Seasonal inundation dynamics and water balance of the Mara Wetland, Tanzania based on multi-temporal Sentinel-2 image classification
Juliette Kool, Stef Lhermitte, Markus Hrachowitz, Francesco Bregoli, Michael E. McClain
Abstract
Anthropogenic hydrologic alternations are a threat to flood dependent wetland ecosystems like The Mara Wetland in Tanzania, which plays an important role in regulating the quality, timing and magnitude of the flow of water into Lake Victoria and provides natural resources and habitat for a variety of species. Baseline information on the spatiotemporal dynamics of the seasonal inundation is critical to manage potential threats such as dam construction and local climate change, but is challenging to acquire due to the heterogeneity of the landscape caused by the interspersion of open water and emergent vegetation. In this study, intra- and inter-annual inundation pattern of the Mara Wetland are reproduced for 2017–2019 using remotely sensed data. A Random Forest (RF) algorithm is trained bi-seasonally, using eight Sentinel-2 bands, five water- and vegetation-indices and a Digital Surface Model (DSM) to classify the land cover of the wetland region in a semi-automated way, by processing a total of 73 Sentinel-2 scenes based on wet- and dry-state training images for 2017. The scenes are classified into seven classes; three wetland- and four dryland-classes. The overall classification accuracy achieved (based on an independent validation set) is 98.6%. The inundation extents are used in combination with available hydrological field-data to estimate the water balance. The seasonal expansion and contraction of the wetland follows a consistent bi-modal regime, and the results from the water balance affirm the importance of river-flow during the dry-season and local precipitation for the seasonal fluctuations in water-availability.