InSAR stacking to detect active landslides and investigate their relation to rainfalls in the Northern Apennines of Italy

Interferometric stacking is a useful technique to detect active slope deformation over vast mountainous area. Compared to the analysis of single interferograms, the stacking approach improves the signal to noise ratio and deformation signals become clearer. In this work, we used stacked interferograms to detect active slow-moving landslides during the years 2015 to 2019 over a 1200 km2 portion of the Northern Apennines of Italy. C-band Sentinel 1 SAR images were used to create short temporal baseline (6 to 24 days) interferograms which limit decorrelation and maximize the range of measurable displacement rates. Then we operated a further selection of interferograms based on coherence and visual inspection, before stacking over multiple time scales (1 month to years). The products of the analysis are ground displacement maps where the presence of residual noise is inversely proportional to the duration of the stack and the deformation signal is clearly recognized. Results show that only a small fraction of the mapped landslide deposits are experiencing deformation that can be detected by differential interferometry. In particular, we identified 118 InSAR deformation signals corresponding to ongoing gravitational slope deformations over the 9916 landslides mapped in the area. Active movements are mostly located on landslides that have undergone catastrophic reactivation in relatively recent times. Annual interferometric stacks proved better suited to detect active slope movements, while <15 % of our deformation signals can only be detected by inspecting monthly stacks. Active slow-moving landslides show variable displacement rates in monthly stacks. Periods of dormancy alternate to accelerations that may lead to actual catastrophic failures or, more often, determine finite periods of sustained slow movement before the displacement rates drop below the detection limit. We compared the evolutionary trends of the phenomena to the occurrence of rainfall events. For this purpose, we use the probability of landslide occurrence which is associated with a rainfall event based on a territorial alert threshold. Results show that, at increasing landslide occurrence probabilities, an increasing fraction of actively deforming landslides can be detected by InSAR.