MACHINE LEARNING ENHANCED SNOWFALL RATE RETRIEVALS FROM PASSIVE MICROWAVE SENSORS
Marzo 25, 2026IMPLEMENTATION OF SELF-SIMILAR RAYLEIGH-GANS APPROXIMATION INTO PASSIVE MICROWAVE RADIATIVE TRANSFER SIMULATIONS
Marzo 25, 2026B. Pospichal1, T. Marke1, T. Böck1, U. Löhnert1
1University of Cologne
Within ACTRIS—the European Research Infrastructure for Aerosol, Clouds, and Trace Gases—a network of about 25 microwave radiometers (MWR) has been established across Europe, together with Doppler cloud radars, ceilometers and Doppler lidars. The goal of ACTRIS is to provide continuous and comprehensive observations of cloud properties and water vapor, harmonized across the entire network. The resulting data are comparable between sites and are used for various applications, such as process studies, model evaluation, or satellite validation. To ensure high-quality data, ACTRIS places a strong emphasis on rigorous quality control and continuous instrument monitoring.
MWRs have been operating continuously at many sites throughout the world and have proven to be a valuable tool for observing cloud liquid water, integrated water vapor, and temperature profiles. However, for high quality observations and data assimilation, it is essential to understand and quantify the uncertainties associated with state-of-the-art MWRs.
Regarding uncertainties for MWR observations, we need to take into account two sources of errors: first the instrument related errors, i.e. the errors of the observed brightness temperatures, and second, the error of retrieval products.
Two main sources of uncertainty must be considered in MWR observations: (1) instrument-related errors, such as inaccuracies in the observed brightness temperatures, and (2) retrieval-related errors when deriving atmospheric parameters from the raw measurements. To assess instrument-related uncertainties, we analyzed long time-series of co-located MWR observations, focusing on radiometric noise, long-term drifts and jumps, calibration repeatability, systematic biases between instruments, and radome degradation. Within ACTRIS, we implement spectral consistency checks, long-term calibration monitoring, and statistical comparisons with models (observation-minus-background).
Another key development within ACTRIS is the generation of harmonized atmospheric products from MWR observations across the entire network. Retrieval algorithms—currently under development—are based on statistical approaches (neural networks), using model reanalysis profiles as input.
In this presentation, we will first give a short overview of the latest developments within the ACTRIS network and MWR data processing. We will present an overview of the uncertainty assessment, quantify the associated errors, and propose mitigation strategies. Firnally, we will provide an overview of the common retrieval framework along with associated uncertainty estimates.
