Falling snow is a key component for the global atmospheric, hydrological and energy cycles, and its retrieval from space-based observations represents the best current capability to evaluate IT…

The University of Maryland, College Park received a $360,839 grant award (21-PMMST21-0086) from a civilian federal agency to conduct research on snowfall type classification for improving passive microwave falling snow estimation and retrieval. The project, which began on March 1, 2022 and concludes on December 31, 2026, addresses critical gaps in the Global Precipitation Measurement (GPM) mission's Goddard Profiling (GPROF) algorithm for detecting snowfall from space-based observations. Current precipitation retrieval algorithms struggle with certain snowfall types, particularly shallow convective snow, and suffer from contamination issues caused by snow-covered surfaces affecting passive microwave signals. The research aims to develop a classification system that can distinguish between different snowfall types and enable the GPROF algorithm to adjust its characteristics accordingly, potentially improving the accuracy and reliability of global snowfall detection critical for understanding Earth's energy and hydrological cycles.

This grant is structured as a research contract with no set-aside designation, reflecting the university's competitive technical qualifications in remote sensing and atmospheric science research. The work will be performed at the University of Maryland, College Park campus in College Park, Maryland, with the university operating through its Office of Research Administration & Advancement. As a research-intensive land-grant institution, the University of Maryland brings significant expertise in satellite-based monitoring systems and scientific research to address this complex algorithmic and data representativeness challenge within the GPM mission framework.

Falling snow is a key component for the global atmospheric, hydrological and energy cycles, and its retrieval from space-based observations represents the best current capability to evaluate IT globally. Precipitation retrieval algorithms have been developed and refined over the few last years and their accuracy and reliability are becoming increasingly more important for earth s energy and radiation budgets, as well as for human activities. Unfortunately, many issues still affect snowfall retrievals and algorithms performances cannot be considered accurate and reliable enough yet. In particular, previous work showed that, within the global precipitation measurement (GPM) mission, the goddard profiling (gprof) algorithm snowfall retrieval performances strongly depend on the snowfall type. Some type of snowfall, such as the shallow convective one, are detected with extreme difficulties and in most of the cases completely missed. Moreover, since the retrieval is based on passive microwave (PMW) data, the well-known issues of snow-covered surfaces contaminating the signal arise, inducing false alarms in the precipitation estimates. The main reasons for gprof behavior have been identified in the bayesian nature of gprof which relies on a-priori databases for its estimates. If the a-priori databases are not representative enough for a particular type of precipitation event, the estimates cannot actually find a perfect match and miss or underestimate precipitation. Is representativeness the problem or is non-uniqueness i.e., the same TB signature make mean something in stratiform snow but another in convective snow? Some other details of gprof (i.e. channel sensitivity, surface classification, etc.) could be modified to improve the retrievals and a classification flag that recognizes the snowfall type and makes the algorithm switching its characteristics would be particularly helpful. Many efforts have been made to partition rainfall between convective and stratiform, but very few efforts have been spent for snowfall and none of the methods developed can be applied to PMW retrievals. Given the importance of convective snowfall type both globally (i.e., northern atlantic region, sea of japan, antarctic belt, etc.) and locally (i.e. lake-effect snow over the great lakes), this proposal aims to partition snowfall type providing a flag able to activate a specific and dedicated setup of gprof. This will be achieved through development of a new deep neural network (DNN) model for snowfall class detection. Relying on the available cloudsat data, combined cloudsat-GPM dataset, and relevant ancillary products, GMI observations will be linked to a number of snowfall classes to offer a complementary information content to the existing PMW algorithms. In this effort, a newly developed GMI non-rain retrieval with surface-specific emissivity constraints will provide the necessary skill for precipitation detection. An evaluation exercise will be done, using part of the cloudsat dataset (mainly over ocean) and ground-based sensors (multi-radar multi-sensor dataset over US, finland and great lakes region radars).