Day 4

Detailed paper information

Back to list

Paper title ESA CCI High Resolution Land Cover Products
  1. Lorenzo Bruzzone University of Trento Speaker
  2. Angelo Amodio Planetek Italia s.r.l.
  3. Francesca Bovolo Fondazione Bruno Kessler - FBK
  4. Maria Antonia Brovelli Politecnico di Milano
  5. Marco Corsi e-GEOS - an Italian Space Agency and Telespazio company
  6. Pierre Defourny UCLouvain-Geomatics Belgium
  7. Paolo Gamba University of Pavia
  8. Gabriele Moser University of Genoa
  9. Catherine Ottlé Laboratoire des Sciences du Climat et de l'Environnement (LSCE)
  10. Lluís Pesquer Centre de Recerca Ecològica i Aplicacions Forestals (CREAF)
  11. Michael Riffler GeoVille Information Systems GmbH
Form of presentation Poster
  • A5. Climate
    • A5.02 The role of Earth Observation in climate services
Abstract text 1. Introduction
The ESA-CCI High Resolution (HR) Land Cover(LC) project [1] has focused on the study of the spatial resolution in analyzing the role of the land cover in climate modeling. The project has designed a methodology and developed a processing chain for the production of high resolution land-cover and land-cover change products (10/30 m spatial resolution) by using both optical multispectral images and SAR data. The HRLC Essential Climate Variable (ECV) is derived over long time series of data in the period 1990-2019 by considering sub-continental and regional areas. Images acquired by ESA Sentinel-2, Landsat 5/7/8 multispectral sensors, and Sentinel-1, Envisat and ERS 1/2 SAR sensors have been processed for the generation of the final products. Given the spatial resolution and the long time period, this resulted in a big data problem characterized by a huge amount of images and a very large volume of data that have been considered and processed.
This contribution presents the primary products generated by the project that consist of: (i) HR land-cover maps at subcontinental scale derived in a given target year, (ii) a long-term record of regional HR land cover maps, and (iii) land-cover change maps.

2. Generated Products
The HR land-cover maps at subcontinental level have been generated using time series of images acquired by Sentinel 1 and 2 in 2019 at a resolution of 10m. The processing has been organized to exploit monthly composite of images that can properly represent the seasonality of the classes. With respect to the previous ESA-CCI Land Cover (LC) project [2], the resolution is improved of more than one order of magnitude (from 300m to 10m). Accordingly, the legend of classes has been re-designed to catch the capability of the most recent sensors in capturing smaller objects (e.g., single trees) and their evolution over time. The legend has been defined over 2 levels, where the second one captures the class seasonality, for a total of 20 classes (see figure 1). The HR land-cover maps at subcontinental level behaves as reference static input to the climate models representing the context at high resolution and high quality given the large quantity of available data.

The long-term record of regional HR land cover maps includes 5 maps generated every 5 years in the period 1990-2015. The spatial resolution is of 30m in the regions of interest for the historical analysis (included in but smaller than the regions covered by the sub-continental one). In this time span, the number of yearly-based images available in archives dramatically reduces. This makes the classification problem more challenging. The processing can rely on few images per-year only (in some areas there is only one image or no images) that are thus organized in seasonal or yearly composites depending on data availability. Accordingly, depending on data availability, a higher level legend consistent with the one of the static map has been considered that does not include the seasonal class information when no seasonal information is available (see figure 1).

Land-cover change information is computed yearly at 30 m spatial resolution and is consistent with historical HR land-cover maps. Change information is provided as presence and absence of change, and for changed samples the year of change is provided together with change probability. The change legend considers the climatic most relevant transitions among the possible ones given the LC legend.

All the products are associated with a measure about their uncertainty. The land-cover products also provide information of the second most probable class identified by the classifier on each pixel. This allows to better capture the complexity of the land-cover RCV in input to climate models.

3. Study Areas
The above-mentioned products have been generated over 3 test areas identified by the Climate User Group as of particular interest to study climate change and the related effects in terms of land-cover and land-cover changes. The areas are in three continents involving climate (tropical, semi-arid, boreal) and complex surface atmosphere interactions that have significant impact not only on the regional climate but also on large-scale climate structures. The three regions are in Amazon basin, the Sahel band in Africa and in the northern high latitudes of Siberia as detailed below (see figure 2).

Amazon. This region has been selected due to large deforestation rates, fire drought and agricultural expansion. Those phenomena are potentially associated to large-scale climate impacts and agents of disturbance including losses of carbon storage and changes in regional precipitation patterns and river discharge with some signs of a transition to a disturbance-dominated regime. An example of LC maps for Amazon is given in Figure 3.

Africa. This region is associated to Sahel band including West and East Africa, which is a complex climatic region which experiences severe climatic events (droughts and floods) for which the future predictions are very uncertain. In this area HRLC impact can be evaluated on better modeling the position and seasonal dynamics of the monsoons (the West African and the Indian ones) and surface processes; and on the explanation of the role of El Nino in the initiation of dramatic drought events (eastern part of the Sahelian band).

Siberia. The third region is expected to be strongly affected by climate changes (polar amplification). Mapping LC changes can document the displacement of the forest-shrubs-grasslands-transition zone to the north and the impact on the carbon stored in permafrost, which in turn will affect long-term terrestrial carbon balance and ultimately climate change.

The generated products have been systematically validated both qualitatively and quantitatively (in terms of overall, producer and user accuracy), and intercomparison analysis has conducted with other land-cover products. Sample collection for quantitative analysis has conducted by photointerpretation on very high resolution images (higher than the 10/30m resolution of products) and intercomparison relies on other existing maps for the considered study areas. The products and the related validation will be presented at the symposium.

[1] L. Bruzzone et al, "CCI Essential Climate Variables: High Resolution Land Cover,” ESA Living Planet Symposium, Milan, Italy, 2019.
[2] P. Defourny et al (2017). Land Cover CCI Product User Guide Version 2.0. [online] Available at:

List of the other HRLC team members: M. Zanetti (FBK), C. Domingo (CREAF); K. Meshkini (FBK), C. Lamarche (UCLouvain), L. Agrimano (Planetek), G. Bratic (PoliMI), P. Peylin (LSCE), R. San Martin (LSCE), V. Bastrikov (LSCE), P. Pistillo (EGeos), I. Podsiadlo (UniTN), G. Perantoni (UniTN), F. Ronci (eGeos), D. Kolitzus (GeoVille), T. Castin (UCLouvain), L. Maggiolo (UniGE), D. Solarna (UniGE).