Development of New Snowstorm Indices and A New GIS Snowstorm Database at the National Climatic Data Center Mike Squires, Jay Lawrimore, Richard Heim National Climatic Data Center David Robinson, Mathieu Gerbush, Thomas Estilow Rutgers University Clay Tabor University of North Carolina at Asheville Anna Wilson STG
36
Embed
New Snowstorm Indices and New GIS Snowstorm Database€¦ · Northeast East North Central Central Southeast South. Index Value. 0 5 10 15 20 25 30 35. Distribution of Regional Indices.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Development of New Snowstorm Indicesand
A New GIS Snowstorm Databaseat the
National Climatic Data Center
Mike Squires, Jay Lawrimore, Richard HeimNational Climatic Data Center
David Robinson, Mathieu Gerbush, Thomas EstilowRutgers University
Clay TaborUniversity of North Carolina at Asheville
Anna WilsonSTG
Overview• Goals:
– Bridge the gap between Wxr/Climo and Societal Impacts
– Feedback and identify stakeholders
• Northeast Snowfall Impact Scale
• Data processing issues
• New Regional Snowstorm Indices
• New GIS Snowstorm Database
Northeast Snowfall Impact Scale (NESIS)
• Area of snow, amount of snow, and population affected by >4” of snow
• Societal Impacts
• Puts Northeast snowstorms into historical perspective
Data Processing Using a GIS
• Extract snowfall data from the Global Historical Climate Network – Daily dataset
• Identify stations associated w/ a particular storm
• Quality Control
Unedited Data:January 15-19, 1978
Day 1:January 15, 1978
Presenter
Presentation Notes
The analyst looks at “Daily Weather Maps” to get a general sense of the daily progression of the storm.
Delineated Storm:January 15-19, 1978
Original Delineated
Unrelated Snowfall:January 15, 1978 (Day 1)
Presenter
Presentation Notes
The highlighted stations are the locations that received snow on Day 1. The snowfall in the west is part of the storm being analyzed; the snowfall in the east is part of another storm and must be removed.
Final Dataset:January 15-19, 1978
Delineated Final
Presenter
Presentation Notes
The map on the left shows the final spatial distribution of stations after unrelated snowfall has been removed.
Final Dataset:January 15, 1978
Presenter
Presentation Notes
This slide and the next four slides show the daily evolution of the January 15-19, 1978 snowstorm.
Final Dataset:January 16, 1978
Final Dataset:January 17, 1978
Final Dataset:January 18, 1978
Final Dataset:January 19, 1978
Manual Quality Control of GHCN-D
• Quality Control (QC) is a quasi-subjective process– Want a reliable, robust, repeatable method
• Making QC decisions in the face of uncertainty– Type I error – false positive – throwing out good data (unacceptable)– Type II error – false negative – keeping bad data – The goal is to minimize Type I errors at all costs, so:
• Assume all data is correct unless proven otherwise• In general snowfall is spatially continuous, however
– topography– Lake effect snow– Differences in observation technique– Small scale spatial variability
• Do not change values• Tools: ArcMap, Moran’s I, snowfall grid (bull’s eyes)• Typically ~1% of stations are removed
January 15-17, 1945
Presenter
Presentation Notes
This slide and the next slide show an example of a case where the analyst consulted with the actual observation form to determine the validity of a questionable snowfall total.
Presenter
Presentation Notes
It looks as though the observer meant to report five inches here on day 2; a 7 inch total would be much more reasonable than the 2 inches that we are looking at. This observation was removed.
Regional Snowfall Impact Scale (ReSIS)
• Collaborating with Rutgers University
• Differences from NESIS;– NCDC Regions
– Snowfall and population are confined to specific region
• Finalizing algorithm to compute indices
Intra-Regional Differences
ReSIS = 19.648
ReSIS CalculationMarch 12-13, 1993
Snowfall Population Area (sq mi)2 6,311,630 46,6665 8,110,060 41,237
This slide shows the ranked indices and the relative contribution in percent of each of the four terms from the ReSIS algorithm. Using this method, indices for storms in the top of the distributions are being driven by the third and fourth terms. Indices for storms in the bottom of the distribution are being driven by the first and second terms. In the middle of the rankings, the final index value is comprised of contributions from all terms. This pattern of how the individual terms contribute to the index values is a desirable attribute to this method. All of the regions behaved this way.
This table and map combination shows how a high ReSIS value and a medium ReSIS value relate to a conventional snowfall map. The stars on these maps represent metropolitan areas with populations over 500,000 people.
Northeast
East North C
entral
Central
Southeast
South
Inde
x Va
lue
0
5
10
15
20
25
30
35
Distribution of Regional Indices
Presenter
Presentation Notes
These box plots represent the distributions of regional indices. In these box plots, all of the outliers are displayed. The boxes encapsulate the central 50% of the distribution, the whiskers define the central 90% of the distribution. The median is represented by the solid horizontal line and the mean is depicted by the dashed horizontal line. Although there are some differences between individual regions, the distributions of all the regional snowfall indices are quite similar. The Wilcoxon-Mann-Whitney rank sum test was applied to this data to formally test if these distributions are similar. The null hypothesis is that the two data samples are from the same distribution. The test was constructed so that each individual regional distribution was compared to the pooled distribution of all regions. The test was repeated five times; once for each of the five individual regions. In all cases the results were found to be insignificant. These results give more evidence to the observation that these distributions are similar to each other.
New National Snowfall Index
• Created any time a regional index is calculated
• Societal impacts and historical perspective
• Allows discrimination between national and regional impacts
• Algorithm under development
New GIS Snowstorm Database
• Collection of top 100-200 U.S. snowstorms from 1900 to present
• “GIS format”
• Will be used to calculate the new national snowfall index
• Could be used by other researchers, policy makers, and the general public
• Societal impact
GIS Snowstorm Database
Miles of Interstate Highway Affected by Snowfall Above a Threshold
Storm Date > 4" > 10" > 15" > 20" > 30"
1977-01-08-11 16,318 2,221 668 58 0
1993-03-12-15 17,249 12,485 7,311 3,295 277
1996-01-06-09 16,063 11,418 8,095 4,490 143
2005-12-07-10 13,039 1,674 43 0 0
Miles of Railroad Affected by Snowfall Above a Threshold
Storm Date > 4" > 10" > 15" > 20">
30"
1977-01-08-11 42,505 5,190 1,478 91 0
1993-03-12-15 38,972 29,357 18,799 9,445 620
1996-01-06-09 32,805 22,219 15,568 9,837 446
2005-12-07-10 33,299 2,795 73 0 0
Pre-Summarized Tables
Presenter
Presentation Notes
This is an example of how the available tools will transmit information to the user via attribute tables. There will be similar information for airports, hospitals, retail centers, and schools.
Potential Interactive Queries
• How many square miles were affected by snowfall >10” during the Jan 1996 storm?
• Which schools in Illinois and Wisconsin have experienced snowstorms >20” in November?
• How many miles of interstate received >30” in New York during the Mar 1993 storm?
• Which hospitals in Pennsylvania have been affected by > 15” of snowfall in any month?
St Louis University & National Weather ServiceAnalog Snowfall Guidance
NCDC GIS Snowstorm DatabaseNumber of Hospitals Affected by Snowfall
Above Specified Threshold
Analogs for 2008/12/19
Rank Date Score
1 1994 12 07 7.741
2 1990 12 15 6.859
3 2000 02 18 6.813
4 1993 02 21 6.758
Presenter
Presentation Notes
St. Louis University and the National Weather Service are developing an Analog Snowfall Guidance product. The purpose is to identify historical snowstorms that are similar to the 48h or 72h GFS forecast. The North American Regional Reanalysis (NARR) is searched using various statistical techniques to identify the top 15 analogs. The goal is not to produce a deterministic forecast but rather give guidance about the magnitude and scale of impending snowstorms. This product can be used with the GIS Snowstorm Database (GSDB) to give an indication of the societal impacts the storm will produce. In this case, analogs are identified for a forecast valid at Dec 19, 2008. The top four analogs are listed. The fourth rated storm is an event that occurred in February of 1993, which has an entry in GSDB. The table for hospitals is shown, indicating that approximately 150 hospitals experienced over 10” of snow during that storm. If the forecaster has confidence in the forecast, it is reasonable to assume that there will be similar impacts for this storm. Of course the actual location of heavy snow in relationship to the locations of the hospitals will affect the actual number of facilities affected. There are also tables available for interstates, railroads, airports, retail centers, and schools.
Issues
• Actively seeking out stakeholders and users• “GIS format”