Land Cover Characteristics and Tree Cover Dynamics in the Mekong Basin. Analysing one decade of satellite data

Kurzfassung

The Mekong Basin in Southeast Asia is one of the largest international river basins in the world. Its abundant natural resources are shared by six riparian countries, i.e. China, My-anmar, Laos, Thailand, Cambodia, and Vietnam, and provide the basis for more than 72 million people, many who are directly dependant on the resources for their subsistence. Rapid socio-economic growth throughout the last decades, though, has had substantial effects on the rate and intensity of anthropogenic interventions in the region’s ecosystems and on the respective modifications in land cover and land usage, first and foremost on the highly valuable remaining forests in the region. Although socio-economically very diverse, economic reforms towards market liberalisa-tion and intensified transboundary economic flow among the countries have increasingly transformed the Basin into an integrated region. In the course of such continuous coales-cence many transboundary environmental issues have arisen that require interregional approaches as well as regionally consistent environmental information to be adequately assessed and accounted for. In this respect, remote sensing has evolved as a key tool for the spatially continuous and consistent documentation of environmental characteristics, particularly those related to the terrestrial land cover, beyond physical or political bounda-ries, and at various temporal and spatial scales. In view of the absence of such adequate and up-to-date information products for this re-gion, this dissertation contributes towards an improved knowledge base on the current land cover-related characteristics and change processes in the Mekong Basin. More spe-cifically, the aim has been firstly, to present a comprehensive and detailed depiction of the current land cover distribution in the entire Mekong Basin and secondly, to give insight on the most dominant process of land cover change in the Basin and Southeast Asia generally, namely the unprecedented rate of deforestation. This has been achieved by producing regional-specific information products on current and historic land cover characteristics on the basis of remote sensing data from the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the platforms Aqua and Terra. However, on account of almost persistent cloud cover in the rainy season, the application of optical remote sensing data poses a challenge that demands special data processing and classification methods. In the early stages of data processing, existing quality prod-ucts with regard to the detection of clouds in MODIS satellite data, so-called cloud masks, are compared and assessed with respect to their validity. It could be shown that due to faulty cloud flags, the availability of suitable observations differs widely depending on the respective cloud mask that is referred to. In this respect, a cloud mask enhancement al-gorithm is presented for increasing cloud flag reliability when working with daily MODIS satellite data. Since cloud masks are essential variables with respect to the conversion of cloudy satellite images into cloud-free image composites or the production of vegetation-index time series, the findings from this comparison are fundamental to the generation of a high-quality data basis that is used for all subsequent applications and analyses. Intra-annual vegetation dynamics (phenology) for the entire Mekong Basin are derived and interpreted for the year 2010. Therefore, an Enhanced Vegetation Index (EVI) time series is produced, with dropouts and noise being effectively reduced by applying an adaptive Savitzky–Golay filter in combination with a harmonic analysis that allows for ad-justments adapted to the respective growing cycles in the time series. Information on land cover is derived by a hierarchical unsupervised classification approach, optimised for re-gions with frequent cloud cover. Moreover, the environmental heterogeneous conditions throughout the region are addressed by a regionally-tuned clustering approach, based on pre-defined physiographic subregions and the use of auxiliary geodata. In this way a re-gional-specific land cover map, the Mekong LC2010 product, is produced that to the au-thor’s best knowledge, provides the scientific community with the first regionally-optimised land cover characterisation for the entire Mekong Basin. This map differentiates 22 land cover classes and includes not only a broad range of standard classes, but also very specific land use types such as classes on double, or triple season croplands, or on dif-ferent aquaculture intensities. The forest cover dynamic in the Lower Mekong Basin is analysed by deriving vegetation structure as per pixel fractional cover of woody vegetation, herbaceous vegetation and barren land in annual time intervals between 2001 and 2011. A multi-scale approach is applied based on a non-linear regression tree algorithm that relates MODIS reflectance data as explanatory variables to subpixel canopy cover as response variable. The latter is derived by the classification and aggregation of high-resolution Landsat data. A particular methodological focus in this application is directed to minimising inter-annual fluctuations in model predictions originating from noisy input data and algorithm limitations, to improve the identification of real land cover changes on the ground. Significant reductions in tree cover undergone within the observation period were identi-fied by utilizing the long-term statistics on inter-annual prediction variability as a function of tree cover. Furthermore, new methodological approaches are tested to utilize the full canopy cover trajectory throughout a period of eleven years to perform a deeper, analyti-cal examination of forest cover dynamics that goes beyond differentiating a simple change-no-change situation. The specific temporal patterns in the tree cover trajectory enable the differentiation between permanent forest cover conversions and temporary forest losses, as well as characterising change processes according to their level of ab-ruptness. In this way, it is shown that the reconstruction of forest disturbance histories can be used to reveal information about the underlying causes of forest cover reduction or may be indicative of the land’s respective anthropogenic usage. Derived results indicate that particularly the dense evergreen forests in Cambodia and the forests in the Central Highlands of Vietnam are at high risk, suffering losses at rates of 0.8% and 1.2% per year. While deforestation in Cambodia is observed as being of mostly permanent nature, in the Vietnamese regions of the Basin a high concentration of temporary forest loss could be identified in addition to the permanent forest losses. Laos is endowed with the highest proportion of woody cover in the LMB and, thus, despite very large absolute forest losses shows a comparatively low deforestation rate of ‘only’ 0.4% per year. However, extensive forest areas in Laos can be considered as degraded forests with lower canopy cover estimates as a result of shifting cultivation practices. This tradi-tional form of agriculture causes widespread patterns of temporary canopy cover reduc-tion, observed particularly in northern Laos and along the Annamite mountain chain. Thai-land shows with 0.1% the lowest annual forest loss rate within the Basin as well as being characterised by an almost absence of temporary forest losses.