Retrieving atmospheric precipitation from synthetic aperture radar imagery at X and Ka bands for high-spatial resolution hydrometeorological applications

The use of spaceborne microwave synthetic aperture radars (SARs) is becoming a well‐established tool in several Earth remote sensing disciplines. This is mainly due to the SAR versatility, its global coverage and its capability of high spatial resolution imaging (on the order of meters) in almost all‐weather conditions. Indeed, for frequencies above C‐band, the impact of precipitating clouds may significantly impair the backscattered signal from the ground (e.g. Ferrazzoli and Schiavon 1997). However, it must be recognized that the chaotic and intermittent nature of atmospheric precipitations, associated with a low orbit duty cycle of space‐based SARs, and a limited coverage, limits the probabilistic ccurrence of precipitation signature on SAR image acquisitions (Danklmayer et al. 2009). Nonetheless, the impact of precipitating clouds on both amplitude and phase‐compressed SAR signals cannot be neglected, as recently reassessed by using Xband SARs (X‐SAR) currently in orbit, such as COSMOSkyMed (CSK) and TerraSAR‐X (TSX) (e.g. Marzano et al. 2010, Baldini et al. 2014). This work aims to propose innovative and enhanced XSARs products for final users in the field of hydrology and water‐related risk management. The basic idea is to exploit the high performance of the recent X‐SARs in terms of spatial resolution, revisit time and global coverage, to assessing the impacts of the atmospheric precipitations in the post‐event scenarios as well as in a pre‐warning configuration. Post‐event scenario analyses are more related to automatic recognition of flooded areas and its quantification whereas the prewarning stage, on the other hand, is related to the analysis of heavy atmospheric precipitation, which can cause floods and then make necessary the activation of post‐event analysis. X‐SARs give an unprecedented opportunity to final users (e.g. hydrologists) to test observations of rain fields at catchment scale and at a spatial resolution as never before, into their models for flood forecasting. Nevertheless, research is still at an early stage and several issues have to be addressed. Several quantitative retrieval algorithms of rainfall precipitation from SAR acquisitions have been presented (Marzano et al. 2010 and 2011). In this work we will apply an enhanced retrieval algorithm which can take into account the a‐priori information of the wet‐land covered areas. Indeed, the precipitation estimation by means of spaceborne X‐Band SARs could be improved by means of a pre‐processing step in which precipitating areas are distinguished by flooded and permanent water surfaces or wet snow absorption (Schellenberger et al. 2012). All these areas look dark (low backscattering) in the SAR images, so they might overlap each other in terms of radar signature then causing an overestimation of SAR‐detected flooded areas or an underestimation of SAR detected rain precipitation. The identification of such areas is then critical and necessary both for floods analysis and 61 precipitation applications. To tackle this problem, we have proposed an automatic method to distinguish, in an X‐band SAR image, water surfaces (either flooded or permanent water bodies) from artifacts due to heavy precipitation and wet snow (e.g. Pulvirenti et al. 2012 and Mori et al. 2012). The algorithm is mainly based on image segmentation techniques and fuzzy logic. Ancillary data, such as a local incidence angle map, a land cover map, and an optical image are also used. The performances of the SAR‐based proposed retrieval methodology have been evaluated through simulated datasets also. For this purpose we have developed a multiband fully‐polarimetric spaceborne SAR‐response 2D numerical simulator, derived from the extension of the SAR model firstly described in Weinman and Marzano 2008. The proposed model framework accounts for the SAR slant observing geometry and it is able to characterize the fully‐polarimetric response both in amplitude and phase. Ground surface can be given through semi‐empirical models (e.g. Oh et al. 2002) or canonical targets, while the atmospheric hydrometeor distributions through realistic highresolution Cloud Resolving Models (e.g. Blossey et al. 2007) or simplified synthetic scenarios. The application of this methodology to CSK and TSX study cases, acquired within the European FP7 project Earth2Observe, shows an appreciable improvement of the estimated precipitation maps. The list of the case studies we have selected is mainly dependent on the availability of ground‐based rain data (e.g. from weather radars or raingauge networks) and includes the TSX cases: Orleans (F) March 16, 2008; Mississippi (USA) April 18, 2008; Florida (USA) August 8 and 19, 2008; Louisiana (USA) September 2, 2008; and the CSK Liguria (I) November 4‐8, 2011. In order to investigate the potential of Ka SARs, the SAR‐response simulator has been recently applied, within an ESA project, for a feasibility study of a spaceborne Ka‐SAR system (e.g. Mori et al. 2015). Indeed, Ka‐band shows a very high sensitivity to atmospheric hydrometeors and this offers interesting possibilities towards atmospheric precipitations analysis. In this work simulated Ka‐SAR numerical scenarios will be also discussed.