Evaluating historical, basin-wide landslide activity in a context of land abandonment and climate change: Effects of landslide visibility and temporal resolution

Drainage basins of the Northern Apennines, particularly in the clayey settings, bear among the highest rates of landsliding worldwide. A history of major land cover changes has left a landscape characterized by sparse, coppice-managed forest, transitional shrubs, and actively eroding badlands. Historical trends of landslide occurrence are examined in the Sillaro River basin (139 km2) in relation to land cover and climatic changes. To this purpose we have compiled a multi-temporal (1954-2018) landslide inventory (n = 1164) across twelve sequential photo sets that bears decadal (7- to 15-yr) and finer (2- to 6-yr) temporal resolution respectively before and after 1996. To account for changes in meteorological forcing, we examine: (i) the total annual precipitation (PRCPTOT); (ii) the annual maximum daily precipitation (RX1day); and (iii) the precipitation fraction (R99pTOT) due to extremely wet days. We find that landslide activity is strongly controlled by lithology, with landslide densities in claystones 3-to-4 times higher than in marl-sandstone alternations. This difference is chiefly associated with badlands, which are the most active land cover type and where new scars at a site could recur up to nine times. To evaluate the influence of varying temporal resolution on inventory completeness, hence on inference about land cover and climatic effects, we constrain the time scales of landslide visibility and assess the relative rates of undersampling. We find that visibility functions decline non-linearly with time, and that an inventory compiled at 5-year resolution would be missing up to 20 % of the landslide scars, with the size of an additional 27 % that would be underestimated due to revegetation. Overall, detection of entire landslide scars, which varies with land cover, becomes rare after 13 years in transitional shrubs, and after 17 years in badlands and managed forest. The historical analysis shows that landslide count: (i) increases in 1955-1976, a period of maximum anthropogenic pressure and wetter conditions; (ii) decreases steadily from 1977 through 2000, during a phase of land abandonment and decline in annual precipitation; and (iii) grows highest in 2000-2014, a period of land cover stability characterized by lesser precipitation although increasingly focused on high-magnitude events. To evaluate the likely reason of this recent increase in landsliding (i.e., R99pTOT vs inventorying resolution), we replicate the post-1996 mapping at coarser resolution. In the simplified inventory, landslide densities drop up to a factor of 2, and the inverse correlation originally linking landslide count with R99pTOT, loses significance. We conclude that, when the bias associated with varying inventorying resolution is removed, dependencies previously attributed to climatic effects become drastically reduced, and in some instances can even disappear.