Robust spatial quality control of road weather sensor measurements

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 18 January 2010: 4:15 PM
B302 (GWCC)
Tressa L. Fowler, NCAR, Boulder, CO; and M. Limber, S. Sullivan, M. B. Chapman, S. D. Drobot, and T. L. Jensen

Presentation PDF (718.2 kB)

Spatial quality control tests compare measurements against neighboring measurements to determine their reasonableness. This paper describes a robust automated method for making these comparisons, along with some preliminary results, based on data from the Clarus network. Algorithm design considerations include appropriate weather variables for spatial testing, defining neighbors, minimum number of neighbors to require, and how to use measurements from the neighbors to define a range of reasonable values for a target measurement. A minimum number of neighboring stations must be available in order to perform any spatial quality control test. Several minimum neighbor sizes were considered, and five was selected as a good compromise. When too few neighbors are used by the test, the results are poor. When too many neighbors are required, many stations cannot be tested as they lack sufficient neighbors. This minimum size should vary according to the density of the network and the spatial continuity of the measurement.

The real-time nature of this test causes difficulties in neighbor comparisons. Measurements from neighboring stations cannot be quality checked by basic tests prior to their inclusion in the spatial quality check since all tests are run simultaneously. Though a gross error in measurement at neighbor will almost certainly be flagged by this or some other test, that bad neighbor can also skew the test on good measurements at its neighboring sites. In this situation, the spatial algorithm must be somewhat robust to gross errors. Thus, the median and inter-quartile range, both robust statistics, are used to define the range of reasonable values for a target measurement. The algorithm was tested on a set of cases. Overall performance is good, though it varies somewhat by location, type of measurement, and weather condition.