Jump to Main Content
A highly automated algorithm for wetland detection using multi-temporal optical satellite data
- Ludwig, Christina, Walli, Andreas, Schleicher, Christian, Weichselbaum, Jürgen, Riffler, Michael
- Remote sensing of environment 2019 v.224 pp. 333-351
- algorithms, anthropogenic activities, automation, climate change, cost effectiveness, ecosystem services, endangered ecosystems, flood control, food availability, monitoring, probability, processing time, remote sensing, selection methods, topography, wetlands, Algeria, Austria, Kenya, Uganda
- Wetlands are valuable ecosystems providing a variety of important ecosystem services such as food supply and flood control. Due to increasing anthropogenic influences and the impact of climate change, wetlands are increasingly threatened and degraded. An effective monitoring of wetlands is therefore necessary to preserve and restore these endangered ecosystems. Earth Observation (EO) data offer a great potential to support cost-effective and large-scale monitoring of wetlands. Current state-of-the-art methods for wetland mapping, however, require large training data and manual effort and can therefore only be locally applied. The focus of this study is to evaluate a methodology for large-scale and highly automated wetland mapping based on current EO data streams. For this purpose, an algorithm for water and wetness detection based on multi-temporal optical imagery and topographic data is presented. Suitable spectral indices sensitive to water and wetness were identified using feature selection methods based on mutual information between optical indices and occurrence of water and wetness. In combination with the Topographic Wetness Index (TWI), these were used to derive monthly water and wetness masks using a dynamic thresholding approach. Aggregating all observations corrected for seasonal bias yielded flooding and wetness frequencies and the Water Wetness Presence (or Probability) Index (WWPI) as an indicator for wetland occurrence or a pre-inventory. To demonstrate the applicability of the proposed method, the algorithm is demonstrated at three study sites with different wetland types in Kenya/Uganda, Algeria, and Austria using Sentinel-2 MultiSpectral Instrument (MSI) imagery. For all sites, the overall accuracy was above 92%. User's and producer's accuracies were higher for water (>96%) than for wetness (>75%). Due to the high degree of automation and low processing time, the proposed method is applicable on a large scale and has already been applied during the production of the Copernicus High Resolution Water-Wetness Layer and within the European Space Agency (ESA) project GlobWetland Africa.