Mapping Vegetation Types in a Savanna Ecosystem in Namibia: Concepts for Integrated Land Cover Assessments

Kurzfassung

The characterization and evaluation of the recent status of biodiversity and land-cover in Southern Africa’s Savannas is a major prerequisite for suitable and sustainable land management and conservation purposes. The knowledge of the spatial distribution of vegetation types is an important information source for all social benefit areas. Remote sensing techniques are essential tools for mapping and monitoring of land-cover. The development and evaluation of concepts’ for integrated land-cover assessments attracted increased interest in the remote sensing community since evolving standards for the characterization of land-cover enable an easier access and intercomparability of earth observation data. Regarding the complexity of the savanna biome in terms of the spatiotemporal heterogeneity of the vegetation structure and rainfall variability, the main research needs are addressing the assessment of the capabilities and limitations of Moderate Resolution Imaging Spectroradiometer (MODIS) time series data for mapping vegetation types in Namibia. The Random Forest framework was evaluated for mapping MODIS time series metrics in a Kalahari test site in north-eastern Namibia. In regard of the necessity to report the usefulness of the FAO and UNEP Land Cover Classification System (LCCS) in regional case studies, LCCS is evaluated in terms of the applicability in open savanna ecosystems and as ontology for the semantic integration of an in-situ vegetation database in a coarse scale mapping framework based on MODIS data. The results of the integrated use of in-situ-, Landsat, and MODIS data in a standard mapping framework are used to assess the capabilities of the methodological setups of global land-cover mapping initiatives. In order to assess the existing accuracy uncertainties of mapping savannas at global scales, the effects of composite length and varying observation periods were compared in terms of mapping accuracy. The implications for global monitoring were discussed. The determinants of precipitation amount and mapping accuracy were evaluated by comparing MODIS and Tropical Rainfall Measuring Mission (TRMM) time series. The synergistic use of multi-scale land-cover information, such as life form, cover, and height of vegetation types (in-situ), vegetation physiognomy and local patterns (Landsat), and phenology (MODIS) in an integrated ecosystem assessment framework resulted in a flexible land-cover map including a broad structural-physiognomic and a hytosociological legend. The principle of classifiers and modifiers in LCCS proved to be applicable in dry savanna ecosystems and can be confirmed as overarching land-cover ontology. Analyses of time series classifications showed that mapping accuracy increases with increasing observation period. Small composite period lengths lead to increased mapping accuracies. The relationship between mapping accuracy and observation period was observed as a function of precipitation input and the magnitude of change between land-cover stages. The integration of in-situ data in a multi-scale framework leads to improved knowledge of the regionalisation of Namibian vegetation types. On the one hand, the case study in the north-eastern Kalahari showed that multi-data mapping approaches using in-situ to ‘coarse’ MODIS time series data bear the potential of the wall-to-wall update of existing vegetation type maps. On the other hand, the global remote sensing community can extend the reference databases by integrating regional standardised biodiversity and ecotype assessments in calibration and validation activities. The studies point on the uncertainties of mapping savannas at global scales and suggest possible solutions for improvements by adapting the remotely sensed feature sets, classification methods, and integrating dynamic processes of semi-arid ecosystems in the mapping framework.