FTRE HISTORY TN THE OHIO RTVER VALLEY AND TTS R.ELATION TO CLI]IfATEI
Daniel A. Yaussy and Elaine Kennedy Sutherland2
ABSTRACT: Annual wildfire records (1926-77) from the national forests in states bordering the Ohio
fuver (lllinois, Indian4 Kentucky, Missouri, Ohio, and West Virginia) were compared to climate records
to assess relationships. Summaries of spring and fall fire seasons obtained for the Daniel Boone National
Forest in Kentucky (1970-92) and for the State of Ohio (1969-84, including national forest lands) allowed
for more detailed analysis. A wet spring followed by the onset of drought conditions, as measured by
Palmer's Drought Severity Index (PDSI), was significantly conelated to the area burned by wildfire the
following year (p < 0.0009). We hypothesize this was caused by a buildup of fine fuel followed by fuel
desiccation in the 6r* stages ofdrought. Extremely severe fall fire s€asons were weakly related to El Nifros.
Keywords: ENSO, ignition source, southern oscillatioq teleconnections, wildfire.
TNTRODUCTION
Teleconnections between large-scale atmospheric circulation patterns and regional weather patterns
might provide forest managers with an important planning tool. For example, in a broad, regionally
based study, Simard and others (1985) found a high correlation with fires in the southern states and El
Nifros, but found little or no correlation in the northeastern or southwestern United States. Simard and
Main (1987) then used about 50 years of data to develop a regression equation to predict severity ofthe
fire season in the south using the El Nifro/Southern Oscillation (ENSO) climate anomaly. Although
these preliminary studies showed only a weak relation between the ENSO and wildfires in a region that
includes the Southwest, Swetnatn and Betancourt (1990) found significant teleconnections between the
ENSO and the occulrence of wildfires in the National Forests of Arizona and New Mexico in 200 years
of data. We investigated these relationships for'the Ohio River Valley (OR$, defined as the states of
Illinois, Indiura, Kentucky, Mssouri, Ohio, and West Virgini4 and found littte association between the
ENSO events and Pdmer Drought Severity Index (a mea$rre of drought) in 93 years of dat4 but r
significant correlation beween spring and fall fire seasons from 1970 through 1992 in Kentucky and
variables constructed &om PDSI and ENSO (Yaussy and Sutherland, in press). In this paper, we have
added data from Ohio and looked at variables constructed from PDSI to capture the drought events.
I
l2th Conference on Fire and Meteorology: Fire, Meteorology, and the
Landscape; Jekyll Island, GA Oaober 26-28,1993.
Presented at the
2 USOA Forest Servicg Northeastern
Delaware OH 43015.
Forest Experiment Station, 359 Main Roa4
DATA and METHODS
Annual Nationd Forest fire records
(1925-77) for the ORV show that fires
in these states are a cultural artifact
(figure
l).
AJthough humans are the
primary ignition source for fires in the
east (gy/o) and culrural factors should
determine the number of fires, the area
burned by these fires should respond to'
climatic influences. For example, many
fires set during wet periods woutd burn
only small iueas; however, several 6res
set during dry periods could quickly
cover large areas. Therefore, we chose
to study the ar.ea burned by wildfires,
which is indicative of the severity of a
fire season as opposed to the number of
fires reported.
E'
co
:'
co
v,
tu
€
IJ
o
ddd-td- "C-#ru.uu
Flgur. f. Area burned by cause of fire
for the national forestl in the oRV
(Le4O-771.
we chose the PDSI to measure drought intensity (PaLner 1965). pDSI is based
on climEtc factors
(temperature' s€asol\ region), water
stored
soil
moisture),
@recipitatioq
and
wal€r demsnd
ryqntv
(orapotranspiration'
percoluion) posi islot"t respond to sudden changes in the weather,
PPq
so th€
value from one month is highly conelated to the-value
of the jrevious month. PDSJ is a standardized indea
so negative numbers indicate a drought-condition and positive values
are wet periods. A pDSI of -l
irdi.cates a slight drought while a -4 signifies a severe arought.
we are using pDSI as an indicator of fuel
moisture, which determines the intensity of the fire seasons]
Monthly pDSI measurements were obtained
from data supplied by the National Ctimatic Data Center for 1925
through 1992 forhydrologic regions of
the states in the ORV.
The Southern oscillation lndex (soD is the calibrated difference
between atmospheric sea-twel pressure
Potynesia ard Darwiq Ausilrali& when rhe SoI is low, the atmospheric
$.r{Alrench
sea-tw"t
is low in Tahitiand high in Dsrwin. This is called an El
Mflo .".nt -d *rrrrponis to w€t weather in th€
southwestern united states (Swefium and Betancourt 1990).
when the sol is rtigrr it is termed a Ia Mfla
tt'd qeiqry rssociated with droughts in the Southwest. Monthly
sol rieasurements used in this
I
study are tlnsc daidogod and gandardized by wright (1989)
for lg25 through lggt. we used a 3-month
moving average to rerooth the SOI time series toiti*in"t. noise witho"t
riLi"g trends, srch that:
ir*r,
*T
SOIAV, = (SO[ + SOLr + SOLJ3.
Transformations of PDSI and
solAv
also were used in this analysis.
Several types of fire data were used for this analysis: annual records
of area burned by wildfire for the
Nationd Forests in the oRV (1925-17); records olinai"iaua fire euenJl
for the Daniel Boonc
National Forest (DBhIF) in Kentucky; annual_fue records
Q9a2-92) fo, u. rrgio-ns of ohio protected by
ohio Department of Natural Resources (ODflR); and iecords orspring
!h9
v'u'i '
'v' Ohio
' e '---- A1 fire seasons for
tiiiiil
-i
(1969-84).
778
16,000
14,000
!q) 12,000
c
5 10,000
co
th
(u
8,000
C'
(J
6,000
c,
4,000
2,000
0
JAN FEB [4AR APR MAY JUN JUL AUG SEP OCT NOV DEC
Flgurr 2. Area burned by wildfire on the
fron 1970 to L992.
DBNF
accumulated by month
When the DBNF data are summarized by month, the two 6re seasons of the ORV, spring and frll, ue
apparent (figure 2). From figure 2 it appears that the fall season is more s€vere than the sprin& but 6gure
3 demonstrates that 1981, 1987, and l99l were exceptional fall 6re s€asons. To combine the DBNF ard
ODNR seasonal data sets, we summarized the individual event data from the DBNF into spring (January
through June) and fall (July tkough December) observations to coincide with the ODNR data. The DBNF
and ODNR seasonal observatiors of area burned (AB) were then divided by the area under 6re protectiorl
scaled by 1,000, to obtain AB per 1,000 ha protecled. A log transformation was taken of these values to
normalize the variables, LAB:
LAB = ln( (AB / (area protected) )
'
1000).
E
5
$
1970
1975
l9a5
l9AO
1990
1995
ll,gurr 3. Tlneline of area burned by nonth on the DBNF.
with the standard tests of ascertaining whaher the
vuiables were indeed normal, we interpreted normal probability plots (figure 4) of the data. These give no
Becanse the resrlting data sets were too snull
to
analy?E
indication that the variables are not distributed normally.
779
3
Cl)
.2
3
~
>
2
a.
"'0
j
Cl)
b.
2
+..J
i
Cl)
I
i
1
J
I
u
a.
r
w><
0
~
E
.._
I
0
·1
I,
·2
I
·1
0
z
I
·2
I
·3
·3
1.
·2
·1
0
-3
2
3
3
Cl)
.2
-3
I
'
>
2
"'0
c.
2
Cl)
u
Cl)
3
d.
.
0
·1
0
z
2
:
0
~
E
.._
0
1
a.
w><
·1
3
~
\
·2
·1
·2
·2
·3
-3
-2
·1
0
Standardized LAB
1
2
·3
3
·3
·2
·1
0
Standardized LAB
1
2
3
Fi gu re 4. P ro ba bi
(b .) DBNF fa ll , (cli ty pl ot s of st an da rd iz ed LAB fo
.) ODNR sp ri ng , (d
r (a .) DBNF sp ri ng
.) ODNR fa ll .
,
Since we were trying to
find variables that would
predict severity of fire se
SOIAV, PDSI, and transfo
ason, we analyzed value
rmations of these variable
s of
s from 1 to 12 months pr
seasons. Relationships
ec
eding the spring or fall fir
between LAB and these
e
lagged values were quan
statistics.
tified with Pearson corre
lation
RESULTS
Prior to World War II,
the value of wood produc
ts as war materials prom
fire campaigns on preven
pted the initiation of na
tion (Pyne 1982). Figure
tional
5 demonstrates the effec
and following World W
t of fire suppression du
ar II on the area burned
ring
in national forests of the
most severe fires coincide
ORV. Yet the years wi
with or precede droughts
th
the
that follow wet periods
lush growth of fine fuels
. This can be explained
during the wet periods an
by
d a rapid drying of these
fuels at the onset of drou
ght.
780
60,
1 40,
I
4
120,
H€ctares
loo,
Bumed
ao,
60,
40,
20,
2
PDSI
o
.
-4
s..-
SOLAV
4
€
€
.t
- / I,
19 75
Figurr 5. Tineline of the SOI, PDSI averaged for all
hydrologic.regions in the oRv, and area burned by wirdfire
in the national forests of the oRV.
The DBNF data show the same relationship as found for the annual records of the ORV large fire years
P,leceding the worst periods of drought, evident for 1988 and l99l when plotted against ttre ipst for thc
climatological region conraining the DBNF (figure 6). Ginger Brudwold, a member olthe Fire and Avirtion
staff of the Forest Service's Southern Regior\ assured us that these were s€vere fire seasons ard not a
change in fire suppression policy which woutd allow manageable fires to burn to lessen the fuel load.
Hectares
t,rr":O
2
PDSI
-2
1975
l9a5
4
€
€
tlgur. a. Tinerine of the pDsr Kentucky hydrologic region 4 and
area burned by wildfire in the DBNF.
The ODNR records do not begin before WWII nor show the increased fire suppression indicated for the
national forests (6gures 7-8) while still showing a relationship with fluauating iost a"eraged for the l0
hydrologic regiors in Ohio. Figure 9 demonstrates that the ODNR and DBNF data s€ts are bom different
di*ributions and must be analyzed separately; however, they do appear synchronous. The difference may
be attributable to diferences in land use and ownerships.
7El
12,000
8
I
10,000
6
I
,!
Hectares
4
1!
8,
ri
Eumed
2
I
6
o
lrr
4,000
tl
rf
I
I
2,W0
,
-2
ll
I
lr
,l
I
u
I
?
#rr::io''
'l
/r 11
/'
lr
: ^ u
,/J
ot
plguro
PDSI
..,
|
4
I
1
1
\
6
€
Tinerine of the
' and
area
uurnea--uv
"o_rT.
averaged
"irdfi;;'ii
hydror?grc
fgl."rr
iu'os profecteo oy regions
the
6,000
5,000
tto
E
€=
4,000
.t, 3,000
g
a(J
{u
2,W
1,000
q
1981
;iffi
1983
3;nfiltn"rison of area burned bv uildfire in areas protected
782
6
3
4
2
2
I
0
3o
'V/.i,,1'r
rigurr
1975
9. Conparison
DBNF and
the
4
1980 1985
\t
5 r'
1965 1970 r975 1980 1985 1990
tt
-1965 1970
-)
t\,1 -,,/
-2
-l
1990
of transforned fire-season variables for the
ODNR.
The most significant truuformation, PDDIF, was devised to reflect the buildup of fine fuels in wet periods
which is then followed by drought (figure l0):
PDDIF,{ = Ma\,*r..r,+.r2)(PDSI) - PDSL{
where:
PDSI,* = the lagged value of PDSI
index of month beginning the fire season (t = Ianuary for spring t = July for
fall)
k= index of lagged month (k = 1,..,12)
MDt,*r.t*-,2)eDSD = the ma"rimum PDSI meas,rement contained within the preceding 12 months
of PDSI,n.
t-
For example, when Ma\,*t..r*.,2)eDSI) is 3 (a wet period) and PDS[* is -3 (drought conditions) then
PDDIF* is 6. Thus, PDDIF,* should be positively conelated to LAB.
3500
I
3000
I
4
J
I
Hcctrror
Burned
2500
l!
2000
i!
2_
rl
'l
o
1500
-2
1000
1
500
OL
1979
lgal
1gao
tlgrur. 10. Calculation of
783
r9a2
PDDIF.
€
PDSI
so|Av
PDDIF of the preceding
luly (t = January,.k = 6) was the most significant variable
spring LAB on the D,BNF
pDDtr exprains about
'wtil;
4r%oof the
= 0 64, p 0 000e,
i
i: r)
ilryr iii-lrrv
u-ltT*.t
#*,*r*lr
;illiJffiSy,ff,lffiJil::;r*
6Z
prediaing the
il*r'l fJ.li.ill::fi,f
ro#'irrn, lornv of the
-o-.,*f,i0 1370, n = rq) _d *as the most
signincant
Explaining 32o/o of the variation
in the spring LAB for
the oDNR dat4 PDDIF of the
the nrcst significantlv conetatedv#;
previous Jury was
that the trend is opposite
n l-o.rrrr,n = l6). Ho*euer, figure r2 shows
*,rr. DINF tes,,r-, r iy *a the hypothesized
Tt'e PDSI ortne
positive correration.
iievlout 'r'"t
for
in falr rr*b or,r,. qDNR
data (r = 0 6r,
such as the rowest three orthe
rar sapr,
ii::"^,:::;r,
"ippii;
eugu't;.t
4ta/o;ffi;;on
,l,il;?'iil,.,,f]r;?Tm;'rll**#:;:f#fis
.t
3 .t,
1
a.t
2
Arr
a
l.l
rootlJ'
Flgrurr 11. correrations
of t|:nost sisnificant vaiiiurJ"
wirh
ofLAB for the Dry
totll"tM;ffbirnr.,
?!
rqr aa.a
rrgurr.12. correlation of the
yith
"oJi"ig"iriJi"i=iiriarres
ina iaii-L''of the oorii
j*i:;
;Bii:r and farr r"es-Ji-ffi
The relationstrips betrveen
the annual totals
the period 1940 to 1992
ll
in area
and crimate were investigated.
uuiJ a*
lffJiifl ,,?:3g,1ig,gi;,;um.ur"
t",ti!,
3
al2
51
a a
O.
r-O
-
a a
we
chose
ro $ppressioi Juring and
after wwII.
joj, p <
=b
0 0076, n = 53),
a
l
0
-1
-2L
-2
pDDtF Juty
rlgrurt 13. Relationshio of_annual
total.IAB of the
fron 19{O to 19e2 ana irooir-"'e-ir,'""iilrrro,rs
Jury.
DBIIF
From tlrese res.rltg PDDIF of rhe prwious July appean to be a variable associated with LAB. We analyzed
the coneluion of this variable with the LAB observations of all the national forests in the ORV from igcO
to 1977. The resrlts in table I indicate that PDDIF is not conelated to LAB for states in the ORV with the
possible exception of Kenrucky.
l.- Pearson conelation statistics for LAB of the national forests in
the ORV and PDDIF of the preceding July from 1940 to 1977 (1940-92
Trblc
folKgntuckv)
State
p-level
Number
Illinois
0 060
0.7t9
38
Indiana
-0 0r6
0 926
38
Kentucky
0 363
0 008
53
Missouri
0 015
Ohio
-0.039
9e3
0.815
38
38
CONCLUSIONS
Visually and logically, there should be a sound relationship betrrreen area burned and climatic factors
such as drought. The most severe fire seasons on the DBNF were in the fall of 1981, l9g7, and 1991.
These are the types of events we would like to be able to predict. With the methods used in this study,
these seasons were the least conelated with the variables we chose to depict climate.
The reason for the lack of conelation may be that the relationships in climate data look good for a short
time but tend to degenerate when the time &ame is ercended. We have been ignoring interference 6'om
other sources, such as the North Atlantic Oscillatioq and that may explain this degeneration; or the
apparent relationships may well be spurious @urroughs l99Z).
As any statistician would say, "We need more data.' We are in the process of obtaining records of
individual events from the national forest-fue data base and &om state agencies. However, much of
these data cover thc 1970's and 1980's, which had few exreme fire events. We atso will er*end the SOI
data through 1992 to irrclude the most recent El Nifro events. Since climate in the ORV is driven by a
combination of thc prcv"ailing westerlies, inlluenced by Et Nulos, and moisture from the Gulf of Mexico,
we will to explore thc effects of the North Atlantic
Oscillation.
The effect of drought following a period of growth on 6re s€ason severity warrents further attempts to
quanti$ and investigate the relationship.
785
/
LITERATURE CITED
BtrRRoUgFIfuYJ,l?i;,:rTiH
PA'ME&
cycles: Rear or rmaginary cambridge
University press, New
X;i;"i?1;#;:fg"i* l;:"*,
Res pap No 65 u
s
weather Bureau, office
or
PYIJE, s.J. 1992. Fire in America:
A currurar
s History
ruJtvrt of
ur wirdrand
wuoliUl(t ttre.
Fire. pri
PnnCAon University preSS,
PrinCetOn, NJ, 654 pp.
SIN'{ARD'
AJ;
and MAIN' w'A' l987.. Globat
climate change, the potentiar for
changes in wirdrand
actiuity in the southeast' Proceedingt
fire
on^crimate change in the southern
United
;lffi *ffi.,T',ffi
STMARD'
AJ'; FIAINES' D'A;
oiri; t ily
#,[T,1tr[*J:t*;ifi .1'#]$.*o,r,-i,ln-ls
and
anomalies and wildland fire
36:93- I04.
lv{AlN,
w.A
1985. Relations benveen
Er Nr.f,o/Southern oscilration
aaivity in tne
i,tJrJJ",.r. egri*i*rJ
-a
Forest Meteororory
SWETNAM T'w'; and BETANcouRI
J'L' 1990 El Mfro-southern oscilration
(ENsoy pheronreru
and forest fires in the southwesT
stateslu
no.*aing,
or*"
si"tr, annuat
(PACL&O workshop, Marctr
YTg
s-e, tqrq,
paiti Uy a.*Ly1 J Lpacific ctimarc
urd MacKey,
watei
rnteragencv-E*
i#i
tuil*ita.
it[H,?:iTT:,11or
il;;;
i"E."r
s
tudies prograrn
SWETNAM T w; ard BETANCOURT,.J.L.
1992. Temporal panerns
of El M.fio/Southern oscillationwildfire teleconnections in trtt
*tr,t o;; u;;ffi;* In Er Mflo:
ruo"ri*r and pareocrimaric
oscillation' Edited
anJ
?:cambridge universitv
btDa;r.
frffir::S-H*
*ooF*t"i;"H13,,!-oSosenizea
YAussY' D'A';
"J;"{
long-period southern-oscittarion
indices. tnternationar Journar
of
qI q
and SLJTHERLAND'
press). Relationships of
the southern oscilration of
Pacifc ocesn rnd drorghts ana
the
witaeres irirr,r tiN" nrrer valrev.
and
managemenr
events' Edited bv lvtevers, w.
andF.ir**r,o, s. university of
h;;;;;"ry
fiffiffid"?s.'optric
786
Proceedings of the
12th International Conference
on Fire and Forest Meteorology
October 2*28, 1993
Jekyll lsland, Georgia
Society of American Foresters
5400 Grosvenor Lane
Bethesda, MD 2081Z|.2198