Abstract
Grasslands provide key ecosystem services comprising food production, water supply and flow regulation, carbon storage, erosion control, climate risk mitigation, pollination, cultural amenity and support a high biodiversity. In the Alpine region, grasslands are used as meadows and pastures across a large elevational gradient, their management ranges from extensive to intensive, varies strongly in space and time and is often unknown for large areas. Due to the large coverage of grasslands and their provision of manifold ecosystem services, information about grassland management intensity is important for a range of stakeholders including agricultural agencies and nature protection. Within the Alpine Regional Initiative of ESA, we address those needs in the Eco4Alps project and develop a cloud-based operational grassland management service to map the timing and number of mowing events as an indicator for grassland management intensity for the Alpine region.
Optical EO-based mowing detection as a proxy for grassland management has gained considerable attention in the last years and several approaches have been developed (Griffiths, et al., 2020) which rely on the detection of mowing events based on an abrupt decline in the intra-annual vegetation signal. In general, there is consensus that a dense timeseries is significantly improving the grassland mowing event mapping performance. While the additional integration of radar data is challenging (De Vroey et al., 2021), the use of combined optical timeseries has shown promising results especially with harmonized Landsat/Sentinel products available at 30m (Griffiths et al., 2020) and 10m (Schwieder at al., 2021).
Whereas our service makes use of existing methods in the optical domain it particularly considers the peculiarities of the Alpine region with its complex topography, small-scaled structured landscapes and high and persistent cloud cover. We show the concept and assessments performed to establish the Alpine grassland mowing event service for the province of South Tyrol, Italy.
We use a curve-fitting approach to model the seasonal vegetation growth based on vegetation indices and identify potential grassland mowing events, where observations significantly differ from this idealized trajectory. These potential mowing events need to fulfill further criteria to be labeled as mowing events such as a minimum timespan to the preceding mowing event and a plausibility check based on elevation-dependent rules. We tested different Sentinel data products to assess the trade-off in spatial-temporal resolution for grassland mowing detection performance including Sentinel-2 only, the Harmonized Landsat-Sentinel product and the newly released sen2like fusion (Saunier et al., 2019) product, which offers data at 10/20 m resolution. We additionally evaluated how the choice of vegetation index affects the detection performance and tested different vegetation indices such as the NDVI and EVI. To validate our grassland mowing results, we setup a webcam-based database with visually interpreted mowing events of overall 300 grassland fields for the years 2017-2020. We applied the method on the pixel level and integrated the results to the parcel-level based on local cadaster information and derived spatial maps of mowing events for the time range 2017-2020.