Agricultural landscape change and its environmental challenges in Catalonia (1945–2023): a field-level deep learning approach using airborne imagery

Abstract

Abstract Agricultural landscapes in Catalonia (northeastern Spain) have undergone profound transformations since 1945, driven by major social, economic, and environmental changes, ranging from the post-Civil War period and Francoist developmental policies to market liberalization and integration into the European common market. To assess these long-term changes, we analyzed five time periods (1945, 1956, 1970s, 2004, and 2023) in two counties with contrasting socio-environmental contexts: the rural Banyoles Lake region and the highly urbanized Vallès Plain. Using black-and-white and red-green-blue high-resolution airborne imagery, we implemented a deep learning workflow to classify land uses and delineate agricultural fields, adapting the Segment Anything Model for field boundary detection. Our results reveal divergent regional trajectories. Agricultural land in the Vallès Plain declined threefold between 1945 and 2023, whereas the Banyoles Lake region experienced a more moderate reduction of approximately 30%. Woody crops also decreased substantially in both regions, along with their associated carbon stocks, with overall reductions of 68% in the Vallès Plain and nearly 90% in the Banyoles Lake region. In contrast, mean field size increased substantially in both regions, accompanied by a loss of geometric regularity and perimeter-to-area ratio. These patterns are consistent with agricultural intensification, mechanization, declining crop diversity, and land consolidation processes initiated in Spain in the mid-twentieth century. These transformations have important environmental implications. Larger and more homogeneous fields reduce edge density, with negative consequences for biodiversity. Furthermore, the decline of woody crops reduces the carbon sequestration capacity of agricultural landscapes and weakens their contribution to climate change mitigation.