Validation of surface water maps in selected Australian floodplains derived from Landsat imagery using hydrodynamic modelling

Abstract

Mapping spatial inundation dynamics during floods is important for environmental management and disaster monitoring. Remote sensing technologies provide an affordable means for the routine monitoring of flood size and dynamics with reasonable spatial and temporal coverage. Landsat imagery (30-m horizontal resolution) can provide the appropriate spatial detail for many water mapping applications, although its temporal frequency is every 16 days. The public release of Geoscience Australia’s Australian Geoscience Datacube (AGDC) offers opportunities for the consistent, repeatable production of surface water maps. The AGDC is based on the entire Landsat archive for Australia, following orthorectification and correction to surface reflectance with associated cloud masks. However, validation of these Landsat derived surface water maps is essential to provide confidence in them. Hydrodynamic (HD) modelling tools are widely used for floodplain inundation modelling. While these tools require large amounts of data for model configuration and calibration, and can take a long time to process, they are highly effective in producing detailed characteristics of floodplain inundation with high accuracy at sub-daily time-steps throughout a flood event. For arid floodplains, antecedent soil moisture conditions and groundwater infiltration may need to be incorporated in such modelling. This study compares two methods used for mapping surface water with Landsat in an Australian catchment and compares them to output derived from a well calibrated/validated, high resolution two-dimensional (2D) HD model. Using the 2D HD model to validate surface water extent derived from remote sensing has a number of advantages compared to the more common methods, which use higher resolution optical imagery or the same Landsat data to manually create a water/non-water dataset. The HD model provides an independent source of information, is available at a high spatial resolution, can be extracted to occur at a similar time as the image acquisition used to derive the water map, and can have full spatial coverage of the floodplain of interest. It is hoped that independent validation methods can help to provide confidence to the end user in the adoption of Landsat derived water maps.