JOINT VARIABILITY OF GLOBAL RUNOFF AND SEA-SURFACE TEMPERATURES
Gregory J. McCabe and David M. Wolock
Abstract
A monthly water-balance model is used with global monthly temperature and
precipitation data on a 0.5 degree by 0.5 degree grid to compute time series of annual
gridded runoff for the period 1905-2002. The runoff time series are subsequently
aggregated to a 5 degree by 5 degree (5x5) grid resolution for analysis in combination
with time series of gridded annual sea-surface temperature (SST) data on a near 10
degree grid resolution. The runoff and SST data are subjected to a principal
components analysis (PCA) with varimax rotation to identify the primary modes of
variability in these data sets. The first three components from the rotated PCA explain
approximately 29 percent (%) of the total variability in the combined runoff/SST data.
The first component explains 13% of the total variance and primarily represents longterm trends in the data. The long-term trends in SSTs are most clearly evident as
warming in the Atlantic and Indian oceans. The associated long-term trends in runoff
suggest increasing flows for much of North America and Australia, and in parts of
South America and Eurasia; decreasing runoff is most notable in Africa and southern
Asia. The second component explains 11% of the total variance and reflects variability
of the El Nino/Southern Oscillation (ENSO) and its associated influence on global
annual runoff patterns. The third component explains 6% of the total variance and
indicates a response of global annual runoff to variability in North Atlantic SSTs. The
association between runoff and North Atlantic SSTs appears to explain an apparent
step-like change in runoff that occurred around 1970 for a number of continental
regions.
(U.S. Geological Survey, Denver, Colorado)
Water Balance Model
Verification
Estimates of annual runoff from the water
balance model were verified by comparing
the estimates with time series of measured
runoff for drainage basins across the
globe.
The water balance model reliably
simulates the temporal variability of annual
runoff.
Locations of drainage basins
used for verification of the water
balance model
Data and Methods
Monthly temperature and precipitation data for the globe (at a 0.5 degree resolution)
and for the period 1901-2002 were obtained from Climate Research United at East
Anglia, United Kingdom [the CRUTS2.1 dataset. These data were used as inputs to a
monthly water-balance model to generate time series of annual runoff for each grid
cell.
Monthly sea-surface temperature (SST) data were
obtained from the Kaplan extended SST dataset of
monthly SSTs. The SST data are on a 5 degree by
5 degree (5x5) grid and span the period 1856 to the
present. The monthly SST data were used to
compute mean annual SSTs for the 5x5 grid.
The runoff and SST data were standardized and
subjected to a principal components analysis to
identify the primary modes of variability.
The principal components analysis (with varimax rotation) resulted in 3 components that explain
30% of the variability in the runoff and SST data.
Component 1 (13%)
Component 2 (11%)
Component 3 (6%)
The score time series for
component 1 (RPC1) indicates
a long-term trend in the data.
The pattern of SST loadings for
RPC1 indicate increasing
SSTs for most of the oceans,
with the largest increases in
the Indian and Atlantic Oceans.
The SST loadings for RPC2
indicate the influences of
ENSO on global annual
runoff. A comparison of
NINO3.4 SSTs with the
score time series for RPC2
indicates a large correlation
(correlation coefficient equal
to 0.91 (p < 0.01).
The SST loadings pattern for
RPC3 indicates that the most
significant signal is in the
North Atlantic Ocean,
particularly the tropical North
Atlantic Ocean.
The runoff loadings for RPC1
indicate positive values (i.e.
increasing trends) in runoff for
most of North and South
America, a large part of Asia,
and Australia. The correlations
between the pattern of trends
in runoff and the patterns of
trends in precipitation and
temperature are 0.93 and -0.12
respectively.
The pattern of runoff
loadings for RPC2 indicates
features of the ENSO signal.
For example, the
northwest/southwest U.S.
dipole is apparent, as well
as the ENSO signal in South
America and Australia.
Runoff loadings for RPC3 are
positive over North America,
most of South America, and
Australia. Negative loadings
are found over western and
central Africa, most of Europe
and most of Asia. Previous
studies have indicated that the
North Atlantic SST
associations with global
hydro-climate are particularly
noticeable on decadal to
multi-decadal time scales.
Comparison of measured (black
lines) and water-balance
estimated (gray lines) annual
runoff
Diagram of Monthly
Water Balance Model
Although the CRUTS2.1 dataset begins in 1901, the
first few years of the water balance simulations were
not analyzed so that the effects of prescribed initial
conditions would be minimized. In addition, for
computational efficiency, the 0.5 degree resolution
time series of annual runoff were aggregated to a 5
degrees of latitude by 5 degrees of
longitude (5x5) grid, resulting in 638 grid cells, each
with a unique time series. The time series values for
the 5x5 grid were used in subsequent analyses.
The SST grid cells were re-sampled to provide a
relatively equal number of grids as that of the runoff
data. The re-sampling of the SST data provided 604
SST grid cells for analysis.
Principal Components Analysis
Summary
Runoff
Grid
SST
Grid
B.
Results indicate three modes of variability that explain approximately 30% of the variance in global
runoff and SST data. The first mode is related to long-term trends in runoff and SSTs, the second
mode reflects ENSO variability, and the third mode represents associations between variability of
tropical Atlantic SSTs and annual runoff for some regions.
Trends in global annual runoff generally indicate increasing runoff in the mid- to high-latitudes and
decreasing runoff in low-latitudes. However, only about 25% of the trends are significant at a 95%
confidence level.
Results also suggest that SSTs in the tropical North Atlantic Ocean may be a source of hydro-climatic
variability that requires additional research. Hydro-climatic effects of variability in topical North Atlantic
SSTs may be especially important on decadal to multi-decadal time scales.