From cultural landscapes to forest expansion: 70 years of habitat change in the European Alps

Mountain areas host several complex ecosystems that are not immune to biodiversity loss. In the European Alps, traditional agro-pastoral activities have shaped semi-natural cultural landscapes with unique ecological features, such as terraced landscapes and wood pastures. The peak in the abandonment of traditional practices in the last century led to intense natural forest expansion on montane slopes. This translated into changes in habitat structures resulting in the homogenisation of landscape features and the loss of open areas. These dynamics have rarely been investigated at appropriate ecological scales, as existing studies are constrained by trade-offs between spatial and temporal extent and resolution, and often rely on non-harmonised data.

The aim of this study is to analyse the trajectories and drivers of habitat changes across the Alps over the last 70 years, using three temporal steps, i.e., the 1950s, 1980s, and 2020s. We hypothesise that: i) natural forests expansion patterns are mostly associated with gap filling between the 1950s and the 1980s, but with treeline upshift in the last 40 years; ii) grasslands have decreased over the entire time period, but primary successions on unvegetated areas have recently occurred at higher elevations; iii) the relative importance and direction of effect of drivers differ between time periods and regional administrations, reflecting land-use legacies, history, and territorial policies.

A database of 393 historical aerial photographs from the 1950s and the 1980s, and satellite images from the 2020s was assembled for 138 alpine landscapes, each with an area of 9 km², encompassing more than 1000 km². Landscape areas were selected between 1000 and 3000 m a.s.l. to represent the topographic, climatic, and socio-economic diversity of the montane, subalpine, and alpine belts of the Alpine region. Images were obtained from multiple national and regional geoportals, orthorectified when needed, and harmonised at a spatial resolution of 1m. Deep-learning architectures were trained to classify the landscapes into 10 land-cover classes, including grasslands, croplands, forests, unvegetated or anthropic areas, and shadows. A semi-automatic post-processing procedure was implemented to ameliorate classification results. The spatiotemporal trajectories of habitat changes were assessed through landscape and class metrics, as well as morphological spatial pattern analysis. A machine-learning approach was adopted to quantify the importance and direction of effect of several topographic, socioeconomic, soil, climatic, and vegetation drivers.

Preliminary results confirmed the hypotheses regarding habitat transitions over time. Forests have increased everywhere, leading to a widespread landscape homogenisation dominated by closed-canopy habitats. By further investigating the evolution of spatial patterns of grasslands and open areas, we seek to better decipher landscape and habitat changes at different elevations. Moreover, the identification of the most relevant socio-ecological factors shaping semi-natural cultural landscapes can inform biodiversity conservation and rewilding agendas.