The influence of forest cover on landslide occurrence explored with spatio-temporal information

Multi-temporal landslide inventories in widely forested landscapes are scarce and further studies are required to face the challenges of producing reliable inventories in woodland areas. An elaboration of valuable empirical relationships between shallow landslides and forest cover based on recent remote sensing data alone is often hampered due to constant land cover changes, differing ages of landslides within a landslide inventory and the fact that usage of different data sets for mapping might lead to various systematic mapping biases. Within this study, we attempted to overcome these difficulties in order to explore the effect of forest cover on shallow landslide occurrences. Thus, forest dynamics were examined on the basis of 9 orthophoto series from 1950s to 2015, distinguishing 3 forest classes, based on the wood type. These classes were furthermore distinguished in 12 subclasses, considering stand density and age. A multi-temporal landslide inventory was compiled for the same period based on the aerial photography, 2 airborne LiDAR imageries, 8 field surveys and archive data. We derived topographical parameters (slope, topographical positioning index and convergency index) from the digital elevation model for areal correction and accounting for topographical confounders within a logistic regression model. Empirical relationships were assessed by means of (a) areal changes of forests and logged areas, (b) spatio-temporal distribution of shallow translational landslides, (c) frequency ratios and (d) logistic regression analysis. The findings revealed that forests increased by 16.2% from 1950s to 2015. 311 landslides of 351 in total that where mapped in total could be assigned to the observed time series and were considered for our analyses. Frequency ratios and odds ratios indicated a stabilising effect of all forest classes on landslide occurrences. Odds ratios observed for the models based on aggregated data sets (3 forest classes) indicated provided evidence that forest was constantly estimated to be less prone to slope failure than their non-forested counterparts. The chances for forest classes to be affected by shallow landslides were estimated to be considerably lower whenever topographic predictors were as well included in the model. A detailed inspection of the statistical results suggests that the obtained empirical relationships should be interpreted with care. Challenges in the mapping procedures of forests and landslides, implications of the applied methods and potential pitfalls are discussed.