References

1
Detecting climate signals in river discharge and
precipitation data for the central Georgia
coast Joan Sheldon and Adrian Burd Department of
Marine Sciences, University of Georgia, Athens,
GA, email jsheldon_at_uga.edu
Abstract Identifying the effects of global change
requires identifying global-scale signals in
local data. We are seeking evidence of climate
signals such as the Southern Oscillation Index
(SOI) and the North Atlantic Oscillation Index
(NAOI) in long-term data for the Georgia coast.
The NAOI leads the SOI by 1 month and the two are
very weakly correlated, so these two signals
could explain different patterns in local data.
Monthly standardized anomalies of Altamaha River
discharge and Georgia coastal precipitation were
constructed by normalizing, deseasonalizing, and
detrending those data. Climate indices were also
transformed if necessary, and all series were
prewhitened prior to cross-correlation analyses
to determine the most appropriate lags between
series. Altamaha discharge and coastal
precipitation are weakly negatively correlated,
suggesting that coastal and inland precipitation
patterns are different. However, climate signals
explain very little of the variability in the
local data (R2 lt 0.02). This analysis will be
extended to other datasets for coastal Georgia.
Georgia Coastal Ecosystems Long-Term Ecological
Research (GCE-LTER) Project The project study
site, located on the Georgia coast in the
vicinity of Sapelo Island, encompasses the lower
Altamaha River estuary, Doboy Sound, and Sapelo
Sound. The Altamaha River watershed is one of the
largest watersheds on the east coast of the U.S.
(36,718 km2) and delivers freshwater to the coast
at an average rate of 400 m3 s-1, whereas Sapelo
and Doboy Sounds receive little or no direct
riverine inflow and are influenced more by local
rainfall and runoff. The first six years of the
project have focused on the differences and
interactions among these three estuaries,
especially with regard to the effects of these
differences in freshwater inputs.
Results and Conclusions Many cross-correlations
among the data have very low correlation
coefficients (R) but are statistically
significant only because of the large sample
sizes. These cases are described as weak
correlations with low explanations of variability
(low R2 but plt0.05) as opposed to nonexistent
(pgt0.05). Only a few cross-correlations are both
statistically significant and possibly
substantial in effect. R and R2 values are
reported for the transformed variables.
Cross-correlations prior to whitening series
Cross-correlations after whitening series
Introduction Large-scale patterns in atmospheric
pressure, circulation and oceanic temperatures
have important influences on weather at global
and regional scales. El Niño / Southern
Oscillation (ENSO) The atmospheric component of
this well-known pattern, the Southern
Oscillation, is reflected in air pressure
differences between the western and eastern
tropical Pacific. The Southern Oscillation Index
(SOI), one measure of this pattern, is usually
calculated based on the anomaly (difference from
normal) in air pressure between Tahiti and
Darwin, Australia and corresponds well with
changes in eastern tropical Pacific Ocean
temperatures.
As expected, the two NAO indices are positively
correlated, with no lag, when they overlap
(R0.7). This is not as high as might be expected
but is probably related to the fact that the NCAR
index uses two stations whereas the NCEP index
uses a RPCA of many stations. The SOI is not
highly correlated with either NAO index
(0gtRgt-0.1). Slight but significant negative
correlations indicate that the NAO indices may
lead the SOI by 1 month. The lack of substantial
correlation between the SOI and the NAOI means
that these two indicators could potentially
explain different patterns of variability in
local data.
The ENSO and NAO may be expected to affect the
GCE study site by different pathways. Pacific
phenomena (ENSO) would propagate primarily via
weather fronts from the west, reaching the
Altamaha River watershed before reaching the GCE
site itself. We might expect the SOI to be better
correlated with Altamaha River discharge (as an
integrator of watershed effects) than with
coastal precipitation. The NAO indices (NAOI),
based on Atlantic phenomena from the east, might
be better correlated with coastal precipitation.
However, the GCE study site is not in the higher
correlation regions for either index. The purpose
of this study is to determine if these two
large-scale climate signals (SOI, NAOI) can be
detected in observational data from the GCE site
and surrounding area.
Altamaha discharge and Brunswick precipitation
are weakly negatively correlated, with coastal
precipitation leading the river discharge series
by up to 1 month (R(-0.3)-(-0.4)). Coastal
precipitation has an opposite pattern from
freshwater delivery to the coast via the river
(these data and D. Di Iorio, pers. comm.). This
suggests that coastal and inland precipitation
patterns are different and that both these
weather patterns may influence freshwater
delivery to the GCE site.
Methods Climate signal indices are often
calculated in slightly different ways depending
on the sources and intended use of the data, but
comparisons show that competing indices are
generally well correlated (Elliott and Angell
1988). We used the SOI from the Australian Bureau
of Meteorology (SOIaust) rather than one from
NOAA because of its longer time series. We used
two NAO indices the shorter series from the NOAA
National Centers for Environmental Prediction
(NAOIncep) uses a more robust algorithm involving
Rotated Principal Component Analysis (RPCA) of
data from several stations, whereas the longer
series from the National Center for Atmospheric
Research (NAOIncar) uses only two stations in a
simple difference of station pressure anomalies.
We used data from the NWS station at Brunswick,
GA, (BrunsPrecip) rather than the one on Sapelo
Island for coastal precipitation because of its
longer time series. Altamaha River discharge data
(DoctDisch) was from the most downstream USGS
station at Doctortown, GA. The time series were
processed for analysis of causality according to
Hipel and McLeod (1994). Causality in this case
is defined as X causes Y if the present Y can be
better predicted by using past values of X than
by not doing so. River discharge and
precipitation data were transformed to
approximate normality, then each data value was
deseasonalized by differencing from its
respective monthly mean and dividing by its
monthly std. deviation. These anomalies had
standard normal distributions. The climate
indices were already normalized anomalies, but
with different variances due to different methods
of calculation. These were transformed to
standard normal as necessary. Series were then
detrended, and finally autocorrelations were
removed (prewhitening) to obtain residuals or
innovations in the series. Cross-correlations
between pairs of series were evaluated both
before and after removing autocorrelations.
Apparent cross-correlations of non-whitened
series could be spurious due to autocorrelations
within the individual time series (Elliott and
Angell 1988 Hipel and McLeod 1994), but similar
autocorrelations in two series could also reflect
a common source of variability. Assuming that
large-scale climate patterns would affect local
observations within one year, we evaluated
autocorrelations out to lags of 24 months, and
cross-correlations out to 12 months. The SOI was
more highly autocorrelated at short time scales
than the NAOI, with a memory of 5 months vs. 1
month. River discharge was autocorrelated over 3
months, but coastal precipitation was not
autocorrelated at the monthly scale.
Altamaha discharge and Brunswick precipitation
are only weakly explained by the SOI (R20.02,
lags 2 and 1 month, respectively), using the
cross-correlation prior to whitening the series.
Using the whitened series, these relationships
become even weaker (R20.01). NAO indices explain
even less of the variability in discharge and
precipitation (R2lt 0.01).
North Atlantic Oscillation (NAO) This prominent
large-scale pattern describes fluctuations in air
pressure between the higher and central latitudes
of the North Atlantic Ocean, with one region over
Greenland and the other spanning the central
North Atlantic, eastern U.S. and western Europe
between latitudes 35N and 40N. The NAO is
associated with changes in the intensity and
location of the North Atlantic jet stream, storm
tracks, and patterns of temperature and
precipitation from eastern North America to
western and central Europe.
In a few cases, prewhitening the data clarified a
correlation that was apparent between the
non-whitened series, but in other cases broad
correlations at multiple adjacent lags in the
non-whitened data disappeared in the analysis of
residuals. This suggests that some apparent
correlations may have been due to the interaction
of two different autocorrelations. However, even
if the cross-correlations of non-whitened series
are real, the amount of variability in the local
variables (discharge and precipitation) that is
explained by climate indices is very low (2 by
the SOI, less by NAOI). This supports the general
observation that the GCE study site is not in the
zones of greatest effects for either the ENSO or
the NAO.
Future work Looking at monthly data and
short-term lags are a starting point but they may
not be the best frequency and time scale for
detecting patterns in these data, if they exist.
For example, similarly high correlations at
multiple adjacent lags between some variables
(even using the prewhitened data) suggest that it
may be valuable to examine the data at a
different frequency such as quarterly, in order
to combine optimum lags spread over several
months and to reduce the noise in the data
(Trenberth 1984). Such broad correlation peaks
were evident in several comparisons involving the
Altamaha discharge data. Furthermore, some
connections may be strong only in certain
seasons for example, the winter NAOI is often
examined rather than the whole year. It may be
beneficial to examine the data with regard to
seasonal subsets of the indices.
Time series processing
Map from NOAA, National Weather Service, National
Centers for Environmental Prediction, Climate
Prediction Center web site.
References Australian Government, Bureau of
Meteorology web site, http//www.bom.gov.au/climat
e/. Accessed January 2007. Elliott, W. P. and J.
K. Angell. 1988. Evidence for changes in Southern
Oscillation relationships during the last 100
years. Journal of Climate 1729-737. Hipel, K. W.
and A. I. McLeod. 1994. Time Series Modelling of
Water Resources and Environmental Systems.
Elsevier, Amsterdam. National Center for
Atmospheric Research, Climate Global Dynamics
Division, Climate Analysis Section web site
http//www.cgd.ucar.edu/cas/jhurrell/indices.html.
Accessed January 2007. NOAA, National Weather
Service, National Centers for Environmental
Prediction, Climate Prediction Center web site
http//www.cpc.noaa.gov/. Accessed January
2007. Trenberth, K. E. 1984. Signal versus noise
in the Southern Oscillation. Monthly Weather
Review 112326-332.
Acknowledgments We thank Kristin Meehan and
Sylvia Schaefer for providing maps of the GCE
study area and the Altamaha River watershed, and
Merryl Alber and Mark Ohman for discussions
related to these analyses. This work was funded
by the National Science Foundation through the
Georgia Coastal Ecosystems LTER project (NSF
Award OCE 99-82133).