A multi-sensor approach for detecting the different land covers of tidal flats in the German Wadden Sea — A case study at Norderney

The tidal flats of the Wadden Sea are a highly dynamic and largely natural ecosystem with high economic and ecological value but which are also at risk due to climate change, rising sea levels, algae blooms, invasive species and marine pollution. There is a need for the detection of emerging changes and the potential loss of the natural or semi-natural ecosystems accompanied by a decrease in water quality. Accessibility both from sea and land is very poor, which makes the monitoring and mapping of tidal flat environments from in situ measurements very difficult. In this study, a multi-sensor concept for the classification of intertidal areas in the Wadden Sea is developed. The basis for this method is a combined analysis involving RapidEye (RE) and TerraSAR-X (TSX) satellite data combined with ancillary vector data about the distribution of vegetation, shellfish beds and sediments for the accuracy assessment. The overall methodology is based on a hierarchical decision tree. First, the water coverage is separated from the tidal flats by using the normalized difference water index (NDWI). Second, the shellfish beds are estimated with the textural features of the TSX data and morphologic filters (MFs). Third, the classification of vegetation (salt marsh, sea grass/algae) is based on the modified soil-adjusted vegetation index (MSAVI), object-based features and exclusionary criteria. The remaining area is then separated into different sediment types with an algorithm that uses a thresholding technique applied to radiometric values, the MSAVI and a majority filter. The results show that we are able to identify the location and shape of salt marsh and shellfish beds (a true positive rate of 0.63 and a precision of 0.55) by using multi-sensor remote sensing data. A detailed shellfish bed classification can only be done with radar sensors, like TSX. The extraction of the sea grass areas from the multi-sensor approach is difficult. Sea grass often grows very sparsely in the study area which, with respect to the spatial resolution of RE, leads to a mixture of the spectral signatures of sediment and sea grass. The classification of the sediment types in tidal flats is a challenge compared to vegetation and shellfish beds. An overall accuracy of 64.53%, 66.85%, 68.22% and 68.2% was achieved. The results emphasize that a sediment-type classification cannot be achieved with high accuracy using spectral properties alone due to their similarity, which is predominately caused by their water content.