Comment on amt-2023-94

Abstract

<strong class="journal-contentHeaderColor">Abstract.</strong> Snowfall detection and quantification are challenging tasks in the Earth system science field. Ground-based instruments have limited spatial coverage and are scarce or absent at high latitudes. Therefore, the development of satellite-based snowfall retrieval methods is necessary for the global monitoring of snowfall. Passive Microwave (PMW) sensors can be exploited for snowfall quantification purposes because their measurements in the high-frequency channels (&gt; 80 GHz) respond to snowfall microphysics. However, the highly non-linear PMW multichannel response to snowfall, the weakness of snowfall signature and the contamination by the background surface emission/scattering signal make snowfall retrieval very difficult. This phenomenon is particularly evident at high latitudes, where light snowfall events in extremely cold and dry environmental conditions are predominant. ML techniques have been demonstrated to be very suitable to handle the complex PMW multichannel relationship to snowfall. Operational microwave sounders on near-polar orbit satellites such as the Advanced Technology Microwave Sounder (ATMS), and the European MetOp-SG Microwave Sounder in the future, offer a very good coverage at high latitudes. Moreover, their wide range of channel frequencies (from 23 GHz to 190 GHz), allows for the radiometric characterization of the surface at the time of the overpass along with the exploitation of the high-frequency channels for snowfall retrieval. The paper describes the High lAtitude sNow Detection and Estimation aLgorithm for ATMS (HANDEL-ATMS), a new machine learning-based snowfall retrieval algorithm developed specifically for high latitude environmental conditions and based on the ATMS observations. HANDEL-ATMS is based on the use of an observational dataset in the training phase, where each ATMS multichannel observation is associated with coincident (in time and space) CloudSat Cloud Profiling Radar (CPR) vertical snow profile and surface snowfall rate. The main novelty of the approach is the radiometric characterization of the background surface (including snow covered land and sea ice) at the time of the overpass to derive multi-channel surface emissivities and clear-sky contribution to be used in the snowfall retrieval process. The snowfall retrieval is based on four different artificial neural networks for snow water path (SWP) and surface snowfall rate (SSR) detection and retrieval HANDEL-ATMS shows very good detection capabilities - POD = 0.83, FAR = 0.18, and HSS = 0.68 for the SSR detection module. Estimation error statistics show a good agreement with CPR snowfall products for SSR &gt; 10^&minus;2 mm h^&minus;1 (RMSE 0.08 mm h^&minus;1, bias = 0,02 mm h^&minus;1). The analysis of the results for an independent CPR dataset and of selected snowfall events evidence the unique capability of HANDEL-ATMS to detect and estimate SWP and SSR also in presence of extreme cold and dry environmental conditions typical of high latitudes.