VCHILL 2: A System for Semantic Processing of Radar Data

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 7 January 2015: 1:45 PM
132AB (Phoenix Convention Center - West and North Buildings)
Joseph Hardin, Colorado State Univ., Fort Collins, CO; and V. Chandrasekar, M. Wiesner, and S. Derbyshire

Current radar archive systems are primarily date based, in that one searches through files by accessing a particular time period, and manually inspecting each file to determine whether that file satisfies some set of criterion. This is a very laborious process, and usually devolves into asking someone involved with an instrument whether they remember a satisfactory case. This work presents a new methodology and implementation of a system designed to allow users to intuitively work with the semantic meaning of radar data. Semantic meaning involves both context and relationship meanings about what the data actually represents. Instead of manually searching each file, a user is able to search the data by metadata. This metadata includes both low level information such as type of sweep and coverage, as well as higher level semantic attributes such as presence of a bow echo, or the cumulative distribution function of velocity.

This system is based around a series of interconnected components leveraging geospatial databases, HTML5 web servers, and a RESTful interface implemented in python. The system currently contains over 14 years worth of data from the CSU-CHILL radar, however the backend implementation accepts multiple different radar formats from a wide variety of instruments including NEXRAD, IRIS, and UF based systems.

This system allows for higher-level search functionality than is currently possible with operational systems. An example query we empower is a user requesting all tornado possible days, defined as days with at least 30 % of the scan having 30 dbz or higher, and maximum measured velocities of 40 m/s or higher. Another example would be requesting flood prevalent conditions defined as days with rainfall coverage in at least 50% of the scan, and maximum rain rates of greater than 10 mm/hr.

These are just a few possible examples and the user is in full control of how they want to define their criterion. The backend design is very modular and it is very easy to add new types of data. Plugins just need to accept a radar common data model object that is shared between multiple python libraries(pyart, pydisdrometer, etc.).

This paper will detail both the implementation of our system, as well as discussing the new types of workflows this methodology allows.