The utility of satellite remote sensing for flood prediction in sparsely gauged catchments

Abstract

Floods are one of the most damaging natural disasters, occurring worldwide and causing loss of life and property. However, in many catchments flood prediction is difficult because of the lack of data including topographic data, river water levels and river discharge. Satellite data provides large-scale and near-real time remotely sensed information that can be used to monitor climatic and hydrological parameters. The proliferation of remote sensing data provides the opportunity to develop new research methods for flood modelling. However, robust, accurate and fast approaches for use of freely-available satellite data in data-scarce regions are difficult to identify. This thesis aims to address these gaps by using freely-available satellite data in ungauged or sparsely gauged basins for flood forecasting. The overall goal of the research is to improve the use of satellite-based data as inputs to flood models and to develop simple model structures that are useful in data-scare regions. To achieve this objective, this thesis discusses multiple sources of freely-available satellite data which form boundary conditions to flood models, and describes the application and validation of new methods at different sites with varying data availability and at contrasting scales. Firstly, the thesis develops a linear combination method to combine SRTM and ASTER GDEMs to generate improved DEMs for flood models at sites without high quality terrain data. Secondly, daily MODIS land surface temperature (LST) data is used to improve the 10-day temporal resolution of satellite altimetry Jason-2 based on the relationship between water levels and difference in LST between daytime and nighttime. This model is then extended to incorporate satellite precipitation and soil moisture to improve the predictive skill of the model as well as considering the uncertainties that result from multiple sources of satellite data. Finally the conditions under which satellite-based LST is useful to quantify streamflow during floods is assessed for Australia and globally. As part of this assessment a method to fill missing LST data due to cloud cover is developed. In summary, this thesis contributes original approaches to enhance the use of freely-available satellite data as inputs to flood models by providing information on topography, LST, water levels and river discharge.