This study aimed at evaluating and mapping Ground Subsidence Susceptibility (GSS) in the Grosseto plain (Tuscany Region, Italy) by exploiting multi-temporal satellite InSAR data and by applying two parallel approaches; a bivariate statistical analysis (Frequency Ratio) and a mathematical probabilistic model (Fuzzy Logic operator). The Grosseto plain experienced subsidence and sinkholes due to natural causes in the past and it is still suffering slow-moving ground lowering. Five conditioning subsidence-related factors were selected and managed in a GIS environment through an overlay pixel-by-pixel analysis. Firstly, multi-temporal ground subsidence inventory maps were prepared in the study area by starting from two inventories referred to distinct temporal intervals (2003–2009 and 2014–2019) derived from Persistent Scatterers Interferometry (PSI) data of ENVISAT and SENTINEL-1 satellites. Then, the susceptibility modelling was performed through the Frequency Ratio (FR) and Fuzzy Logic (FL) approaches. These analyses led to slightly different scenarios which were compared and discussed. Results show that flat areas on alluvial and colluvial deposits with thick sedimentary cover (higher than 20 m) on the bedrock in the central and eastern sectors of the plain are the most susceptible to land subsidence. The obtained FR- and FL-based GSS maps were finally validated with a ROC (Receiver Operating Characteristic) analysis, in order to estimate the overall performance of the models. The AUC (Area Under Curve) values of ROC analysis of the FR model were higher than the ones of FL model, suggesting that the former is a better and more appropriate predictor for subsidence susceptibility analysis in the study area. In conclusion, GSS maps provided a qualitative overview of the subsidence scenarios and may be helpful to predict and preliminarily identify high-risk areas for environmental local authorities and decision makers in charge of land use planning in the study area. Finally, the presented methodologies to derive GSS maps are easily reproducible and could also be applied and tested in other test sites worldwide, in order to check the modeling performance in different environmental settings.
Read full abstract