The use of Envisat ASAR Global Monitoring Mode data to map rapid broad-scale flood events

This thesis seeks to enhance our ability to map the extent of large floods in near real time using coarse resolution C-band radar remote sensing. The microwave part of the electromagnetic spectrum has a great advantage over visible and infrared light in its ability to penetrate cloud cover, and as radar is an active system, it does not rely on daylight hours for reflected solar radiation. The European Space Agency's Advanced Synthetic Aperture Radar aboard the Envisat satellite, operating in Global Monitoring Mode (GM), is targeted for particular consideration due to its high temporal frequency, comprehensive coverage and ease of acquisition. Challenges are identified which relate both to the use of radar generally, and also in particular to GM data, in the demarcation of water and land, as well as to the practical business of data processing. These challenges relate to the way that water is identified, which can be by a low signal where specular surface reflection away from the sensor occurs, or by a high signal where multiple interactions occur between the water surface and emergent structures such as vegetation. Thresholds must make the distinction between the two cases, and as such, some prior knowledge of land cover is needed in the segmentation process. With such coarse data as GM, mixed pixels comprising both high and low water signals are often encountered, which result in a mid-range pixel value that masks the presence of water. The thresholding process is further complicated by the relationship of the signal returned to the sensor with incidence angle, which varies between about 14-44° with GM data. Under some wind conditions, waves of a particular pitch and orientation on the surface of open water cause resonance effects, returning a very high signal - sometimes even a gain - to the sensor. In particular circumstances, where Flood waters flow through arid land, the low signal returned from open water due to specular reflection cannot be distinguished from the low signal returned from desert due to attenuation and absorption. In literature surrounding research in this field, results from observations of radar response in wetlands and flooded grasslands are mixed, pointing to the importance for further work in this area. In Australia, the need for a better understanding of the expected backscatter response from inundated areas in tropical savanna, which covers one third of its landmass, is clear. The computational framework was set up for the efficient download, registration and orthorectification of GM data using scripting and open source software. Full advantage was made of the parallel processing capabilities of James Cook University's High Performance Computing network, scripts were tailored to GM data's characteristics and test results proved the method appropriate for the high volume processing required by the large GM dataset. This capability was used to carry out regression on a pixel-wise basis across a year's worth of GM data, categorised by seasonal rainfall periods, in order to normalise backscatter values with respect to incidence angle. Correlation of the resulting characteristics with surface parameters, such as regolith, vegetation and soil type were observed. The potential confusion between absorption in dry, homogeneous soils, and specular reflection on surface water was predicted. It was observed that the degree of change of backscatter with incidence angle on open water appeared independent of the presence of Bragg Resonance, despite absolute values being at opposite ends of the scale, depending on whether resonance did, or did not, occur. A major flood event in Pakistan was successfully mapped and made available in near-real time for the disaster relief effort. An image differencing technique allowed the successful separation of low backscatter response from open water with that from the immediately surrounding desert. GM data were found to fill a gap in the period where the flood was obscured to visible and infrared sensors, during the crucial first week of the event. Definition of the extremities of the flood were tackled with a spatial threshold using a region growing algorithm, and the radiometric backscatter threshold was established using an incremental convergence technique, employing multiple κ -statistic calculations with contemporaneous MODIS SWIR data. Both the stability of the radar threshold, and the instability of the MODIS SWIR reflectance threshold, were highlighted. The backscatter responses to two large flood events in the tropical savanna of northern Australia were investigated, showing markedly different results. One flood, in the floodplain of Queensland's Flinders River, involved total inundation of tussock grasslands over an area of 9000 km², allowing accurate classification using GM data (κ = 0.7), with predictable dihedral scattering returns as the flood receded and the emergent tussock grasses caused multiple interactions between the radar signal and the surface water. Inundated areas covered by emergent vegetation in the other flood, in Cape York's Staaten River floodplain, were almost completely indistinguishable from the surrounding wet vegetation. Data from water height loggers established in the neighbouring Mitchell floodplain over a dry/wet season period provided an insight into the interaction of these particular vegetation conditions under flood. Results concurred with the work of others, that backscatter response is a complex combination of effects depending on relative water height, vegetation spacial density, biomass, and verticality, or enmeshment, of super-surface grasses. The need for further work is discussed, together with spin-off opportunities, in the context of current and planned alternative C-band satellite data sources. The planned contribution of C-band data, along with contemporary visible/infrared products in the upscaling of current and ongoing JCU research into greenhouse gas emissions in the Mary River in the Northern Territory is outlined, together with the possible use of C-band radar to gauge fuel moisture content and fire potential, in the light of our findings in the tropical savanna. The potential use of GM data to explore correlation between Gravity Recovery and Climate Experminent (GRACE) data and surface water and soil moisture over time is discussed