Hedgerow mapping with high resolution satellite imagery to support policy initiatives at national level

Hedgerows provide habitat and food for a wide range of species and play a crucial role for biodiversity in agricultural landscapes. In addition, hedgerows render an important carbon stock, above and below ground, and protect agricultural soils from erosion. However, comprehensive, standardized and area wide information regarding the distribution of hedgerows is often lacking, which makes it hard to incorporate them in nature conservation plans and national carbon balance models. We evaluate the potential of high-resolution PlanetScope multitemporal satellite data and semantic segmentation approaches to map the distribution of hedgerows across the entire agricultural landscape in Germany. Based on a comprehensive set of independent reference data from the federal state of Schleswig-Holstein, we evaluate the performance of different loss functions and different combinations of spectral and temporal input feature sets. We assess the transferability of the final model using independent test data from three additional German Federal states. Additionally, we compare our results against the Copernicus Land Monitoring Service High Resolution Layer Small Woody Features, and a recently published biomass map of trees outside forests. All loss functions tested offered similar performance, but the binary-cross entropy function allowed for overcoming sensor artifacts to some extent. Visible and near-infrared imagery from all four monthly mosaics (April, June, August and October) of PlanetScope data was found to yield better results (F1-score 0.65) than different combinations of months and only red-green-blue inputs. We estimate a total surface of 4081 (± 1425) km2 of hedgerows across Germany, which represent 2.3 % of the agricultural land in Germany. By combining our results with a digital landscape model, we reveal heterogenous estimates of hedgerow height across municipalities. Our findings highlight that semantic segmentation approaches are well-suited for area-wide hedgerow mapping, especially in combination with multitemporal high-resolution satellite data. Furthermore, we underscore the relevance of using application-specific models over post-processing existing products, and provide for the first time a spatially explicit and comprehensive overview of the distribution of hedgerows and their structure across agricultural landscapes in Germany. Our methodology and product can be incorporated into landscape biodiversity models, carbon balance estimations and soil protection policies at national, regional and local scale.