Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Computer Science

Supervisor

John Barron

2nd Supervisor

Robert Mercer

Joint Supervisor

Abstract

Severe weather forecasting is one of the most important and urgent tasks in the meteorology field. This thesis builds on previous work by Barron and Mercer and their graduate students, concerning the use of 3D optical flow to retrieve 3D wind velocity from 3D Doppler radial velocity datasets and tracking 3D severe weather storms using fuzzy points realized as ellipsoids to represent storms and a fuzzy algebra machinery in a relaxation labeling framework to track storms in Doppler precipitation datasets.

We first extend the original 3D optical flow (both least squares and regularization methods) for recovering 3D wind velocity from the multiple overlapping Doppler radial velocity fields. The enhanced methods exhibit improved performance, especially in overlapping radar areas. We also add 3D windprofiler data into our framework. We show that windprofiler data allows the vertical component of 3D velocity to be more accurately recovered. We perform a quantitative analysis on synthetic Doppler data and a qualitative analysis on real Great Lakes Doppler datasets and show that both multiple Doppler data and windprofiler data significantly improve the performance. Our optical flow general frameworks lends itself to adding new sources of data and new constraints on that data.

We also use a "pseudo" storm concept to solve the tracking problems caused by merging and splitting of severe weather storms over time. We first modify the original tracking algorithm to add a pseudo storm definition to it. Then, an advanced storm tracking algorithm taking full advantage of pseudo storms is presented. We compare the results using the original storm tracking algorithm, the original storm tracking algorithm with pseudo storms added and the final advanced pseudo storm tracking algorithm. The advanced pseudo storm tracking algorithm outperforms the other storm tracking algorithms for Great Lakes Doppler precipitation datasets.


Share

COinS