Automated, Multi-parameter Identification and Tracking of Dry Lines: Algorithm Development and Evaluation

An automated, multi-parameter pattern recognition algorithm is being devised to locate and track dry lines present in WRF model output. As the temporal resolution and ensemble sizes of NWP models increases, the amount of information available increases as well. Consequently, studies that require manual identification or examination of meteorological features must begin to narrow their scope to maintain the same pace. However, by creating a system in which automated, objective analysis is possible, the scale of various studies can be greatly expanded, whether by increasing the number of parameter modifications attempted, the cases studied, or days examined, to name only a few. This project is designed to detect dry lines, a feature common to the US High Plains. Dry lines are a boundary between two air masses, one characterized by relatively warm and moist air, the other by relatively hot and dry air. This boundary is often an important feature in the initiation of storm cells, including those that develop into tornadic supercells. The identification of dry lines is achieved through the examination of multiple parameters, some of which are, to an extent, shared with similar meteorological features (e.g., wind shift along a cold front). To be able to accurately identify and track these boundaries, an automated method must be capable of incorporating multiple parameters as well as compare against similar meteorological features. Additionally, it will need to account for the inherent diurnal and seasonal variability of the dry line. The algorithm being developed will automatically analyze model output grids and identify dry lines by searching for regions with sharp moisture gradients accompanied by near constant virtual temperature. The initial approach is to identify moisture gradients using a Laplacian-of-Gaussian operator, cluster the gradients into candidate dry lines, and examine the variation in virtual temperature along the strong gradients detected. Preliminary results and limitations of this approach are presented, and future avenues of exploration are explained.