Sensitivity of the northeast Asian summer monsoon to tropical sea surface temperatures



[1] Observations indicate increasing trends of summer precipitation amount, intensity, and frequency of extremes over northeast Asia since the 1960s. Climate models are generally able to simulate such increases of precipitation over northeast Asia over the 2nd half of the 20th century, and project continuations of these trends in response to the projected warming of the tropical Indo-Pacific Warm Pool, especially around the Philippines and the South China Sea. The principal basis for confidence in these projections is the simplicity and robustness of the mechanisms involved. In essence, the warming of these waters enhances the northward moisture transport from the tropics to northeast Asia, leading to an increase of the northeast Asian precipitation.

1. Background

[2] Changes in regional hydroclimates associated with global climate change are expected to have serious socio-economic consequences. Droughts are among the costliest natural disasters, causing $6 billion to $8 billion (U.S. dollars) annually in global damages [Wilhite, 2000; Trenberth et al., 2003]. Monitoring and understanding regional hydrological cycles are essential prerequisites for predicting the forthcoming changes, especially with regard to water availability and adaptation to the associated socio-economic risks. Changes in the Asian monsoon over the Indo-Pacific Ocean and surrounding land masses, comprising one of the largest regional hydrological cycles over the globe, are of particular concern because of their implications for heavily populated areas.

[3] The northeast Asian summer monsoon (hereafter NEASM) over eastern China, Japan, and Korea (110°E–160°E, 30°N–40°N; see box in Figure 1a) is of great interest because of its influence on over one-third of the world's population resident in those countries [Lau and Li, 1984; Tao and Chen, 1987; Wu et al., 2008]. Recent observational studies suggest that there have been upward trends of summer precipitation amount, intensity, and frequency of extremes over the region since the 1960s [Yang and Lau, 2004; Endo et al., 2005; Fujibe et al., 2005; Su et al., 2005; Chang and Kwon, 2007]. Future climate projections also suggest that East Asia is likely to experience a warmer and wetter climate in the 21st century, with relatively larger magnitudes than globally averaged projections [Hu et al., 2000; Giorgi and Mearns, 2002]. However, there are large inter-model disagreements among such projections, especially of precipitation changes, and the physical mechanisms of such changes are often unclear [e.g., Min et al., 2006].

Figure 1.

(a) The leading pattern (EOF) of detrended JJA-mean precipitation variability in the region 25°N–50°N, 100°E–180°, which explains about 19.5% of the total precipitation variance, and (b) its corresponding amplitude (PC) time series, estimated using the 28-year (1979–2006) CMAP precipitation dataset [Xie and Arkin, 1997]. (c) The linear trend of 28-year JJA-mean CMAP precipitation. (d) Map of linear regression coefficients of the leading PC of precipitation shown in Figure 1b regressed on the JJA-mean tropical SST anomaly time series at each tropical gridpoint during 1979–2006. Regions below the 95% statistical significance based on a t-test are hatched in Figures 1c and 1d.

[4] It is now well recognized that many aspects of the global as well as regional hydrological cycles are sensitive to changes of tropical sea surface temperature (SST) [e.g., Barsugli et al., 2006; Shin et al., 2006, 2010]. This sensitivity is not just to the overall magnitude of the tropical SST changes but also to their detailed spatial structure [e.g., Shin et al., 2010; Xie et al., 2010; Shin and Sardeshmukh, 2011]. The pattern and magnitude of the changes in tropical SST determine the pattern and magnitude of the tropical precipitation (and atmospheric heating) anomalies that subsequently impact the extratropical hydrological cycle through changes in the large-scale atmospheric circulation.

[5] Given the well-recognized role of tropical SST anomalies in regulating NEASM precipitation variability [Wu et al., 2003], it would be useful to determine the most important areas of tropical SST forcing to improve our monitoring and prediction skill of NEASM precipitation. In this study, we investigated the sensitivity of summer (JJA mean) NEASM precipitation to anomalous tropical SST at each tropical location using several atmospheric General Circulation Models (GCMs). A tropical map of such sensitivities may also be interpreted as an optimal tropical SST pattern for forcing large changes in NEASM precipitation. Indeed, we show below that the projection coefficient of an arbitrary tropical SST anomaly field on such a sensitivity map serves as a useful tropical forcing index of NEASM precipitation associated with that SST field.

[6] The principal characteristics of NEASM precipitation variability are summarized in Figure 1 using the 28-year (1979–2006) Climate Prediction Center (CPC) Merged Analysis of Precipitation dataset (CMAP) [Xie and Arkin, 1997]. Figure 1a shows the dominant pattern of the variability of detrended JJA-mean precipitation (i.e., the dominant Empirical Orthogonal Function, EOF) in the region 25°N–50°N, 100°E–180°. The amplitude time series of this pattern, i.e., the Principal Component (PC-1), shows large interannual variations (Figure 1b) superimposed on a generally upward trend (Figure 1c) of precipitation in the region. Although such trends estimated over only 28 years are susceptible to sampling uncertainties, and are indeed not statistically significant at the 95% confidence level according to a t-test, we present evidence drawn from historical climate model simulations that they are robust, and also that they may continue over the next century. The simulations used are summarized in Table 1.

Table 1. The Uncoupled Atmospheric GCM Simulations (GOGA) Used in This Studya
ModelsAtmospheric Model ResolutionNReferences
  • a

    All model outputs were interpolated to a common T42 grid (about 2.8° in latitude and longitude). Uncoupled atmospheric GCM simulations (GOGA) are available at the International Research Institute for Climate and Society (IRI) Climate Data Library (, except for NCAR-CAM3 runs.

  • b

    The T85 and T42 runs were combined before performing the analysis, as very similar results were obtained over our region of interest using the T85 and T42 simulations.

NCAR-CCM3.6T42 × L1824Kiehl et al. [1998]
NCAR-CAM3T42 (T85) × L265(5)bDeser and Phillips [2009]
MPIM-ECHAM5T42 × L1924Roeckner et al. [2006]
GFDL-AM2.14(2.5° × 2°) × L2410Delworth et al. [2006]
NASA-NSIPP-1(2.5° × 2°) × L349Pegion et al. [2000]
COLA-C2.2T63 × L1810DeWitt [1996]

[7] The association between NEASM precipitation and tropical SST variability may be quantified by regressing the PC-1 time series in Figure 1b on the detrended summertime tropical SST anomaly time series at each tropical location for the years 1979–2006 using the HadISST dataset [Rayner et al., 2003]. Figure 1d shows a map of the regression coefficients. NEASM precipitation variability is significantly associated with SST variability over the eastern tropical Pacific, western Indian Ocean, and the South China Sea. The large regressions with the eastern tropical Pacific SSTs reflect the well-known association of NEASM precipitation with the El Niño-Southern Oscillation (ENSO). What is less expected is that NEASM precipitation variability is apparently also strongly tied to SST variability over the western Indian Ocean and the South China Sea. This raises the question of which of these regions is most influential in the NEASM precipitation variability. Note that Figure 1d by itself does not answer this question because, apart from being susceptible to sampling uncertainty due to the use of a relatively short 28-yr observational record, it is not a map of sensitivities but of local regressions that do not account for the correlations among SSTs at different tropical locations. For instance, given the well known ENSO-related positive correlation between Indian ocean and Eastern Pacific SSTs, it is unclear to what extent the positive regressions in Figure 1d over the Indian Ocean merely reflect that ENSO connection rather than a direct influence of Indian Ocean SSTs on NEASM precipitation.

[8] To assess the direct influence of SSTs at different tropical locations on the NEASM precipitation trend over the 2nd half of the 20th century (1951–2000), we examined that trend in atmospheric GCM simulations of that period with prescribed observed global SSTs, and assessed to what extent it could be attributed to a positive projection of the 50-yr tropical SST trend pattern on the pattern of NEASM precipitation sensitivities to tropical SSTs, i.e., on the “optimal” tropical SST forcing pattern of NEASM precipitation. This tropical SST sensitivity pattern was derived from the NEASM precipitation responses obtained in separate uncoupled atmospheric GCM experiments with idealized SST anomaly “patches” prescribed at 43 regularly spaced locations throughout the tropical oceans. The details of our simulations and derivation of the sensitivity pattern are given in section 2, and a discussion and summary follow in section 3.

2. Sensitivity of NEASM Precipitation to Tropical SST Changes

[9] Considering the chaotic nature of climate system, not all of the observed NEASM precipitation trend shown in Figure 1c is attributable to SST changes. To estimate the SST-forced portion of trend, a 50-year (1951–2000) ensemble mean precipitation trend was derived from a multi-model ensemble of 87 multi- atmospheric GCM simulations with prescribed observed time varying global SSTs (a.k.a. GOGA simulations; see Table 1 for the models used). By averaging over the ensemble members, the noise in the individual simulations is reduced by as much as a factor of 9, leading to a much more confident estimate of the SST-forced response.

[10] In Figure 2, the multi-model ensemble-mean NEASM precipitation trend and time history of precipitation anomalies are shown, along with the accompanying tropical SST trend and its detrended interannual standard deviation in the HadISST dataset [Rayner et al., 2003]. The simulated NEASM precipitation trend is about 0.5 mm/day over the 50-yr period, consistent with previous observational studies. As evidence of the robustness of this result, all 6 individual atmospheric GCMs show similar trends (Figure 2b), despite the diversity of the dynamical and physical parameterizations implemented in them.

Figure 2.

(a) 50-year linear trend and (b) time series of JJA-mean NEASM precipitation anomalies (thick red curve) over 1951–2000 derived from multi-model ensemble mean uncoupled atmospheric model simulations with prescribed observed time varying global SSTs (see Table 1). For reference, the time series derived from each atmospheric GCM is also shown (blue curves). Observed (c) 50-year (1951–2000) linear trend of JJA-mean tropical SSTs and (d) standard deviation of linearly detrended JJA-mean SSTs during 1950–2000. SST observations are from the HadISST dataset [Rayner et al., 2003]. Regions below the 95% statistical significance based on a t-test are hatched in Figures 2a and 2c.

[11] The warming trend of tropical SSTs is the largest in the Indian and western Pacific Oceans including the South China Sea, while interannual SST variability is dominated by eastern tropical Pacific ENSO activity. All of these ocean basins were suggested to be associated with NEASM precipitation variability in Figure 1c. To determine the part of the tropical SST changes most critical for NEASM precipitation variability, we derived a map of NEASM precipitation sensitivities to tropical SST anomalies at each tropical location. This was accomplished by synthesizing the NEASM precipitation responses to 43 localized cosine-squared shaped tropical SST anomaly “patches” prescribed in the Max Planck Institute for Meteorology (MPIM) atmospheric GCM ECHAM5 [Roeckner et al., 2006], which has a horizontal resolution of T42 (∼2.8° in latitude and longitude) and 19 vertical levels. Our choice of the MPIM-ECHAM5 GCM was based in part on its realistic representation of interannual NEASM precipitation variability in GOGA simulations (see Figure S1 in the auxiliary material). In these “patch” integrations, a warm as well as a cold SST anomaly patch with a central extremum of 2°C (and a patch average of about 0.66°C) was added to the climatological annual cycle of SSTs at each of the 43 locations shown in Figure 3a as dots. For each patch, 20-member ensemble integrations were performed for 25 months starting 1 October.

Figure 3.

(a) Sensitivity of the JJA-mean NEASM precipitation to tropical SSTs derived from the MPIM-ECHAM5 “patch” integrations. The black dots represent the centers of the prescribed SST anomaly patches used to estimate the sensitivity pattern. The patterns of the imposed Indo-Pacific (left) and Atlantic (right) SST patches are shown in the gray shaded rectangles at the bottom of the figure. (b) The ensemble-mean JJA precipitation anomalies in the GOGA simulations over 1950–2004 (thick black curve) and the corresponding linearly reconstructed precipitation anomalies obtained as the weighted sum of the responses to the individual patches (blue curve), with the weights proportional to the amplitude of JJA SST anomaly over each patch in each year. A second series of reconstructions (red curve) were obtained using the projection of the observed JJA tropical SST anomaly fields on the smoothed patch-experiment-derived sensitivity pattern shown in Figure 3a. All three time series have been smoothed with 1-2-1 binomial filter.

[12] We first estimated a raw sensitivity to the j-th patch (Sj) as the linear response of ensemble-mean NEASM precipitation,〈Pj〉 = (〈Pj+〉 − 〈Pj〉)/2, to a unit SST forcing as Sj = 〈Pjequation imagej, where equation imagej = {equation imageTj (xk) dAk}−1. Here 〈Pj+〉 and 〈Pj〉 are the ensemble-mean NEASM precipitation responses to the warm and cold patches, respectively, and Tj(xk) is the j-th patch pattern, shown in the bottom of Figure 3a, at the grid location xk. These raw sensitivity values were then assigned to the corresponding center points of the patches (dots in Figure 3a), and spatially smoothed using a smoothing spline procedure based on the signal-to-noise-ratio as described by Barsugli et al. [2006].

[13] The resulting smoothed sensitivity map (Figure 3a) indicates that NEASM precipitation has the largest positive sensitivity to SSTs at the northern edge of Pacific Warm Pool, especially around the Philippines and the South China Sea. It has almost equally large negative sensitivity (whereby a warm SST anomaly leads to a decrease of NEASM precipitation) to SSTs in the north-central tropical Pacific and the Gulf of Mexico and Caribbean Sea (Figure 3a). The sensitivity value plotted in Figure 3a at each location represents the NEASM precipitation response to a unit (+1°C) SST anomaly over a 106 km2 area centered at that location. Note that the actual NEASM precipitation response to an SST anomaly at the location is the product of the sensitivity value and the SST anomaly. Thus, though large, the regions of negative sensitivity in Figure 3a likely did not contribute substantially to the observed NEASM precipitation trend and interannual variability over the 2nd half of the 20th century, since the summertime SST trends as well as interannual SST variability were relatively weak in those regions, as shown in Figures 2c and 2d.

[14] To confirm the relevance of the tropical SST sensitivity map, we projected the observed summer SST anomaly fields for each of the years 1950–2009 on the sensitivity pattern in Figure 3a. The resulting linearly reconstructed NEASM precipitation series (red curve in Figure 3b) represents a linear response to only the tropical (as opposed to global) SSTs, and may be compared with the ensemble mean precipitation response time series obtained in the fully nonlinear MPIM ECHAM5 GOGA simulations (thick black curve). The high correlation (0.82) of the two time series demonstrates 1) that NEASM precipitation is highly sensitive to tropical (compared to extratropical) SST anomalies, 2) our sensitivity map correctly identifies the spatial variation of this sensitivity, and 3) the response of NEASM precipitation to tropical SST changes is approximately linear. These results generate confidence in our ability to assess the impact of an arbitrary SST anomaly field on NEASM precipitation by determining the spatial projection of that SST anomaly field on the smoothed sensitivity map.

[15] Very consistent results were obtained by reconstructing the NEASM precipitation anomaly for each summer in 1950–2009 as the weighted sum of the raw precipitation responses to the individual SST patches in the patch integrations, with weights proportional to the amplitudes of observed tropical SST anomaly at the patch locations (blue curve in Figure 3b; see Shin et al. [2010] for details). Such a raw linearly reconstructed NEASM precipitation series is also highly correlated with the GOGA series (0.84), and confirms that the NEASM precipitation response to tropical SSTs is essentially linear.

[16] We have further confirmed the robustness of our sensitivity map in Figure 3a by repeating the entire patch experiment with a different model, the NCAR CCM3 atmospheric GCM [Kiehl et al., 1998], and following identical sensitivity map construction procedures. The resulting sensitivity map, shown in Figure S2, is very similar to that in Figure 3a and has pattern correlation with it of 0.77.

[17] The basic reason that SST changes near the Philippines and the South China Sea are particularly efficient at generating a large NEASM precipitation response is that the atmospheric circulation response to those SSTs effectively enhances the northward moisture transport into the NEASM region, increasing the precipitation there. To establish this dynamical link, we first linearly reconstructed the summertime precipitation and sea level pressure responses to the “optimal” tropical SST anomaly pattern for forcing NEASM precipitation, identical to Figure 3a but with a 1°C r.m.s. amplitude over the tropical oceans, and show them in Figure 4a. The figure shows that the NEASM precipitation response is associated with a surface pressure low centered near the Philippines and South China Sea region of largest SST sensitivity (indicated by the red star). Figure 4b is in a similar format to Figure 4a, but shows the precipitation, sea level pressure, and lower tropospheric (850 hPa) wind responses to the patch centered at that red-star location of largest SST sensitivity. The response to the localized patch warming is a classic Matsuno-Gill type atmospheric response to off-equatorial heating [Matsuno, 1966; Gill, 1980], including a classic equatorial Kelvin wave response to the east and a Rossby wave response to the northwest of the forcing. It is this cyclonic Rossby wave response that enhances the northward moisture transport over the far western Pacific Ocean into the NEASM region and increases the precipitation there (see Figure S3 for the vertically integrated tropospheric moisture transport and its convergence and divergence).

Figure 4.

(a) Linearly reconstructed JJA extratropical precipitation (colored rectangles) and sea level pressure (contoured) responses to the “optimal” tropical SST anomaly field (gray shaded, same pattern as Figure 3a) with r.m.s. amplitude of 1°C over the entire tropical oceans. (b) JJA extratropical precipitation (colored rectangles), sea level pressure (contoured), and 850-hPa wind (vectors) responses to the “patch” centered in the South China Sea. The location of patch is indicated by the red star in Figure 4a. Note that the average SST anomaly in the South China Sea in Figure 4a is 4 times larger than that in Figure 4b, as are the responses.

3. Summary and Discussion

[18] Both observations and multiple atmospheric GCM simulations with the prescribed observed time history of global SSTs (i.e., GOGA simulations) indicate that summer precipitation over the NEASM region increased over the 2nd half of the 20th century. In this study we constructed a sensitivity map of NEASM precipitation to tropical SST anomalies, by synthesizing the MPIM ECHAM5 atmospheric GCM's responses to 43 regularly spaced localized SST anomaly “patches” prescribed over the tropical oceans. This sensitivity map (Figure 3a) shows that the most critical SST changes influencing NEASM precipitation occur in the far western Pacific Ocean around the Philippines and the South China Sea. A classic Matsuno-Gill type circulation response to surface warming in this region enhances the northward moisture transport into the NEASM region and increases the precipitation there. We also demonstrated the utility of this sensitivity map (Figure 3a), as a whole, in predicting NEASM precipitation responses to SST anomaly fields, determined simply as the projections of those SST anomaly fields on the sensitivity map.

[19] Observations and coupled atmosphere-ocean climate model simulations used in the Fourth Assessment Report (AR4) of the Intergovernmental Panel on Climate Change (IPCC) consistently indicate that the tropical oceans, including the area around the Philippines and the South China Sea, warmed over the second half of the 20th century and will continue to warm over the 21st century [Intergovernmental Panel on Climate Change, 2007]. To the extent that the tropical ocean warming continues to project positively on our SST sensitivity map for NEASM precipitation, we may expect that East Asia will likely experience a wetter climate over the 21st century, as suggested also in some previous studies [Hu et al., 2000; Giorgi and Mearns, 2002]. An important reason for confidence in this prediction is that the basic mechanism of this forthcoming change, associated with enhanced northward moisture transport into the NEASM region, is remarkably simple and robust as shown here.

[20] In summary, our study provides compelling evidence that SSTs near the Philippines and the South China Sea play a critical role in NEASM hydrology, and suggests that monitoring and prediction of those SSTs will greatly aid the monitoring and prediction of NEASM changes.


[21] The authors thank the two anonymous reviewers for their helpful comments. This work was partially supported by NOAA's Climate Program Office (SS and PDS), the Korea National Institute of Meteorological Research (SS), and the Korean Meteorological Administration Research and Development Program under Grant RACS_2010-2006 (SY). Our own simulations were performed at the NOAA ESRL High Performance Computing Systems (HPCS) facility.

[22] The Editor thanks Vasubandhu Misra and an anonymous reviewer for their assistance in evaluating this paper.