Access to global land cover reference datasets and their suitability for land cover mapping activities Brice Mora, N.E.Tsendbazar, M.Herold LPVE meeting, Frascati, January 29, 2014
Outlook 1. Background 2. Suitability GLCR datasets for land cover mapping activities 1. Existing GLCR datasets 2. Users of GLCR datasets and their requirements 3. Suitability of GLCR datasets for different users 3. Access to GLCR datasets 4. Conclusions
1. Background CEOS Cal/Val-LPV addresses user-oriented validation of global land products. Global land cover (GLC) maps and their accuracy information are useful to different scientific communities. These communities have varying requirements for GLC datasets. To date, several GLC reference datasets have been produced and used for production and accuracy assessment of the specific maps. GLC reference datasets IGBP- DIScover GLC 2000 Globcover 2005 and 2009 GLCNMO LC- CCI Boston/GOFC- GOLD FROM- GLC Other datasets that can be used for validaeon training datasets for MODIS and GLCNMO Volunteer- based Geo- Wiki and View- IT datasets Other datasets e.g. FAO- FRA- RSS
1. Background Despite significant efforts put into generating them, their availability and role in applications outside their intended use have been very limited. This is mainly due to the limited accessibility to these datasets. A thorough analysis on how these datasets can be used beyond their original scope, and what the implications would be for specific use cases is lacking. Objectives: - Analyse the published literature to provide information on GLCR datasets and their user requirements. - Assess the potential uses and limitations of different GLCR datasets for four targeted user groups. - To provide access to available GLCR datasets and guide the user to the most appropriate dataset based on their specific needs.
2.1. Existing GLCR datasets which are assessed Dataset Abbreviation Current state Source of information IGBP-DISCover IGBP-DIS (Scepan et al., 1999) GLC 2000 GLC2000 (Mayaux et al., 2006) GlobCover 2005 GlobCov5 (Defourny et al., 2011b; Fritz et al., 2011a) GlobCover 2009 GlobCov9 (Bontemps et al., 2011a) GLCNMO validation and GLCNMO-tr/val training dataset Existing (Tateishi et al., 2011) MODIS training (STEP dataset) FAO-FRA GEO-WIKI VIEW-IT LC-CCI Boston U. /GOFC-GOLD MODIS-tr (Friedl et al., 2010; Strahler et al., 2003) FAO-FRA (Kooistra et al., 2010; Potapov et al., 2011) Geo-Wiki (Fritz et al., 2009; Fritz et al., 2011a) View-IT (Clark and Aide, 2011b) LC-CCI On-going (Achard et al., 2011) GOFC-GOLD (Olofsson et al., 2012)
2.1. Existing GLCR datasets: Legend and Sampling design Classification schemes of the datasets Datasets IGBP LCCS Other Number of classes IGBP-DIS P 16 GLC 2000 P 22 GlobCov5 P 22 GlobCov9 P 22 P 20 GLCNMO-tr P 14 MODIS-tr P 17 FAO-FRA P 9 LC-CCI P 22 GLCNMOval GOFC- GOLD P 12 GEO-Wiki P P 17-22 VIEW-IT P 7 Sample unit type and size Sample selection scheme and sample size
2.1. Existing GLCR datasets: Response design Data source for interpreta7on Time series Spot- VGT, MERIS and MODIS Temporal coverage of reference data sources for the valida7on datasets Landsat, Sport, Aster imagery Quickbird, Geo- Eye imagery Open source Google- Earth, Open street, Bing and Yahoo maps Geo- tagged photos, Confluence photos Other regional maps and aerial photographs Means of reference data classifica7on (le;), their quality flag and verifica7on (right) Number of datasets 6 5 4 3 2 1 0 Number of datasets 5 4 3 2 1 0 verified and confidence recorded verified only confidence recorded only none available/no informaeon
2.1. Existing GLCR datasets: Current use Dataset Intended application (map accuracy assessment) Other application Pre-processing Estimates Source IGBP-DIS IGBP map FAO Global forest cover Translated into 4 general classes GLC2000 GLC2000 MAP Validating IGBP, GLC2000, MODIS maps and their synergy GlobCov5 GlobCover map 2005 Some samples fed to the validation datasets for GlobCover 2009 GlobCov9 GlobCover map 2009 GLCNMO-val GLCNMO map GLCNMO-tr GLCNMO training MODIS-tr MODIS GLC map MODIS: Global Urban Area mapping FAO-FRA LC-CCI GOFC-GOLD GEO-WIKI VIEW-IT Forest resources assessment Land change of Latin America and the Caribbean Validating MODIS IGBP map in Europe African hybrid cropland map; Biofuel land availability map Land change of Bolivia; deforestation and reforestation of Latin America and the Caribbean; forest change of Guatemala; land use, land cover map of Uruguay Translated into 5 and 11 generic classes; quality and consistency were checked Re-interpretation Overall, class specific accuracy, standard error, also for continental level Overall accuracy, class specific accuracy Overall accuracy, class specific accuracy (FAO, 2001) ( Göhmann et al., 2009) ( Bontemps et al., 2011a) Revised for training ( Schneider et al., 2009) Re-stratification The percentage of cropland within a 1 km pixel; extent of human impact and abandoned land as well as land cover type were recorded with confidence levels Generalized into 5 classes Overall and class specific accuracies Overall accuracy, error of omission and commission for cropland delineation of 5 GLC maps Overall and class specific accuracies (Potapov, et al., 2011) ( Stehman et al., 2012) (Fritz et al., 2011c; Perger et al., 2012) (Aide et al., 2012; Clark and Aide, 2011a; López-Carr et al., 2011; Redo et al., 2012)
2.2.Users of GLCR datasets and their requirements 1. Climate modelling community (ECV) 3. GEO Global Agricultural Monitoring CoP Requirements Statistically rigorous (accurate and precise) 1 overall and class specific accuracy estimates Compatibility in combining and augmenting the 2 samples 3 Flexible thematic land cover characterization 4 Stability over multi-date datasets 5 Spatial resolution requirement 6 Qualify flag information available Requirements 1 Statistically rigorous 2 Affords area estimation Thematic detail and classification scheme for 3 agricultural land covers 4 Definition of cropland and pasture 5 Majority classes and its fraction cover 6 Spatial resolution 7 Quality flag information is available 2. Global forest change analysts 4. Producers of improved GLC map Requirements 1 Statistically rigorous and high sampling density 2 Sample selection scheme is suitable 3 Affords area estimation 4 Thematic detail (forest and non-forest classes) 5 Forest definition is suitable 6 Multi-date sample 7 Suitable spatial detail for change detection 8 Quality flag information available Requirements 1 High sampling density and spatial coverage Detailed representation of thematic classes 2 especially rare classes 3 Flexible thematic land cover characterization 4 Temporal coverage 5 Spatial resolution 6 Ability to characterise heterogeneous areas 7 Quality flag information available
2.3. Suitability of GLCR datasets for different users Metadata of GLCR datasets User requirements for reference datasets User requirement criteria Evaluation of GLCR datasets for each user requirement criteria Combining criteria performance using multi-criteria approach Criteria performance of GLCR datasets for each user requirement criteria of Climate modelling users Criteria Probability sampling Class representatio n Easily combined and augmented sampling scheme Classification scheme and thematic details of the legend Hierarchical classifiers provided Temporal coverage Stable multi- date sample Spatial resolution IGBP-DIS ++ - + +/- ++ - -- ++ + ++ GlobCov5 ++ +/- + ++ ++ + - ++ ++ ++ GlobCov9 ++ + + ++ ++ + - ++ - ++ GLC2000 ++ +/- + ++ ++ + -- - ++ - GLCNMOval Verified ++ +/- + ++ - + -- + - - GLCNMO-tr -- +/- -- +/- - + -- +/- +/- - MODIS-tr -- +/- -- +/- ++ + - +/- + - FAO-FRA ++ ++ +/- +/- - + ++ ++ + - LC-CCI ++ ++ +/- ++ ++ ++ ++ ++ + ++ GOFC-GOLD ++ + ++ +/- ++ + - ++ + ++ GEO-WIKI -- + -- + +/- + - + - ++ VIEW-IT +/- + - - - + - -- + ++ (++ Highly suitable, + Very suitable, +/- Moderately suitable, - Marginally suitable, -- Not suitable ) Interpreta tion confidenc e recorded
2.3. Suitability of GLCR datasets for different users The LC-CCI, GOFC-GOLD, FAO- FRA and Geo-Wiki datasets were generally more suitable for reuse than the other datasets. The analysed datasets are generally more suitable for agricultural monitoring and improving GLC maps. Climate modelling community and forest change analysis require stable multi-date sample which could not be met by many of the GLCR datasets.
3. Access to GLCR datasets GOFC-GOLD reference data web portal Available datasets Name G L C 2000 Glob Cover 2005 Sampling design 2 stage straefied cluster sampling StraEfied random sampling Sample size 1265 4258 STEP straefied 1780 VIIRS straefied random sampling 2500 Sample unit size 3 X 3 pixels 5X5 pixel 1 km pixel Source or reference data Landsat 2000, aerial photographs, NDVI profile SPOT VGT- NDVI profile, Google Earth, Landsat, high and low resolueon images (Google Earth) Google earth, VHSR images Legend LCCS 22 classes LCCS 22 classes IGBP 17 classes IGBP 17 classes Reference Mayaux et al 2006 Defourny et al 2009 Friedl et al., 2000 Sulla- Menashe et al., 2011 BU
Conclusions and recommendations The potentiality of existing and forthcoming GLCR datasets for multiple use cases, particularly for agricultural monitoring and improved map GLC map production. Providing systematic information about GLCR datasets and their reusability in different use cases is useful for guiding appropriate usage of datasets for specific uses. Coordinated international efforts are working to increase the integrity of GLC maps and reference datasets and to support the re-usability of available GLCR datasets by making them accessible to the public. To increase the general usability of GLCR datasets for multiple users, the use of probability sampling scheme, LCCS-based legend, sample selection and sample unit area which are independent of any GLC maps and providing quality assurance information are recommended.