Daniel A. Yaussy and Elaine Kennedy Sutherland
           Research Forester and Research Ecologist, respectively,
         USDA Forest Service, Northeastern Forest Experiment Station,
                      359 Main Road, Delaware, OH 43015

ABSTRACT:  Changes in global climate might alter the frequency, duration, and
intensity of catastrophic weather events such as droughts, floods, and violent
storms.  As the frequency and severity of drought increase, so does the
probability of wildfires in forest ecosystems and the area burned.  This study
investigates the effect of the El Nino/Southern Oscillation (ENSO) climatic
phenomenon on the occurrence of drought and wildfire in the Ohio River Valley,
USA.  The strength of the ENSO phenomenon is measured by the Southern
Oscillation Index (SOI) and the Palmer Drought Severity Index (PDSI).  We used
the area burned by wildfires to associate PDSI and SOI with fire because it is
less dependent on human population density than the numbers of fires reported.
Although droughts could not be related to the Southern Oscillation, 25 percent
of the variation in area burned by spring and fall fires could be explained by
SOI and variables computed from PDSI measured more than 5 months before the
fire season.


   Changes in global climate might alter the frequency, duration, and
intensity of catastrophic weather events such as droughts, floods, violent
storms, etc.  As the frequency and severity of drought increase, so does the
probability of wildfires in forest ecosystems and the area burned.  The
ability to foresee these catastrophic events would better equip forest
managers to make contingency plans and develop fire management policies.

   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 et al.
(1985) found a high correlation with fires in the southern states and the El
Nino/Southern Oscillation (ENSO), but found little or no correlation in the
northeastern or southwestern United States.  Simard and Main (1987) used the
ENSO to develop a regression equation to predict the severity of the fire
season in the South.  Although these preliminary studies showed no relation
between the ENSO and wildfires in the southwest, Swetnam and Betancourt (1990)
found significant teleconnections between the ENSO and the occurrence of
wildfires in the National Forests of Arizona and New Mexico.  In this study we
investigated the interactions of the ENSO climatic phenomenon, the occurrence
of drought, and the area burned by wildfire in the Ohio River Valley, USA.  We
have loosely defined the Ohio River Valley to include the states of Illinois,
Indiana, Kentucky, Missouri, Ohio, and West Virginia.


   Our analysis was based on graphical interpretation and the use of the
Pearson Correlation test.  Correlations were calculated among the Palmer
Drought Severity Index (PDSI), Southern Oscillation Index (SOI), and the area
burned by wildfire.  To provide predictive capabilities, we calculated the
correlations between the variable of interest and values of the other
variables in the preceding 12 months.


   We chose the PDSI to measure drought intensity (Palmer 1965).  Based on
climate factors (temperature, season, region), water supply (precipitation,
stored soil moisture), and water demand (evapotranspiration, runoff,
percolation), the PDSI is slow to respond to sudden changes in the weather, so
the value from one month is highly correlated to the value of the previous
month.  Monthly PDSIs were obtained from data supplied by the National
Climatic Data Center for 1895 through 1988 for the Hydrologic regions of the
states in the Ohio River Valley ( Figure 1).

   The SOI is the calibrated difference between atmospheric sea-level pressure
at Tahiti, French Polynesia and Darwin, Australia.  When the SOI is low, the
atmospheric sea-level pressure is low in Tahiti and high in Darwin.  This is
called El Nino and corresponds to wet weather in the southwestern United
States (Swetnam and Betancourt, 1990).  When the SOI is high, it is termed a
La Nina and is associated with droughts in the Southwest.  Monthly SOIs in
this study are those developed and standardized by Wright (1989) for 1895
through 1988.  We used a 3-month moving average to smooth the SOI time series
to eliminate noise without masking trends, such that:

                 SOIAV(t) = {SOI(t) + SOI(t-1) + SOI(t-2)}/3.

This corresponds to 3-month seasonal averages used by others, such as Swetnam
and Betancourt (1992).

   In recent history, fires in the eastern United States have been considered
a cultural artifact ( Figure 2); that is, the number of fires is heavily
influenced by human population dynamics.  For example, the higher the
population density, the larger the number of fires set and reported.  In an
area of constant population, the number of fires should be distributed
uniformly, and the area burned by these fires should respond to climatic
influences and be less uniform in distribution.  Therefore, we chose to study
the area burned by wildfires, which is more indicative of the severity of a
fire season.

   Daily records of fire events were available for the Daniel Boone National
Forest (DBNF) in Kentucky from January 1970 through November 1992.  These data
were summed to monthly totals to assess the correlations with the SOIAV and
PDSI data sets.


   PDSIs for each hydrologic region for each month were compared to the SOIAV
time series lagged from 1 to 12 months.  The resulting correlations were low
(r < 0.1) but highly significant (p < 0.0001) due to the large number of data
points (n = 49,631).  On inspecting the data during extreme drought years
(1931 and 1953-54), we concluded that the hydrologic regions bordering the
Ohio River may have been more affected by drought than those farther from the
river ( Figure 3).  We recalculated the correlations, including only those
regions adjacent to the river.  This raised the correlations slightly (maximum
r = .18), but there was no meaningful relationship between PDSI and
combinations of lagged SOIAV, latitude and longitude by region, or month.

   The highest correlation between PDSI and SOIAV was revealed when
considering extreme drought events (PDSI more than two standard deviations
from the mean).  There was a 0.56 correlation (n = 39) between an extreme PDSI
and SOIAV from the previous 2 months.  Considering  Figure
4 and the small
number of observations, we doubt the usefulness of this statistic.

   Values for acreage burned obtained from the daily record of fire events for
the DBNF were summed by month to correspond to the monthly records of SOIAV
and PDSI ( Figure 5). 
 We correlated acreage burned (AB) with the SOIAV at lags
up to 12 months.  The most significant correlation was at 5 months' lag, r =
-0.20, p < 0.0018 with n = 228 (only the years 1970 through 1988 were present
in all data sets).  The 5-month lag corresponds to a similar lag found by
Swetnam and Betancourt (1992) for the Southwest.  They found a relationship
between the June fire season in the Southwest and SOIAV from February.  The
correlation between lagged PDSIs and AB decreased steadily as lag increased;
that is, the relationship between PDSI and AB was most significant in the
month preceding the month of the fires (r = -0.20, p < 0.0026, n = 228).  This
relationship is of little long-range predictive use.

    Figure 6 shows that the DBNF has two fire season each year -- March-April
and October-November.  Since the AB for the fire seasons is of greatest
interest in fire management planning, these values for March and April and
those for October and November were summed and analyzed as representing the
spring and fall fire season, respectively.  We found that the fall AB was most
highly correlated to the SOIAV of the previous June (r = -0.61, p < 0.0075,
n = 18), the 5-month lag reported earlier.  However, the spring AB was most
significantly correlated to the SOIAV from 10 months previous -- again, June
(r = -.039, p < 0.0879, n = 20).

   While re-examining  Figure 5, we noticed that the large values of AB
preceded the lowest values of PDSI.  This indicated that only several months
of declining PDSIs were needed to create conditions favorable for fire, which
is quite logical.  The larger values of AB followed periods of high PDSI,
which would stimulate lush periods of foliage growth (fine fuels).  The severe
fire seasons of the fall of 1981 and the spring of 1988 preceded the most
severe periods of drought in 1982 and 1988 ( Figure
5).  To capture this
relationship, a variable (DUR) was created that measured the length of time
(in months) that the PDSI had been increasing or decreasing.  If the trend
changed from increasing or decreasing or vice versa, DUR was set to 0.  If
PDSI was increasing (decreasing), 1 (-1) was added to the previous month's
DUR.  DUR should represent the length of the trend in PDSI, but does not
account for the high initial PDSI that would stimulate growth.  DURMAX was set
equal to DUR times the maximum PDSI of the preceding 12 months to quantify the
potential fine-fuel growth.

   Correlating these variables, lagged from 1 to 12 months, with spring AB and
fall AB provided the link between PDSI and AB.  Fall AB was most highly
correlated with the previous June's DUR (r = -0.34, p < 0.1212, n = 22) and
DUR from 12 months earlier, November (r = -0.34, p < 0.1170, n = 22).  Spring
AB was most significantly associated with the DUR from the previous June
(r = -0.50, p < 0.0160, n = 23), DURMAX of the previous October (r = -0.50,
p < 0.0213, n = 21), and DURMAX of the previous June (r = -0.61, p < 0.0025,
n = 22).   Figure 7 shows that these correlations are due to one or two highly
influential observations.  We usually deleted these obvious outliers, but
these extreme fire seasons are the events we are trying to predict.

  Table 1 lists the results of multiple linear regressions using the variables
mentioned.  Two independent variables explain 33 percent of the variation in
the amount of AB during the fall fire season, and three variables explain 51
percent of the variation of AB in the spring fire season.

Table 1.  Regression coefficients and measures of fit for area burned (AB),
          in hectares, for spring and fall fire seasons.

Variable      Estimate       Std. error of estimate        Significance
--------      --------       ----------------------        ------------
          Fall AB:  Adj. R-sq = 0.370, F = 6.002, p < 0.012, n = 18
Intercept      464.214              555.388                   0.416
SOIAV Jun    -1571.743              551.436                   0.012
DUR Nov        495.446              346.032                   0.173

         Spring AB:  Adj. R-sq = 0.512, F = 7.647, p < 0.002, n = 20
Intercept      459.947               91.134                   0.000
SOIAV Jun     -204.159               92.762                   0.043
DURMAX Oct     -18.655               10.649                   0.099
DURMAX Jun     -69.369               24.513                   0.012


   Although fires in the Ohio River Valley almost always are started by
humans, there is a relationship among severe fire seasons, climate, and the
SOI.  These connections offer some potential for predicting the severity of a
fire season.  A larger data set is needed to further test these connections,
and we soon will be gaining access to the large USDA Forest Service fire data
base and other data bases at the state level.

   Weather patterns and systems that develop and generally move from west to
east are influenced by the Southern Oscillation and the earth's surface
(mountains, bodies of water, soil moisture, vegetation, and so on).  Weather
and climate in the Ohio River Valley is strongly influenced by this, the
Atlantic Ocean, and Gulf moisture.  We hope to acquire a data base of the
North Atlantic Oscillation to enhance our ability to predict droughts and fire


   We would like to thank Kathleen Kennedy, Ginger Brudevold, Warren Heilman,
and Thomas Swetnam for providing data and insight for this project.

                               LITERATURE CITED

Palmer, W. C.  1965.  Meteorological drought.  Res. Pap. No. 65, U.S. Weather
   Bureau, Office of Climatology, Washington, DC.  58 p.

Simard, A. J. and W. A. Main.  1987.  Global climate change, the potential for
   changes in wildland fire activity in the southeast.  Proceedings of the
   symposium on climate change in the southern United States:  future impacts
   and present policy issues, May 28-29, 1987, New Orleans, LA.  U.S.
   Environmental Protection Agency, Washington, DC.  pp. 280-308.

Simard, A. J., D. A. Haines and W. A. Main.  1985.  Relations between El Nino/
   Southern Oscillation anomalies and wildland fire activity in the United
   States.  Agricultural and Forest Meteorology 36:93-104.

Swetnam, T. W. and J. L. Betancourt.  1990.  El Nino-Southern Oscillation
   (ENSO) phenomena and forest fires in the southwestern United States.  In:
   Proceedings of the sixth annual Pacific climate (PACLIM) workshop, March
   5-8, 1989, Asilomar, CA.  Edited by J. L. Betancourt and A. M. MacKay.
   California Department of Water Resources, Interagency Ecological Studies
   Program Technical Report 23.  pp. 129-134.

Swetnam, T. W. and J. L. Betancourt.  1992.  Temporal patterns of El Nino/
   Southern Oscillation -- wildfire teleconnections in the southwestern United
   States.  In:  El Nino -- historical and paleoclimatic aspects of the
   Southern Oscillation.  Edited by H. F. Diaz and V. Markgraf.  Cambridge
   Univ. Press.  pp. 259-270.

Wright, P. B.  1989.  Homogenized long-period southern-oscillation indices.
   International Journal of Climatology 9:33-54.