Towards national-scale characterization of grassland use intensity from integrated Sentinel-2 and Landsat time series

Highlights

• Current EO data sources provide unprecedented temporal data at parcel level detail.

• Opportunity for improved characterization of large-area grassland management

• Novel approach exploits phenology signal from integrated S2 & L8 time series.

• Algorithm presented and first national scale assessment for Germany performed.

• Approach is sensitive to parcel level mowing regimes, regional insights provided.

Abstract

The increased availability of systematically acquired high spatial and temporal resolution optical imagery improves the characterization of dynamic land surface processes such as agriculture. The use of time series phenology can help overcome limitations of conventional classification-based mapping approaches encountered when, for example, attempting to characterize grassland use intensity. In Europe, permanent grasslands account for more than one third of all agricultural land and a considerable share of the EU Common Agricultural Policy (CAP) budget is devoted to grasslands. The frequency and timing of mowing events is an important proxy for grassland use intensity and methods that allow characterizing grassland use intensity at the parcel level and over large areas are urgently needed. Here we present a novel algorithm that allows detecting and quantifying the number and timing of mowing events in central European grasslands. The algorithm utilizes all imagery acquired by Sentinel-2 MSI and Landsat-8 OLI for one entire year as available from the NASA Harmonized Landsat-Sentinel dataset. Cloud-free observations from both sensors are first synthesized through compositing at 10-day interval. Machine learning algorithms are then used to derive a grassland stratum. The intra-annual growing season profiles of NDVI values are subsequently assessed and compared to an idealized growing season trajectory. Residuals between the idealized trajectory and a polynomial model fit to the observed NDVI values are then evaluated to detect potential mowing events. We demonstrate and evaluate the performance of our algorithm and utilize its large area analysis capabilities by mapping the frequency and timing of grassland mowing events in 2016 on the national-scale across Germany. Our results suggest that 25% of the grassland area is not used for mowing. Validation results however suggest a relatively high omission error of the algorithm for areas that only experienced a single mowing event. The date ranges of detected mowing events compare overall well to a sample of interpreted time series points and to farm level reports on mowing dates. The mapped mowing patterns depict typical management regimes across Germany. Overall, our results exemplify the value of multi-sensor time series applications for characterizing land use intensity across large areas.

Introduction

Grasslands ecosystems represent considerable proportions of land systems in the majority of biomes around the world. Globally, grasslands serve as grazing land and provide fodder for livestock in form of silage or hay, which is used for meat, milk and dairy production. Over the last decades, also bioenergy production from grassland biomass has become an important revenue source for European farmers (Rösch et al., 2009). Apart from production, grasslands provide valuable habitats for many species, often supporting high levels of biodiversity. Many important ecosystem services are provided by grassland ecosystems, including carbon sequestration, water filtration and flood water retention. In Europe, grassland accounts for more than one third of all agricultural lands and its primary purpose is the support of milk and meat production (Smit et al., 2008). A considerable share of the European Union (EU) Common Agricultural Policy (CAP) payments are devoted to maintaining the ecological value of grasslands, in appreciation of grasslands as part of European cultural landscapes and in support of European agriculture in general. Detailed knowledge on grassland management and its intensity is therefore required on the level of individual farms to ensure the envisioned evolution of the CAP subsidies inform of direct payments (1st pillar of the CAP, European Commission, 2013a) and for the conceptual design of targeted agro-ecological and climate-focussed measures within the second pillar (European Commission, 2013b).

Grassland use intensity and management are therefore prime factors affecting ecosystem services. Factors that primarily influence grassland use intensity include fertilization, grazing and the frequency and timing of mowing events for silage or hay production. Several of the payments that farmers receive are bound to extensive management such as no-fertilizer application or late mowing dates to safeguard ground breeding birds. While several of these land management practices cannot directly be assessed with remote sensing (Kuemmerle et al., 2013), mowing event frequency and timing can be detected and serve as a proxy for grassland use intensity for an agro-ecological und economical evaluation of individual farms. Finally, controlling the compliance of farmers to climate change mitigation measures is an integral component of the CAP Integrated Administration and Control System (IACS, European Commission, 2013c).

The Geospatial Aid Application (GSAA) process within the CAP provides one of the few spatially explicit European datasets on grassland use. It represents self-reported information provided by farmers on their grassland production and management on individual parcels. Some European countries have begun making such data accessible to the public. Claims for subsidy payments made within the CAP are traditionally verified based on a yearly 5% sample of parcels per EU member state or administrative region. A two-step control procedure involves remote sensing-based interpretation of VHR imagery (so called Control with Remote Sensing, CwRS) and subsequent field visits for a reduced sample for ground-based verification (Rösch et al., 2009). While in this procedure crop type and acreage can be determined with relatively high certainty, verification of grassland management and use intensity are much more challenging. Moreover, with the next renewal of the CAP for 2020, the control mechanism shall evolve into a full-area monitoring mechanism underlining the need for consistent assessments on the national level.

Detecting and quantifying the timing of mowing events in grassland requires a high temporal observation frequency at decameter-scale and measurements in the major optical spectral domains in order to cope with the heterogeneous spatial patterns, small parcel sizes and rapid growth of perennials in summer. The EU Copernicus program with its fleet of Sentinel satellites offers high-quality, open-source data from several platforms, after the USGS-based Landsat archive has paved the way towards free and open data policies (Berger et al., 2012; Woodcock et al., 2008). The twin platform Copernicus Sentinel-2 (S2) constellation has reached its full operational capacity in 2018 and now provides systematic five-day repeat observations globally. The S2 Multi Spectral Imager (MSI) instrument provides key improvements compared to previous missions, including three novel spectral bands in the red edge wavelengths, 10 and 20 m resolutions and an swath width of 290 km, leading to higher observation frequencies and hence better spatio-temporal coverage (Gascon et al., 2017). While the temporal repeat frequency is unprecedented for multi-spectral data at 10 m to 20 m spatial resolution, persistent cloud cover can still seriously impede its usability. Sensor constellations with similar observation characteristics can therefore further improve the effective observation frequency. Landsat is presumably the most suited mission to exploit such synergies given continuity considerations, free and open data policy and corresponding measurements in most spectral domains. Sensor differences in spectral band pass, spatial resolution and observation geometry have to be taken into account for a successful integration of both data sources towards an improved grassland use characterization (Flood, 2017; Storey et al., 2016).

For optical data cloud coverage, cloud shadows and other atmospheric constituents are limitations for grassland monitoring. These need to be carefully addressed in order to detect a temporally transient feature such as mowing events in grassland. In the optical domain, a sequence of growing season Rapid Eye imagery has recently been used to distinguish three grassland use intensity classes for the Canton of Zurich, Switzerland (Gómez Giménez et al., 2017). The authors utilized the spectral contrast between different acquisitions in comparison to the mean spectral appearance of grassland areas to differentiate intensity classes. They combined results with modelled livestock grazing density information and conclude that the combined use provides a good indicator of grassland use intensity. The approach, however, has limited potential to be applied to large areas. Another study has recently underlined the importance of high quality cloud screening and the authors conclude that the observation frequency from optical time series of S2 currently still leads to relatively high omission rates (Kolecka et al., 2018).

Overall, published approaches for mapping the timing and frequency of mowing events in grassland at the parcel level are rare and most existing approaches lack capabilities to cover large areas (Schmidt et al., 2018; Gómez Giménez et al., 2017). This is unfortunate, as national or regional scale monitoring approaches can provide information that is not available from other data sources and that is crucial to inform policies and improve management of natural resources. Here we present an approach that allows mapping the number and the timing of mowing events in grasslands at high spatial resolution by combining a novel mowing detection algorithm coupled with dense temporal interval compositing of multi-sensor optical data.

Our specific objectives were to:

• Present and evaluate the mowing detection algorithm

• Map the frequency and timing of mowing events in Germany during 2016.