ADVANCED SPECTRAL RETRIEVALS FOR AUTOMATIC MONITORING OF DATA QUALITY AND SYSTEM STATUS OF GROUND-BASED RADIOMETERS

H. Czekala1

¹RPG Radiometer Physics GmbH

Ground-based microwave radiometers provide data with high time-resolution already to a couple of operational networks around the world. Direct assimilation of brightness temperature (TB) is gaining focus beyond the Level-2 atmospheric data created by retrieval methods. While quality control and the ability to self-diagnose has always been a strong demand in the research community, the shift to operational networks emphasizes the need for fully automatic diagnostics of the radiometer’s health status. Beyond instrumental status, any adverse conditions in the atmosphere which render the data unfit for any retrieval of reliable meteorological information need to be detected as well.

Beyond simplistic range checks and plausibility tests, there exist sophisticated methods to compare observations to expected behavior by “observation minus background” statistics. Such O-B statistics require elaborate data processing involving modeled atmosphere and simulated measurements derived thereof. The advantage of the presented approach is to provide real time analysis of the radiometer system with neither time delay nor using external data or computer resources beyond the inherent capabilities of the observation system itself. The performance evaluation relies on a retrieval product which uses the observed microwave brightness temperature data of a multi-channel radiometer as input and equally as output. Ideally, this self-consistent solution should not generate differences. Differences between retrieval output to input indicate that problems with either the suitability of the radiative transfer model to the observed atmosphere, or problems with the technical integrity of the radiometer system. In both theoretical and practical evaluation, the observed spectral difference between measured TB and retrieved instrumental-TB (TB minus INS) allows detection of data which is not fit for evaluation.

Deviations due to the inability to precisely model the observed brightness temperatures from the atmospheric state include the bias due to errors of gas absorption models, but also the modeling of hydrometeors. The numerical model for radiative transfer used in this study utilizes the Rosenkranz-2024 gas absorption model. The absorption cross sections for hydrometeors are modeled according to Lorenz-Mie theory. Comparison with high-quality observation data reveals less than 0.5 K bias in a few channels even in light to moderate rain. In case of heavy rain, the deviations increase and show a systematic and monotonic trend in frequency. The treatment of scattering particles, which was neglected in the radiative transfer simulation, leads to large biases. Although the TB readings are technically correct, this observed spectral difference reveals conditions in which the retrieved atmospheric data is no longer accurate.

When technical problems occur, like channel dropouts, receiver instability in certain channels, or radio frequency interference (RFI) by artificial signal sources, the self-consistent spectral difference indicates strong deviations in the affected channel. This strong difference is partially counterbalanced by differences with opposing sign in neighboring channels. Such behavior clearly indicates that the measured TB vector is not a physically consistent solution within the modelled database. The spectral pattern of such deviations is different to the spectral dependence originating from insufficient hydrometeor modeling. Therefore, it is possible to separate quality disturbance by heavy rain from technical malfunctions of the radiometer.

For the best performance of the quality monitoring by self-consistent spectral retrievals, it is mandatory to minimize bias errors in the radiative transfer models. Any systematic bias in specific channels, or at certain climate subregions, will masquerade the deviations which indicate a low performance of the observation. Future improvements in our radiative transfer model shall improve the microphysical representation of hydrometeors. Such efforts will extend the range of rain rates at which the quality assessment still yields sufficiently small deviations in the self-consistent spectral difference. In consequence, the applicability of unbiased Level-2 data retrieval will also be extended.