Multi-scale and Multi-temporal Wetland Monitoring in Sub-Saharan West-Africa Using Time Series of Medium and High Resolution Optical Satellite Data

Abstract

Surface water is a critical resource in the Sahel region. The Sahel has been found to be extremely vulnerable to climate change and variability, and has suffered severe droughts in the mid-1970s, mid-1980s and the last ten years, among them in 2012. Farmers, pastoralists and villagers strongly depend on the availability of water in wetlands, particularly during and towards the end of the dry season. A case study carried out in Burkina Faso where more then 2000 small dams have been built, shows the strong human influence on water management driven by land use change, agricultural intensification and urban growth. Three different wetlands in the study area are part of the Ramsar List of Wetlands of International Importance. Wetland detection and monitoring is commonly applied using satellite imagery and is of particular interest in areas where ground sampling and data are not very dense, or larger time periods or regions need to be covered. However, there are still many uncertainties and challenges due to the spatio-temporal dynamics of wetlands in semi-arid areas. Strong seasonal variations account for limitations when using spatially high resolution (HR) imagery, and small extension of significant wetlands cause limitations for spatially medium resolution (MR) imagery. Since spatial and temporal monitoring requirements cannot easily be satisfied using solely one sensor or method, in this work a multi-temporal and multi-scale approach is investigated. The three parameters: wetland type, surface water area, and flooding regime are monitored by image to image change detection of HR images in combination with MR time series analysis. In a further step it is examined how wetland changes are related to the occurrence of droughts. Time series imagery used in this study origin from MODIS (250m, since 2000) and the recently developed BioPar Surface Water Body product (1km, since 1998) based on SPOT VEGETATION. On the high resolution scale, two Landsat time series, one acquired at the end of the rainy season (Oct/Nov, where wetlands reach their maximum water level) and one at the end of the dry season (Mar/Apr), are used. For both MODIS and Landsat, spectral bands in the range of red and NIR serve to identify wetlands and build indices where NIR is employed as a proxy for flooding/standing water and the Normalized Differenced Vegetation Index (NDVI) as a proxy for dry season vegetation. Long-term seasonal variability is computed by automatic derivation of surface water using a dynamic NIR threshold on MODIS time series supported by topographic information from the SRTM digital elevation model (DEM); the BioPar Surface Water product serves for validation. Anomalies of surface water coverage are determined using time series of computed water layers (8-day and monthly means) based on MODIS, and Landsat images covering wet and dry seasons from several years. These bi-seasonal Landsat data are also used for wetland type classification based on approaches building dynamic classes in ongoing wetland and land cover mapping projects (GlobWetland-II, Landcover CCI), focusing on distinction between natural and artificial wetlands. The final classification is a combined result of the Landsat analysis and the derived flooding regime (number of inundated months) from MODIS. Preliminary results show an increase of artificial wetlands, differences in flooding regime between northern and southern Sahel and also that wetland parameters would be suitable to monitor water stress. As a result of this, the study can show how spatio-temporal wetland dynamics can be captured using multi-scale remote sensing imagery and thus can provide a better understanding of the relationship between available water in wetlands, drought occurrence and agricultural development as critical elements of livelihoods in semi-arid areas.