Forecast of El Niño Event by
Extended Associate Pattern Analysis *
Contributed by CUI Maochang1,2, Mo Jun1,
Yu Yongqiang2
1Institute of Oceanology, Chinese Academy of Sciences, Qingdao 266071, P. R. China
2LASG, Institute of
Atmospheric Physics, Chinese Academy of Sciences , Beijing 100080, P. R. China
1.
Introduction
El Niño event is the strongest climate signal on annual time scale, which affects both of regional and global climate (1-2). Previous studies showed that if SST could be predicted rather well and seasonal-to-interannual climate could also be foretold in a great success. In other words, to large extent, the climate is predictable (CLIVAR-1996).
Interactions between ocean and atmosphere contribute to climate fluctuations over a broad spectrum of time scales. Studies of those interactions have thus far focused on El Niño events, which have a period of 3 to 7 years and whose principal signal is in the tropical Pacific(3). Superimposed on this natural mode of the coupled ocean-atmosphere system are interdecadal fluctuations that contribute to the irregularity of El Niño events not only in time regime, for example, a new feature of El Niño events is found, which is of shorter interval and warm events are dominant, during 1990s but also in space regime, the anomalous warm water covers the most regions of the tropical Pacific from Nino1 to Nino4 (NCEP, Climate Diagnostics Bulletin. 1997.1. 79pp) and for most El Niño events before 1980, warming of the equatorial Pacific usually started from the coast of South America and then extended westward, while for those after 1980, it appeared first in the western or central equatorial Pacific and then propagated eastward.
In terms of predictive capability, depending on both the predictive capability of the models and the predictability of climate system, most models exhibit obvious dependence of skill on decade as reviewed by Latif et al.(4). Chen et al.(5) pointed out that the predictive capability for 1990s El Niño events is considerably poorer than for 1980s’. And the predictive capability for 1980s’ is much higher than for 1970s. Strictly speaking, the potential predictability is an intrinsic characteristic of climate system and does not depend on what models are used. However, it can be reflected from the model results.
The research on El Niño events in China started in the middle of 70’s(6). It connected to the subtropical high, which has strong influence on the summer rainfall in China, as the major feature. Li(7) pointed out that more frequent and stronger cold waves in the east of Asia, associated with a strong winter monsoon, can enhance cumulus convection over the equatorial western Pacific, which, in turn, may strengthen the 30-60 day oscillation in the western Pacific and trigger an El Niño onset. The sea surface temperature (SST) in the west of Indian Ocean is usually higher than that in the east with weak seasonality. Bjerknes(8) indicated that such a seawater temperature distribution in the equatorial Indian Ocean arouses the Walker circulation reversed to that in the equatorial Pacific. Wu(9) further indicated that such a positive correlation is associated with the strong gear-like coupling between monsoon zonal circulation over the equatorial Indian Ocean and the Walker circulation over the Pacific with an anomalous gearing point near the Indonesian Islands. After 80’s and before most El Niño onsets, the anomalous gearing point takes place first and then propagats eastward into the Pacific and may trigger on the occurrence of El Niño events.
To further study the formation mechanism of El Niño events and make its forecasts, extended associate pattern analysis is set up with the combined observed monthly anomalous SST and sea level pressure (SLP) in or over the Pacific and related seas.
2.
Data and method
The data used are the Gisst and rebuilt monthly SST by Kaplan et al. and NCEP SLP or 1000 hPa Dynamic high. The Gisst monthly SST with a 1°×1°grid length (for mechanism study) for the period of 1949.1~1997.12 and the XBT monthly SST with a 5°×2°grid length (for forecast) for the period of 1955.1~2002.12 covering the region (40°S~60°N, 40°E~80°W), NCEP monthly SLP or geopotential hight at 1000 hPa with a 2.5°×2.5°grid length for the period of 1949.1~2002.12 covering the same region, and the El Niño index for the period of 1950.1~2002.12 down loaded from the internet. Prior to analysis, the anomalous datasets some times need smoothed by running average and properly weighted to remove short period noise and improve the precision of simulation and forecast and divided by its space average of standard deviation before combination.
Given time series X and variable field Y
X = { x ( j ) | j = 1,……, n }
Y ={ y( i, j ) | i = 1,……, m; j = 1,……, n }
with zero mean (<X> = <Y> = 0 ). A space vector A
A = { a(I)| i = 1,……, m }
can always calculated by the least square method to suit the condition
( i = 1, ……, m) (1).
It physically means that the information given by X can be explained by its regression value a(i) at any space point i of variable field Y, therefore space vector A can used for the study on formation mechanism of time series X . A is called as X ‘s associate pattern in field Y(10). If field Y is projected onto the direction of A, a new time series X’ can be obtained, which is usually well correlated to the time series X, defined as X ‘s associate time series in field Y, so that the former can be used for the latter’s simulation and forecast. The r (X, X’) represents the precision of the simulation and forecast. The matrix product Y’ of the column vector X’T and unified row vector of A is defined as the associate field separated from the field Y by the time series X. At any space point i in the fields Y and Y’, the standard deviation σy (i)、σy’ (i) can be easily calculated, The variance explained by associate pattern A can be written as
(2)
which represents how much real variations of the field Y are contained in the field Y‘. This method is a natural extension of the associate pattern analysis and should be called as the extended associate pattern analysis.
3. Results
Taken the standardized El Niño index as time series X and combined monthly SST or SLP anomalies as field Y, the major results from the extended associate pattern analysis are listed in Tables 1 and 2.
The in phase SLP and SST
associate patterns (not shown), corresponding associate time series of which
are closely related to Nino3 index with the correlation of 0.79 or 0.89,
respectively, represent the essential spacial distribution of anomalous SLP and
SST, when Nino3 index reaches its maximum. At this time the SST associate
pattern appears as a typical El
Niño pattern (not
shown) and the negative
SLP anomaly over the North Pacific merged into the Aleutian low. The period of
9 months before Nino3 index peaks is defined as starting and developing period. During this period, a positive SLP
anomaly from the meddle-high latitudes shifts into the Asia-Australia land bridge; a negative SLP anomaly over middle latitudes appears
as the North Pacific Oscillation in an anti-phase;
the negative SLP anomaly composing SO gets stronger, and then the negative SLP anomaly in middle of the
North Pacific moves northward (not shown). At the mean while, a weak (with the standard deviation of
0.16℃ only) positive SST anomaly
appears along Peruvian coast, in the east
equator, east tropics and Northeast Pacific; negative in China sea and west of
Australia and then the positive SST anomaly signals along equator become
stronger and extend from Nino1-3 to Nino4 region. By this way, El Niño event comes. Negative SST
anomalies appear in China sea and west of
Australia first,
and then get stronger, extend northeastward and southeastward respectively,
forming a pincers movement centered at the West Pacific (not shown) Since variations of the SST
anomalies can be roughly explained by the anomalous geostrophic wind driven
effect in the most regions, the ocean is mainly driven by the atmosphere in
this period. The period of 9 months after the Nino3 index peaks is defined as overdeveloping and ending period.
During this period, a similar positive SLP anomaly pincers movement is formed
over the negative SST anomaly pincers movement. Its northeastward branch forces
the negative SLP
anomaly in the middle of North Pacific moving northwestward, enhancing the
subtropical high over the North Pacific; and its southeastward branch makes the
negative SLP anomaly disappear, then change into a positive one and finally
ending El Niño
events. Since the SST anomaly changes ahead the SLP anomaly in the most regions
except the eastern tropical region, the atmosphere is mainly driven by the
ocean in this period.
Fig.1 simulation and forecast (
dashed line ) of standardized Nino3 index (real line)
Table 1 Major parameters of the extended associate pattern analysis with SLP and Nino3
|
Running
months |
Leading
months |
Corre- lation |
Strong
signal position and feature in the leading SLP associate pattern of Nino3
index |
Explained
variance |
|
7 |
9 |
0.52 |
Positive
over Aleutian, Asia and negative over meddle of the North Pacific, tropic of
the South Pacific |
14.29% |
|
7 |
6 |
0.57 |
Positive
one extends to Indian Ocean and Australia; the negative one over the South
Pacific becomes stronger |
22.57% |
|
7 |
3 |
0.70 |
Center
of Positive one moves southeastward; both negative ones become stronger |
31.96% |
|
7 |
0 |
0.79 |
Positive
one in the east is divided into two branches; the north negative one merged
into Aleutian low |
38.00% |
|
7 |
-3 |
0.77 |
Two
positive branches extend eastward; the south negative one is weakened |
36.60% |
|
7 |
-6 |
0.64 |
Two
positive branches extend eastward continually; the south negative one disappeared
totally |
27.94% |
|
7 |
-9 |
0.54 |
The
south positive branch covers the whole tropic of the South Pacific |
18.27% |
Table 2 Major parameters of the extended associate pattern analysis with SST and Nino3
|
Running
months |
Leading
months |
Corre- lation |
Strong
signal position and feature in the leading SST associate pattern of Nino3
index |
Explained
variance |
|
0 |
9 |
0.56 |
Positive
along Peruvian coast, in the east equator, east tropics and Northeast
Pacific; negative in China sea and west of Australia |
8.90% |
|
0 |
6 |
0.47 |
Positive
one covers Nino1-3 regions; negative ones extend eastward and form two
branches |
12.77% |
|
0 |
3 |
0.69 |
Positive
one cover Nino1-4 regions; two negative
branches become stronger |
20.09% |
|
0 |
0 |
0.89 |
The
positive one in Nino regions becomes strongest; the west part of two negative
branches are weakened |
27.22% |
|
0 |
-3 |
0.77 |
The
positive one in Nino regions becomes wider and weakened; two negative
branches are separated |
26.97% |
|
0 |
-6 |
0.60 |
The
wider positive one is weakened, some appear along California coast; two negative
ones weakened |
22.74% |
|
0 |
-9 |
0.46 |
The
wider positive one is separated into two, more positive in west Pacific, SCS
and Indian Ocean |
15.76% |
4. Discussion and conclusion
Because the negative SLP anomaly in the middle of North
Pacific holds the region where the explosive
cyclones take place most frequently over the temperate North Pacific during the starting and
developing period, it is most likely produced by the average
air pressure decreasing
effect of explosive cyclones through
precipitation over the temperate North Pacific.
If only the data sets after the 1977 climate shift is used, the results
(not shown) are quite similar to each other. The major difference is that the
positive SLP anomaly over the tropical Asia-Australia land bridge comes from
the western Asia or southern Indian Ocean and Australia instead of the East or middle
Asia (not shown) and the positive SST anomaly appears in the meddle of
equatorial Pacific first instead of the east of equatorial Pacific. And
corresponding to Nino3 index peaks, the positive SLP anomaly comes from the tropical
land bridge and El
Niño disappears about 3 months later in this case. The results
are consistent with the fact: the positive SST anomaly usually appears in the
meddle of equatorial Pacific first during El Niño onsets before
the climate shift
but in its meddle first after it.
If the standardized associate time series of Nino3 index and
combined GH1000/XBT
SST fields, processed by 13 month running mean and weighted by 1:3, are used to
simulate and predict its variation, the simulations and forecast (with
correlation of 0.67; significant level of 0.999; explained variance 44.8%) can
be easily carried out. The simulations are good and the forecast shows that 2003
would be a normal year (Fig. 1).
The conclusion comes as following. a negative SLP anomaly over middle latitudes composing the North Pacific Oscillation in an anti-phase, a positive SLP anomaly over the Asia-Australia land bridge, formed by the positive SLP anomaly shifting from the high-middle latitudes, and a negative SLP anomaly composing SO are the major causes for El Niño onset. During the starting and developing period of El Niño, the ocean is mainly driven by the atmosphere. A positive SLP anomaly in the East Asia accompanying with negative SLP anomalies over the middle of North Pacific and the tropic of South Pacific produces the east type El Niño; a positive SLP anomaly over the West Asia or southern Indian Ocean and Australia
accompanying with
the same two negative SLP anomalies produces the middle type El Niño. During the overdeveloping
and ending period, the atmosphere is mainly driven by the ocean. A negative
anomalous SST pincers movement drives a similar positive anomalous SLP over it,
its northeastward branch forces the negative SLP anomaly over the middle of
North Pacific moving northwestward and its southeastward branch makes the
negative SLP anomaly composing SO disappear and change into a positive one and
finally stops the El Niño
events.
The hypothesis that the negative SLP anomaly over the middle of North Pacific is produced by the average air pressure decreasing effect of explosive cyclones through precipitation over the temperate North Pacific during the starting and developing period of El Niño needs proved with qualified rainfall data in the North Pacific. Although there are some data sets (such as ECMWF reanalysis, NCEP reanalysis and Pentad CMAP) available, no one is qualified.
REFENCES
1
Wallace J, Gutzler D (1981) Teleconnection in the geopotential height
field in the northern hemisphere. Mon
Wea Rev 109: 784-812.
2 Trenberth K, Branstator G, Karoly D, Kumar A, Lau N, Ropelewski C (1998) Progress during TOGA in understanding and modeling global teleconnections associated with tropical sea surface temperatures. J Geophy Res 103(C7): 14291-14324.
3 Philander S G H (1990) El Niño, La Niña and the South Oscillation. Academic Press, New York.
4 Latif M, Barnett D Cane M, Kleeman R, Leetmaa A, O’brien J, Rosati A, Schneider E (1998) A review of the predictability and prediction of ENSO. J Geophy Res 103(C7): 14375-14392.
5 Chen D, Zebiak S, Basalachi A, Cane M (1995) An improved procedure For El Niño forecasting: Implications for predictability. Science 269: 1699-1702.
6 Fu C. et al. (1977). The influence of the tropical oceans on the long-range variability of the subtropical high. Chinese Science Bulletin, 21(B),313-317.
7 Li C. (1988). The strong and frequent activities of the East Asian trough and El Niño onset. Science in China (B), 18, 667-674.
8 Bjerknes, J. (1969), Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev., 97, 163-172.
9 Wu G., Meng W. (1998). Gearing between the Indo-monsoon circulation and the Pacific -Walker circulation and the ENSO. Part I: Data analyses. Scientia Atmospherica Sinica, 22(4): 470-480. (in Chinese with English abstract)
10 Latif M, Barnett T P, 1994. Causes of Decadal Climate Variability over the North Pacific and North America. Science, 266, 634-637.