Long and Short Term View of U.S. Tornado Risk

28th Severe Local Storms, Portland OR

Transcript

1. Statistical Models for Tornado Climatology: Long- & Short-Term Views James B. Elsner (@JBElsner) Department of Geography, Florida State University Tallahassee, FL November 11, 2016 28th AMS Conference on Severe Local Storms Help: Thomas Jagger, Tyler Fricker, Victor Mesev Financial Support: RPI2.0 (Mark Guishard, John Wardman)
4. Why are we doing this? Dynamical models can’t predict tornadoes They can predict conditions necessary But necessary does not imply sufficient Statistical models are needed, but how should they be made? Here I show one way that is quite flexible Spatial model (county level) ⇒ Space-time model (grid level)
5. Model must be spatial Kansas Tennessee 1 10 100 El Nino Neutral La Nina Annual Number of Tornadoes [1954−2014]
6. Model must handle pathological records Enhanced Fujita Scale Mobile Radar Doppler Radar Discovery of Microbursts Fujita Scale Tornado Spotter Network Tornado Watch and Warning Program Systematic Tabulation of Tornado Data 1900 1925 1950 1975 2000 2025 Year
7. 0.8 1.0 1.2 1.4 0 20 40 60 Distance to Nearest City Center (km) Tornado Reports/10 km2
8. 1961−1970 1962−1971 1963−1972 1964−1973 1965−1974 1966−1975 1967−1976 1968−1977 1969−1978 1970−1979 1971−1980 1972−1981 1973−1982 1974−1983 1975−1984 1976−1985 1977−1986 1978−1987 1979−1988 1980−1989 1981−1990 1982−1991 1983−1992 1984−1993 1985−1994 1986−1995 1987−1996 1988−1997 1989−1998 1990−1999 1991−2000 1992−2001 1993−2002 1994−2003 1995−2004 1996−2005 1997−2006 1998−2007 1999−2008 2000−2009 2001−2010 2002−2011 0.1 0.2 0.3 0.1 0.2 0.3 0.1 0.2 0.3 0.1 0.2 0.3 0.1 0.2 0.3 0.1 0.2 0.3 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 Distance From Nearest City Center (km) Tornado Reports/100 km2 per Decade
9. Population (2012) 64 512 4096 32768 262144 2097152 16777216
10. Annual number of tornado reports 533 718 566 922 798 687 650 712 591 669 697 642 904 734 772 567 623 544 556 639 930 906 1035 893 814 949 873 881 1009 851 850 1017 758 1156 1350 1047 911 973 1441 980 1115 1394 776 780 732 1070 0 500 1000 1500 1970 1980 1990 2000 2010 Year Annual Number of EF0+ Tornadoes
11. Number of EF0+ Tornadoes (1970−2015) 0 40 80 120 160 200 240
12. The number of tornado reports in each cell (Ts) is assumed to follow a negative binomial distribution (NegBin) with mean (µs) and parameter rs. Ts|µs, rs ∼ NegBin(µs, rs) µs = As exp(νs) νs = β0 + β1 lpds + β2 (t − t0) + β3 lpds (t − t0) + us rs = As n where the mean of the distribution is linked to a structured additive response νs and the county area (As). The base-two log of county population density is lpds , t is the year, t0 is the base year set to 1991 (middle year of the record), n is the dispersion parameter, and us is the random effects term.
13. To account for spatial correlation the random effects term follows an intrinsic Besag formulation with a sum-to-zero constraint. ui |{uj,j=i , τ} ∼ N   1 mi i∼j uj , 1 mi τ   , where N is the normal distribution with mean 1/mi · i∼j uj and variance 1/mi · 1/τ where mi is the number of neighbors of cell i and τ is the precision; i ∼ j indicates cells i and j are neighbors. Neighboring cells are determined by contiguity (queen’s rule).
14. Expected Annual Tornado (EF0+) Occurrence Rate [per 100 km square region] 0.03 0.06 0.12 0.25 0.5 1 2 4 8
16. .01 1 10 .01 1 10 Raw Annual Rate [Tornadoes/100 km square region] Expected Annual Rate [Tornadoes/100 km square region]
17. Expected Annual Tornado (EF1+) Occurrence Rate [per 100 km square region] 0.008 0.016 0.031 0.062 0.125 0.25 0.5 1 2
18. Number of Tornadoes 30°N 35°N 40°N 2000 2001 2002 2003 2004 2005 2006 2007 30°N 35°N 40°N 2008 2009 2010 2011 100°W 90°W 2012 2013 100°W 90°W 2014 2015 1 2 4 8 16 32 64 128
19. −2 −1 0 1 2 1960 1980 2000 Year ENSO Index (s.d.) A −2 −1 0 1 2 1960 1980 2000 Year NAO Index (s.d.) B 24.0 24.5 25.0 25.5 26.0 26.5 1960 1980 2000 Year Western Caribbean SST (WCA) (°C) C 3 4 5 6 1960 1980 2000 Year Gulf of Alaska SST (GAK) (°C) D
20. Space-time model Ts,t|µs,t, rs,t ∼ NegBin(µs,t, rs,t) (1) µs,t = As exp(νs,t) (2) νs,t = β1 (t − t0) + β2 IDs + β3 GAKt + β4,s ENSOt + β5,s NAOt + β6,s WCAt (3) rs,t = As n (4)
21. ENSO Effect on Tornadoes [% Change in Rate/S.D. Increase in ENSO Index] 30°N 35°N 40°N 100°W 95°W 90°W 85°W −20 −10 0 10 20
22. Significance Level of ENSO Effect [s.d.] 30°N 35°N 40°N 100°W 95°W 90°W 85°W 0.0 0.5 1.0 1.5 2.0 2.5 3.0
23. According to Aon, the costliest U.S. thunderstorm outbreak on record occurred in late April 2011 across the Lower Mississippi Valley and cost insurers $7.7 bn in today’s dollars. Hindcast of Tornado Activity for 2011 [% of Long−Term Rate] 30°N 35°N 40°N 100°W 95°W 90°W 85°W 80% 90% 100% 110% 120% 130% 140% 150%
24. Why does it matter? 1 10 El Nino Neutral La Nina Annual Tennessee Tornado Fatalities by ENSO Phase [1954−2014]
25. Summary Tornado climatology needs improvement. Predictive climatology is possible with spatial statistical models. Models are built from historical tornado observations. Modern fitting techniques handle clustered and pathological records. Models produce long- and short-term views of regional tornado risk. Code for both views are available at rpubs.com/jelsner The most important factor in the short-term view is ENSO. New work will attempt to predict the risk of insured loss & casualties given the climate variables.