This is a repost of my post from the now defunct Whatever-Weather.com.
In my last post relating Geographic Information Systems (GIS) and Meteorology, I mentioned that GIS is a way to relate the real world to the raw and derived meteorological output. It is far from a perfect relation of course. Many GIS datasets have much larger scales than meteorological output (e.g. roads in the misoscale/meso-gamma scale and WSR-88D data at the mesoscale), so they are too precise relative to meteorological datasets.
I first want to do a quick review of precision. Precision is the level of measurement and exactness of a dataset. It often gets confused with accuracy, which is the degree in which the data matches the true or accepted values. It should be noted that high accuracy does not imply high precision and vice versa (Foote 2000). The terms are often used interchangeably and are often confused with each other.
Let’s set up a basic overlay with a long range WSR-88D radar scan and a road Shapefile dataset from the US Census Bureau’s Tiger Line Shapefile set as an example. The current dataset has an expected error of around 7.6m or 25ft. (By road engineering standards, this dataset is considered a cruder dataset, but in this case it is the more accurate dataset.) For the radar data at that range, the beam height and width values are many orders of magnitude greater than the road file. For example, at 150mi from the radar the beam width is about 15,000ft wide and at 250mi it is 25,000ft wide (Letxa.com 2000), and the height at that distance varies by tilts, from anywhere from 15,000-70,000ft.
This can be shown clearer when trying to compare actual tornado tracks to the locations of the radar returns. In 2006 Douglas Speheger and Richard Smith of the Norman, Oklahoma Weather Forecast Office compared the two. They found significant errors at both close and distant ranges between the range gate value locations and track locations as shown in Figure 1.
At closer ranges, (30mi) radar returns were off by 1-2 miles of the post-event surveyed track and at longer ranges (110mi) the returns were off by 8 miles. They pointed out that location error of 1-2mi is unacceptable for real time pathcasting of tornado tracks as shown in Figure 2.
Some of the error is from tilted mesocyclone structure. The weakness is in the beam height at that distance which can only be mitigated with a higher density radar network.
The Warning Decision Training Branch (WDTB) has an online course related to this paper called “Pathcasts in Severe Local Storms”. The course also mentions from Speheger and Smith’s paper that when producing warnings, cities and notable locations are represented as points in AWIPS when many urban locations are best represented by polygons showing the outer boundary of the area, and right now it mentions how best to adjust for the locations for the maximum clarity in each warning product.
Another example would be surface observations. ASOS/AWOS locations at airports are spread out somewhere between the synoptic and mesoscales, that if they were gridded and interpolated, a useful grid would be around 30km by 45km (UW Extention). You can never have too many observation sites, but it good to know the scale limitations with the current system.
Error, Accuracy, and Precision, Foote 2000
On the Imprecision of Radar Signature Locations and Storm Path Forecasts
Technical Explanation of NEXRAD, Letxa.com
Tiger/Line Shapefiles 2009 Technical Documentation (Page 31)
(c)2010-2012 Charles Schoeneberger