannual variability in ocean LWP is tied to sea-surface temperature, large-scale circulation patterns and aerosol loading, the last through the first and second aerosol indirect effects. Detecting trends in LWP at the few g/m² level over decades requires careful inter-satellite calibration, because each instrument has its own bias. The Remote Sensing Systems multi-satellite merged product addresses this, though residual inter-sensor offsets of 5–10 g/m² remain a known source of uncertainty in trend analyses.

Precipitation onset and the LWP threshold question

One applied use of LWP retrievals is identifying when a cloud is close to precipitating. Observational studies using collocated CloudSat radar and passive microwave data have shown that warm rain onset in marine stratocumulus tends to occur when LWP exceeds roughly 150–200 g/m², though the threshold varies with cloud-drop number concentration and therefore aerosol environment. A high-aerosol cloud can sustain higher LWP before precipitation forms, which is one observational signature of the second aerosol indirect effect.

For operational forecasting, the spatial resolution of AMSR2 and GMI (footprints of 15–25 km at the primary retrieval frequencies) is coarse relative to convective organisation. A mesoscale convective system evolves on scales of 1–10 km and over timescales of tens of minutes. The satellite provides a snapshot, not a continuous watch. Geostationary microwave sounders, still in early deployment, may eventually close that gap, but the current passive microwave constellation is better suited to climatological analysis than to storm-scale nowcasting.

What Satellize runs on these datasets

Satellize ingests the AMSR2 Level-2 ocean LWP product and the GPM GMI retrievals through standard open distribution channels, combining them with ERA5 reanalysis fields to produce basin-scale cloud water anomaly maps for climate model validation and maritime weather-risk assessments.

The honest boundary of the service mirrors the honest boundary of the physics. Over land, we do not offer LWP products derived from passive microwave data because the retrieval is not reliable there. Clients with continental cloud-property requirements are directed to ground-based radiometer networks or active spaceborne sensors, depending on their geographic focus and latency needs.

Typical figures

Primary retrieval frequencies	18.7 and 36.5 GHz (window channels); 23.8 GHz used for water vapour separation on some instruments
Spatial resolution at 18.7 GHz	Approximately 19–22 km (AMSR2, GMI); ~20 km footprint for Sentinel-3 MWR
Swath width	885 km (GMI) to 1,450 km (AMSR2); Sentinel-3 MWR is nadir-only (~20 km strip)
Revisit (ocean, mid-latitudes)	Daily to twice-daily with AMSR2 alone; 4–6 passes per day when AMSR2, GMI and SSMIS are combined
Detection floor (LWP over ocean)	10–20 g/m²; retrievals below this threshold are unreliable
Upper valid range	Approximately 300–500 g/m²; heavier precipitation contaminates and is screened
Coverage	Ocean surfaces globally; land retrievals are not operationally valid
Archive depth	SSM/I from July 1987; AMSR2 from May 2012; GMI from March 2014
Typical retrieval uncertainty	10–25 g/m² (random); inter-sensor biases of 5–10 g/m² in merged climatologies
Latency (near-real-time products)	3–6 hours for AMSR2 and GMI standard NRT products via JAXA/NASA distribution

Analytics Satellize can run

Ocean LWP anomaly maps	Dual-frequency brightness temperature retrieval (Wentz-type algorithm) applied to AMSR2 L2 swath data, anomaly computed against ERA5 monthly climatology	Monthly GeoTIFF grids at 0.25° resolution, with anomaly magnitude and confidence flag layers
Multi-sensor merged LWP climatology	Inter-calibrated combination of AMSR2, GMI and SSMIS retrievals using published cross-calibration offsets; temporal averaging to reduce swath gaps	Annual and seasonal NetCDF climatology files covering the satellite era from 1987 to present
Marine stratocumulus LWP time series	Spatial subsetting over defined ocean regions (e.g. southeast Atlantic, northeast Pacific), extraction of daily mean LWP with precipitation-contaminated pixels screened	CSV time series with uncertainty bounds, suitable for climate model evaluation or reinsurance risk assessment
Precipitation-onset probability layer	LWP threshold exceedance analysis (>150 g/m²) combined with drop-number proxies from aerosol optical depth fields, following published observational relationships from CloudSat collocations	Gridded probability raster updated per satellite overpass, delivered as a GIS layer with overpass timestamp
Inter-satellite calibration diagnostics	Double-differencing of collocated AMSR2 and GMI retrievals over stable ocean scenes to quantify and flag inter-sensor bias drift	Quarterly bias report with time-series plots, flagging periods where merged products exceed stated uncertainty thresholds

Sources

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.