831 A Technique for Automated Hail Pad Processing

Wednesday, 9 January 2013
Exhibit Hall 3 (Austin Convention Center)
Peter Marinescu, Stony Brook University, Stony Brook, NY; and B. Windschitl

Handout (330.7 kB)

This study will focus on a method to automate hail pad analysis using image processing programming. The capability to measure hail pads efficiently, accurately, and objectively can create a new set of observational data that may have numerous applications in weather and climate studies.

Hail pads are one-inch thick Styrofoam blocks that are often created to have a twelve-inch length and a twelve-inch width. These blocks are also wrapped in heavy duty aluminum foil. The purpose of hail pads is to measure the characteristics of hail that strikes and leaves dents in the pad. A volunteer network of weather observers, Community Collaborative Rain, Hail & Snow Network (CoCoRaHS), was organized in 1998 to help document rain, hail and snow measurements. CoCoRaHS has created an extensive network of volunteers throughout the United States and Canada and has increased the number of hail pads used and recorded. Furthermore, CoCoRaHS generates hail reports regardless of the size of hail. Most standard hail reports from other organizations solely focus on severe hail, which the National Weather Service defines as greater than one inch (2.54 centimeters) in diameter. CoCoRaHS provides an additional data set that includes both severe and non-severe hail observations through its hail pads and reports.

Currently, hail pads are processed by a CoCoRaHS staff member. First, the staff member determines sample regions on the pad; the size of the sample regions depends on the frequency of hail markings on the hail pad. Then, the staff member studies, interprets, measures and counts hail markings in the sample regions. Markings considered to be caused by hail are then categorized by size (small, medium, large or jumbo). The counts from each size category for all the sample regions are averaged and scaled up to the size of the hail pad to provide an estimate of the total number of small, medium, large and jumbo hail stones that impacted the hail pad. This process takes into account numerous assumptions, including that the sample regions are accurate representations of the entire hail pad and that human judgment can objectively determine hail markings from one pad to the next.

Various research articles have attempted to calibrate hail pad data and have created adjustments for some of the errors in hail pad processing. However, limited efforts have been undertaken to automate the current subjective and manual task of processing hail pads, especially with respect to the CoCoRaHS hail pad network. In this study, an algorithm was developed using MATLAB ® that is based on the Circular Hough Transform and aims to detect circular hail markings on hail pads. Simple photography techniques using multiple illumination sources and angles were initially used to determine whether photograhphy can improve the algorithm's results. The algorithm measures and counts detected hail markings and categorizes these markings into size groupings similar to the CoCoRaHS processing method to test the accuracy of the algorithm. The researchers also counted and measured all markings considered to be caused by hail on each hail pad (Research count) as a check value in this analysis.

Initial results using four hail pads with varying hail characteristics provide promise in the development of an automated algorithm that can objectively and accurately detect, count and measure hail dents on hail pads. From these four pads, there was no clear benefit from using a specific photography technique. Therefore, sunlit images were used for algorithm development, with the hopes that volunteers from across the widespread CoCoRaHS network can submit photographs of hail pads themselves. An example of the algorithm's results and output from one of the four hail pads can be seen in Figure 1. The results using all four hail pads demonstrate the algorithm's high accuracy in detecting small and medium size dents, which are associated with markings less than 0.75 inch (1.9 centimeter) in diameter. Small dents detected by the algorithm had less than 20% percentage differences from both the CoCoRaHS and Research counts for three of the four hail pads. However, the algorithm displayed less accurate results for large and jumbo size dents. Further testing of the algorithm on additional hail pads in the upcoming months will lead to more information on how to improve the algorithm's results.

Figure 1. A sunlit photograph of the CO-LR-290 hail pad from a hail event on July 13, 2011 near Fort Collins, Colorado (top left) and the algorithm results plotted atop the hail pad image (top right). One of the main benefits of such an algorithm is the production of hail size distributions (bottom left), which would provide additional data regarding hailfall. The algorithm is compared to both CoCoRaHS and Researcher counts to determine the algorithm's accuracy (bottom right).

The benefit of such an algorithm is two-fold; one benefit is more narrowly defined, while the other is more comprehensive. First, automated hail pad processing will provide CoCoRaHS with an efficient and objective method of analyzing hail pads, which with the growth of hail pad use will be important for CoCoRaHS operations. Second, the capability to create accurate hail size distributions with high frequency over expansive regions will benefit the scientific and operational meteorological community in the form of a new set of observational data regarding hailfall characteristics. A few examples of the uses of this data may include better understanding hail formation in convective thunderstorms, creating hail climatological studies and improving hydrometeor classification algorithms that may aid hail forecasting efforts.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner