The Copernicus Climate Change Service (C3S) is one of the six thematic information services provided by the Copernicus Earth Observation Programme of the European Union (EU). C3S, which is implemented by ECMWF on behalf of the European Commission, will provide past, present and future Climate Data Record (CDR) and information on a range of themes, freely accessible through the Climate Data Store (CDS). It benefits from a sustained network of in-situ and satellite-based observations, re-analysis of the Earth climate and modelling scenarios, based on a variety of climate projections.

Within the Land Biosphere component of C3S, satellite-based observations will be used to provide the longest possible, consistent and mature products at the global scale for the following Essential Climate Variables (ECVs): Surface albedo, Leaf Area Index (LAI), the fraction of Absorbed Photosynthetic Active Radiation (fAPAR), Land Cover (LC), Fire, Burnt Areas (BA), and Fire Radiative Power (FRP). State-of-the-art algorithms that respond to GCOS requirements will be used and the products will be quality assured following the protocols, guidelines and metrics defined to be consistent with the Land Product Validation (LPV) group of the Committee on Earth Observation Satellite (CEOS) for the validation of satellite-derived land products.

To reach this goal, the following approach is proposed.(i) consolidate the CDR and secure continuation of the products by moving towards Copernicus mission Sentinel-3 as primary data source,(ii) make an important steps towards cross-CDR consistency by harmonizing the pre-processing for all CDRs and implement operational tools to monitor the quality of the CDRs and their cross-consistency (iii) apply an extensive quality assessment of the delivered CDRs and benchmark with other existing datasets.

The surface albedo products will be extended in time to obtain the single-sensor CDRs for the period 1982 until present. In addition, the surface albedo CDR will be further consolidated by moving towards a multi-sensor approach to derive a fully consistent CDR. We propose to change the algorithm to derive LAI and fAPAR using a physically-based approach (Two-stream Inversion Package TIP method), that ensures cross-consistency with surface albedo. The TIP model derives the LAI and fAPAR from the surface albedo dataset. For Land Cover and Fire Burned Areas, maximization of the exploitation of the scientific expertise from the ESA CCI Land Cover and Fire projects was one of our primary goals. This will be achieved by (i) brokering the already existing BA and Land Cover CDRs from these projects, (ii) adapting them to Sentinel-3 SLSTR and OLCI sensors, and (iii) bringing these processing chains into fully operational and agile production lines. The Land Cover maps of the ESA CCI project will be further extended from 2016 until 2019 in this proposal. Likewise, the Burned Areas products from ESA CCI derived from MODIS will be brokered. These CDRs will be extended in time using the same approach adapted to Sentinel-3A and B satellite. Finally, the Fire Radiative Power product will be introduced in the service using only Sentinel-3 data as input. A strategy towards its further evolution within C3S will be prepared to include possibly also other FRP datasets and extend it to the past.

Essential Biodiversity Variables (EBV’s) have become an accepted nomenclature in the remote sensing community. However, finding an agreement on what biodiversity elements should actually be monitored is critical to assessing how biodiversity can be monitored using Earth Observation and the Copernicus services. Here we retain the original EBV classes, but reduce the number of EBVs by coalescing terms into logical biological groupings which are also widely referred to in the remote sensing literature. For each EBV class we group EO products specifically derivable from remote sensing. In this way, we propose a structure that represents a major step forward by defining EBVs in terms understandable for both biologists and remote sensing experts, reducing confusion between these two communities interact, and stimulating a new era of rapid development of remote sensing biodiversity products from satellites.

Land surface phenology (LSP) characterises recurrent biological events in the annual profile of vegetated land surface, at ecosystem scale as observed from remote sensing. LSP has been widely used as measure for terrestrial ecosystems’ responses to changes of environmental conditions or for characterising species composition and biodiversity of an ecosystem. Observation of LSP requires a dense time series of observation in order to image rapid changes in vegetation activity such as during green-up and senescence phases. With the availability of free and open Sentinel-2A/B and Landsat 8 optical satellite images, we are now able to have a sufficient dense time series to image LSP on global scale in high-resolution and therefore also to narrow the scale gap between plan phenology and Land Surface Phenology.
In the frame of ESA’s GlobDiversity project, we developed a new approach to derive LSP from satellite images with the aim to contribute to the framework of Essential Biodiversity Variables (EBVs). We can demonstrate that satellite observations can contribute to a global biodiversity monitoring in a similar way as they already do for the Essential Climate Variables (ECVs).
The algorithm for LSP retrieval was developed on several pilot sites located in different vegetation types such as temperate forest, arctic tundra, Mediterranean wetland, and sub-tropical savannah and shrublands. A trade-off analysis of existing approaches was made with focus on a fully automated and computational efficient algorithm with minimal human interaction and a global usability. The algorithm includes a control and flagging system to allow quality measures and validation with ground based phenocams installed at the pilot sites. Phenocams observe the vegetation ground-based with a visual camera installed for instance on a flux tower and allow an independent measure of the annual growing-cycle of the vegetation.
We will present LSP products generated for different pilot sites located around the world and put them in relation of a general usability and possible applications. The data sets will be set into a larger context of the EBV framework and other RS-enabled EBVs and the potential for policy makers and other users will be discussed. We will illustrate with an outlook the next steps towards a large – even global – spatial scale and its contribution to the development of data sets towards a global biodiversity monitoring and conservation.

Integrated Biodiversity Conservation and Carbon Sequestration in the Changing Environment (IBC-CARBON) is a major research project funded by the Strategic Research Council of Finland with 3.8 million € funding for the first phase and it aims at integrative research and planning to identify feasible options for securing forest biodiversity, important ecosystem services and sustainable forest use in Finland (Fig 1). Here we present some key findings related to the integration of various Earth Observation techniques (WP4) to estimate biodiversity in boreal forests.

The aims and selected findings presented here are to: (i) collect cutting-edge EO data at various spatial and temporal scales for IBC-CARBON studies; (ii) develop and produce novel remotely sensed indicators of BD variables and variables for C-sequestration and forest radiation studies. Different variables describing biodiversity and ecosystem properties are produced using a multi-sensor approach. We will utilize 1) optical satellite images (e.g. Sentinel, Landsat), 2) airborne laser scanning and hyperspectral data, and 3) unmanned aircraft systems (UAS). Optical remote sensing covers large geographical areas at 10-30 m spatial resolution and temporal span of several decades. Laser scanning is a superb method to capture the 3D structure of forested ecosystems with sub-meter accuracy and has been used in growing numbers to study wildlife habitats and BD. Hyperspectral data (HySpex) is used to characterize tree species recognition and identification, and to collect information from spectral traits. Integrating multisource and multiscale remote sensing (RS) data and methods allows studying forest BD and C-sequestration related questions in great detail. Novel UAS methods contribute for bridging the gap in EO between field and airborne measurements and providing ultra-high spatial and temporal resolution imagery for detailed assessment of different ecosystems properties. Because of the potential for rapid deployment, spatially explicit data from UASs can be acquired irrespective of many of the costs, scheduling, logistic and weather limitations to satellite or piloted aircraft missions. Field data (e.g. species diversity, canopy foliar traits) has been collected especially from the airborne and UAS-data acquisition sites.

We have developed further on the concept of spectral traits in boreal environments. Biotic traits, especially functional traits, are becoming increasingly important concept in ecology, conservation biology and sustainable resource management. They can be biochemical, physiological, morphological, structural, phenological or functional characteristics of plants, populations or communities. Spectral traits are traits that can be directly or indirectly recorded using RS. Deriving spectral traits from various RS data can provide detailed valuable information for BD research. In addition, the use of lidar data allows studying traits related to forest and vegetation structure in 3D. Thus, as forest and vegetation structure is a primary determinant of habitat quality and species diversity it is imperative to assess this kind of structure in 3D. Multi-temporal EO data also enable detection of changes in spectral traits, i.e. spectral trait variations which can be related to physiology and phenology as well as stress and disturbances.

We will show preliminary results of the developed and produced remotely sensed indicators of BD variables. These variables will be available in multiple spatial resolutions (~ few cm to 30 m) and temporal resolutions (years to decades) depending on the boundaries of the examined area. Indicators and proxies are based on spectral features of the targets (satellite images, airborne hyperspectral data and UAS) and 3D structural variables (laser scanning data). A wide range of different variables can be produced for various needs. For example, laser scanning data is analyzed and processed with tailored geospatial algorithms, which enables calculation of variables that are considered the most topical and important ones in specific studies.

Fig 1. Graphical presentation of the IBC-CARBON project. This presentation describes the actions and selected results of the WP4 Improved Earth Observation Techniques.

Palm oil is nowadays the most common vegetable oil used globally. As the demand grows, the expansion of Palm Oil plantations in the tropical regions will remain one of the main drivers of deforestation, having a considerable impact on biodiversity and disrupt the carbon cycle. Although the current hot spot of production is located in South-East Asia, future expansion is expected for African and South American tropical regions. The monitoring of its current extent and future expansion is crucial for validating the recent efforts towards a sustainable production on a global scale.
One of the main difficulties in mapping palm oil on a global scale is the regular cloud coverage over tropical regions and its spectral similarity to natural forests in the optical domain, which both hinders its applicability on a global scale. In this study we overcome both limitations by using Sentinel-1 C-band SAR, ALOS-2 Palsar-2 L-Band SAR and SRTM elevation data. Traditionally L-Band SAR data is considered to be most efficient in mapping forested areas due to its long-wavelength signal that is capable of penetrating the canopy. In contrast, the strength of our method capitalises mainly on the dual-polarised VV/VH C-Band SAR data from Sentinel-1, which is particularly well suited to distinguish palm trees from other forest types.
In addition, the 12-day revisit cycle of Sentinel-1 allows for the creation of dense time-series. The temporal behaviour of the backscatter is then captured by using the timescan method, which depicts the usage of descriptive statistical parameters for every pixel in the full time-series. This helps in reducing the influence of environmental conditions and adds further strength in discriminating palm oil plantations. Simultaneously, the amount of data used for classification is reduced when compared to the full time-series and the number of input channels into the subsequent machine learning algorithm is standardised.
The drawback of this method is the huge amount of data pre-processing. To overcome this obstacle, we utilise the online platform Google Earth Engine that allows for the planetary-scale application of Earth Observation data.
For the collection of ground truth data, we introduce a mobile app in which a pre-defined sampling scheme is imported. The app automatically points to the training sites of the sampling scheme. Different background maps are made available via APIs to Google Earth and Google Earth Engine.
More than 10000 collected ground truth samples are then used to train and validate a Random Forest classifier, which is applied to the full dataset. As a result, the first global map of palm oil dated to the year 2017 is presented in this study.

Mapping habitat suitability for target species is a key component of wildlife management and conservation planning and today is typically achieved using correlative species distribution models (SDMs). Species’ distributions are also recognized as Essential Biodiversity Variables, which can serve as important indicators for assessing biodiversity change, given they can be mapped across larger areas. Satellite images are an important input to SDMs, providing spatially and temporally continuous information on land cover. Yet, traditionally, remote sensing information in SDMs was mainly limited to the use of categorical land-cover maps. More recently, the potential of continuous variables which can be derived directly from satellite imagery has gained increasing attention.
The Landsat archives are particularly well-suited for characterizing wildlife habitat and species’ distributions, because they are globally available, provide imagery at relatively high spatial resolution, and go back multiple decades in time. Yet, their use in SDMs has been mostly limited to spectral indices derived from single images or image pairs. Integrating multi-temporal Landsat imagery recorded within a given timeframe allows for the calculation of spectral-temporal metrics summarizing the distribution of spectral responses measured over time. Such metrics have shown to improve land cover classifications, and should also harbor considerable potential for mapping habitat suitability directly from the Landsat image record.
We developed a new approach for mapping habitat suitability based on Landsat habitat metrics, i.e., continuous variables characterizing key aspects of habitat that can be derived directly from the Landsat archive. For testing our approach, we mapped habitat suitability for three large mammal species (Eurasian lynx, red deer and roe deer) in the Bohemian Forest Ecosystem located in Central Europe, for which extensive GPS tracking data were available. We derived metrics summarizing spectral-temporal responses of based on spectral indices (central tendency, scale, rate of change, seasonal values), as well as information on snow-cover derived from pixel-quality bands. For calculating the spectral-temporal metrics, we compared two different approaches, using non-robust statistics (e.g., mean, standard deviation), and robust statistics (e.g., median, mean absolute deviation). Then, we tested the usefulness of these metrics for mapping habitat suitability for lynx, red deer and roe deer by fitting SDMs using continuous Landsat habitat metrics (using non-robust and robust metrics), and compared them with models based on categorical land-cover maps.
Our results highlight that using Landsat habitat metrics consistently outperformed land cover-based models, across all species, different temporal subsets of wildlife data, and different modeling algorithms (average increase in cross-validated model AUC: 0.14). Moreover, habitat metric-based models allowed capturing considerable variability of habitat suitability within land cover types, e.g., associated with forest stands of different age. While the use of robust statistics for the calculation of spectral-temporal metrics did not improve discrimination ability in our performance assessment, the derived habitat suitability maps were smoother than those based on non-robust metrics, showing more realistic and considerably less noisy patterns of habitat suitability.
More broadly, our study shows how combining newly available rich wildlife tracking data with the Landsat image archives opens up new possibilities for improving spatial information on the distribution of wildlife species and their habitat. Moreover, using metrics that can be derived directly and flexibly from the Landsat image record in SDMs also harbors great potential for mapping habitat across large areas, and potentially also reconstructing habitat dynamics back in time. Such approaches enabling assessing wildlife habitat dynamically in space and time based on remote sensing data can be an important component of global biodiversity monitoring, and can provide information essential to conservation planning.