Addressing spatio-temporal resolution constraints in Landsat and MODIS-based mapping of large-scale floodplain inundation dynamics

Abstract

Recent studies have developed novel long-term records of surface water (SW) maps on continental and global scales but due to the spatial and temporal resolution constraints of available satellite sensors, they are either of high spatial and low temporal resolution or vice versa. In this study, we address this limitation by exploring two approaches for generating an 8-day series of Landsat resolution (30 m) SW maps for three floodplain sites in south-eastern Australia during the 2010 La Nina Floods. Firstly, we applied a generalized additive regression model (GAM) that empirically relates Landsat-based SW extent to in-situ river flow to then predict additional time steps. Secondly, we used the STARFM and ESTARFM blending algorithms for predicting the Open Water Likelihood at 8-daily intervals and 30 m resolution from Landsat and MODIS imagery. ESTARFM outperformed STARFM and best blending accuracies were achieved in the floodplain site with the slowest changes in inundation extent through time. There was good agreement between the blended and statistically-modeled 8-day SW series and both series provided new and temporally consistent information about changes in inundation extent throughout the flooding cycles. After careful consideration of accuracy limitations and model assumptions, these SW records hold great potential for assimilation into hydrodynamic and hydrological models as well as improved management of terrestrial freshwater ecosystems.