Download files
Abstract
Accurate spatial and temporal distribution of rainfall is important for hydrological applications. Rain gauges and weather radars are most widely used sensors for rainfall measurement. This research is a step towards improved rainfall estimates by reducing the weaknesses that restrict their ability to represent the rainfall field properly. This thesis focuses on the methods to minimize the uncertainty of radar rainfall estimates in parametric approach. Considering the limitations of the parametric approach, a nonparametric method has been developed. This research also demonstrates a new technique to merge radar and gauge rainfall. Finally, studying several existing merging methods, the optimal method has been suggested.
In parametric approach, the uncertainty in reflectivity to rainfall conversion parameters originates from the inaccurate representation of spatial distribution of rainfall from rain gauge at radar grid resolution, and from incompatible temporal resolutions of radar-gauge pairs. An error model has developed to compute the spatial uncertainty of gauge. Then SIMEX method uses this error model to determine the uncertainty present in the conversion parameters. The uncertainty related to temporal resolutions is addressed by using the optimal temporal resolutions of radar reflectivities for a specified gauge temporal resolution that results best radar estimate across ten Australian weather radars.
A nonparametric (NPR) method for converting radar reflectivity into rainfall is suggested that uses the bandwidth of reflectivity and rainfall pairs. Compared to parametric method the NPR method performs better in reducing error at around 90% of the locations. The NPR has smaller errors than spatially interpolated gauge rainfall in regions with low gauge density.
Methods to merge radar and gauge rainfall are investigated to avoid the concerns about the biases in radar rainfall and difficulties of gauges to capture spatial representation of rainfall. The proposed dynamic combination method uses weights that vary in space and time and is calculated from the error covariance matrix. The performance of the new method is better than the individual radar or gauge estimates and is about 20% more accurate than the traditional parametric method. Comparing this new and other existing merging methods, an optimal method is suggested depending on the attributes of interest.