Séminaire
Broad-scale monitoring of wetlands is essential for researches on carbon or water cycles, water resource management and disaster monitoring such as flood and droughts, on both spatial and temporal scales. Satellite remote sensing with spatiotemporal data fusion (STF) has recently attracted attention as a way of simultaneously describing spatial heterogeneity and tracking the temporal variability of earth surface. However, existing approaches still have limitations when describing abrupt temporal changes, integrating “unlike” datasets (i.e. fusions between microwave and optical data), and compiling long-term, frequent STF datasets. Here, lookup-table based STF (database unmixing: DBUX) is newly developed and evaluated using multiple types of satellite data (AMSR series, MODIS and Landsat) to overcome these limitations, and applied to semi-arid seasonal wetlands in Namibia. The result shows that 1) DBUX is flexible in integrating optical data with microwave data; 2) able to generate long-term, frequent Landsat-like dataset; and 3) more reliable than an existing approach (STARFM) when tracking dynamic temporal variations over seasonal wetlands. Water maps retrieved from the resulting STF dataset had a 30-m spatial resolution and a temporal frequency of every 1 or 2 days, and covered from 2002 to 2015. Based on the water maps, hydrological variables (i.e. water storage volume and evapotranspiration) over three test sites in Namibia were also estimated and evaluated. Further investigations are required for applications in other regions over other landscapes with different satellite sensor combinations.