Global Warming Shifts the Monsoon Circulation, Drying South Asia

Transcription

1 VOLUME 26 J O U R N A L O F C L I M A T E 1 MAY 2013 Global Warming Shifts the Monsoon Circulation, Drying South Asia H. ANNAMALAI, JAN HAFNER, K.P.SOORAJ,* AND P. PILLAI 1 International Pacific Research Center, University of Hawaii at Manoa, Honolulu, Hawaii (Manuscript received 13 April 2012, in final form 4 October 2012) ABSTRACT Monsoon rainfall over South Asia has decreased during the last 5 to 6 decades according to several sets of observations. Although sea surface temperature (SST) has risen across the Indo-Pacific warm pool during this period, the expected accompanying increased rainfall has occurred only in the tropical western Pacific. The above changes noted in observations are also seen in a coupled climate model, but only when the model includes the recent increase in greenhouse gas concentration. The hypothesis that the robust rise in SST over the warm pool, perhaps anchored by an increase in greenhouse gas concentrations, is instrumental in the east west shift in monsoon rainfall (enhanced rainfall over tropical western Pacific and decreased rainfall over South Asia) is proposed. A suite of controlled experiments with an atmospheric general circulation model has been performed to isolate the impact of regional SST warming trends on the dryness over South Asia. Model experiments support the hypothesis that the rising SST trend over the tropical western Pacific has changed the atmospheric circulation: over the Bay of Bengal more dry and cool air is advected from the northeast than previously. Moist static energy budget diagnostics on the model solutions identify the sources for this east west shift. SST warming over the warm pool has accelerated in recent decades. Therefore, a close monitoring of that warming is important for long-term variations of monsoon rainfall. The inconsistency in the amplitude of drying over South Asia among the various land-based rainfall observations and lack of sustained rainfall observations over the open oceans, however, poses constraints in the results. 1. Introduction a. Background Rainfall observations over land only show a trend toward less rainfall from 1951 to 2000 in the subtropics and tropics but more rainfall in the midlatitudes of the Northern Hemisphere, a signal that has been attributed to the increase in greenhouse gas (GHG) concentrations (Zhang et al. 2007). Models also capture the large-scale tropical drying patterns (Neelin et al. 2006). This drying trend has also been noted in the seasonal-mean (June September) regional rainfall associated with the South * Current affiliation: Pusan National University, Pusan, South Korea. 1 Current affiliation: Indian Institute of Tropical Meteorology, Pune, India. Corresponding author address: Dr. H. Annamalai, International Pacific Research Center/School of Ocean and Earth Science and Technology, University of Hawaii at Manoa, 1680 East West Road, Honolulu, HI hanna@hawaii.edu Asian summer monsoon. We show here that these changes are noted in a coupled climate model that has demonstrated skill in capturing the present-day mean and spectrum of monsoon precipitation variability (Annamalai et al. 2007; Stowasser et al. 2009). An unprecedented increase in GHG concentrations has been observed during the last six decades (Knutson et al. 2006). While the drying trend may be due to factors other than increased GHG concentration, such as aerosols (Ramanathan et al. 2005), we show here that anthropogenic forcing through sea surface temperature (SST) warming over the tropical western Pacific likely causes the drying trend over South Asia. The All-India Rainfall (AIR) index is widely used as a measure of the strength of the Indian summer monsoon (Parthasarathy et al. 1994). A simple running mean applied to AIR seasonal anomalies (see Fig. 1a, red line) shows dry and wet periods: dry periods occurred from 1900 to 1930 and from 1965 to 1995, and a wet period from 1930 to 1960, consistent with earlier studies (Mooley and Parthasarathy 1984). If the changes shown in Fig. 1a are periodic variations associated with multidecadal drivers (Zhang and Delworth 2006), then a wet DOI: /JCLI-D Ó 2013 American Meteorological Society 2701

2 2702 J O U R N A L O F C L I M A T E VOLUME 26 b. Present study FIG. 1. (a) Running mean (31 yr) applied on boreal summer [June September (JJAS) or only July August (JA)] observed all- India rainfall (AIR) anomalies (in percent departures) from two data sources (red line, AIR_JJAS_IITM; blue line, AIR_CRU; black line, AIR_JA_IITM). (b) 31-yr running mean applied on AIR (black line) and JJAS observed sea surface temperature (SST, 8C) averaged over the tropical western Pacific (5 258N, E) from three different products (color lines). (c) Running mean (31 yr) applied on the ensemble mean of the coupled model, CM2.1, simulated JJAS rainfall anomalies averaged over South Asia (58 258N, E; red line), tropical western Pacific ( N, E; blue line), and SST over the tropical western Pacific (brown line); In (a), decadal means of AIR are shown while in (b) and (c) the changes in AIR and SST for two different periods are provided. The three SST products examined are 1) HadISST2, 2) ERSST.v3, and 3) Kaplan SST V2. period would be anticipated to begin in the 1990s. Yet, a reduction in AIR occurred since 1951 with a steady decline in its decadal means and the last decade experiencing a seasonal mean of only mm of rainfall (AIR decadal means are provided in Fig. 1a). Such a decline is evident during July August, the peak monsoon period (Fig. 1a, black line). The AIR index based on Climatic Research Unit (CRU) data (Mitchell and Jones 2005) captures the drying, but its amplitude is stronger (Fig. 1a, blue line). The coherent spatial pattern obtained from regression analysis between the AIR index and rainfall over other regions (not shown) suggests that the AIR index represents rainfall variations across South Asia. For the period , the trend in AIR_IITM is significant at about 90% but we show that the drying tendency is robust across central India. Here, to assess the robustness in this drying trend, we diagnosed trends in multiple observational data and compared them to coupled model integrations. The robust result (section 3) is an increase (decrease) in monsoon rainfall and circulation over the tropical western Pacific (South Asia), implying an east west shift in recent decades. We propose a hypothesis that the rise in SST over the tropical Indo-Pacific warm pool is instrumental in causing this shift. To validate this, we performed sensitivity experiments with an atmospheric general circulation model (AGCM) and applied moist static energy (MSE) budget to the model solutions (section 4) to isolate the processes that contribute to dryness (wetness) over South Asia (tropical western Pacific). Our results suggest that in response to SST warming over the warm pool rainfall changes within the broader Asian monsoon and the associated dynamical feedbacks appear responsible for the drying tendency over South Asia (sections 4 6). Salient findings and their implications are summarized in section Data, methods, models, and numerical experiments a. Observations and reanalysis products The AIR index is based on area-weighted average of 306 rainfall stations covering the entire Indian subcontinent (Parthasarathy et al. 1994) and has been widely used to assess natural variations of monsoon rainfall and also to detect long-term trends (e.g., Gautam et al. 2009). The drying trend noted in the AIR index was examined in detail by comparing the monsoon climatology to the long-term trends in seasonal-mean anomalies in the following datasets: CRU land-based gridded observation version 2.1 (CRU-TS_2.1; Mitchell and Jones 2005); high-resolution ( ) data compiled by Delaware University (Willmott and Matsuura 2001), and the Precipitation Reconstruction data over Land (PREC/L) of Chen et al. (2002). Three SST products the Hadley Centre Global Sea Ice and Sea Surface Temperature version 2 (HadISST2; Rayner et al. 2003), the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed SST version 3 (ERSST.v3; Smith et al. 2008) and the Kaplan extended SST version 2 (Kaplan SST V2; Kaplan et al. 1998) are diagnosed to assess the consistency in the magnitude of SST trends. HadISST2 has a resolution of , whereas the resolution is in ERSST.v3 and in Kaplan SST V2. To infer changes over the oceanic regions, the total cloud amount from the International Comprehensive

3 1MAY 2013 A N N A M A L A I E T A L Ocean Atmosphere Data Set (ICOADS) Release 2.4 (Woodruff et al. 2008) is analyzed. To ascertain if the changes are dynamically consistent, monthly sea level pressure (SLP) data from the Hadley Centre (Basnett and Parker 1997) and 850-hPa wind from the National Centers for Environmental Prediction (NCEP) National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al. 1996) are diagnosed. In all observations, reanalysis products, and the coupled model integrations (described below) the period of analysis is The only exception is the results presented in Figs. 1a and 1b where the AIR and SST indices up to 2010 are examined. b. Models 1) COUPLED MODEL Of the coupled models that participated in phase 3 of the Coupled Model Intercomparison Project (CMIP3; Meehl et al. 2007) the version of the climate model developed at the Geophysical Fluid Dynamics Laboratory (GFDL CM2.1) has a realistic simulation of monsoon precipitation climatology including the three regional heat sources (Annamalai et al. 2007), synoptic systems (Stowasser et al. 2009), certain aspects of the active-break cycles associated with the intraseasonal variability (Sperber and Annamalai 2008), and the ENSO monsoon relationship (Annamalai et al. 2007). Here, we will show its ability in capturing longer time scale variability associated with the monsoons. A brief description of CM2.1 is provided here, and formulation details are documented elsewhere (Delworth et al. 2006). The atmosphere component is AM2.1 (Anderson et al. 2004), which has a finite volume dynamical core with 24 vertical levels and horizontal resolution of 2.58 longitude and 28 latitude. The ocean component is based on the Modular Ocean Model version 4 (MOM4) and has 50 vertical levels (10-m resolution in the upper 220 m). The horizontal resolution is but its meridional resolution reduces to 1 /3 equatorward of 308 (Griffies et al. 2003). Different components of the CM2.1 are coupled through the Flexible Modeling System, without applying flux corrections. The atmosphere, ocean, land, and sea ice exchange fluxes every 2 h. The simulations follow the protocol set by the Intergovernmental Panel on Climate Change (IPCC) in its Fourth Assessment Report (AR4). The model was forced with estimates of changes in well-mixed GHGs, tropospheric sulfate and carbonaceous aerosols, volcanic aerosols, ozone, solar irradiance, and land use. Direct forcing from black carbon aerosols is incorporated. All five members of the ensemble are analyzed. In addition, we also diagnosed the 200-yr-long preindustrial control run to confirm that the recent drying is due to anthropogenic forcing. The detailed model description and its ability in simulating global and regional temperature trends, as well as the current and future Sahel drought, are documented elsewhere (Knutson et al. 2006; Held et al. 2005). 2) AGCM AND EXPERIMENTAL DESIGNS The AGCM used for the sensitivity experiments is AM2.1. In the Control run, the model was forced for 52 years with seasonally varying climatological SST. In two sensitivity experiments, the monthly increment in observed SST from 1949 to 2000 (i.e., only the trend and climatology) was used to force the model selectively in two different regions: EXP1, only in the tropical western Pacific (308S 308N, E); and EXP2, only in the tropical Indo-Pacific warm pool (308S 308N, E). The differences between these two experiments are expected to indicate the tropical Indian Ocean SST role. Each sensitivity experiment was run five times with initial conditions on 1 January, 15 January, 1 February, 15 February, and 1 March, all taken from the control run. To avoid the monsoon response to interannual or decadal SST signals, and to isolate the sole effect due to the monotonic warming trend, AM2.1 is forced only with the monthly SST trend in conjunction with the climatology. To see if the SST warming along the equatorial Pacific (108S 108N, 1308E 1008W) has any impact on the monsoon weakening over South Asia, one additional experiment, EXP3, with only one ensemble member was carried out. c. Methods 1) TREND AND SIGNIFICANCE For each variable in observations and reanalysis products, the seasonal (June September) anomalies were obtained by subtracting its respective climatology. A similar procedure was applied to CM2.1 simulations. In CM2.1, the AIR index is constructed by averaging the anomalous rainfall over the region N, E. The results are robust if only the land points over India are considered. For ICOADS cloud amount data, following Deser et al. (2010) we applied spatial smoothing (5-point longitude and 3-point latitude binomial filter) to capture large-scale features before forming the seasonal anomalies. The long-term linear trend is estimated by simple regression analysis as in past studies (Deser et al. 2010). To insure that trends are above natural variability, their significance is verified against the local interannual standard deviation of the variable of interest. In addition, trend significance is examined by applying a Student s

4 2704 J O U R N A L O F C L I M A T E VOLUME 26 t test, taking into account serial correlations. In CM2.1, the ensemble-mean trend is based on the averages of the trends seen in each individual run. For CM2.1, the significance was also tested against trends (due to internal variability) obtained from 20 overlapping 50-yr chunks of the coupled model 200-yr control run. For assessing significance in AM2.1 sensitivity experiments, the differences between the ensemble mean and the Control run were subjected to a t test. To appreciate the application of the t test on modelsimulated rainfall, the probability distribution of the AIR index was examined and found to be normally distributed (not shown). Further, a nonparametric test based on Monte Carlo was applied to assess the trend significance. The results are consistent with those of the t test and hence not shown. 2) MOIST STATIC ENERGY ANALYSIS Here, we briefly provide the formulations; readers are referred to Su and Neelin (2002) for details. Recently, we employed MSE budgets to understand the moist processes involved in the interannual (Annamalai 2010; Pillai and Annamalai 2012) and intraseasonal (Prasanna and Annamalai 2012) variations of South Asian summer monsoon rainfall. The vertically integrated anomalous MSE equation used here is given by hd T T 0 i1hd q q 0 i1hv p h 0 i5 (g/p T )(E 0 1 H 0 )1R 0, (1) where H 0, E 0, and R 0 are the anomalous surface sensible heat flux, latent heat flux, and net radiative flux into the atmospheric column, respectively; g is acceleration due to gravity; P T is pressure depth; T 0 is temperature perturbation; q 0 is anomalous moisture; v 0 is anomalous vertical velocity; and h 0 is anomalous MSE. Here, anomalies are trends with respect to the Control simulation. The symbol represents vertical integration over the troposphere. The operators D T (for temperature) and D q (for moisture) include horizontal advection and horizontal diffusion terms: D 5 y $ 2 K H = 2. (2) Note that T 0 and q 0 are in units of joules per kilogram (i.e., T 0 and q 0 absorb the heat capacity at constant pressure and latent heat of condensation, respectively). 3. Robustness in the drying over South Asia a. East west shift in monsoon rainfall and circulation To ascertain the spatial extent of drying over South Asia, the linear trend was estimated in gridded observational rainfall and analyzed circulation fields. In land-based rainfall products (e.g., Fig. 2a), the declining pattern over the plains of central India and Indo-China is consistent with other studies (e.g., Ramanathan et al. 2005). The spatial extent of drying is captured in other rainfall observations [not shown here, but see Bollasina et al. (2011)] but the estimated intensity is higher in CRU (Fig. 2a). In their analysis of 18 gridded rainfall data produced by the India Meteorological Department, Bollasina et al. (2011) noted the drying pattern in conjunction with regional differences among the various products. From the results presented here and elsewhere, one robust feature is the declining tendency in monsoon rainfall over central India but because of observational uncertainties and different algorithms employed in regridding station rainfall observations, the magnitude of this declining varies among the products. To reduce uncertainty, Turner and Annamalai (2012) suggested reprocessing observational rainfall products and verifying them against independent observations such as crop yields. In seeking attribution for this declining tendency, next we estimate trends in SST and circulation fields. During , it is clear from Fig. 2c that SST rose in both the tropical Indian Ocean and the western tropical Pacific (a region extending over 108S 308N, E) in the former region by ;0.758C, and in the latter by ;0.58C. This is a robust result in all three SST products (Figs. 1b and 3b). Yet, in the Indian Ocean, despite the SST rise, SLP has increased over the western Indian Ocean, and the climatological southwesterly low-level monsoon winds have weakened (Fig. 2b). The circulation features are dynamically consistent with the drying over India that extends into the Bay of Bengal, as suggested by decreasing tendency in total cloud amount (Fig. 3a). In the tropical western Pacific ( N, E), in contrast, SLP has dropped and the cross-equatorial flow emanating from the Australian high, the low-level westerlies, and anomalous cyclonic circulation have strengthened (Fig. 2b). Direct rainfall observations are unavailable; nevertheless, the decrease in sea surface salinity from 1955 to 2003 (Delcroix et al. 2007), the increased atmospheric water vapor content from 1988 to 2006 (Santer et al. 2007), and the drop in SLP (Copsey et al. 2006) all imply this region has had more rainfall in recent decades. These indirect observations are confirmed by an increase in observed total cloud amount (Fig. 3a). Thus, observations point to a change in the monsoon circulation that has resulted in less rainfall over South Asia and more over the tropical western Pacific. In the three reconstructed SST products (not shown), the overall trend patterns and amplitude over the tropical western Pacific are similar and consistent with others (Deser et al. 2010; L Heureux et al. 2013).

5 1MAY 2013 A N N A M A L A I E T A L FIG. 2. (a) Linear trend in observed rainfall [mm month 21 (52 yr) 21 ] during boreal summer (JJAS) from the CRU gridded dataset for the period (b) As in (a), but for sea level pressure [hpa (52 yr) 21 ] from Hadley Centre data and 850-hPa wind [m s 21 (52 yr) 21 ] from the NCEP NCAR reanalysis. (c) As in (a), but for SST data from HadISST2 [8C (52 yr) 21 ]. In each of the panels, negative trend values are shown in blue and purple while positive values are in red. At each grid, only trend values greater than local interannual std dev are shown. The elliptical region highlighted in (a) represents South Asia, and in (b) the boxed regions (South Asia solid, and tropical western Pacific dotted) are highlighted. In (a), contours represent significant values. Among the three products, details in SST trends over the near-equatorial Indian Ocean differ but the area average shows consistent warming in recent decades (Fig. 3b). The far-eastern equatorial Pacific shows warming only in ERSST.v3 (Deser et al. 2010), but this signal is not reflected in total cloud amount (Fig. 3a). Observational studies suggest a shift toward more central Pacific versus eastern Pacific El Niños during recent decade (McPhaden 2012), which also accounts for weakening of the monsoon rainfall (Pillai and Annamalai 2012). While SST trends suggest a mild warming around the date line, lack of observed rainfall there is a major impediment to associating the long-term changes in monsoon rainfall (50-yr trend) to changes in El Niño features. If the ENSO-related signal is removed from the tropical SST, the estimated trend retains the warming in the Indo-Pacific warm pool but a cooling tendency is noticeable along the equatorial eastern Pacific (Compo and Sardeshmukh 2010). Although at interannual time scales SST cooling over the equatorial eastern Pacific tends to enhance the South Asian monsoon rainfall and circulation, at longer time scales the reduction in the South Asian monsoon circulation identified here implies the possible role of factors other than ENSO. Is there support from a coupled model simulation for this observed east west shift in monsoon rainfall? During , barring regional details, CM2.1 simulated the observed drying over South Asia (Fig. 4a), the increased SLP and the reduced monsoon circulation over the northern Indian Ocean (Fig. 4b), the increased SST in the Indo-Pacific warm pool, and the insignificant warming along the equatorial eastern Pacific (Fig. 4c). In the tropical western Pacific, increased rainfall (Fig. 4a) is dynamically consistent with stronger low-level cyclonic circulation and lower SLP (Fig. 4b). While the coupled model captures the broad features noted in observations and reanalysis (Fig. 2), there are certain limitations including lack of simulation of positive rainfall trend over southern China and negative rainfall trend over the plains of Indo-China (Fig. 4a). Like the observed AIR (Fig. 1a), the model ensemble members capture multidecadal epochs in monsoon rainfall (Figs. 5a,b), and the ensemble-mean plots show the current drying over South Asia [see Fig. 1c (red line) and Fig. 5a (thick line)] and the wetter tendency over the tropical western Pacific [see Fig. 1c (blue line) and Fig. 5b (thick line)]. The shift appears to occur shortly after a systematic rise in SST or a consistent period of mean state change in the tropical western Pacific latter (Fig. 1c, brown line). Like in observations (Fig. 1b), in CM2.1 the tropical western Pacific SST also rises at a higher rate after the mid-1970s than during (Fig. 1c), but the changes in rainfall over South Asia during those

6 2706 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 3. (a) Linear trend in observed total cloud amount during boreal summer [okta (52 yr) 21 ] and (b) temporal evolution of observed boreal summer SST anomalies averaged over the equatorial Indian Ocean ( E, 108S 58N) from three SST products. In (a), negative values are light and positive values are dark. two periods remain the same; we return to this point in section 5. A 200-yr control CM2.1 simulation without anthropogenic forcing shows decade-long dry and wet periods, but not the declining trend over South Asia or increasing trend over tropical western Pacific (Fig. 5c). In the absence of such trends in the control run, the shift in the monsoon circulation seen in model runs (e.g., Fig. 4a) in which GHGs and aerosols are included implies that man-made emissions are likely the cause of these changes, but the lack of sustained observed rainfall over the oceanic regions places a severe constraint in our findings. In the CM2.1 control simulation (Fig. 5c), the two regional rainfall indices are roughly out of phase except for a brief period during years b. Proposed hypothesis Results presented so far indicate that the centers of monsoon rainfall have shifted eastward in recent decades. Is the SST warming in the Indo-Pacific warm pool the cause for the monsoon declining over South Asia? In a previous study, we have shown that the rainfall pattern over the Asian monsoon (ASM) region is determined by three heat sources, or convergence zones: the western tropical Pacific, the Bay of Bengal extending into the Indian subcontinent, and the equatorial Indian Ocean (Annamalai 2010). The rise in the Indo-Pacific SST may have affected these heat sources differently. In other words, within the major mean convergence zones of the ASM, mutual interactions among the regional heat sources can substantially affect rainfall. Our hypothesis is that the rise in Indo-Pacific SST increases tropical western Pacific monsoon rainfall, which incites a Rossby wave that forces descending air to the west, drying South Asia. 4. Validation of hypothesis a. AGCM results and budget diagnostics The differences between the ensemble mean and the control run for the entire period (52-yr trends) are

7 1MAY 2013 A N N A M A L A I E T A L FIG. 4. Boreal summer (JJAS) linear trend estimated from GFDL CM2.1 coupled model simulations: (a) rainfall [mm month 21 (52 yr) 21 ], (b) SLP [hpa (52 yr) 21 ] and 850-hPa wind [m s 21 (52 yr) 21 ], and (c) SST [8C (52 yr) 21 ]. In each of the panels, negative trend values are shown in blue and purple while positive values are in red. Note that in the coupled model, the trend is estimated in each of the ensemble members separately and then a grand ensemble mean is estimated and shown here. At each grid, only trend values greater than local interannual std dev are shown. In (a) and (b), the boxed regions [solid, South Asia; dotted, tropical western Pacific] are highlighted. shown. Figures 6a f show solutions obtained from EXP2. Results from EXP1 are presented in Fig. 7. In spite of their different SST forcing, both the experimental model solutions show greater rainfall over the tropical western Pacific and drying over South Asia (Figs. 6a and 7a). The model SLP and winds at 850 hpa (Figs. 6b and 7b) are dynamically consistent [i.e., climatological monsoon winds have weakened (strengthened) and SLP has risen (dropped) in South Asia (western Pacific)] with changes in rainfall. Given that the SST trend used to force AM2.1 is largely of anthropogenic origin (Stott et al. 2000; Knutson et al. 2006; Santer et al. 2007), consistency among observations and CM2.1 and AM2.1 solutions implies that the east west shift in monsoon rainfall is likely due to anthropogenic forcing through changes in SST warming over the tropical western Pacific. Because column-integrated MSE (Fig. 6c) closely resembles rainfall anomalies over the deep tropics and illustrates energy export from deep convective regions by overturning circulations (Neelin et al. 2006; Annamalai 2010), we calculated the MSE budget to understand how the SST rise affected the three heat sources and their individual contributions to rainfall anomalies over the other regions [see Eq. (1)]. The MSE equation is particularly useful because the large terms in the individual moisture and temperature equations cancel each other out, revealing the sensitivity of convective rainfall to forcing from other important sources in tropical climate dynamics. Over South Asia, the net effect of reduced evaporation and dry air advection (boxed area with solid line in Fig. 6) dominates the MSE import, contributing about 40%. In addition, cold air is advected into South Asia (Fig. 6f) and radiative cooling over the northern Indian Ocean promotes descent (Fig. 8b), reinforcing the drying. Although the reduced cloud cover increases sensible heat flux, its contribution over the ocean is small (Fig. 8a). Over the tropical northern Indian Ocean, despite the prescription of warm SST, reduced wind speed reduces evaporation (Fig. 6d). Quantitatively, over South Asia both dry advection and reduced evaporation make up the major contribution (;40%) to MSE divergence followed by sensible heat flux, temperature advection, and net radiation. Over the tropical western Pacific (boxed area with dotted lines in Fig. 6), increased evaporation (Fig. 6d) dominates the MSE export. Here, the steady rise in SST increases surface heat and moisture fluxes, increasing the moist entropy of the boundary layer and promoting rainfall. A positive feedback between rainfall and circulation is generated, strengthening the low-level wind (Fig. 6b), which in turn enhances evaporation (Fig. 6d).

8 2708 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 5. (a) 31-yr running mean applied on GFDL CM2.1 coupled model simulated seasonal (JJAS) rainfall anomalies averaged over South Asia (58 258N, E) from all the five members and ensemble mean (solid black). (b) As in (a), but averaged over the tropical western Pacific ( N, E). (c) 31-yr running mean of boreal summer rainfall anomalies averaged over South Asia (solid line) and the tropical western Pacific (dashed line) from CM2.1 control run performed without GHG forcing. Results from AM2.1 experiments forced with the tropical western Pacific SST trend only (EXP1) suggest that both the spatial pattern and strength of the simulated rainfall (Fig. 7a) and associated circulation anomalies (Fig. 7b) over the ASM closely resemble those in which the SST trend is prescribed over the entire Indo-Pacific warm pool region. One notable difference is that in EXP1 no appreciable positive precipitation anomalies are simulated over the equatorial Indian Ocean. Also in EXP1, an examination of budget diagnostics indicates that dry advection (Fig. 7e) is the primary cause for reduction in MSE (Fig. 7c) or rainfall over South Asia while evaporation (Fig. 7d) contributes substantially to MSE export or increase in rainfall over tropical western Pacific. One caveat is that AM2.1 uses sigma coordinates and the output is interpolated to standard pressure levels, and MSE can be sensitive to interpolations. Also, the different numerical schemes employed for advection in the original model codes versus the one employed by us

9 1MAY 2013 A N N A M A L A I E T A L can introduce errors. Therefore, the budget is not closed exactly. In future, we plan to compute the budget quantities in the native model coordinates. Because of differing amplitude in SST trend along the equatorial Pacific, the ensemble mean of the three SST products was constructed for forcing AM2.1 in EXP3. While near-equatorial Pacific SST warming promotes enhanced rainfall over tropical western Pacific (e.g., Pillai and Annamalai 2012), the response over South Asia is too weak (Fig. 9a). To highlight the importance of tropical Indian Ocean SST, Fig. 9b shows the difference in simulated rainfall between EXP1 and EXP2. In response to local warming, modest rainfall increase along the tropical northern Indian Ocean is compensated with a decrease over South China Sea western Pacific, consistent with the results of Annamalai and Sperber (2005). A closer examination of model-simulated rainfall trends suggests some regional inconsistencies compared to observed rainfall trends. Specifically, both in AM2.1 (Figs. 6a, 7a) and CM2.1 (Fig. 4a), drying over South Asia is more pronounced over the Arabian Sea than over continental India. This is, perhaps, associated with systematic errors in the model basic state, particularly higher rainfall over the Arabian Sea (Pillai and Annamalai 2012). The suggested east west dynamical linkage between the tropical western Pacific and South Asia is interpreted as follows. The Rossby wave response to increase in tropical western Pacific rainfall (Fig. 6a) strengthens the in situ low-level cyclonic circulation (Fig. 6b). The northerly component of this Rossby wave on its poleward flank advects dry and cold air from the continental subtropics into the South Asian monsoon region, particularly into the Bay of Bengal and the plains of Indo- China (Figs. 6b,e,f; see also Figs. 2b and 4b). This reduces rainfall over the Bay of Bengal and dampens one of the three heat sources controlling the South Asian monsoon rainfall and weakens the moisture laden cross-equatorial monsoon flow. Evaporation is reduced along the flow path, maintaining the dryer cycle over South Asia. b. Moisture advection role Since moisture advection makes a substantial contribution to budget diagnostics, we examine the contributions from individual terms. The governing equation of anomalous moisture advection is hy $q 0 i 5 hy 0 $q c i 1 hy c ($q) 0 i 1 hy 0 ($q) 0 i 1 hy 00 ($q) 00 i 0. (3) In the above equation, angle brackets represent vertical integration. The first term on the right-hand side, hy 0 $q c i, denotes advection associated with anomalous wind acting on the climatological moisture gradient, and the second term, hy c ($q) 0 i, represents advection due to climatological wind acting on the anomalous moisture gradient. The other terms, hy 0 ($q) 0 i and hy 00 ($q) 00 i 0, represent anomalous wind acting on anomalous moisture gradient and a residual term due to transient variability, respectively. Figures 8c e show the individual terms of the moisture advection equation. Over the plains of South Asia, dry advection from all the three terms contributes to the total dry advection noted earlier (Fig. 6e). The anomalous low-level circulation with associated northeasterlies (Fig. 6b) advects dry air or air of lesser MSE from the subtropics into parts of Southeast and South Asia. A notable feature also is that the easterly anomalies over the southern Arabian Sea force anticyclonic vorticity through Ekman pumping and result in dry air advection from the north into western parts of South Asia. While all the terms in the moisture advection equation contribute, the contribution from the anomalous winds acting on climatological moisture gradient dominates, consistent with the Rossby wave interpretation offered here. The role of South Asian monsoon heat source forcing absolute climatological descent over the eastern Mediterranean through Rossby waves has been demonstrated elsewhere (Rodwell and Hoskins 1996). Here, at longer time scales, we identify the moist processes responsible for descent anomalies over South Asia that in turn are forced by enhanced rainfall anomalies within the broader Asian monsoon region. Further support for our results can be found in recent studies. Two atmospheric models forced with observed SST variations for the period capture the enhanced rainfall trend and associated vertical velocity over the tropical west Pacific (58 258N, E) and reduced rainfall tendency and associated descent over South Asia (58 258N, E; Kumar et al. 2004). Of the five atmospheric models forced with observed SST warming trends over the Indo-Pacific two reproduced the east west precipitation shift (Zhou et al. 2009; Sanchez-Gomez et al. 2008). While our results may be model dependent, AM2.1 captures the regional precipitation climatology and the moist mechanism through which tropical SST variations influence the South Asian monsoon (Pillai and Annamalai 2012), implying the role of the basic state. In the deep tropics, SST governs the lower tropospheric MSE (Neelin et al. 2006) and the high-mean SST and stronger gradients in the vertical profile of potential temperature result in a highly unstable atmosphere. Figure 10 shows vertical structure of area-averaged equivalent potential temperature (or MSE) over the

10 2710 JOURNAL OF CLIMATE VOLUME 26 FIG yr trend during boreal summer (JJAS) simulated by the GFDL AM2.1 forced by SST trend over the tropical Indo-Pacific warm pool region (EXP2). The trend is obtained by subtracting the experiment (EXP2) values from the Control run that was forced by climatological SST: (a) precipitation (W m22), (b) SLP (hpa) and 850-hPa wind (m s21), (c) vertically integrated moist static energy divergence (W m22), (d) surface evaporation (W m22), (e) vertically integrated moisture advection (W m22), and (f) vertically integrated temperature advection (W m22). Note that precipitation unit conversion is 28 W m mm day21. In the above panels, the region outlined in solid (dotted) lines represents South Asia (tropical western Pacific). In Fig. 6c, positive values indicate MSE export due to enhanced rainfall, and negative values indicate air of low MSE import; in Fig. 6e, moist advection into the region is positive and dry advection is negative. three regional heat sources from the Control integration. In the basic state, the rate of decrease of equivalent potential temperature with height in the lowto-middle troposphere (a measure of atmospheric stability) is sharper over the tropical western Pacific. Thus, compared to South Asia and the equatorial Indian Ocean, because of both higher surface and boundary layer MSE and its vertical gradient, the atmosphere is very unstable over the tropical western Pacific region, and therefore sensitive to SST perturbations. At higher tropospheric levels, equivalent potential temperature gradients increase primarily due to temperature. In summary, during summer, the atmosphere over the tropical western Pacific is unstable than over the tropical Indian Ocean to begin with. Thus the rise in SST and the accompanying increased moisture in the lower

11 1MAY 2013 A N N A M A L A I E T A L FIG yr trend during boreal summer (JJAS) simulated by the atmosphere model forced by SST trend over the tropical western Pacific region (EXP1). The trend is obtained by subtracting the EXP1 values from the Control run that was forced by climatological SST: (a) precipitation (W m 22 ), (b) SLP (hpa) and 850-hPa winds (m s 21 ), (c) vertically integrated moist static energy divergence (W m 22 ), (d) evaporation (W m 22 ), (e) vertically integrated moisture advection (W m 22 ), and (f) vertically integrated temperature advection (W m 22 ). In the above panels, the region outlined in solid (dotted) lines represents South Asia (tropical western Pacific). Note that precipitation unit conversion is 1.0 mm day W m 22. troposphere can more readily destabilize the atmosphere and increase the vertical velocity anchoring convection in this region than over the tropical Indian Ocean. c. CM2.1 budget diagnostics MSE divergence (Fig. 11a) and moisture advection (Fig. 11b) trends estimated from CM2.1 ensemble mean integration, the difference between averages and averages, supports the results from AM2.1 sensitivity experiments. In particular, dry advection (Fig. 11b) contributes to MSE divergence or rainfall reduction (Fig. 4a) over South Asia. Over the tropical western Pacific, MSE export indicative of enhanced rainfall (Fig. 4a) dominates. The alternating positive and negative values in MSE poleward of 208S are accounted for by similar signatures in temperature advection (not shown). Analysis of all the budget terms (not shown) indicates the dominance of dry advection to MSE divergence over South Asia (note the same shading interval in Figs. 11a and 11b). If climate models are to simulate these changes, they need to capture the basic

12 2712 J O U R N A L O F C L I M A T E VOLUME 26 FIG yr trend during boreal summer (JJAS) simulated by the atmosphere model forced by SST trend over the tropical Indo-Pacific warm pool region (EXP2). The trend is obtained by subtracting the EXP2 values from the control run that was forced by climatological SST: (a) sensible heat flux (W m 22 ), (b) net radiative flux into the column (W m 22 ), and (c) vertically integrated moisture advection (W m 22 ) due to anomalous wind acting on climatological moisture gradient [term 1 in Eq. (3)]. (d) As in (c), but for climatological wind acting on anomalous moisture gradient [term 2 in Eq. (3)]; (e) as in (c), but anomalous wind acting on anomalous moisture gradient [term 3 in Eq. (3)]. state of the regional distribution in the monsoon precipitation. The lack thereof explains why some models do not capture the east west dynamical linkage (Zhou et al. 2009; Sanchez-Gomez et al. 2008). Steady-state solutions from a linear AGCM forced with the observed SST trend in the tropical western Pacific capture the rainfall shift and indicate the role of dry advection over South Asia (not shown). 5. Discussion The results presented point to the rise in SST over the tropical western Pacific as a possible factor for the drying over South Asia. Figure 1b shows the observed AIR and SST over the tropical west Pacific. While a negative relationship and a broad coherency exist between the two indices over the last 5 6 decades, SST rise is greater

13 1MAY 2013 A N N A M A L A I E T A L FIG. 9. (a) 52-yr precipitation trend during boreal summer simulated by the GFDL AM2.1 forced by SST trend over the equatorial Pacific (108S 108N, 1308E 1008W; EXP3). The trend is obtained by subtracting the experiment (EXP3) values from the control run that was forced with climatological SST. (b) AM2.1-simulated difference in precipitation trend (mm day 21 ) between EXP1 and EXP2. In both panels, positive values are shaded progressively while negative values are shown in contours. The contour interval is 0.2 (0.5) in the top (bottom) panel. (0.138C) during the latter half while the decrease in AIR (4%) is slightly greater during the first half when the SST rise is only about 0.078C. Should we expect a one-to-one relationship between SST warming in the tropical western Pacific and AIR drying? In a warming scenario and based on MSE principles, both coupled modeling studies (Knutson and Manabe 1995; Williams et al. 2009; Sud et al. 2008) and observations (Johnson and Xie 2010) show that the SST threshold for the occurrence of tropical deep convection rise. An examination of

14 2714 J O U R N A L O F C L I M A T E VOLUME 26 observed SST and outgoing longwave radiation (OLR) over the tropical western Pacific for two different 15-yr periods ( vs ) supports this (Fig. 12a): during the first 15 years the threshold for deep convection (OLR values between 235 and 220 W m 22 ) was in the SST range of C (blue squares) while in recent 15 years the threshold for deep convection rose to C (red squares), an increase of about 0.48C. Further support comes from EXP1, in which AM2.1 is forced only with the observed SST trend over the tropical western Pacific for the period A steady increase in rainfall over the tropical western Pacific (Fig. 12b) is accompanied with a steady rainfall decline over South Asia (Fig. 12c). In summary, it is noted that under the influence of strong ascending motion (such as in the tropical western Pacific during boreal summer), convection increases with increasing SST even at SST higher than 29.58C, and the rate of increase of convection can be very large for the SST range of C (Lau et al. 1997). Thus, despite a slow warming during the 1950s 70s, the SST anomalies are positive and sufficient enough to anchor the in situ convection. Compared to recent decades, prior to 1945 the relationship between AIR and tropical western Pacific SST is inconsistent (Fig. 1b). Compo and Sardeshmukh (2010) suggest that once the ENSO-related variations are removed, tropical Indo-Pacific warm pool shows a cooling tendency from ;1900 to ;1960s (see their Fig. 11a). Our future work will revisit the monsoon western Pacific SST relationship for the available record following the Compo and Sardeshmukh (2010) approach. Climatologically, monsoon rainfall over the Indo- Pacific warm pool is intense (.6 mm day 21 ) with strong cross-equatorial flow that turns southwesterly over the northern Indian Ocean and tropical western Pacific. According to thermodynamics (Held and Soden 2006), the rise in Indo-Pacific SST should have promoted more moisture convergence and rainfall over the whole region. Yet, the expected increase in rainfall over the tropical Indian Ocean is not seen, and the reasons for this will be examined in a separate article. 6. Summary Examination of millennial-length data of oxygen isotopes from Dandak Cave in east central India provides compelling evidence for a persistent multidecadal cycle in monsoon rainfall (Berkelhammer et al. 2010). Since reliable rainfall observations over India (the AIR) became available 140 years ago, the first breakdown in the alternating multidecadal dry and wet phases seems to be taking place, with the entry to the expected wet phase FIG. 10. Vertical profiles of equivalent potential temperature (or MSE) over the three regional heat sources estimated from the Control simulation: tropical western Pacific averaged over N, E (gray line with open squares), South Asia averaged over N, E (dark line with closed squares), and the equatorial Indian Ocean averaged over 108S 58N, E (solid line with open circles). not occurring. What is causing the current observed drying trend in South Asia monsoon rainfall? Modeling studies have suggested that the Atlantic multidecadal oscillation impacts the decadal swings in AIR, but based on its current warm phase we should be in a wetter monsoon period (Zhang and Delworth 2006). While some studies suggest black carbon (Ramanathan et al. 2005) or sulfate (Bollasina et al. 2011) aerosols as a factor, other modeling studies indicate otherwise (Wang 2004). CM2.1 sensitivity experiments comparing naturalonly and anthropogenic-only forcing suggest the latter for SST warming over the tropical western Pacific (Knutson et al. 2006). In a warmer world, increased temperature promotes increased tropospheric moisture and rainfall over mean convective regions. While this thermodynamic effect work on global, hemispherical, and annual averages, over the ASM region dynamical feedbacks among the regional heat sources complicate the above picture. No single forcing is strong enough to impact the heating over the entire ASM region, equatorial and off-equatorial (108S 308N, E), to produce a single-signed anomaly (Annamalai 2010). The circulation that develops in response to rainfall

15 1MAY 2013 A N N A M A L A I E T A L FIG. 11. (a) The difference ( minus ) in vertically integrated moist static energy divergence during boreal summer (JJAS) in GFDL CM2.1 historical simulations. (b) As in (a), but for vertically integrated moisture advection (W m 22 ). The budgets were estimated in each of the ensemble member, and then a grand-ensemble mean is estimated and shown here. Negative values are shaded progressively and positive values are shown in contours with an interval of 4. changes in the warm pool interacts with moist processes, influencing other heat sources (i.e., the small signal in the MSE equation, indicative of dry advection), can be multiplied through positive feedbacks and have a large impact on precipitation. The SST trend together with effect of aerosols may have amplified the drying tendency. The frequent drier monsoon over South Asia since the 1960s raises concerns about water and food security issues. The anthropogenic-induced drying trend can be anticipated to continue. Although CMIP3 models driven with doubled or quadrupled CO 2 concentrations project a slight increase in rainfall over South Asia after the models have reached equilibrium (Annamalai

16 2716 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 12. (a) Scatterplot between observed SST and outgoing longwave radiation (OLR) during boreal summer (JJAS) averaged over the tropical west Pacific ( N, E) for two different epochs. (b) JJAS rainfall (mm day 21 ) over the tropical western Pacific simulated by AM2.1 forced with observed SST trend over the tropical western Pacific (EXP1). (c) As in (b), but over South Asia. In (b) and (c), the estimated linear trend is shown (red line) and is significant at 95% level. et al. 2007; Turner et al. 2007) that is, toward the end of twenty-first century the results here point to declining monsoon rainfall during the transient phase. The results presented here are being verified in CMIP5 models. Acknowledgments. The authors acknowledge the National Energy Research Scientific Computing Centre (NERSC), which is supported by the Department of Energy, for the computational resources. This work is supported by the Office of Science (BER), U.S. Department of Energy, Grant DEFG02-07ER6445, and also by the three institutional grants (JAMSTEC, NOAA, and NASA) of the IPRC. Dr. Gisela Speidel is acknowledged for editorial assistance. Comments from the two anonymous reviewers are appreciated. REFERENCES Anderson, J. L., and Coauthors, 2004: The new GFDL global atmosphere and land model AM2-LM2: Evaluation with prescribed SST simulations. J. Climate, 17, Annamalai, H., 2010: Moist dynamical linkage between the equatorial Indian Ocean and the South Asian monsoon trough. J. Atmos. Sci., 67,

17 1MAY 2013 A N N A M A L A I E T A L. 2717, and K. R. Sperber, 2005: Regional heat sources and the active and break phases of boreal summer intraseasonal (30-50 day) variability. J. Atmos. Sci., 62, , K. Hamilton, and K. R. Sperber, 2007: The South Asian summer monsoon and its relationship with ENSO in the IPCC AR4 simulations. J. Climate, 18, Basnett, T. A., and D. R. Parker, 1997: Development of the global mean sea level pressure data set GMSLP2. Met Office Climate Research Tech. Note CRTN79, 16 pp. Berkelhammer, M., and Coauthors, 2010: Persistent multidecadal power of the Indian summer monsoon. Earth Planet. Sci. Lett., 290, Bollasina, M. A., Y. Ming, and V. Ramaswamy, 2011: Anthropogenic aerosols and the weakening of the South Asian summer monsoon. Science, 334, Chen, M., P. Xie, J. Janowiak, and P. Arkin, 2002: Global land precipitation: A 50-yr monthly analysis based on gauge observations. J. Hydrometeor., 3, Compo, G. P., and P. Sardeshmukh, 2010: Removing ENSO-related variations from the climate record. J. Climate, 23, Copsey, D., R. Sutton, and J. R. Knight, 2006: Recent trends in sea level pressure in the Indian Ocean region. Geophys. Res. Lett., 33, L19712, doi: /2006gl Delcroix, T., S. Cravatte, and M. J. McPhaden, 2007: Decadal variations and trends in tropical Pacific sea surface salinity since J. Geophys. Res., 112, C03012, doi: / 2006JC Delworth, T. L., and Coauthors, 2006: GFDL s CM2 global coupled climate models. Part I: Formulation and simulation characteristics. J. Climate, 19, Deser, C., A. S. Phillips, and M. A. Alexander, 2010: Twentieth century tropical sea surface temperature trends revisited. Geophys. Res. Lett., 37, L10701, doi: /2010gl Gautam, R., N. C. Hsu, K.-M. Lau, and M. Kafatos, 2009: Aerosol and rainfall variability over the Indian monsoon region: Distributions, trends and coupling. Ann. Geophys., 27, Griffies, S. M., M. J. Harrison, R. C. Pacanowski, and A. R. Rosati, 2003: A technical guide to MOM4. GFDL Ocean Group Tech. Rep. 5, 295 pp. Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, ; Corrigendum, 24, , T. L. Delworth, J. Lu, K. L. Findell, and T. R. Knutson, 2005: Simulation of Sahel drought in the 20th and 21st centuries. Proc. Natl. Acad. Sci. USA, 102, ; Corrigendum, 103, Johnson, N., and S. P. Xie, 2010: Changes in the sea surface temperature threshold for tropical convection. Nat. Geosci., 3, Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, Kaplan, A., M. Cane, Y. Kushnir, A. Clement, M. Blumenthal, and B. Rajagopalan, 1998: Analyses of global sea surface temperature J. Geophys. Res., 103, Knutson, T. R., and S. Manabe, 1995: Time-mean response over the tropical Pacific to increased CO 2 in a coupled ocean atmosphere model. J. Climate, 8, , and Coauthors, 2006: Assessment of twentieth-century regional surface temperature trends using the GFDL CM2 coupled models. J. Climate, 19, Kumar, A., F. Yang, L. Goddard, and S. Schubert, 2004: Differing trends in the tropical surface temperatures and precipitation over land and oceans. J. Climate, 17, Lau, K.-M., H.-T. Wu, and S. Bony, 1997: The role of large-scale atmospheric circulation in the relationship between tropical convection and sea surface temperature. J. Climate, 10, L Heureux, M., D. C. Collins, and Z.-Z. Hu, 2013: Linear trends in sea surface temperature of the tropical Pacific Ocean and implications for the El Niño Southern Oscillation. Climate Dyn., 40, , doi: /s McPhaden, M. J., 2012: A 21st century shift in the relationship between ENSO SST and warm water volume anomalies. Geophys. Res. Lett., 39, L09706, doi: /2012gl Meehl, G. A., C. Covey, K. E. Taylor, T. Delworth, R. J. Stouffer, M. Latif, B. McAvaney, and J. F. B. Mitchell, 2007: The WCRP CMIP3 multimodel dataset: A new era in climate change research. Bull. Amer. Meteor. Soc., 88, Mitchell, T. D., and P. D. Jones, 2005: An improved method of constructing a database of monthly climate observations and associated high-resolution grids. Int. J. Climatol., 25, Mooley, D. A., and B. Parthasarathy, 1984: Fluctuations in all- India summer monsoon rainfall during Climate Change, 6, Neelin, J. D., M. Münnich, H. Su, J. E. Meyerson, and C. E. Holloway, 2006: Tropical drying trends in global warming models and observations. Proc. Natl. Acad. Sci. USA, 103, Parthasarathy, B., K. Rupakumar, and A. A. Munot, 1994: All- India monthly and seasonal rainfall series: Theor. Appl. Climatol., 49, Pillai, P. A., and H. Annamalai, 2012: Moist dynamics of severe monsoons over South Asia: Role of the tropical SST. J. Atmos. Sci., 69, Prasanna, V., and H. Annamalai, 2012: Moist dynamics of extended monsoon breaks over South Asia. J. Climate, 25, Ramanathan, V., and Coauthors, 2005: Atmospheric brown clouds: Impacts on South Asian climate and hydrological cycle. Proc. Natl. Acad. Sci. USA, 102, Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Globally complete analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108, 4407, doi: / 2002JD Rodwell, M. J., and B. J. Hoskins, 1996: Monsoons and the dynamics of deserts. Quart. J. Roy. Meteor. Soc., 122, Sanchez-Gomez, E., C. Cassou, D. L. R. Hodson, N. Keenlyside, Y. Okumura, and T. Zhou, 2008: North Atlantic weather regimes response to Indian-western Pacific Ocean warming: A multi-model study. Geophys. Res. Lett., 35, L15706, doi: / 2008GL Santer, B. D., and Coauthors, 2007: Identification of humaninduced changes in atmospheric moisture content. Proc. Natl. Acad. Sci. USA, 104, Smith, T. M., R. W. Reynolds, P. C. Peterson, and J. Lawrimore, 2008: Improvements to NOAA s Historical Merged Land Ocean Surface Temperature Analysis ( ). J. Climate, 21, Sperber, K. R., and H. Annamalai, 2008: Coupled model simulations of boreal summer intraseasonal (30-50 day) variability, Part I: Systematic errors and caution on use of metrics. Climate Dyn., 31, , doi: /s Stott, P. A., S. F. B. Tett, G. S. Jones, M. R. Allen, J. F. B. Mitchell, and G. J. Jenkins, 2000: External control of 20th century temperature by natural and anthropogenic forcings. Science, 290,