The Copernicus Programme, managed and coordinated by the European Commission (EC) and implemented in partnership with the European Space Agency (ESA), member states and various EU agencies, offers services mainly based on Earth Observation (EO) data provided by ESA through the Copernicus Space Component. In complement to the operational-phase implementation, the Horizon 2020 project ECoLaSS (Evolution of Copernicus Land Services based on Sentinel data) targets to improve existing and develop next-generation Copernicus Land Monitoring Service (CLMS) products of the pan-European and global component, with a suggested implementation schedule from 2020 onwards.

One of the key thematic products of the pan-European component of the CLMS are the so-called High Resolution Layers (HRLs), which are thematic products targeting land cover characteristics of 5 main thematic classes (Imperviousness, Forest, Grassland, Water/Wetness, Small Woody Features). These layers are produced in regular 3-year intervals from multi-temporal EO data of 10-20 meter spatial resolution, in a consistent manner for 39 European countries (EEA 39). In ECoLaSS, innovative methods and algorithms, based on innovative dense time series processing of optical (Sentinel-2) and SAR (Sentinel-1) satellite data, are applied to develop, test and demonstrate various prototypes of improved or novel HRL products. The datasets are made fully publically available together with extensive respective documentation.

This work introduces the ECoLaSS project setup, together with the so-far collected user and stakeholder requirements, developed processing methodologies and the suggested prototypes. A special focus is laid on the recently established prototypes for potential next-generation HRL products on forest, grassland and agriculture, based on dense Sentinel-1 and -2 time series evaluation. The ECoLaSS prototypes of the evolved HRL Forest and HRL Grassland products of the reference year 2015 will be shown as implemented in various representative large sites, covering diverse European biogeographic regions. These comprise, amongst others, prototypes of improved forest types and forest cover change layers and their yearly updates, as well as improved permanent grassland classification and categorisation together with yearly change. The project also targets entirely novel products, such as a possible future HRL on agricultural/arable land, for which a prototype is presented, featuring a pan-European crop mask of high precision and representative crop types. All prototypes are developed with a view to potential future European-wide implementation and fitting into the operational Copernicus service infrastructure, in case of being selected for service roll-out by the procuring entity(ies).

From a methodological point of view, these prototypes are implemented applying temporal analysis to a multitude of time features (multiple temporal statistical metrics), as well as derived indices (e.g. VV/VH ratio, NDVVVH from SAR data and NDWI, NDVI, Brightness and IRECI from optical data) extracted from dense Sentinel-1 and -2 time series. Using a random forest-based classification approach on several hundred time features, a feature selection is performed to identify the most appropriate input features for classification, whilst ensuring a cost and time efficient processing with high accuracy and quality. The prototype implementation fully builds upon the capabilities of state-of-the-art cloud processing solutions, exploring implementation on, amongst others, the Copernicus Data and Information Access Service (DIAS) platforms.

The Copernicus Global Land Service is the component of the European Copernicus service which ensures a global systematic monitoring of the Earth’s land surface. It provides bio-geophysical variables in near real time describing the daily state, and changes in state, of vegetation and land surface processes.

Lately, the service provides dynamic yearly global land cover maps at 100m spatial resolution from 2015 onwards with a focus on real-time environment monitoring targeting a large set of different land management user communities (i.e. forest, biodiversity, crop monitoring, desertification and Sustainable Development Goals monitoring). This variety of different user groups and their different needs make it hard to create a one fits all land cover map. As a result most detailed land cover maps end up with a lot of mixed land cover classes which are hard to be used in different applications. This is where our novel mapping approach fits in as the classified discrete map is provided with a set of continuous cover layers (forest, shrub, bare and grassland) which allows the customization of the land cover map to the specific needs. Moreover, near-real time land cover changes are provided by yearly updates and consolidated maps covering confirmed changes are provided the following year.

The classification system is setup as an operational component of the service, where the reference data collection is being aligned with the Copernicus Pan-European land cover, and is compatible and able to use higher spatial resolution data (i.e. Landsat 30m, Sentinel1/2 10-20m). The training and validation datasets are continuously collected as part of this operational service. Moreover, their collection sheme is setup independent of each other and any land cover legend to allow the flexibility of being used for various 10-100m resolution mapping and validation approaches. The initial maps are derived from the PROBA-V time-series archive which was fully reprocessed to improve accuracy at high latitudes, to align with the Sentinel-2 UTM tiling grid and to generate an Analysis Ready Data (ARD) archive. The workflow is fully automated, able to handle the millions of tiles and runs in the Proba-V Mission Exploitation Platform (MEP). A solution has been developed to deploy this workflow from the private MEP cloud onto a public cloud (i.e. DIAS) to deal with the required exponential expansion of the required resources when applying in the large on the Sentinel datasets. This solution could enable a more flexible, user-oriented on-demand regional mapping in the cloud.

This presentation presents the system and results to generate the Copernicus Global land cover map at 100m spatial resolution, its yearly change detection updates as well as some preliminary 20m regional mapping results.

The space segment of the Copernicus program, the Sentinels, defines a whole new playground where to explore the limits and the potentialities of diverse technologies to generate updated and precise land-cover maps all around the globe. In this context, Synthetic Aperture Radar (SAR) technology is able to provide data during the day and night and independently of the weather. Consequently, these systems deliver systematic and reliable temporal data records of an observed area of the Earth surface. Since its launch in April 2014, the C-band Sentinel-1A system provides SAR data time series with a 12-day repeat cycle. The launch of twin C-band satellite Sentinel-1B system, in April 2016, has reduced it to a 6-day repeat cycle. Nevertheless, due to the interferometric capabilities, the temporal information is contained in the time series of SAR images, as well as in the interferometric coherence from pairs of SAR images. The availability of these global data favours the emergence of alternative approaches to the mapping scene where mostly the optical, but also the radiometric, data have established their niche. In this regard, the ESA SEOM SInCohMap project aims to develop, analyse and validate novel methodologies for land cover and vegetation mapping by using time series of Sentinel-1 data and by exploiting the temporal evolution of the interferometric coherence. The project aims to quantify the impacts and benefits of using Sentinel-1 InSAR (Interferometric Synthetic Aperture Radar) coherence data relative to traditional land cover and vegetation mapping approaches such as those using optical data (especially Sentinel-2) and SAR (Synthetic Aperture Radar) intensity. In general, interferometric coherence is affected by a combination of terms derived from the system, the observation geometry and the properties of the observed scene. In previous studies, coherence has already proven to be a good parameter for inferring land cover.

In this regard, the purpose of this work is to explore and to evaluate the performance of using the interferometric coherence from SAR, and specifically Sentinel-1, time series and interferometric data for land cover and vegetation mapping. The first objective is to quantify the impact of using Sentinel-1 InSAR (Interferometric Synthetic Aperture Radar) data relative to traditional land cover and vegetation mapping using SAR-based approaches. Previous studies in the literature have already indicated the usefulness of using InSAR information for land cover classification. In addition to the impact of interferometric coherence, this work also evaluates the impact, in terms of land-cover classification and vegetation mapping, of SAR intensity and SAR polarimetry. Classification performances depend also on the methodology employed to process the data. Then, the second objective of this work is also to evaluate the impact of the classification technique employed for data classification, as well as the different aspects associated with them. Consequently, seven different classification techniques are evaluated, where aspects like the use of pixel or segment-based approaches or the impact of the sampling strategy for training are also considered. The third and last objective of this work is to assess the impact of the temporal decorrelation in the proposed classification schemes. It is well known that the temporal decorrelation is one of the main parameters driving the quality and information content of the interferometric data, especially under evolving scenarios such as vegetated areas. The repeat cycle, i.e. the rate of image acquisitions over a particular area, is an essential parameter that must be assessed in the classification framework to properly define a feasible context. In this regard, the focus in this work is set on the different acquisition scenarios of the Sentinel-1 systems (6 days, 12 days and 24 days, depending on the region) and how the classification is affected or limited by this parameter.

Monitoring the state and condition of our land-based capital to achieve sustainable resource management is paramount to advancing ecological sustainable development. Reliable, consistent, standardised and continental scale mapping of land cover and changes over time remains a significant challenge for both scientists and practitioners. Using such information to add value to environmental policy and decision making is also a common ambition. Through development of its open source analysis platform Digital Earth Australia (DEA), as part of the Open Data Cube (ODC) initiative, is making significant progress towards this goal. This platform prepares and facilitates the analysis of the vast volumes of Earth Observation (EO) data to produce a range of information products for the Australian government and business.

This study demonstrates how land cover and change classifications can be generated by integrating environmental variables (EVs) accessible through DEA within the Food and Agriculture Organisation (FAO) Land Cover Classification System (LCCS) taxonomy. For four sites, land cover and change classifications were generated from time-static layers and annual summaries of EVs, and for two time periods. Changes identified included mangrove dieback in the southeastern Gulf of Carpentaria (between 2014 and 2016), reduced dam water levels in Lake Ross, Burdekin catchment, and associated expansion of aquatic vegetation, and development of cropping practices (e.g., irrigation) in the Gwydir catchment (2009 and 2016). Changes were determined and mapped based on evidence provided through an integrated time-series comparison of LCCS component codes and EVs. The extent of detected changes corresponded with that observed using time-series of RapidEye data (2014 to 2016; for the Gulf of Carpentaria) and Google Earth imagery (2009-2016; all other sites).

The study demonstrated capacity to use EVs within DEA to generate land cover and change classifications within the Australian context and the potential for mapping and monitoring land cover dynamics and changes driven by natural and/or anthropic pressures. This novel approach, which has been applied to European environments, is able to be implemented within the ODC environment, and applied to other countries with this capacity. As such, it provides the opportunity to routinely generate land cover maps at high (<= 30 m) spatial resolution (based on those of Landsat or Sentinel-1/2), at least annually and using a consistent and internationally recognised taxonomy.

The diverse applications of land cover (LC) and land use (LU) data have led to the development of many classification systems. Most of these contain a mixture of land cover and land use information. Each application emphasizes particular aspects of land cover and land use, related to specific requirements and purposes. Furthermore, data collection methods, scales, tailored-to-purpose definitions and lack of completeness hamper the data transfer from one application / classification system to another, and thus form an obstacle for cross-border and interdisciplinary data integration.

The technical circumstances and conditions including quality of affordable satellite imagery, computing power, data storage capacities and methodologies have developed rapidly, which now opens new approaches to the extraction of land information from remote sensing data. In parallel, thematic requirements and political reporting obligations have also evolved and reach a higher level of urgency. Responses need to be found to address climate change and other environmental challenges caused by local actions having global consequences. Therefore, a common understanding of local phenomena within the global context is crucial to provide the fundament for monitoring and decision making processes.

The EAGLE concept embodies a new approach for land monitoring initiatives following an object-oriented approach in landscape modelling. It promotes the paradigm shift from single value-coded conventional classification techniques to a multi-variable descriptive characterization of the land. The EAGLE concept comprises a data model that stores information about land cover components (LCC), land use attributes (LUA) and further characteristics (CH). Each land cover object then contains a set of descriptive attributes from which an understanding of the area can be derived, but which can also be exploited to derive thematically details properties of the landscape that would be impossible with conventional classification schemes. In this way, it aims at providing a basis for an integrated multi-use European Land Monitoring Framework.

Two strong constraints in knowledge creation can be identified: one is access to data, the other is understanding of data, and subsequent extraction of information. Once a decision maker or the citizenship has gained access to spatial information relevant to their needs and duties, still the issue of data and product comparability and interoperability remains.

The EAGLE concept with its data model can be applied in two ways:
(1) to use it as a vehicle to analyze and show comparability and differences between different LC/LU classification systems and legends on an abstract semantic level,
(2) to use it as a guidance document for LCLU mapping programs to directly capture parameterized information and describe real landscape situations as an instantiated application of the model.

The more flexible and comprehensive object-oriented approach avoids the loss of information about landscape details, which on the other hand is inevitable when applying a strict classification approach with fixed class definitions or fixed thresholds.

On European level, the European Environmental Agency (EEA) contracted the EAGLE group to develop the concept for a 2nd generation CORINE Land Cover program. Based on the conceptual ideas behind the EAGLE model this initiative comprises a number of phases and development steps:
1. CLC-Backbone is a spatially detailed, large scale inventory in vector format providing a geometric spatial structure attributed with raster data for landscape features with limited, but robust, EO-based land cover thematic detail on which to build other products.
2. CLC-Core is a consistent multi-use grid database repository for environmental land monitoring information populated with a broad range of land cover (including but not limited to CLC-Backbone), land use and ancillary data from the CLMS and other sources, forming the information engine to deliver and support tailored thematic information requirements.
3. CLC-Instance is the “nominal” end point or final product in the establishment of CLC+ product suite. There may potentially be multiple “instances” with different specifications and for different purposes within the Copernicus Land monitoring program. It is a derived raster / grid product from the CLC-Core and will be a LULC monitoring product with improved spatial and thematic performance, relative to the current CLC, for reporting and assessment. It can be tailored towards different types of application domains.
The final element of the conceptual design, although not strictly a new product, is the ability to continue to produce the existing CLC, which may be referred to as CLC-Legacy in the future, which already has a well-established and agreed specification.
The grid approach, which is planned to form the spatial reference framework for CLC-Core, will enable the user to combine various thematic data domains beyond LC and LU information (e.g. environmental monitoring, habitat types, meteorology, population census data, hazards and risks) which all refer to the standardized and geometrically stable and scalable grid cells.

As data modelling architecture, multi-dimensional data cubes and linked data are appropriate approaches to build up a multi-purpose structure for data storage, information analysis and statistical correlations.

A further perspective can be to broaden the application of the EAGLE concept to the global perspective. With the view on global challenges like the monitoring of SDGs, and two existing abstract modelling concepts of land cover, namely the EAGLE concept and the ISO standard Land Cover Meta Language (LCML), the ISO technical committee TC 211 opened a procedure of harmonizing these two concepts into a revised and enhanced standard. Once harmonized, the outcome of a revision process can provide the semantic fundament for common global understanding of LC and LU data and its interrelated processes and human activities on the Earth´s surface.
With its thematically neutral approach and enough application flexibility within the architecture, the EAGLE concept addresses different user communities and working domains.

In the current decade, the production of global land cover products has been proliferated significantly thanks to the availability of higher spatial and temporal resolution satellite data and increased computation capabilities. As such, global land cover mapping is being operationalized as a service with a production of yearly and long-term maps. Complementing to the current production of global land cover maps, the Copernicus Global Land Service (CGLS) provides a yearly Dynamic Land Cover 100m (CGLS-LC100) resolution product from 2015 onwards. The product includes discrete (with a fixed legend) and fractional (vegetation continuous field layers: e.g., for trees, shrub and herbaceous vegetation) at the global scale. For the myriad of applications that use global land cover maps, the availability of these discrete and fractional land cover maps opens the possibility of creating user-tuned global land cover maps with different legends.
As an intrinsic part of an operational land cover mapping service by the CGLS, the validation of the CGLS-LC100 product is subject to certain requirements. These include (I) applicability to maps with different legends (including discrete and fractional maps); (II) suitability to operationalize the validation for subsequent yearly products. This study addresses the two requirements and presents CGLS Land Cover validation approach for operational and user-oriented land cover mapping.
To provide user-oriented accuracy assessments of the CGLS-LC100 global land cover map, we developed an independent validation dataset for global land cover maps by using the Geo-Wiki web-application for reference data collection (https://geo-wiki.org/). At each validation sample location, land cover element information for a block of 100 subpixels at 10m resolution was recorded. A team of 30 experts from different regions of the world contributed to collect a global land cover validation data consisting of around 20 000 validation sample sites. Land cover element information at 10m resolution allows the validation data to be suitable for multiple maps with different legends and resolutions (10-100m resolution). We demonstrate the applicability of this validation data by assessing the accuracy of a prototype version of a global land cover map of CGLS that is currently under production. The sampling design allows us to provide statistical accuracy of the map at global and as well as continental levels.
To operationalize the validation for subsequent yearly products, the validation dataset is designed using a stratified sampling design with a stratification independent of any land cover maps. Thanks to the flexibility of stratified random sampling, additional sample points can be added without compromising the statistical rigorousness. For assessing the subsequent maps of the CGLS-LC100, temporary sets of sample sites are added to the original sample to better represent modified or change recorded areas. We present preliminary results of operational validation for subsequent land cover maps for the African continent. Here validations of CGLS-LC100 Africa maps for 2015, 2016 and 2017 are demonstrated.
As part of Copernicus Global Land Service’s operational mapping of global land cover, we demonstrate a user-oriented and operational validation approach. By developing a validation system that is flexible in terms of legend, spatial resolution and sampling design, the efficiency and consistency of validation dataset are improved for more operational and user-oriented land cover validation. We emphasize further the efforts required for maintaining a validation dataset up-to-date.