Consistent Global Land Cover Maps for Climate Modeling Communities: Current Achievements of the ESA's Land Cover CCI
Bontemps, Sophie1; Defourny, Pierre1; Radoux, Julien1; Van Bogaert, Eric1; Lamarche, Celine1; Achard, Frederic2; Mayaux, Philippe2; Boettcher, Martin3; Brockmann, Carsten3; Kirches, Grit3; Zülkhe, Martin3; Kalogirou, Vasileos4; Arino, Olivier4
1Université catholique de Louvain, BELGIUM; 2Joint Research Center, ITALY; 3Brockmann Consult, GERMANY; 4ESA, ITALY
In order to define the need for information in support to climate science, the Global Climate Observing System established a list of Essential Climate Variables (ECV), selected to be critical for a full understanding of the climate system and currently ready for global implementation on a systematic basis. In response to the ECV list, ESA initiated a new program - namely the Climate Change Initiative (CCI) -to develop global monitoring data set to contribute in a comprehensive and timely manner to the need for long-term satellite-based products in the climate domain. Among the 14 ESA-CCI components respectively addressing the atmospheric, oceanic and terrestrial domains, the ESA CCI Land Cover (LC_CCI) project is dedicated to land cover (LC) characterization. LC is indeed referred to as one of the most obvious and commonly used indicators for land surface and the associated human induced or naturally occurring processes, while also playing a significant role in climate forcing. This project builds on the ESA-GlobCover projects experiences (Arino et al. 2008, Defourny et al. 2009). It aims at revisiting all algorithms required for the generation of a global LC product from various Earth Observation (EO) instruments and that matches the needs of key users’ belonging to the climate modeling community.
First, a user requirements analysis was established with this community to identify its specific needs in terms of satellite-based global LC products. This analysis highlighted a set of requirements in terms of thematic content, spatial and temporal resolution, stability and accuracy that are not met by existing global products. One finding of particular interest was the priority for both stable and consistent LC products over time. Some interest was also expressed for more dynamic information reflecting LC change and vegetation phenology.
Since the early nineties, several global LC products have been delivered, all based on ''single-year'' and ''single-sensor'' approaches. More recently, the accumulation of global long-term time series of EO data has allowed the delivery of several global maps derived from the same sensor (e.g. MODIS land cover and GlobCover products). However, these products are affected by significant year-to-year variations in LC labels not associated with land cover change, thus demonstrating that current land cover mapping approaches are not able to efficiently extract the stable land cover component (Friedl et al. 2010).
Specific analysis illustrated that, using multi-year EO dataset as input contributes to make the classification less sensitive to the period of observation assuming that no LC change has occurred during the multi-year period (Bontemps et al. 2012). Capitalizing on these global LC experiences and targeting the climate community priorities, the land cover concept was revisited introducing an important distinction between the stable and dynamic components of any terrestrial surface. These were mapped using the full archive of several European instruments, such as ENVISAT MERIS, ENVISAT ASAR (see Santoro et al. - also submitted for the same conference) and SPOT-Vegetation.
The methodology for mapping the stable component of the land cover relies on complex pre-processing and classification chains. Pre-processing includes the following steps: geometric and radiometric corrections, cloud and snow identification, land and water re-classification, atmospheric correction, map projection and temporal compositing. Classification was organized into four main steps, including stratification, compositing, spectral supervised and unsupervised classification and temporal classification. Major effort was required to optimize the integration of a multi-year EO time series into a single land cover product. Multiple years are combined either during the compositing procedure or after the classification step. In the first case, it is assumed that multi-year composites show particularly good properties in terms of spatial and radiometric consistency. Using them as input therefore increases the classification algorithms' performance. In the second case, several maps are derived from annual EO dataset and are then summarized in accounting as efficiently as possible for natural variability.
Based on this innovative new LC mapping approach, a set of three successive global LC products are being generated, related to the 1998-2002, 2003-2007 and 2008-2012 periods. EO time series are mostly produced from MERIS Full Resolution imagery. MERIS Reduced Resolution and SPOT-Vegetation dataset are used to fill in gaps and to increase the MERIS spectral resolution by providing a short-wave infrared channel. The legend counts 21 classes defined using the FAO Land Cover Classification System (LCCS). The products are delivered with an aggregation tool, enabling the users to adjust the map projection and spatial resolution to their needs and to convert the LCCS classes into Plant Functional Types.
A quantitative and independent product validation exercise is under implementation through (i) a confidence-building procedure to reduce macroscopic errors, (ii) a statistical accuracy assessment which should allow users to determine the ''map's fitness for use'' for his/her application, (iii) a comparison with other global land cover products and (iv) a temporal consistency assessment. The statistical accuracy assessment aims at assessing the accuracy of the CCI land cover maps using an independent reference dataset, built by a network of international experts. Experts were asked to interpret validation samples overlaid in Google Earth thanks to a dedicated on-line environment for ''ground truth'' data collection.
The public release of the three CCI global LC products is planned for October 2013. These products corresponding to the stable component of the land cover will be delivered with land cover condition products (see Lamarche et al. - also submitted for the same conference) which will depict the dynamic component of the land cover.