Dottorato in INFRASTRUTTURE E TRASPORTI

Titolo della tesi: CloudSat based Assessment of ATMS Snowfall Detection Capabilities

The aim of this PhD activity is the assessment of the micro-wave cross-track scanner Advanced Technology Microwave Sensor (ATMS) snowfall detection and retrieval capabilities based on CloudSat CPR observations and the development of a snowfall detection and retrieval algorithm for ATMS. Precipitation measurement is a challenging question in the atmospheric sciences. Ground-based instruments, such as rain-gauges and ground-based radars, provide precise measurements but they are limited in their spatial coverage, often scarce or sparse over remote areas such as Polar regions and unavailable over the sea. On the contrary, space-borne sensors provide indirect measurements of the precipitation parameters but offer wide spatial coverage and a good temporal resolution. In particular, Passive Microwave (PMW) radiometric data are very effective for remote sensing of weather and precipitation phenomena because the upwelling radiation is responsive to the precipitation structure. In this context, snowfall detection and quantification are challenging tasks in the atmospheric science field. Snowfall precipitation represents only 5 % of the total precipitation, but it is predominant above 60-70 °N/S. These areas are among the most affected by global climate change, and therefore it is important to develop methods to obtain a global snowfall quantification, with a particular focus on the high latitude areas. The snowfall detection using satellite microwave data is based on the scattering effect of the snowflakes visible using the high-frequency channels (>80 GHz) on the upwelling radiation; however, the detection is made difficult by the weakness of this signature. This phenomenon is particularly evident at the high latitudes, where the coincidence of the prevalence of very light snow precipitation events and cold/dry environmental and boundary conditions - low humidity level, low temperature, presence of snowpack over the background surface - make the detection very difficult. In this context, the development of the last generation microwave radiometers - such as the ATMS – characterized by a wide range of frequency channels allows to exploit the multi-channel microwave observations for both a radiometric characterization of the background surface - using the low-frequency channels- and the detection and the retrieval of the snowfall precipitation - using the high-frequency channels. Moreover, the use of the so-called observational “coincidence datasets” - i. e. datasets built from the coincident observations (in time and space) of a space-borne microwave radiometer and a space-borne cloud and precipitation radar - allows to analyze in a direct way the relationship between the vertical cloud structure - and the related precipitation rate- observed by the radar and the microwave observations obtained by the microwave radiometers, without using simulated cloud-radiation databases. Since the multi-channel microwave observations generally have a non-unique response to precipitation intensity, advanced statistical methods are used for precipitation detection and retrieval algorithms, such as Bayesian methods and the new machine learning methods. The research focused on the ATMS instrument because: 1) it is carried by near-polar orbiting satellites, providing global coverage and, in particular, an high temporal resolution over the high latitude areas; 2) it is equipped with several channels (from 23 GHz to200 GHz) suitable not only for snowfall retrieval but also for the characterization of the frozen background surface; 3) it is onboard operational satellites guaranteeing to continue observations for the next decades; 4) it has similar characteristics to the European Microwave Sounder (MWS) which will be one of the main PWM radiometers onboard the Meteorological Operational Second Generation (MetOp-SG) satellite mission, planned for 2024. The research activity has been based on a coincidence dataset between ATMS and the 94 GHz Cloud Profiling Radar, carried by the satellite Cloud-Sat. This instrument has proven to be the most sensitive for a global snowfall estimation. In particular, the extreme conditions typical of high latitudes - low humidity levels, low temperatures and presence of snowpack on the background surface, light snowfall events - have been analyzed. These conditions are typical of most snowfall events at a global scale and are characterized by an ambiguous and hardly detectable signal, so the characterization of the radiometric signal derived from the background is fundamental for robust detection and retrieval of the snow precipitation. During the research activity, the Passive Empirical cold Surface Classification Algorithm (PESCA) has been developed to exploit ATMS low-frequency channels for background surface characterization. The algorithm performances, estimated using the AutoSnow product as reference, appears to be very robust in discriminating sea ice from open water and snow-covered land from snow-free land (POD=0.98 and FAR=0.01 for the sea module, POD=0.98 and FAR=0.01 for the land module). However, the radiometric characterization of the surfaces shows some limits, because, especially for snow cover classes, the uncertainty linked to the estimated spectra results to be too high. Therefore, a refinement process based on Self Organizing Maps and Linear Discriminant Analysis to obtain better emissivity estimations has been developed. An analysis of the MW snowfall signature has been carried out. A comparison between the observed MW signal in presence of snowfall and in clear-sky conditions shows the high ambiguity and the strong dependence on environmental and boundary parameters of MW snowfall signature. In addition, a comparison between the simulated clear-sky signal - based on the emissivity spectra obtained with the PESCA refinement process - and the observed signal in presence of snowfall shows the difficulty to discriminate the snowfall signature from the noise in the simulated clear sky TBs, that in cold and dry conditions can be relatively high. Finally, the High lAtitude sNow Detection and rEtrieval aLgorithm for ATMS (HANDEL-ATMS) has been developed. This algorithm is composed of four modules (for snow water path detection and retrieval and surface snowfall rate detection and retrieval) and has as inputs ATMS TBs and the differences between simulated clear-sky TBs (obtained from PESCA) and ATMS observed TBs. It shows very good detection capabilities (POD=0.85, FAR=0.15 and HSS=0.7 for SWP module, POD=0.84, FAR=0.17 and HSS=0.69 for SSR detection) and manages to detect snowfall also in very cold/dry conditions and for low snowfall rates. Retrieval error statistics show an overestimation of very light snowfall events, but a good agreement for more intense events. The algorithm can be applied only to the typical environmental conditions of high latitudes - dry atmosphere, low temperature, snow-covered background surface. So, the development of a global algorithm could be obtained by the integration with other approaches.