Regional-scale InSAR investigation and landslide early warning thresholds in Umbria, Italy

In this work, we use conventional two-pass differential satellite interferometry for the investigation of active landslides at regional scale in the Umbria Region of Italy. First, we use InSAR for the detection of active slope movements during the study period (October 2019 to January 2021). Then, we characterize the displacement rates and their variation through time for selected active slopes, by analysing 6-days interferograms. The evolutionary trends of the phenomena are compared to the pertaining climate and soil parameters that are used by the regional landslide early warning system (LEWS) to predict the possibility of landslide occurrence over the territory. We aim to explore the potential of differential interferometry to detect and monitor active landslides and, through that, to validate a geographical LEWS: Through the analysis of stacked interferograms, we identified 256 InSAR deformation signals (IDS) corresponding to active slope movements. Their distribution is compared to the geological map and the landslide inventory, illustrating how active landslides are favoured by weak lithologies and pre-existing slope instability. Also, the relative orientation of the satellite's line of sight with respect to the slope influences the results indicating a bias in the completeness of remotely sensed data. However, in the context of land management and civil protection, the results constitute a valuable information dataset for plans managed by local authorities. The analysis, at the interferogram scale (6-days), of 13 clearly recognizable IDSs illustrates the dynamics of the slopes through the variation of the displacement rates. Due to decorrelation, only about half of the interferograms can be used for this purpose, and interpretation is required. The comparison to the alert thresholds of the Umbria region shows the remarkable relationship between InSAR-derived landslide activity and the parameters used for landslide prediction (i.e., 48 h rainfall and soil moisture conditions). The maximum 48 h cumulative rainfall and the soil saturation index prove effective to predict landslide reactivations and accelerations. We provide uncertainty measures of the landslide prediction through binary classification metrics. Such results are only preliminary, given the limited number of landslides and timespan of our dataset. However, it represents one of the few attempts to validate the forecast of a regional LEWS quantitatively. © 2023 The Authors