Grasslands cover approximately 40% of the Earth's land area, encompassing nearly 70% of the global agricultural land area, and are distributed on all continents and across all latitudes (Suttie et al., 2005; White et al., 2000). Grassland dynamics influence global ecosystem functioning, and their impact is widely modulated by management practices intensity on these landscapes (Zhao et al., 2020). Management practices are primarily driven by grassland landscape maintenance, as well as by ecosystem service of provisioning offered by the grasslands. Grasslands are subject to management practices such as mowing or grazing or a combination of both. Therefore, monitoring grassland management practices is essential for assessing management intensity level, which in turn plays a critical role in studies related to biodiversity (XXXX), water (XXXXX) and carbon (XXXXX) cycling and others topics (XXXX). In France, the National Observatory of Mowed Grassland Ecosystems conducts birdlife monitoring in mowed grasslands, with a particular focus on the rise in breeding failures attributed to increasingly early mowing. Early mowing intercepts birds' reproductive period and interrupts their breeding process (Broyer et al., 2012). Usually, responsible agencies conduct occasional observation campaigns to support ecosystem-related public policies, but ground observations are not spatially exhaustive and are time-consuming. As an alternative source, synoptic remote sensing data enables regular and global-scale monitoring, enabling tracking of vegetation dynamics. Currently, Sentinel-2 mission provides cost-free high resolution data at 10m spatial resolution with a 5-day temporal frequency (10 days before 2017), allowing intra-plot level observations. Grassland mowing events timing and intensity have already been mapped using remote sensing-based time series, mainly from features sensitive to vegetation status, such as Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Leaf Area Index (LAI) and more. There have been several methods used to detect mowing events from satellite time series. These methods were mainly based on temporal changes in time series using threshold-based methods and anomalies detection approach. More recently, deep learning-based architectures were also used to detect mowing events timing. Estel et al. (2018) assessed grassland use intensity spatial patterns across Europe. To extract annual mowing frequency, a temporal change analysis based on spline-adjusted MODIS NDVI time series was used. Their approach involved identifying mowing events as instances where a local minima exhibited a change, relative to its preceding peak, exceeding 10% of growing season amplitude. The results showed an overall accuracy of 80%, which decreases as the frequency of events increases. In northern Switzerland, Kolecka et al. (2018) also estimated mowing frequency employing similar temporal change analysis, but based on raw Sentinel-2 NDVI time series. Here, a drop in NDVI greater than 0.2, between two consecutive cloud-free acquisition dates, was counted as a mowing event. Their method accurately identified 77% of observed events and highlighted that false detection can occur due to residual cloud presence, while sparse time series led to the omission of mowing events. Regarding Griffiths et al. (2020), mowing events frequency and timing were mapped in Germany using 10-day composite Harmonized Landsat-Sentinel NDVI time series. Discrepancies between a hypothetical bell-shaped curve and the current polynomial-fitted curve were evaluated. An event was counted when the difference exceeded 0.2 NDVI. Findings revealed consistent spatial patterns in mowing frequency (indicating extensive and intensive management). However, estimated dates exhibited significant discrepancies compared to observed dates (MAE > 50 days), which could be due to lower temporal resolution of Sentinel-2 before 2017 and the absence of reliable ground data for calibration and validation. Stumpf et al. (2020) mapped grassland management (grazing or mowing) and its intensity based on biomass productivity and management frequency, respectively. The latter were extracted from n-day composite Landsat ETM + and Landsat OLI NDVI time series. As in previous cases, a management event was counted when NDVI loss is higher than a threshold, which was based on the probability density function of all NDVI changes across the time series and was specified for p = 0.01. Their approach yielded management patterns consistent with several management-related indicators (species richness, nutrient supply, slope, etc). Recently, Watzig et al. (2023) estimated mowing events in Austria, using Sentinel-2 NDVI time series and implementing discrepancy analysis between a idealized unmowed trajectory and actual NDVI values. An event was recorded if the difference exceeded-0.061. Commission errors due to residual clouds were reduced via a subsequent binary classification of each estimated event using a gradient boosting algorithm trained over cloudy plots. Findings indicated an overall accuracy of 80% in correct event detection, with estimated dates closely aligning with observed dates (MAE < 5 days). Vroey et al. (2022) developed a algorithm for detecting mowing events across Europe. Here, raw Sentinel-2 NDVI and Sentinel-1 VH-coherence time series were used separately.