Extensive Use of Permanent Grasslands
The use case focuses on the development of a methodology for detecting candidates of non-managed grassland parcels. This is crucial for the further decisions on meeting parcel eligibility criteria in various countries. Each demonstrative country, including the Netherlands, Sweden, and Spain (Castilla y Leon), defines its specific grassland management rules. For instance, in the Netherlands, grassland must be present for five years and managed without ploughing, while in Sweden, semi-natural pastures must be grazed yearly without mowing. Spain’s management involves restrictions on mowing frequency and grazing obligations.
Principle of the NDVI profile analysis: time series of NDVI (black crosses) is extracted from the source Sentinel-2 and Landsat-8 imagery. RBF function is used to interpolate and smooth the NDVI profile to the regular 1-day step (black line). Continuum (turquoise line) is fitted to the interpolated NDVI profile between start-of-season (green dotted line) and end-of-season (blue dotted line) dates. Feature extraction and analysis of their significance.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
The developed methodology primarily employs Sentinel-2 and Landsat-8/9 satellite imagery, specifically focusing on the Normalized Difference Vegetation Index (NDVI) profiles. The NDVI data was collected for multiple vegetation seasons (2020-2022), and additional datasets such as tree masks and geospatial data from the GSAA were also used to support the analysis. The NDVI profiles were extracted from the source data followed by using a radial basis function (RBF) to smooth and interpolate the data on a one-day step basis. A continuum was fitted to simulate a “non-management scenario” by forming a convex hull between the start and end of the vegetation season. Features detected within the NDVI profile indicating potential management events were assessed for significance using metrics like the Exponential Moving Average (EMA), MADC momentum, and relative depth to the continuum. The Gross Productivity Index (GPI) based on a ratio of real grassland productivity (calculated as the area under the NDVI curve using the real profile) and theoretical no-management scenario productivity (calculated as the area under curve using the fitted continuum) was used to quantify the biomass removed from the site due to management activities. The methodology was then applied iteratively, gradually eliminating parcels with clear signs of management to focus on those potentially non-managed.
Comparison of intensively managed (upper line) and non-managed (lower line) grasslands. Several significant features can be identified in case of the managed grasslands with whereas no features are identified in case of the non-managed grasslands making the real NDVI profile very similar to the no-management scenario simulated by the fitted continuum.
In Sweden, two demonstration sites (one in the North and one in the South) were analysed over three vegetation seasons. Out of 24,417 parcels, 53% showed signs of management in 2022. The percentage of parcels suspected of non-management decreased with multiple years of data—37% of parcels in both 2021 and 2022, and only 32% for parcels across all three years (2020-2022). Increasing the GPI threshold reduced the number of candidate parcels for non-management to 7% by 2022 when a threshold of GPI > 0.97 was applied.
For Spain (Castilla y Leon), two demonstration sites were also analyzed. In 2022, out of 39,064 parcels, 16% were flagged for potential non-management, with a lower number (8%) showing the same result across both 2021 and 2022. The number of candidate parcels decreased to 5% when data for 2020 was added. When the GPI threshold was increased to 0.98, only 0.2% of parcels were flagged for non-management. Grazing was identified as the dominant form of grassland management, making it challenging to detect non-management due to consistently low NDVI levels.
In the Netherlands, the entire country was analyzed, with 422,950 grassland parcels included in the assessment. Of these, 93.9% were technically monitorable, and 11% were identified as potential candidates for non-management in 2022. This percentage decreased to 5% when parcels were analyzed across multiple years (2021-2022). When a GPI threshold of >0.98 was applied, only 1% of the parcels remained as candidates for non-management. Natural grassland parcels were found to have a higher risk of non-management compared to permanent grassland parcels.
In conclusion, the methodology proves to be an effective filter for identifying parcels that may need additional compliance verification, helping authorities ensure that farms maintain the required levels of grassland management. This approach can be adapted to different countries and farming conditions, providing a scalable solution to support the sustainable management of grasslands across Europe.
Example of the final categorization of the grassland management intensity.