Yet these instruments have a well-known Achilles heel: clouds. In the microwave region, gaseous absorption is governed mainly by temperature, pressure, and water-vapor density, and under clear skies the atmosphere is nearly transparent in the window channels, with the K-band dominated by water vapor and the V-band by oxygen. When clouds and precipitation appear, however, hydrometeors, including cloud water, rain, ice, snow, and graupel, add their own absorption and scattering to the signal, rapidly increasing the optical thickness and perturbing the observed brightness temperatures. For years, that complexity meant many cloud-affected radiometer observations were simply set aside, their information value lost because no suitable forward model existed to interpret them. Now a team led by Ziyue Huang of Fudan University and the China Meteorological Administration, together with Yi-Ning Shi, Fuzhong Weng, and Jun Yang, has changed that equation. In a paper published in Geoscientific Model Development, they present ARMS-gb v2.0, an upgraded fast radiative transfer model that extends ground-based microwave modeling from clear-sky to all-sky conditions, validated against six months of field measurements from two sites in China.

Radiative transfer models provide the essential mathematical link between the state of the atmosphere and the radiances an instrument actually records. Fast versions of these models are indispensable for variational data assimilation in modern numerical weather prediction, where each observation must be compared with a model background within seconds. Mainstream fast models such as CRTM, RTTOV, and ARMS have already been extended to all-sky capabilities for satellites, enabling the operational assimilation of cloud-affected satellite radiances. Ground-based applications have lagged behind, owing to the lack of suitable forward models and effective methods for handling water vapor and hydrometeors. High-accuracy monochromatic tools such as MonoRTM, together with absorption models like MPM, serve as reference standards but are computationally far too expensive for routine broadband simulations at the temporal resolution radiometers demand. Approximating each broadband channel by a single calculation at its center frequency introduces systematic brightness-temperature biases of roughly two to three kelvin. To the authors’ knowledge, only two dedicated fast radiative transfer models for ground-based microwave radiometers currently exist: RTTOV-gb in Europe and ARMS-gb in China. RTTOV-gb was a significant development, but recent studies report substantial biases in water-vapor channels and it lacks an operational cloud-scattering option. ARMS-gb v1.0 incorporated a line-by-line trained gaseous absorption scheme and advanced water-vapor interpolation, performing comparably to or better than RTTOV-gb in humid environments, but it remained a clear-sky model.

ARMS-gb v2.0 changes that with two major additions. The first is a cloud particle absorption and scattering module built on a pre-computed look-up table covering five hydrometeor categories: cloud water, rain, ice, snow, and graupel. Single-particle optical properties, including the extinction coefficient, single-scattering albedo, and the full phase function, are calculated with Mie theory for liquid droplets and with a rough-surfaced aggregate model based on the discrete dipole approximation, or DDA, for nonspherical frozen particles. These single-particle attributes are then integrated over a modified gamma size distribution to obtain bulk optical parameters. The look-up table spans liquid hydrometeors at temperatures from 275 to 290 kelvin in five-kelvin intervals and solid hydrometeors from 243 to 273 kelvin in ten-kelvin intervals, with refractive indices drawn from Ray’s 1972 data for liquid water and Mätzler’s ice model, and Maxwell-Garnett mixing applied to represent the effective dielectric properties of ice-air mixtures within frozen particles. Crucially, the table was constructed entirely from physics-based scattering calculations, making it independent of the observational datasets used for evaluation.

The second addition is an accelerated discrete ordinate method, or ADOM, a numerically stable multiple-scattering solver. Ground-based radiometers observe downwelling radiance dominated by atmospheric thermal emission, with the cosmic microwave background entering as the top boundary. Under clear skies, the source function reduces to gaseous Planck emission with absorption. In all-sky scenes, clouds and precipitation introduce extinction, emission, and multiple scattering, requiring the full scalar radiative transfer equation to be solved. ADOM applies the discrete ordinates method exclusively to scattering layers while merging adjacent non-scattering layers based on non-scattering radiative transfer theory, dramatically improving computational efficiency. The atmosphere is divided into plane-parallel layers, the phase function is expanded in Legendre polynomials, and the radiance is decomposed into a Fourier cosine series, yielding a set of independent equations solved with Gaussian quadrature. The lower boundary is determined by surface thermal emission and diffuse reflection, and an additional Gaussian quadrature point with zero weight allows the radiance to be computed at an arbitrary viewing zenith angle. The model also extends its tangent-linear and adjoint modules to hydrometeor processes, providing the Jacobians that variational retrieval and data assimilation systems require.

Sensitivity experiments reveal how strongly the microwave signal depends on hydrometeor microphysics. Using the U.S. Standard Atmosphere as a clear-sky background, the team prescribed individual hydrometeor species at realistic altitudes, with liquid species between roughly two and three kilometers and frozen species between five and eight kilometers, and varied water paths and particle effective radii independently. Cloud liquid water generally produces positive brightness-temperature perturbations through absorption and thermal emission, with the strongest responses in the 22-31 gigahertz water-vapor band and the weak oxygen band near 51-55 gigahertz, increasing approximately linearly with liquid water path. Rainwater behaves more nonlinearly: at small water paths, scattering produces negative perturbations in several weak oxygen-band channels, while absorption and emission grow increasingly important as the water path increases. Frozen species act differently. Cloud ice and snow generally produce negative perturbations in the weak oxygen band because particle scattering redirects radiation away from the viewing direction. Scattering strengthens rapidly with particle size and frequency, and once effective radii reach a few hundred micrometers and frequencies exceed roughly 45-50 gigahertz, the single-scattering albedo approaches unity, marking a scattering-dominated regime most evident in the oxygen band.

The Mie-versus-DDA comparison adds a subtle but important dimension. For ice and snow, differences between the spherical and nonspherical treatments are small at small radii and low frequencies but grow to positive values with size and frequency, meaning the spherical assumption tends to produce stronger extinction and higher albedo than the nonspherical treatment. Differences for graupel remain comparatively small, consistent with its more compact, near-spherical habit. Band dependence is clear: the water-vapor band is only moderately sensitive to frozen-phase scattering, whereas the oxygen band is highly sensitive, amplifying the DDA-Mie discrepancies there. In forward modeling and data assimilation, the Mie-based spherical treatment yields stronger scattering for large frozen hydrometeors in the oxygen band than the DDA-based nonspherical treatment, with the magnitude depending on particle size and frequency.

Validation against real observations is where the upgrade proves itself. The team used two ground-based radiometer systems: the 22-channel YKW-1 at Yan’an Station in northern China and the 14-channel YKW-2 at Karamay Station in the northwest. Hourly ERA5 reanalysis profiles were collocated with radiometer observations from January to June 2025, providing temperature and humidity profiles together with mixing ratios of cloud liquid water, rainwater, cloud ice, and snow. After scene classification and quality-control screening, 2135 hourly cloudy non-precipitating profiles were retained for Yan’an and 2845 for Karamay. The results show that ARMS-gb v2.0 effectively simulates brightness temperatures under cloudy non-precipitating conditions, with the mean observed-minus-simulated brightness temperature generally within one kelvin of the corresponding clear-sky baseline. Compared with ARMS-gb v1.0, which neglects cloud effects, the root mean square error under cloudy conditions decreases by one to two kelvin in the water-vapor band, most notably at 30 gigahertz, where the coefficient of determination rises from 0.55 to 0.77. In the weak oxygen band, where frozen hydrometeors exert stronger scattering effects, the observation-minus-background departure decreases by three to four kelvin. At 51 gigahertz, the RMSE drops from 5.23 to 3.22 kelvin and the coefficient of determination climbs from 0.52 to 0.81. RTTOV-gb improved upon ARMS-gb v1.0 in many hydrometeor-sensitive channels but generally remained farther from the clear-sky baseline than ARMS-gb v2.0. The DDA-based treatment slightly outperformed the Mie model in characterizing frozen hydrometeors at these channels, with maximum differences between the two configurations of roughly 0.33 kelvin at Yan’an and 0.17 kelvin at Karamay.

Jacobian analysis provides a complementary diagnostic of vertical sensitivity. Averaged over the 2135 cloudy profiles at Yan’an, the brightness-temperature Jacobians with respect to hydrometeor water paths show pronounced sensitivity to liquid hydrometeors in the lower troposphere, with cloud-liquid-water Jacobians peaking around 600 hectopascals and rainwater Jacobians closer to the surface near 700 hectopascals. Sensitivity to snow is much weaker, with only modest responses at altitudes of five to nine kilometers, and cloud ice sensitivities are approximately one order of magnitude smaller than those for snow, showing a weak bimodal structure. The response peaks of different channels occur at nearly the same altitudes and differ mainly in magnitude, indicating that ground-based radiometer observations provide limited vertical discrimination of hydrometeor effects, a constraint that retrieval designers must respect.

Precipitation remains the stubborn frontier. During rain events, RMSE values in the water-vapor absorption band reach 30 to 40 kelvin, with mean departures approaching 30 kelvin, reflecting rain-contaminated measurement effects and atmospheric hydrometeor uncertainties. These large departures cannot be attributed solely to radiative transfer errors: the simulations represent atmospheric rainwater absorption and scattering, whereas wet-radome effects and other near-instrument processes remain in the observations and are neither corrected nor represented in the model. Water deposited on the radome alters its effective microwave transmission and emission, while raindrop impacts and evaporation near the instrument introduce further perturbations. Because atmospheric and instrument-related contributions cannot be separated using departure statistics alone, the precipitation cases serve to characterize the practical magnitude of rain-contaminated departures rather than to provide strict model validation. Rain-affected observations should be used in retrieval and data assimilation only after applying rain-specific quality control, wet-radome corrections, and appropriate observation-error characterization. Diurnal analysis of the departures also revealed systematic patterns, with water-vapor channels weakly positive overnight and increasing to one to three kelvin in the early morning, while oxygen-band channels reached midday minima of roughly minus six to minus three kelvin and evening maxima of three to six kelvin, patterns that likely reflect combined contributions from ERA5 representation errors, residual radiative-transfer errors, and temperature-dependent instrumental or calibration effects.

The implications reach well beyond model validation. ARMS-gb v2.0 now provides a fast, accurate forward-model framework with tangent-linear and adjoint capabilities under all-sky conditions, supporting direct assimilation of ground-based radiometer observations in variational systems. Future work will extend instrument coverage to higher-frequency systems, improve the quantitative characterization of atmospheric and instrument-related uncertainties during precipitation, and integrate the model into one-dimensional variational retrieval and data-assimilation frameworks. For a field that has long treated clouds as a reason to discard data, the message is clear: with the right physics, cloudy skies are no longer a blind spot but a source of information.

Subject of Research: Development and validation of ARMS-gb v2.0: extending fast radiative transfer modeling capability to all-sky conditions for ground-based microwave radiometer retrievals

Article Title: Development and validation of ARMS-gb v2.0: extending fast radiative transfer modeling capability to all-sky conditions for ground-based microwave radiometer retrievals

Article References: Huang, Z., Shi, Y.-N., Weng, F., & Yang, J. (2026). Development and validation of ARMS-gb v2.0: extending fast radiative transfer modeling capability to all-sky conditions for ground-based microwave radiometer retrievals. Geoscientific Model Development, 19(20), 9957-9972. https://doi.org/10.5194/gmd-19-9957-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-9957-2026

Keywords: microwave radiometer, radiative transfer, ARMS-gb, all-sky, cloud scattering, data assimilation, boundary layer, numerical weather prediction, Development, validation, extending, fast