Abstract

Recent studies arising from both statistical analysis and dynamical disease models demonstrate a link between the incidence of cholera, a paradigmatic waterborne bacterial illness endemic to Bangladesh, and the El Niño–Southern Oscillation (ENSO). The physical significance of this relationship was investigated by examining links between the regional climate of Bangladesh and western Pacific sea surface temperatures (SST) associated with ENSO using a pacemaker configuration of the Center for Ocean–Land–Atmosphere Studies atmospheric general circulation model. The global SST response to ENSO SST anomalies in the western Pacific alone is found to be relatively weak and unrealistic when compared to observations, indicating that the global response to ENSO is driven primarily by anomalies in the central and eastern tropical Pacific. Despite the weak global response to western Pacific SST anomalies, however, a signal is found in summer rainfall over India and Bangladesh. Specifically, reduced rainfall typically follows winter El Niño events. In the absence of warm SST anomalies in the eastern Pacific, cold anomalies in the western Pacific produce a La Niña–like response in the model circulation. Cold SST anomalies suppress convection over the western Pacific. Large-scale convergence shifts into the eastern Indian Ocean and modifies the summer monsoon circulation over India and Bangladesh.

The probabilistic relationship between Bangladesh rainfall and SST is also explored using a nonparametric statistical technique. Decreased rainfall is strongly associated with cold SST in the western Pacific, while associations between SST and enhanced rainfall are substantially weaker. Also found are strong associations between rainfall and SST in the Indian Ocean in the absence of differences in forcing from the western Pacific. It thus appears that the Indian Ocean may represent an independent source of predictability for the monsoon and cholera risk. Likewise, under certain circumstances, the western Pacific may also exert a significant influence on Bangladesh rainfall and cholera risk.

1. Introduction

Cholera is a waterborne disease endemic to the Bangladesh region. Infection arises from ingesting water contaminated with the bacteria vibrio cholerae and is thus most prevalent in regions with limited or impaired access to clean water. Left untreated, mortality rates as high as 50% are common; however, mortality in even severe cases can be limited to less than 1% with proper treatment (Cook 1996). Although cholera is endemic to the Bangladesh region, there have been numerous worldwide pandemics throughout history and the modern era (Said and Drasar 1996; Parry et al. 2007). Thus, it is a truly global concern. The ability to forecast cholera risk would be of great benefit, particularly in regions with limited or damaged infrastructure, where outbreaks are more likely to occur and medical resources are scarce. Accurate forecasts would allow public health workers and decision makers to prepare for potential epidemics in advance and make the most efficient use of scarce resources.

Cholera incidence in Bangladesh follows a distinct annual cycle, with cases reaching a maximum during the boreal spring and fall. The largest number of cases occur during boreal fall, after the withdrawal of the monsoon rains (Glass et al. 1982). The secondary maxima in the spring occurs prior to the arrival of the monsoon rains and appears to track the rapid increase in temperatures following the winter season (Bouma and Pascual 2001; Ruiz-Moreno et al. 2007). Rainfall and temperature are not the only environmental drivers of cholera risk; other factors, such as salinity and pH, are also thought to play a role.

Numerous studies suggest that it may be possible to forecast cholera risk based on environmental factors (e.g., Colwell 1996; Franco et al. 1997; Pascual et al. 2000; Pascual et al. 2002; Rodó et al. 2002; Koelle and Pascual 2004; Koelle et al. 2005b; Pascual et al. 2008). Of particular interest for cholera forecasting is the observed link between cholera incidence and the El Niño–Southern Oscillation (ENSO). Previous analyses (e.g., Pascual et al. 2000; Koelle et al. 2005b; Pascual et al. 2008) have found significant correlations between the fall (September–October) maximum in cholera incidence and ENSO events during the preceding winter [December–February (DJF)]. The physical significance of this relationship was explored in Cash et al. (2008a, hereafter CRK08a), a companion paper to the current work. Consistent with previous analyses, CRK08a found large regions of statistically significant correlation between September cholera incidence and winter SST that closely resemble the pattern of SST anomalies associated with a warm ENSO event. CRK08a also established that, in both models and observations, winter El Niño events give rise to enhanced precipitation over Bangladesh in the following summer [June–August (JJA)]. Warm SST anomalies in the central and eastern tropical Pacific warm the lower atmosphere throughout the tropics. This warming is accompanied by increased geopotential heights, which persist through the spring and into the summer months, altering the monsoon circulation so as to increase precipitation. This increased precipitation, in turn, leads to greater flooding, breakdowns in sanitation, and enhanced risk of cholera infection (Schwartz et al. 2006).

A physically plausible link between cholera and tropical Pacific SST allows for consideration of whether or not this link can be used to improve cholera forecasts. This question was explored in Pascual et al. (2008), which used a time series susceptible–infectious–recovered–susceptible model (TSIRS; a standard epidemiological model) fit to the observed record of cholera incidence to predict the fall peak in Bangladesh cholera incidence. Including the January Niño-3.4 index improved forecasts measurably in that study, even during periods in which no large ENSO events were recorded.

However, while including the Niño-3.4 index improves the skill of cholera risk forecasts, it is not clear that this is the most physically relevant environmental predictor available. This is true even if the discussion of predictors is limited to ENSO, because the strongest lagged correlations between cholera and SST are not found in the central and eastern tropical Pacific (Fig. 1a). Rather, the largest amplitudes are associated with the negative correlations with SST in the western Pacific “horseshoe” region. This is a region that is known to be associated with the eastern tropical Pacific through the “atmospheric bridge” (e.g., Klein et al. 1999), and CRK08a and other studies have found that models forced with observed eastern tropical Pacific SST alone produce anomalies in the western Pacific similar to those seen in the observations.

Fig. 1.

(a) Lagged rank correlation between September El Tor cholera cases and preceding January. Contours denote regions significant at the 90% level. Shading and interval is 0.1. (b) Incidence of Matlab El Tor cholera cases for 1976–2002 September.

Fig. 1.

(a) Lagged rank correlation between September El Tor cholera cases and preceding January. Contours denote regions significant at the 90% level. Shading and interval is 0.1. (b) Incidence of Matlab El Tor cholera cases for 1976–2002 September.

Because of the high correlations between cholera and western Pacific SST, and the fact that eastern and western Pacific SST anomalies typically coincide, the hypothesis that the western Pacific is the key link between cholera and SST cannot be rejected based on previous results. Cold SST in this region could plausibly alter the monsoon circulation and therefore enhance the chances of a cholera outbreak. Including indices of western Pacific SST in the Pascual et al. (2008) forecast model could potentially improve forecast accuracy beyond the inclusion of Niño-3.4 alone. Alternatively, the correlations between cholera and the western Pacific may simply be a consequence of the well-known anticorrelation between the eastern and western Pacific during ENSO events, rather than representing a physical link. It should also be noted that a region of significant correlation is found in the Bay of Bengal (Fig. 1a), and variations in local SST may also be playing a role in influencing the climate of Bangladesh. However, examination of the monthly, rather than seasonal, correlation between cholera and SST indicates that the significance of the Bay of Bengal center is less robust than the centers in the tropical Pacific, and this feature will not be considered further here.

In this study, we test the hypothesis that the western Pacific “horseshoe” region exerts a significant influence over summer rainfall in Bangladesh and hence cholera. We consider this influence from both probabilistic and deterministic perspectives and address the more general question of how SST anomalies, in the absence of strong variability in the eastern tropical Pacific, affect the regional climate of the Indian Ocean region. Our primary tool in this study is a pacemaker model, similar to the model used in CRK08a, in which the observed record of SST is prescribed for the western Pacific. The model and our analysis methodology are described in section 2. Results from the analysis are described in section 3, and a summary of our results and conclusions are presented in section 4.

2. Data and methodology

a. Observations

There are significant disagreements between the published research-quality rainfall products in the Bangladesh region (Cash et al. 2008c). As a result, defining a “ground truth” rainfall for Bangladesh is not a trivial problem, and results may depend, at least in part, on the choice of rainfall product. We use here the Chen et al. (2002) rainfall product, primarily because it includes multiple rain gauge observations within Bangladesh itself. It has also been shown to agree well with the densely observed rainfall over India reported by the Indian Meteorology Department (Rajeevan et al. 2005). We also make use of the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis product (Kalnay et al. 1996) for observations of the monsoon circulation.

The cholera data used in this study consist of monthly symptomatic cases obtained from a surveillance program in Matlab, Bangladesh, conducted by the International Centre for Diarrheal Disease Research (ICDDR, B). The surveillance site is located 40 km southeast of the capital of Dhaka, in the delta of the Brahmaputra and Ganges rivers. As in CRK08a, we focus here exclusively on September (the month of maximum incidence) cholera cases due to the El Tor biotype from 1976–2002 (see Fig. 1b). We focus on the El Tor biotype for two reasons: El Tor replaced the classical biotype in the mid-1970s as the dominant strain in the environment and is thus responsible for nearly all recorded cholera cases during this period (e.g., Koelle et al. 2005a). The different strains of cholera also exhibit different annual cycles and may be influenced by different aspects of climate variability and thus should be considered separately.

b. Pacemaker experiments

One of the primary tools used in this study is a pacemaker model in which the observed record of SST is prescribed in a limited portion of the ocean domain and represented by a simple ocean model elsewhere. The pacemaker methodology has been employed successfully in a number of previous studies (e.g., Alexander 1992a,b; Alexander et al. 2002; Bladé 1997, 1999; Lau and Nath 2000, 2003; Shinoda et al. 2004; Wu and Kirtman 2004a,b; CRK08a). An important feature of pacemaker models is that the observed SST record in the particular region of interest (the “pacemaker” of the simulation) is included, while a consistent surface energy budget is maintained elsewhere. This last point is of particular concern for simulations of the Indian monsoon, where inconsistencies in the Indian Ocean surface energy budget are known to degrade the quality of the simulation (e.g., Wu and Kirtman 2004a). Thus, the pacemaker methodology allows for a more direct comparison with observations while at the same time maintaining a balanced energy budget throughout most of the domain.

Our model consists of the Center for Ocean–Land–Atmosphere Studies (COLA) atmospheric general circulation model (AGCM), version 3.1 (Misra et al. 2006) with 28 vertical levels and T62 horizontal resolution coupled to a 50-m slab mixed-layer ocean model (thermodynamics only). The AGCM is an update of the COLA AGCM v2.2, which has been used extensively for modeling studies in the past (e.g., Schneider 2002; Kirtman et al. 2002; Kirtman 2003). As described above, the ocean domain is separated into two regions. We prescribe observed SST from the Hadley Centre sea ice and sea surface temperature dataset version 1 (HADISST1; Rayner et al. 2003) in the horseshoe region in the western tropical and central extratropical Pacific (hereafter the Westpac region; Fig. 2a). The prescribed region is selected to coincide with the area of significant negative correlations between SST and cholera (Fig. 1) and also encompasses the typical location of the western SST anomalies associated with ENSO events. The transition between the prescribed and mixed-layer domains is handled through a “blending region,” in which the total SST is calculated from the weighted average of the prescribed and predicted SST (see Fig. 2a for weighting values).

Fig. 2.

(a) Prescribed and blending regions used in pacemaker experiments. Shading denotes weighting of prescribed, observed SST from HadISST dataset. Weighting is set to 1 in tropical central and eastern Pacific and Polar region. (b) RMS error (°C) in SST field.

Fig. 2.

(a) Prescribed and blending regions used in pacemaker experiments. Shading denotes weighting of prescribed, observed SST from HadISST dataset. Weighting is set to 1 in tropical central and eastern Pacific and Polar region. (b) RMS error (°C) in SST field.

To prevent the model SST from drifting away from climatology, an implied ocean heat flux, or q flux, is also prescribed at all mixed-layer points (including those within the blending region). This q-flux field is calculated from a separate 20-yr integration using prescribed climatological SST in the pacemaker region and a 60 W m−2 K−1 relaxation toward climatology in the mixed-layer region. This term represents heat transport by the neglected ocean currents and reaches maxima coincident with the observed western boundary currents. A fixed annual cycle of monthly-mean restoring tendencies from the relaxation term is calculated for each grid point and introduced as an additional term in the pacemaker mixed-layer temperature tendency equation, along with a much weaker (10 W m−2 K−1) relaxation to climatology. Note that only this weaker relaxation term acts on the SST variability generated by the model.

The experiments presented here consist of an eight-member ensemble integrated from 1950–2002. Initial conditions for the individual ensemble members are taken from a separate model run, in which monthly SST from 1949 is prescribed in the pacemaker region for each year. Each member of the ensemble is then initiated from a different December taken from this run. Unless noted otherwise (see below), results presented are for the ensemble mean.

c. Intraevent and intraensemble comparisons

It is naturally of great interest whether we can identify a clear deterministic link between SST in the Westpac region and the ensemble mean Bangladesh rainfall in the model. Determining that a given change in SST leads to a given and repeatable change in Bangladesh rainfall and cholera risk would not only be interesting from an academic perspective but also of the greatest use in forecasting cholera risk. However, the ability to identify and quantify shifts in the probability distribution of cholera and rainfall events based on SST anomalies would also be of use in forecasting cholera risk. Hence, in addition to considering the deterministic links between ensemble mean SST anomalies and Bangladesh rainfall, we also explore the potential for SST anomalies to alter the probability and severity of a cholera outbreak.

We investigate the probabilistic links between SST and Bangladesh rainfall in part by exploiting a unique feature of the pacemaker methodology. In ensembles of pacemaker integrations, as opposed to fully prescribed or fully coupled integrations, each member of the ensemble has identical SST in the prescribed forcing region and freely evolving SST elsewhere. Thus, by construction, differences between pairs of ensemble members in a given field of interest (such as rainfall) cannot be directly related to differences in the forcing region, as each member of the ensemble possesses identical boundary conditions in that limited region. Instead, any differences in the field of interest must arise as a result of a combination of (i) chaotic atmospheric processes; (ii) slowly varying boundary conditions outside of the forcing region; and/or (iii) modulation of the influence of the prescribed forcing region by (i) and (ii).

This same methodology was used in Cash et al. (2008b, hereafter CRK08b) to examine non-ENSO and nondeterministic forcing of Bangladesh rainfall in the CRK08a model configuration (SST forcing in the central and eastern tropical Pacific). Comparing the interannual variations in rainfall response within individual ensemble members, we found the model JJA rainfall response was significantly related to the degree of persistence of warm SST anomalies in the central Pacific into the following spring [March–May (MAM)] and summer. We also found significant differences in the patterns associated with positive and negative rainfall anomalies, indicating a degree of nonlinearity in the response. Negative rainfall anomalies were more closely associated with SST anomalies in the eastern tropical Pacific and positive rainfall anomalies were more closely associated with the central Pacific. We also found that when the effect of variations in the ENSO region is removed by comparing pairs of ensemble members, variations in Indian Ocean SST are significantly associated with variations in Bangladesh rainfall. Hence, for a given ENSO event, the state of the Indian Ocean must also be taken into account when forecasting the response of Bangladesh rainfall.

The results of CRK08b indicate that treating the Niño-3 and Niño-4 regions separately, rather using the Niño-3.4 region alone as in Pascual et al. (2008), might improve our ability to predict cholera incidence and our understanding of the influence of ENSO on tropical climate variability in general. The effect of these variations in ENSO, as well as the presence of independent variations in the Indian Ocean, helps to motivate a similar analysis for the model presented here. Given that considering the central and eastern tropical Pacific individually may provide additional insight into tropical dynamics and forecasting skill, we cannot discount a priori a similar role for the western Pacific in driving variability in the regional climate of Bangladesh.

The model presented here does not include forcing in either the Indian Ocean or the central and eastern tropical Pacific. Different members of the ensemble will, therefore, have slightly different SST anomalies in these critical regions. We can thus examine the effect of SST anomalies in these regions on the simulation independent of the direct effect of western Pacific SST anomalies. Formally, let

 
formula

where Pi and SSTi denote precipitation and SST, respectively, in the ith member of the ensemble and −ac denotes the time-mean annual cycle (for convenience, the prime denoting the anomaly is dropped in the following discussion).

We determine whether or not certain characteristics occur independently between a given set of paired variables (e.g., Pi and SSTi) through a nonparametric paired characteristic test (Kraft and van Eden 1968). Here, we focus on the simple null hypothesis that the signs of the two quantities considered are unrelated. This test can be expressed through a table (see Table 1), where the cell values denote the number of months in which the listed conditions are satisfied. If the characteristics in question are independent, then the individual cell values will be approximately equal. The significance of any deviations from equality can be tested on a chi-squared distribution as

 
formula

(see Table 2 for the list of events). In the following discussion, this test will be used to address the two questions listed next.

Table 1.

Paired characteristics tables. Cells denote number of months with combination of characteristics. Adapted from Kraft and van Eden (1968).Rainfall or SST positiveRainfall or SST negativeBangladesh rainfall positiveN11N12ABangladesh rainfall negativeN21N22BnmN

Paired characteristics tables. Cells denote number of months with combination of characteristics. Adapted from Kraft and van Eden (1968).Rainfall or SST positiveRainfall or SST negativeBangladesh rainfall positiveN11N12ABangladesh rainfall negativeN21N22BnmN
Paired characteristics tables. Cells denote number of months with combination of characteristics. Adapted from Kraft and van Eden (1968).Rainfall or SST positiveRainfall or SST negativeBangladesh rainfall positiveN11N12ABangladesh rainfall negativeN21N22BnmN
Table 2.

El Niño and La Niña years used in this study. Events are based on DJF values of the NINO34 region index and are taken from the CPC (available online at http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml). Years listed below are for January and February; event definitions include December values from the previous year.

El Niño and La Niña years used in this study. Events are based on DJF values of the NINO34 region index and are taken from the CPC (available online at http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml). Years listed below are for January and February; event definitions include December values from the previous year.
El Niño and La Niña years used in this study. Events are based on DJF values of the NINO34 region index and are taken from the CPC (available online at http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml). Years listed below are for January and February; event definitions include December values from the previous year.
  1. For different El Niño events, what precipitation and SST patterns are associated with differences in Bangladesh rainfall?

    We first calculate anomalies in rainfall and SST between all events: 
    formula
    where m and n are labels for the 14 individual El Niño events in Table 2 and anomalies are calculated for all m < n. For example, P112 represents the difference in rainfall between the first and second events for the first member of the ensemble. By applying the paired characteristic test in Table 1 and Eq. (2) to (PBimn, Pimn) and to (PBimn, SSTimn), we can determine the precipitation and SST patterns, respectively, associated with events that produce enhanced (PBimn > 0) or reduced (PBimn < 0) Bangladesh rainfall relative to other El Niño events within the same ensemble member. Statistically significant associations as identified by (2) will strongly suggest that variations in rainfall are driven in part by interannual fluctuations in the boundary conditions. In particular, significant associations with SST anomalies in the Westpac forcing region will indicate a potential influence of the observed record of SST in this region on the climate of Bangladesh.
  2. For identical ENSO events, what precipitation and SST patterns are associated with differences in Bangladesh rainfall?

    For this portion of the analysis, we examine differences between ensemble members for the 29 ENSO events listed in Table 1, that is, 
    formula
    where PBi denotes the precipitation anomaly for the ith ensemble member averaged over the Bangladesh region, i and j range over the eight ensemble members, and k ranges over the 29 ENSO events (all events are considered together and the index k will hereafter be dropped for convenience). Pairs are calculated for all i < j to avoid duplication. As discussed previously, because differences are calculated for pairs of ensemble members, SSTij is zero by construction in the pacemaker region for all k. Thus Pij is independent of the direct influence of SST anomalies in the forcing region. If the differences in rainfall between the ensemble members are entirely due to chaotic atmospheric processes, then we expect to find no statistically significant associations between Pij and SSTij. In contrast, if the anomalies in Pij are driven by differences in SST outside of the forcing region, then we do expect to find statistically significant associations between Pij and SSTij.

3. Results

a. Mean monsoon circulation and rainfall

Considering the idealized nature of the experiments, namely, no ocean dynamics and prescribed SST only in the western Pacific, our first concern is with how accurately the simulation reflects the observations. We find that global mean SST is reasonably well simulated, with root-mean-square (rms) errors generally less than 1° K (Fig. 2b). Note that the rms errors are time dependent, with peaks typically corresponding to large ENSO events. Turning to the question of rainfall, we also find that the simulation of the JJA mean rainfall and monsoon circulation (Figs. 3a and 3c, respectively) to be reasonably accurate. The model produces rainfall maxima over the Western Ghats, Bangladesh, and the Bay of Bengal. Rainfall minima occur over the interior of the Indian subcontinent, consistent with the observed rainfall from the Chen data (Fig. 3b). The most notable error in the simulated rainfall is the westward displacement of the rainfall maximum near Bangladesh. This is presumably due in large part to the low-resolution model topography, which captures neither the abrupt changes in orography in eastern Bangladesh nor the transition to the Tibetan Plateau. The simulated mean monsoon circulation is likewise reasonably accurate, with maxima near the east coast of Africa and over the Arabian Sea (Fig. 3c). A well-organized anticyclonic circulation carries moisture over the west coast of India before turning northward across the Bay of Bengal and Bangladesh. The model westerly winds reach a maximum somewhat farther north than in the NCEP–NCAR reanalysis but are generally reasonably similar to the observations (Fig. 3d).

Fig. 3.

JJA climatology for (a) ensemble mean rainfall, (b) observed rainfall from Chen et al. (2002), (c) 850-mb ensemble mean circulation, and (d) 850-mb reanalysis circulation. Units are in millimeters per day for (a) and (b) and meters per second for (c) and (d).

Fig. 3.

JJA climatology for (a) ensemble mean rainfall, (b) observed rainfall from Chen et al. (2002), (c) 850-mb ensemble mean circulation, and (d) 850-mb reanalysis circulation. Units are in millimeters per day for (a) and (b) and meters per second for (c) and (d).

b. Deterministic influence of SST

Although it can be argued that the mean monsoon circulation in the model is a reasonable approximation of the observations, the composite observed and ensemble mean ENSO SST anomalies (defined as the difference in the mean El Niño and La Niña events) bear relatively little resemblance to each other (see Fig. 4). Note that we consider here only those events from 1976 onward (see Table 2 for the list of all events), which are the same events examined in CRK08a. In each season considered (Fig. 4; JJA, top; MAM, middle; DJF, bottom), the observed warm anomalies in the Indian Ocean and central/eastern tropical Pacific are not reproduced by the model. Instead, we find cold anomalies spread eastward from the pacemaker region, particularly near 20°N, and that these anomalies have significant amplitude in all seasons considered. This tropics-wide reduction in SST has clear implications for the atmospheric circulation during these seasons.

Fig. 4.

Composite model ensemble mean warm–cold (a) JJA, (c) MAM, and (e) DJF SST anomalies and composite observed (b) JJA, (d) MAM, and (f) DJF ENSO SST anomalies from HADSST. Warm and cold events are based on Climate Prediction Center (CPC) DJF Niño-3.4 index. Shading and contour intervals are 0.1°C.

Fig. 4.

Composite model ensemble mean warm–cold (a) JJA, (c) MAM, and (e) DJF SST anomalies and composite observed (b) JJA, (d) MAM, and (f) DJF ENSO SST anomalies from HADSST. Warm and cold events are based on Climate Prediction Center (CPC) DJF Niño-3.4 index. Shading and contour intervals are 0.1°C.

In contrast to the results presented here, CRK08a found that the pacemaker model reproduces most of the observed global ENSO signal when forced with the observed record of SST in the central and eastern tropical Pacific. This inability to reproduce ENSO-related SST anomalies outside of the forcing region is consistent with the correspondence between peaks in rms error (Fig. 2b) and the ENSO events noted above. Thus, as we might expect, warming in the central and eastern tropical Pacific cannot be considered a response to changes in the Westpac region in our model.

Given that the model tends to reproduce the observed SST anomalies only where they are prescribed, the interannual variations of the simulations presented here are significantly further removed from the behavior of the climate system than those in previous pacemaker studies (e.g., Lau and Nath 2003; CRK08a). While this is clearly a deficiency of the simulation, it is not one without benefits. As noted in the section 1, the primary motivation of this study is to quantify the contribution of the Westpac region to variability in the Bangladesh region and to assess its physical relevance as a predictor of cholera risk. The absence of large anomalies in the central and eastern tropical Pacific is beneficial in this case, as it allows us to consider the influence of the western Pacific in isolation. However, any inferences drawn about the observed climate system must be made cautiously as a result. The lack of a strong response outside of the forcing region can also be taken as evidence against a strong role for the western Pacific.

Following CRK08a, we first consider the question of whether ENSO anomalies in the Westpac region significantly effect rainfall in the Bangladesh region. Following cold DJF SST anomalies in the Westpac region, the JJA ensemble mean precipitation (Fig. 5) decreases in the region around Bangladesh. The negative rainfall anomaly is collocated with the model summer rainfall maxima west of Bangladesh (Fig. 3a), indicating a reduction in the overall strength of the monsoon rains. Rainfall also increases slightly over the Bay of Bengal, suggesting a southward shift in precipitation.

Fig. 5.

Model ensemble mean precipitation difference for warm–cold ENSO events for JJA. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading interval is 0.5 mm day−1.

Fig. 5.

Model ensemble mean precipitation difference for warm–cold ENSO events for JJA. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading interval is 0.5 mm day−1.

As a check on the significance of the signal in the ensemble mean, we examine the composite JJA rainfall anomalies in the individual ensemble members (Fig. 6) Negative anomalies appear west of Bangladesh in most members of the ensemble, albeit with variations in location and intensity, indicating that this is a relatively robust response of the model to the forcing in the western Pacific. It is worth noting, however, that the anomalies are less consistent across ensemble members than the corresponding anomalies in CRK08a (see their Fig. 5). The influence of the western Pacific alone on the Bangladesh region is thus not negligible, but it is weaker than that of the central and eastern tropical Pacific.

Fig. 6.

JJA mean precipitation difference for warm–cold ENSO events for ensemble members A–H. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading interval is 1 mm day−1.

Fig. 6.

JJA mean precipitation difference for warm–cold ENSO events for ensemble members A–H. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading interval is 1 mm day−1.

In addition to being weaker in magnitude, the results of this simulation are opposite in sign to those of CRK08a. There we found positive JJA rainfall anomalies typically occur over Bangladesh following DJF El Niño events in both the central and eastern Pacific-forced pacemaker model and the observations. The reduction in monsoon rainfall described here thus warrants further investigation. Do these negative anomalies represent the independent influence of the Westpac region? Alternatively, are they simply a result of the lack of coincident warm anomalies in the central and eastern tropical Pacific in the current study? If the rainfall anomalies in Figs. 5 and 6 reflect the independent influence of the Westpac region, then it follows that the positive anomalies in CRK08a and the observations would be stronger in the presence of weaker negative anomalies in the Westpac region. This relationship, in turn, would have important implications for using the state of the tropical Pacific to predict rainfall and cholera incidence in Bangladesh. Specifically, forecasts of cholera risk might be improved by using separate indices for the eastern, central, and western Pacific regions.

We address the physical basis for an independent link between the Westpac region and Bangladesh rainfall by exploring the dynamical origins of the rainfall anomalies in detail. The changes in ensemble mean precipitation (Fig. 5) correspond to changes in the divergence of the vertically integrated moisture transport (VIMT; Fig. 7a), defined as the vertical average of −∇ · (uq, υq) from 1000 to 500 hPa, where u, υ, and q are the zonal wind, meridional wind, and specific humidity, respectively. Similarly, the weaker positive rainfall anomalies across the Bay of Bengal and India correspond to anomalous VIMT convergence.

Fig. 7.

JJA model ensemble mean vertically integrated (1000–500 hPa) (a) moisture transport convergence and (b) circulation difference for warm–cold ENSO events. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading in (a) denotes magnitude and the shading interval is 10−9 kg m−1 s−1. Shading in (b) denotes convergence of the wind field and the shading interval is 10−7 m−1 s−1.

Fig. 7.

JJA model ensemble mean vertically integrated (1000–500 hPa) (a) moisture transport convergence and (b) circulation difference for warm–cold ENSO events. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading in (a) denotes magnitude and the shading interval is 10−9 kg m−1 s−1. Shading in (b) denotes convergence of the wind field and the shading interval is 10−7 m−1 s−1.

The JJA ensemble mean precipitation anomalies are thus linked to changes in the large-scale transport of moisture. These anomalies in moisture transport, in turn, follow from a reduction in the strength of the overall monsoon circulation (Fig. 7b). Anomalous cyclonic circulation over the northwest Bay of Bengal leads to divergence (shading in Fig. 7b), VIMT divergence (Fig. 7a), and reduced rainfall west of Bangladesh (Fig. 5). The close correspondence between the divergence of the anomalous circulation and VIMT strongly suggests that the anomalous rainfall is being driven by changes to the monsoon dynamics. We also find the JJA composite humidity anomaly field shows little correspondence to the VIMT anomalies (not shown), further indicating that changes in the monsoon dynamics are responsible for the precipitation anomalies.

The circulation anomalies described above are part of the tropics-wide response to the prescribed SST anomalies in the western Pacific (Fig. 8). The easterly anomalies over India and Bangladesh in JJA (Fig. 7b) are due to a region of lowered geopotential heights that stretches from northern India and Bangladesh across Indonesia and into the Southern Hemisphere (Fig. 8a). The negative anomalies in the lower troposphere in the Indian Ocean region during the summer months are a continuation of a pattern established during the winter months (Fig. 8c), which persists through spring (Fig. 8b). It is interesting to note that, in contrast to the lack of a strong remote response in the tropical SST field, cold anomalies in the western Pacific lead to a tropics-wide decrease in lower tropospheric geopotential heights throughout DJF and MAM. This is similar to the response to Pacific SST anomalies in CRK08a (see their Fig. 11), in which warm anomalies in the central and eastern tropical Pacific produce increased lower troposphere heights throughout the tropics. It is not clear why the two experiments produce similar responses in the tropical atmosphere and not in tropical SSTs, although it should be noted that the CRK08a height response is approximately twice the magnitude of the response presented here.

Fig. 8.

Model ensemble mean vertically averaged (1000–500 hPa) geopotential height difference for warm–cold ENSO events for (a) JJA, (b) MAM, and (c) DJF. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading and contour interval is 1 m.

Fig. 8.

Model ensemble mean vertically averaged (1000–500 hPa) geopotential height difference for warm–cold ENSO events for (a) JJA, (b) MAM, and (c) DJF. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading and contour interval is 1 m.

Combining these results with those of CRK08a, we arrive at a more complete picture of the deterministic signal from Pacific SST in Bangladesh rainfall. When the central and eastern tropical Pacific warms and the western Pacific cools, as in the observations and the CRK08a model, large-scale organized convection and associated convergence zones shift to the eastern Pacific (Fig. 9a). This leads to anomalous subsidence and positive height anomalies over Indonesia and the Indian Ocean region and, in turn, to increased convergence over Bangladesh and enhanced rainfall. The same changes in circulation act to decrease rainfall over India, resulting in an out-of-phase relationship between precipitation in India and Bangladesh.

Fig. 9.

Model ensemble mean vertically averaged (1000–500 hPa) convergence difference for warm–cold ENSO events for (a) eastern Pacific–forced model (Cash et al. 2008a) and (b) Westpac-forced model. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading interval is 10−6 s−1. Streamlines denote vertically averaged winds.

Fig. 9.

Model ensemble mean vertically averaged (1000–500 hPa) convergence difference for warm–cold ENSO events for (a) eastern Pacific–forced model (Cash et al. 2008a) and (b) Westpac-forced model. Warm and cold events are based on CPC DJF Niño-3.4 index. Shading interval is 10−6 s−1. Streamlines denote vertically averaged winds.

In the model configuration presented here, there are no significant SST anomalies in the central and eastern tropical Pacific. Cold SST anomalies in the western tropical Pacific still suppress convection and convergence as described above, but the convection does not shift to the eastern Pacific. Instead, convergence is slightly enhanced over Indonesia and the eastern Indian Ocean (Fig. 9b), in a pattern more reminiscent of a typical La Niña year. This leads to enhanced convergence and lowered heights across Indonesia and the Indian Ocean, which appear to interact with the general monsoon circulation to produce anomalous easterlies and divergence over Bangladesh, suppressing precipitation. The potential effect of convergence anomalies in the Indonesian region on the Asian Summer Monsoon has also been demonstrated in a linear model, with prescribed SST anomalies in the western tropical Pacific (Annamalai and Sperber 2005).

c. Probabilistic influence of SST

In the previous section, we demonstrate that a forced, deterministic signal from SST anomalies in the Westpac region is detectable in Bangladesh summer rainfall in our model (Fig. 5). However, the composite ENSO rainfall anomaly is not negative for every member of the ensemble (Fig. 6), nor does it follow every DJF El Niño event (not shown). If indices of Westpac SST are to be used as environmental predictors of cholera risk, similar to the use of NINO34 in Pascual et al. (2008), then it is important that we identify and understand the reasons for these variations between events. One possible explanation is simply that the influence of the Westpac region is relatively weak, and that the deterministic signal is not well distinguished from chaotic fluctuations in the monsoon. Alternatively, variations in SST, both between El Niño events and between ensemble members outside of the forcing region, may also be playing an important role.

We first consider whether fluctuations in Bangladesh rainfall following different El Niño events are dominated by local, small-scale fluctuations in rainfall intensity, or if they are a part of larger differences in the monsoon system. By applying the paired characteristics test to the relationship between rainfall in Bangladesh and in the surrounding region (PBimn,Pimn) to the 14 El Niño events in Table 1, we find that differences in the response of Bangladesh rainfall to different El Niño events do not simply represent local fluctuations in rainfall magnitude (Fig. 10). Instead, they are part of a larger pattern rainfall differences encompassing much of the Indian Ocean region. An out-of-phase relationship between Bangladesh and India is apparent for both increased (Fig. 10a) and decreased (Fig. 10b) Bangladesh rainfall, although the association is significantly stronger in the case of decreased rainfall. There are clear changes in the southern Indian Ocean and the western tropical Pacific, suggestive of a north–south shift in precipitation between 10°S and 10°N. These patterns also closely resemble those found in CRK08b (see their Fig. 5), indicating that this is a robust mode of variability in the COLA AGCM. Note that the difference patterns shown here are relatively insensitive to whether the full period (1950–2002) or recent period (1976–2002) is used (not shown). All results discussed here are hence taken for the full record, both for the sake of robustness and to facilitate comparison to CRK08b.

Fig. 10.

Rainfall patterns (PBimn, Pimn) associated with (a) enhanced (N11 − N12) and (b) reduced (N21 − N22) Bangladesh rainfall following winter El Niño events. See Table 1 and section 2c for definitions of terms. Shading and contour intervals are 50 months.

Fig. 10.

Rainfall patterns (PBimn, Pimn) associated with (a) enhanced (N11 − N12) and (b) reduced (N21 − N22) Bangladesh rainfall following winter El Niño events. See Table 1 and section 2c for definitions of terms. Shading and contour intervals are 50 months.

Having identified large-scale changes in precipitation associated with variations in Bangladesh rainfall, we now ask whether these variations are associated with differences in SST, or if they simply represent chaotic fluctuations of the monsoon system. Applying the paired characteristics test to Bangladesh rainfall and SST (PBimn,SSTimn) for the 14 El Niño events in Table 1, we find some evidence (Fig. 11a) that enhanced JJA Bangladesh rainfall is associated with colder SST in the Indian Ocean and the western Pacific near Indonesia. There are also hints of an association with warmer SST in the central Pacific, although the region of positive associations is fragmentary and may not be significant. There is little to indicate a significant relationship with SST differences during MAM (Fig. 11c). In both seasons, the associations are weaker than for the corresponding case in CRK08a and of questionable significance. However, when we consider the associations between decreased Bangladesh rainfall and JJA SST, a very different picture emerges (Fig. 11b). Reduced rainfall is strongly associated with colder SST in the forcing region, consistent with the analysis of the ensemble mean in the previous section. Reduced rainfall is also clearly associated with warmer SST in the Indian Ocean. Associations with SST in the spring are generally weaker (Fig. 11d), although there is still a strong association with colder SST in the forcing region.

Fig. 11.

JJA SST patterns (PBimn, SSTimn) associated with (a) enhanced (N11 − N12) and (b) reduced (N21 − N22) Bangladesh rainfall following winter El Niño events. MAM SST patterns (PBimn, SSTimn) associated with (c) enhanced (N11 − N12) and (d) reduced (N21 − N22) Bangladesh rainfall following winter El Niño events for (c),(d). See Table 1 and section 2c for definitions of terms. Shading and contour intervals are 50 months.

Fig. 11.

JJA SST patterns (PBimn, SSTimn) associated with (a) enhanced (N11 − N12) and (b) reduced (N21 − N22) Bangladesh rainfall following winter El Niño events. MAM SST patterns (PBimn, SSTimn) associated with (c) enhanced (N11 − N12) and (d) reduced (N21 − N22) Bangladesh rainfall following winter El Niño events for (c),(d). See Table 1 and section 2c for definitions of terms. Shading and contour intervals are 50 months.

Taken together, these results indicate a degree of nonlinearity in the response of Bangladesh rainfall to SST in the Westpac forcing region and the Indian Ocean. Although enhanced Bangladesh rainfall shows some links to SST, the associations are weak and it is not clear that the hypothesis that these are due to random atmospheric fluctuations can be rejected. In contrast, decreased rainfall over Bangladesh appears to be closely tied to the strength of the cold anomaly associated with the El Niño event in the western Pacific.

While the associations between rainfall and JJA SST are notably nonlinear for different El Niño events, when we consider differences between identical ENSO events (PBij,SSTij), we find the associations are mostly linear (Fig. 12). Enhanced JJA Bangladesh rainfall is strongly associated with colder JJA SST anomalies in the Bay of Bengal and in a broad region of the Indian Ocean (Fig. 12a). Enhanced Bangladesh rainfall is also associated with warm southwest Indian Ocean SST anomalies and with colder SST anomalies south of Madagascar. These regions have also been identified as playing an important role in Bangladesh rainfall in other models (CRK08b) and in observations (e.g., Terray et al. 2003, 2005) and clearly merit further investigation. We also find an association between enhanced Bangladesh JJA rainfall and colder SST in the tropical South Atlantic (TSA) near the equatorial African coast. This region has also been identified as being negatively associated with Bangladesh rainfall in other studies (Kucharski et al. 2007) using a different model and different analysis techniques, further supporting their hypothesis that the TSA influences the monsoon circulation.

Fig. 12.

Same as Fig. 11 but for ENSO events.

Fig. 12.

Same as Fig. 11 but for ENSO events.

The associations between decreased Bangladesh JJA rainfall and JJA SST (Fig. 12b) are nearly opposite to those described above, although associations with SST in the eastern Pacific Ocean are, again, stronger in the case of decreased rainfall. For both enhanced and decreased Bangladesh rainfall, the strength of the association between rainfall and SST drops off rapidly with lag. In the case of enhanced Bangladesh rainfall (Fig. 12c), there are some indications of a significant association with MAM SST in the Indian Ocean and the northern tropical Pacific, but the associations are far weaker than in JJA. We once again find signs of nonlinearity in the associations, as reduced Bangladesh rainfall is strongly associated with the southeast Indian Ocean and northern central tropical Pacific.

4. Summary and conclusions

In this work, we further develop the relationship between the regional climate of Bangladesh and tropical Pacific SST to include the effect of the western Pacific “horseshoe” region. We find that forcing a pacemaker model with observed SST anomalies in the western Pacific (Westpac) region produces a relatively weak global SST response that bears relatively little resemblance to the observations. This is in contrast to the results of CRK08a, which found that forcing with observed SST anomalies in the central and eastern tropical Pacific reproduced much of the observed global ENSO signal. From these results we conclude that, in this model, the global SST response to ENSO is driven primarily by SST anomalies in the central and eastern tropical Pacific. Anomalies in the western Pacific appear to play a more passive role and arise primarily as part of the global response. Summer monsoon rains are suppressed over Bangladesh and central India and are slightly enhanced over the Bay of Bengal following winter El Niño events. These changes in precipitation are driven by changes in the overall monsoon circulation, with enhanced easterlies over Bangladesh leading to anomalous divergence in the vertical moisture transport field. The anomalous circulation pattern associated with these anomalies encompasses parts of Eurasia and Indonesia as well as the Indian Ocean and is a continuation of a pattern established during the winter months.

The circulation changes following El Niño winters in the model presented here are opposite in sign and weaker in magnitude than those found in the central and tropical Pacific-forced model presented in CRK08a. The fact that the SST anomalies in the western Pacific are similar in extent and magnitude in both models serves to highlight the importance of the central and eastern tropical Pacific in determining the overall response of the model. In both models, cold SST anomalies in the western Pacific act to suppress convection locally. In the Westpac-forced model, the absence of warm anomalies in the central and eastern tropical Pacific leads convergence to shifts to Indonesia and the eastern Indian Ocean. However, when the cold Westpac anomalies occur simultaneously with warming in the central and eastern tropical Pacific, as we find in CRK08a and the observations, the region of maximum convection follows the warm SST anomalies into the eastern tropical Pacific.

The probabilistic analysis of the influence of the Westpac region on the Bangladesh region also points to a relatively limited effect of the western Pacific on Bangladesh rainfall. Consistent with the deterministic analysis, we find the associations between Bangladesh rainfall and the western Pacific are generally weak and of minimal significance. The case in which decreased rainfall follows a winter El Niño event (Figs. 12b and 12d) is a notable exception to this conclusion, and in this case it appears that the degree to which cold SST anomalies persist into the summer months does effect Bangladesh rainfall.

However, while neither the probabilistic nor deterministic analyses support a strong role for the western Pacific, the state of the Indian Ocean during the summer months does appear to be closely associated with the regional climate of Bangladesh, independent of ENSO. Bangladesh rainfall is strongly associated with SST across a broad swath of the Indian Ocean during the summer months, with colder SST promoting increased rainfall. We also find this relationship when the model is forced with SST in the central and eastern tropical Pacific (CRK08b), consistent with the hypothesis that the Indian Ocean exerts an independent influence on Bangladesh rainfall.

As stated in the section 1, our goal in investigating the influence of the Westpac region on the regional climate of Bangladesh is to understand how, if at all, this region affects the interannual variability of cholera. Given that even the sign of both the tropics-wide response and, more specifically, the monsoon rainfall response over Bangladesh to Westpac SST anomalies is determined by the presence or absence of SST anomalies in the central and eastern tropical Pacific, it seems unlikely that western Pacific SST anomalies play a dominant role in determining the response of cholera to climate. While the influence of the Westpac region on Bangladesh and India is not negligible, it is clearly secondary to the influence of the central and eastern tropical Pacific.

We thus reject the hypothesis posed in section 1 regarding the independent influence of the western Pacific region on cholera. The inclusion of indices of western Pacific SST in the Pascual et al. (2008) model appears unlikely to improve cholera forecasting in a meaningful fashion. However, the state of the Indian Ocean, particularly during the summer months, does appear to represent an additional source of potential predictability for the regional climate of Bangladesh and warrants further exploration from the perspective of both climate and cholera prediction.

Acknowledgments

Support from NSF Grant EF-0429520 and NOAA Grant NA04OAR4600194 is gratefully acknowledged. XR was also in receipt of funds from the Spanish MEC project PANDORACGL-63053. We wish to express our thanks to Md. Yunus and the ICDDR, B for supplying the data on cholera incidence and also to three reviewers, whose comments substantially improved the initial manuscript.

REFERENCES

REFERENCES
Alexander
,
M. A.
,
1992a
:
Midlatitude atmosphere–ocean interaction during El Niño. Part I: The North Pacific Ocean.
J. Climate
,
5
,
944
958
.
Alexander
,
M. A.
,
1992b
:
Midlatitude atmosphere–ocean interaction during El Niño. Part II: The Northern Hemisphere atmosphere.
J. Climate
,
5
,
959
972
.
Alexander
,
M. A.
,
I.
Bladé
,
M.
Newman
,
J. R.
Lanzante
,
N-C.
Lau
, and
J. D.
Scott
,
2002
:
The atmospheric bridge: The influence of ENSO teleconnections on air–sea interaction over the global oceans.
J. Climate
,
15
,
2205
2231
.
Annamalai
,
H.
, and
K.
Sperber
,
2005
:
Regional heat sources and the active and break phases of boreal summer intraseasonal (30–50 day) variability.
J. Atmos. Sci.
,
62
,
2726
2748
.
Bladé
,
I.
,
1997
:
The influence of midlatitude ocean–atmosphere coupling on low-frequency variability of a GCM. Part I: No tropical SST forcing.
J. Climate
,
10
,
2087
2106
.
Bladé
,
I.
,
1999
:
The influence of midlatitude ocean–atmosphere coupling on low-frequency variability of a GCM. Part II: Interannual variability induced by tropical SST forcing.
J. Climate
,
12
,
21
45
.
Bouma
,
M. J.
, and
M.
Pascual
,
2001
:
Seasonal and interannual cycles of endemic cholera in Bengal 1891–1940 in relation to climate and geography.
Hydrobiologia
,
460
,
147
156
.
Cash
,
B. A.
,
X.
Rodó
, and
J. L.
Kinter
III
,
2008a
:
Links between tropical Pacific SST and cholera incidence in Bangladesh: Role of the eastern and central tropical Pacific.
J. Climate
,
21
,
4647
4663
.
Cash
,
B. A.
,
X.
Rodó
, and
J. L.
Kinter
III
,
2008b
:
Non-ENSO variability and the regional climate of Bangladesh: Implications for cholera risk.
COLA Tech. Rep. 260, Center for Ocean–Land–Atmosphere Studies, 39 pp
.
Cash
,
B. A.
,
X.
Rodó
,
J. L.
Kinter
III
,
M. J.
Fennessy
, and
B.
Doty
,
2008c
:
Differing estimates of observed Bangladesh rainfall.
J. Hydrometeor.
,
9
,
1106
1114
.
Chen
,
M.
,
P.
Xie
, and
J. E.
Janowiak
,
2002
:
Global land precipitation: A 50-yr monthly analysis based on gauge observations.
J. Hydrometeor.
,
3
,
249
266
.
Colwell
,
R. R.
,
1996
:
Global climate and infectious disease: The cholera paradigm.
Science
,
274
,
2025
2031
.
Cook
,
G. C.
,
1996
:
Management of cholera: The vital role of rehydration.
Cholera and the Ecology of Vibrio Cholerae, B. S. Drasar and B. D. Forrest, Eds., Chapman and Hall, 54–94
.
Franco
,
A. A.
, and
Coauthors
,
1997
:
Cholera in Lima, Peru, correlates with prior isolation of vibrio cholerae from the environment.
Amer. J. Epidemiol.
,
146
,
1067
1075
.
Glass
,
R. I.
,
S.
Becker
,
M. I.
Huq
,
B. J.
Stoll
,
M. U.
Khan
,
M. H.
Merson
,
J. V.
Lee
, and
R. E.
Black
,
1982
:
Endemic cholera in rural Bangladesh, 1966–1980.
Amer. J. Epidemiol.
,
116
,
959
970
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc.
,
77
,
437
470
.
Kirtman
,
B. P.
,
2003
:
The COLA anomaly coupled model: Ensemble ENSO prediction.
Mon. Wea. Rev.
,
131
,
2324
2341
.
Kirtman
,
B. P.
,
Y.
Fan
, and
E. K.
Schneider
,
2002
:
The COLA global coupled and anomaly coupled ocean–atmosphere GCM.
J. Climate
,
15
,
2301
2320
.
Klein
,
S. A.
,
B. J.
Soden
, and
N-C.
Lau
,
1999
:
Remote sea surface temperature variations during ENSO: Evidence for a tropical atmospheric bridge.
J. Climate
,
12
,
917
932
.
Koelle
,
K.
, and
M.
Pascual
,
2004
:
Disentangling extrinsic from intrinsic factors in disease dynamics: A nonlinear time series approach with an application to cholera.
Amer. Nat.
,
163
,
901
913
.
Koelle
,
K.
,
M.
Pascual
, and
Md
Yunus
,
2005a
:
Pathogen adaptation to seasonal forcing and climate change.
Proc. Roy. Soc. London
,
B272
,
971
977
.
Koelle
,
K.
,
X.
Rodó
,
M.
Pascual
,
Md
Yunus
, and
G.
Mostafa
,
2005b
:
Refractory periods and climate forcing in cholera dynamics.
Nature
,
436
,
696
700
.
Kraft
,
C. H.
, and
C.
van Eden
,
1968
:
A Nonparametric Introduction to Statistics.
Macmillan, 342 pp
.
Kucharski
,
F.
,
A.
Bracco
,
J. H.
Yoo
, and
F.
Molteni
,
2007
:
Low-frequency variability of the Indian Monsoon–ENSO relationship and the tropical Atlantic: The “weakening” of the 1980s and 1990s.
J. Climate
,
20
,
4255
4266
.
Lau
,
N-C.
, and
M. J.
Nath
,
2000
:
Impact of ENSO on the variability of the Asian–Australian monsoons as simulated in GCM experiments.
J. Climate
,
13
,
4287
4309
.
Lau
,
N-C.
, and
M. J.
Nath
,
2003
:
Atmosphere–ocean variations in the Indo-Pacific sector during ENSO Episodes.
J. Climate
,
16
,
3
20
.
Misra
,
V.
, and
Coauthors
,
2006
:
Validating ENSO simulation in coupled climate models.
COLA Tech. Rep. 210, Center for Ocean–Land–Atmosphere Studies, 46 pp
.
Parry
,
M.
,
O.
Canziani
,
J.
Palutikof
,
P.
van der Linden
, and
C.
Hanson
,
2007
:
Climate Change 2007: Impacts, Adaptation and Vulnerability.
Cambridge University Press, 976 pp
.
Pascual
,
M.
,
X.
Rodó
,
S. P.
Ellner
,
R.
Colwell
, and
M. J.
Bouma
,
2000
:
Cholera dynamics and El Niño–Southern Oscillation.
Science
,
8
,
1766
1769
.
Pascual
,
M.
,
M.
Bouma
, and
A. P.
Dobson
,
2002
:
Cholera and climate: Revisiting the quantitative evidence.
Microbes Infect.
,
4
,
237
245
.
Pascual
,
M.
,
L. F.
Chaves
,
B.
Cash
,
X.
Rodó
, and
Md
Yunus
,
2008
:
Predictability of endemic cholera: The role of climate variability and disease dynamics.
Climate Res.
,
36
,
131
140
.
Rajeevan
,
M.
,
J.
Bhate
,
J. D.
Kale
, and
B.
Lai
,
2005
:
Development of a High Resolution Daily Gridded Rainfall Data for the Indian Region. Meteor. Monogr. Climatol.
No. 22/2005, National Climate Centre, India Meteorological Department, 27 pp
.
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
:
Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century.
J. Geophys. Res.
,
108
,
4407
.
doi:10.1029/2002JD002670
.
Rodó
,
X.
,
M.
Pascual
,
G.
Fuchs
, and
A. S. G.
Faruque
,
2002
:
ENSO and cholera: A nonstationary link related to climate change?
Proc. Natl. Acad. Sci. USA
,
99
,
12901
12906
.
Ruiz-Moreno
,
D.
,
M.
Pascual
,
M.
Bouma
,
A.
Dobson
, and
B.
Cash
,
2007
:
Cholera seasonality in Madras (1901–1940): Dual role for rainfall in endemic and epidemic regions.
EcoHealth
,
4
,
52
62
.
Said
,
B.
, and
B.
Drasar
,
1996
:
Vibrio Cholerae.
Cholera and the Ecology of Vibrio Cholerae, B. S. Drasar and B. D. Forrest, Eds., Chapman and Hall, 1–17
.
Schneider
,
E. K.
,
2002
:
Understanding differences between the equatorial Pacific as simulated by two coupled GCMs.
J. Climate
,
15
,
449
469
.
Schwartz
,
B.
, and
Coauthors
,
2006
:
Diarrheal epidemics in Dhaka, Bangladesh, during three consecutive floods: 1988, 1998, and 2004.
Amer. J. Trop. Med. Hyg.
,
74
,
1067
1073
.
Shinoda
,
T.
,
M. A.
Alexander
, and
H. H.
Hendon
,
2004
:
Remote response of the Indian Ocean to interannual SST variations in the tropical Pacific.
J. Climate
,
17
,
362
372
.
Terray
,
P.
,
P.
Delecluse
,
S.
Labattu
, and
L.
Terray
,
2003
:
Sea surface temperature associations with the late Indian summer monsoon.
Climate Dyn.
,
21
,
593
618
.
Terray
,
P.
,
S.
Dominiak
, and
P.
Delecluse
,
2005
:
Role of the southern Indian Ocean in the transitions of the monsoon–ENSO system during recent decades.
Climate Dyn.
,
24
,
169
195
.
Wu
,
R.
, and
B. P.
Kirtman
,
2004a
:
Impacts of the Indian Ocean on the Indian Summer Monsoon–ENSO relationship.
J. Climate
,
17
,
3037
3054
.
Wu
,
R.
, and
B. P.
Kirtman
,
2004b
:
Understanding the impacts of the Indian Ocean on ENSO variability in a coupled GCM.
J. Climate
,
17
,
4019
4031
.

Footnotes

Corresponding author address: Benjamin A. Cash, Center for Ocean–Land–Atmosphere Studies, Institute of Global Environment and Society, 4041 Powder Mill Rd., Suite 302, Calverton, MD 20705. Email: bcash@cola.iges.org