Integrating in-situ, Landsat, and...
Environ Monit Assess DOI 10.1007/s10661-010-1602-5 Integrating in-situ, Landsat, and MODIS data for mapping in Southern African savannas: experiences of LCCS-based land-cover mapping in the Kalahari in Namibia Christian H��ttich �� Martin Herold �� Ben J. Strohbach �� Stefan Dech Received: 31 December 2009 / Accepted: 27 June 2010 �� Springer Science+Business Media B.V. 2010 Abstract Integrated ecosystem assessment ini- tiatives are important steps towards a global biodiversity observing system. Reliable earth observation data are key information for tracking biodiversity change on various scales. Regarding the establishment of standardized environmental observation systems, a key question is: What can be observed on each scale and how can land cover information be transferred? In this study, a land cover map from a dry semi-arid savanna ecosys- Electronic supplementary material The online version of this article (doi:10.1007/s10661-010-1602-5) contains supplementary material, which is available to authorized users. C. H��ttich (B) �� S. Dech Department of Remote Sensing, Institute of Geography, Julius-Maximilians-University W��rzburg, W��rzburg, Germany e-mail: firstname.lastname@example.org M. Herold Center for Geoinformation, Wageningen University, Wageningen, The Netherlands B. J. Strohbach National Botanical Research Institute of Namibia, Windhoek, Namibia S. Dech German Remote Sensing Data Center, German Aerospace Center, Oberpfaffenhofen, Germany tem in Namibia was obtained based on the UN LCCS, in-situ data, and MODIS and Landsat satellite imagery. In situ botanical relev�� samples were used as baseline data for the definition of a standardized LCCS legend. A standard LCCS code for savanna vegetation types is introduced. An object-oriented segmentation of Landsat im- agery was used as intermediate stage for down- scaling in-situ training data on a coarse MODIS resolution. MODIS time series metrics of the growing season 2004/2005 were used to classify Kalahari vegetation types using a tree-based en- semble classifier (Random Forest). The prevail- ing Kalahari vegetation types based on LCCS was open broadleaved deciduous shrubland with an herbaceous layer which differs from the class assignments of the global and regional land- cover maps. The separability analysis based on Bhattacharya distance measurements applied on two LCCS levels indicated a relationship of spec- tral mapping dependencies of annual MODIS time series features due to the thematic detail of the classification scheme. The analysis of LCCS classifiers showed an increased significance of life- form composition and soil conditions to the map- ping accuracy. An overall accuracy of 92.48% was achieved. Woody plant associations proved to be most stable due to small omission and commission errors. The case study comprised a first suitability assessment of the LCCS classifier approach for a southern African savanna ecosystem.
Environ Monit Assess Keywords Harmonization �� Standardization �� Time series �� Random forest �� Remote sensing �� Phenology Introduction Spatially consistent and standardized land cover data provide key information for the monitor- ing of the state of ecosystems and biodiversity conservation at various scales. The integration of environmental databases and the development of interfaces for environmental data from local to global scales (e.g., in-situ field survey data and global archives of satellite imagery) is a forthcoming task for interdisciplinary land-cover and biodiversity monitoring studies. The deliv- ery of spatially explicit land-cover information plays an important role for ecosystem monitoring (Hollingsworth et al. 2005). In this context, the task of the Group on Earth Observations and Biodiversity Observation Network (GEO BON) is to generate a platform to support the standard- ization of top���down observations from satellite systems with bottom���up measurements on ecosys- tem processes and species data (Scholes et al. 2008). Such comparable and standardized mecha- nisms for land-cover data are needed for develop- ing policy-relevant biodiversity indicators (Pereira and Cooper 2006 Scholes and Biggs 2005). The Land Cover Classification System (LCCS) of the United Nation���s Food and Agriculture Or- ganization (FAO) is the currently most accepted land-cover standard, as it is being implemented as an international standard (ISO 19144-1:2009). On global scales LCCS-based land cover legends were used within the framework of the Global Land Cover 2000 (GLC2000, Fritz et al. 2004) and the 300-m ENVISAT-MERIS product (GlobCover, Defourny et al. 2009). Regional Suitability analy- ses of LCCS-based land-cover and land-use char- acterization were conducted within the Africover (Jansen and Gregorio 2002) and BIOTA-Africa (Cord et al. 2010) framework. The aim of LCCS is to provide a standardized description of a certain land cover in a set of pre- defined, diagnostic, and hierarchically arranged generic criteria describing a certain land-cover (Di Gregorio 2005). LCCS uses physiognomic- structural classifiers for the definition of primarily vegetated land cover classes, such as life form, cover, height, and spatial distribution. The uti- lization of the FAO Land Cover Classification system proved an effective tool for a standardized legend definition on global scales (Latifovic 2004 Herold et al. 2008). However, little experiences have been exchanged in the scientific community for adaptations of LCCS in regional case studies and to assess the suitability and capabilities of LCCS on fine thematic scales. Results from legend harmonization analysis of the main global land cover maps (IGBP DIS- Cover, UMD, MODIS 1-km, GLC2000, Loveland et al. 2000 Justice et al. 1998 Hansen et al. 2002, 2003 Fritz et al. 2004) showed limited class agreements in highly heterogeneous landscapes (Herold et al. 2008). Primarily in savanna biomes the problem of mixed classes is evident in coarse scale land-cover maps. In order to overcome exist- ing inconsistencies of class descriptions, research has to be addressed to regional adaptations of LCCS in savanna ecosystems. The world���s semiarid savanna ecosystems cover more than 20% of the earth���s surface. These biomes can be rated as biocomplex systems as the floral and faunal composition is controlled by nonlinear processes (e.g., grazing intensity and fire frequency). Savanna ecosystems are charac- terized as lands with a mixture of herbaceous and woody life forms, often appearing as a structure of a woody tree and shrub layer with herba- ceous understories (Hanan et al. 2006). Compared to more homogeneous vegetation types, savanna classes have the lowest mapping accuracies in the global datasets (Jung et al. 2006). Spatio-temporal variability of rainfall patterns is one of the most prominent uncertainties for semi-arid land-cover mapping based on earth observation data (Scholes et al. 2004 Privette et al. 2004). The data requirements on the spatial, radio- metric, spectral, and temporal resolution of the satellite imagery to be used for specific earth ob- servation tasks are often determined by the scale of the study. Three main fields of applications can be stated: (a) direct mapping of individual plants and plant associations, (b) habitat map- ping, and (c) the analyses of relationships between spectral radiances and in-situ data of species dis-
Environ Monit Assess tribution patterns (Nagendra 2001 Turner 2003 Muchoney 2008). Satellite time series proved to be an efficient tool for monitoring and assessing the state of semi-arid ecosystems. At regional scales, time series of the Moderate Resolution Spec- troradiometer (MODIS) were successfully used to map the southern African Miombo ecosytems based on band pair difference analyses (Sedano et al. 2005), characterized as open forests, thick- ets, and grassland formations (Frost 1996). De- pendencies of the spatial resolution of different environmental datasets were emphasized for the rangelands of the open Kalahari savannas (Trodd and Dougill 1998). A promising concept of repre- senting savanna vegetation structure is the map- ping of the proportional cover of life forms based on satellite time series imagery as done on global (DeFries et al. 2000 Hansen et al. 2003) and regional scales, e.g., for Africa and Australia (Scanlon et al. 2002 Guerschman et al. 2009 Wagenseil and Samimi 2007 Gessner et al. 2009). A well-known problem can be seen in the har- monization of in-situ data with coarse scale satel- lite imagery. LCCS can be used as a common land-cover language (Neumann et al. 2007). On one hand, bottom���up ecosystem assessment ap- proaches provide key information to develop reli- able regionally validated land cover maps and, on the other hand, allow top���down satellite observa- tions for large-area measurements of biophysical variables to be used for ecotype assessments and biodiversity monitoring, such as species, habitat, and plant community mapping. In this context, the main research questions are: What can be ob- served on each scale and how can land cover infor- mation be transferred and included in integrated observation concepts? What is the potential of LCCS to be used as a translation tool of land cover data (in-situ and satellite earth observa- tions) for savanna ecosystems? Which benefits can be achieved for the connected scientific commu- nities related to biodiversity and remote sensing when using LCCS? The aims of this paper are to: ��� assess the applicability of the concept of the LCCS classifiers in semiarid ecosystems and demonstrate first experiences of using LCCS as a ���land-cover language��� in a bottom-up mapping framework ��� present a flexible legend of typical Kalahari savanna vegetation types using the UN-FAO Land Cover Classification System (LCCS) ��� demonstrate a concept for a bottom���up eco- system assessment framework by integrating local scale in-situ botanical survey data with Landsat imagery and coarse scale satellite time series data Mapping framework Study region The mapping area comprises the eastern commu- nal areas in the eastern Kalahari in Namibia from 17���30 E to 21��� E and 19���45 S to 21���45 S. The area is characterized by a sub-continental climate with a long term annual average summer rain period of 324���450 mm and often erratic rainfall events (Mendelsohn and Obeid 2002). The greater part of the study area is characterized by longitudinal Kalahari sand dunes and plains. Within the Kala- hari sand plains, flat calcrete depressions and pans as well as fossil drainage lines (Omiramba) are found. Floodplains and areas subjected to regular flooding are apparent in the western part of the study region (Strohbach et al. 2004). Field data The in-situ reference database includes a num- ber of 422 botanical relev�� samples taken during a reconnaissance survey of soils and vegetation of the eastern communal areas in Namibia from April to May 2004. A stratified sampling scheme was applied with a plot size of 20 �� 50 m after Braun���Blanquet (Strohbach 2001). The sampling included a detailed description of the vegeta- tion composition (species richness and abun- dance), life form composition, and habitat settings within a pedologic assessment (see Table 1). A phyto-sociological analysis was conducted result- ing in a synoptic vegetation type legend. The botanical classification nomenclature was based on the occurrence the two characteristic species (Strohbach et al. 2004) and the related environ- mental setting of each vegetation type (e.g., En- neapogon desvauxii���Eriocephalus luederitzi short