Investigating the feasibility of using Sentinel-1 data to determine the damage state of Ukrainian territory = Indagare sulla fattibilità dell'utilizzo dei dati Sentinel-1 per determinare lo stato di danneggiamento del territorio Ucraino

Since February 24, 2022, the Ukrainian population and territory have been exposed to warlike events that have resulted in victims, destruction, and severe consequences on the European and international levels. Unfortunately, it will not be feasible to resurrect any of the victims, but it will most likely be possible to act in order to recover the damaged or completely destroyed buildings, infrastructures, and parts of urban realities. These interventions necessitate an assessment of the damaged areas in order to plan and fund the necessary actions in an acceptable manner. The maps produced by the United Nations Satellite Center (UNOSAT), part of the United Nations Institute for Training and Research (UNITAR), prove to be a useful support given the very high spatial resolution of the WorldView-3 satellite images to which they refer, which can reach 30cm of spatial resolution. However, these maps derive from rapid photo-interpretation operations mainly affected by human errors and involve economic costs. Research in the field of Remote Sensing therefore attempts to obtain damage maps using free images and innovative algorithms based on the use of classification indexes often compared with UNOSAT sources. The thesis work tries to respond to this general objective by using free Sentinel-1 high-resolution data derived from Synthetic Aperture Radar (SAR) sensors to create the damage maps in the context of the war in Ukraine taking as reference the UNOSAT maps. The choice of this imagery is appropriate since the average spatial resolution of 10m can be compared with the footprints of urban buildings and SAR technology, through the use of microwaves, allows for investigation the territory both during the day and night and in any weather conditions. The work examines three different sites in terms of extension, urban density, geographical position and concentration of damaged areas: the set of areas west of Kyviv, Mariupol and Chernihiv. The effectiveness of 3 polarimetric indices, 6 coherence indices and 6 intensity indices is examined, trying to achieve the secondary objective of identifying the element capable of discriminating the damaged from the undamaged in the best way. A new element in the specific field of damage analysis is represented by the idea of considering the definition of the coherence and intensity indices in a series of 15 calculation boxes increasing around the reference pixel. As a result, 183 rasters data were created for each study area, for a total of 549 rasters data evaluated. The examination of the polarimetric variables largely proved what is frequently stated in the literature about the inadequacy of Sentinel-1 data of the dual-pol type to fully characterize the scatterometric behavior of the investigated surfaces. The relevance of the information content was reduced in each of the three series of rasters data consisting of coherence and polarimetric variables by controlling the signal to noise ratio (SNR).The subsequent analysis of the principal components applied to the residual rasters of the SNR control enabled the generation of damage maps relating to the first principal component with an overall accuracy greater than 75% and to identify the RC VH index defined in the fourth calculation window as the best damaged/undamaged discriminant from a qualitative and quantitative point of view. Some inconsistencies with respect to the accuracy of the UNOSAT sources found in the union of maps of the area west of Kyviv in which elements simultaneously classified as damaged and undamaged were detected, leaving some room for debate about the certainty attributed to the sources comparison of this work and others of its kind.