TMCnet News
Spatiotemporal Variability and Predictability of Relative Humidity over West African Monsoon Region [Journal of Climate](Journal of Climate Via Acquire Media NewsEdge) ABSTRACT Spatial and temporal variability of relative humidity over the West African monsoon (WAM) region is investigated. In particular, the variability during the onset and retreat periods of the monsoon is considered. A K-means cluster analysis was performed to identify spatially coherent regions of relative humidity variability during the two periods. The cluster average of the relative humidity provides a robust representative index of the strength and timing of the transition periods between the dry and wet periods. Correlating the cluster indices with large-scale circulation and sea surface temperatures indicates that the land-ocean temperature gradient and the corresponding circulation, tropical Atlantic sea surface temperatures (SSTs), and to a somewhat lesser extent tropical Pacific SSTs all play a role in modulating the timing of the monsoon season relative humidity onset and retreat. These connections to large-scale climate features were also found to be persistent over interseasonal time scales, and thus best linear predictive models were developed to enable skillful forecasts of relative humidity during the two periods at 15-75-day lead times. The public health risks due to meningitis epidemics are of grave concern to the population in this region, and these risks are strongly tied to regional humidity levels. Because of this linkage, the understanding and predictability of relative humidity variability is of use in meningitis epidemic risk mitigation, which motivated this research. (ProQuest: ... denotes formulae omitted.) 1. Background and motivation This project focuses on the interaction between climate and meningitis disease incidence in eight countries in West Africa: Ghana, Togo, Benin, Nigeria, Chad, Niger, Mali, and Burkina Faso. All or parts of these countries lie within the African ''meningitis belt,'' and their seasonal weather patterns are controlled by the West African monsoon (WAM). The spring transition from dry to monsoon season marks the end of the meningitis season, and the fall transition from monsoon to dry season can influence the start of the preceding meningitis season. Limited health-care networks exist in this region, with international organizations providing logistic and material support for the prevention and treatment of meningococcal meningitis (hereafter referred to as meningitis). This includes the allocation of vaccines to subnational districts given the incidence of meningitis (from the International Coordinating Group on Vaccine Provision: http://www.who.int/csr/disease/ meningococcal/icg/en/). Understanding of the interseasonal variability of the WAM system and its associated dry season could provide more informed decision support in allocating health-care supplies given identified links between meningococcal meningitis epidemics and climate (Seefeldt et al. 2012). TheWAMis a dominant low-level southwesterly flow affecting Subsaharan Africa temperature and precipitation patterns. This seasonal flow advects moisture from the Gulf of Guinea and equatorial Atlantic onshore during the boreal summer in sharp contrast to the dry northeasterly Harmattan winds that exist throughout the rest of the year. Monsoon behavior is linked to the seasonal latitudinal migration of the tropical rain belt (Nicholson 2009) and the intertropical front (ITF), with the latter representing the interface between monsoon winds and Harmattan winds. Lavaysse et al. (2009) identify several dynamical elements of the West African monsoon that represent and influence its behavior including the West African heat low (WAHL), the African easterly jet (AEJ), the tropical easterly jet (TEJ), and African easterly waves (AEWs). It is the pressure gradient between theWAHL and the South Atlantic anticyclone drives the southwesterly monsoon winds. At the same time, the strength and position of the WAHL is dependent on preferential heating controlled by surface albedo and solar heating conditions (Ramel et al. 2006). In addition, work by Drobinski et al. (2005) suggests that the orography of North Africa, the Hoggar massif and Atlas Mountains in particular, aid in the deepening of the WAHL in late spring and describe its location centered over the Sahara Desert. Subsidence to the north of the mountains from the northern branches of the Hadley cell and theWAHL increase the pressure gradient rotating southeasterly winds to northeasterly winds. This behavior strengthens WAHL circulation, deepening the low pressure region. In addition to the large-scale Atlantic-WAHL pressure gradient, Nicholson (2009) also suggests the importance of the AEJ and TEJ to rainfall and moisture advection into the region. The AEJ transports mesoscale convective systems (MCS) westward, which are responsible for large-scale precipitation in the region (Mohr and Thorncroft2006). The speed of the AEJ centered at 650 hPa is controlled by WAHL meridional circulation (Thorncroftand Blackburn 1999). Upliftin the WAHL generates an anticyclone aloftwhose easterly circulation strengthens the AEJ. The strengths of the AEJ and TEJ control the location and propagation of AEWs, which help organize MCS (Jackson et al. 2009). Factors impacting interseasonal variability of WAM rainfall include the number and timing of AEWs, along with the strength of the cross-equator pressure gradient driving the onshore monsoonal flows, with the latter linked to sea surface temperatures (SSTs) in the Gulf of Guinea, with increasing temperatures decreasing precipitation in the Sahel (Lough 1986; Vizy and Cook 2002; Fontaine and Louvet 2006; Caniaux et al. 2011). In terms of variability, the traditional measure of monsoon strength has been through seasonal rainfall. The region experiences extended periods of low and high rainfall, including the Sahelian drought of the 1970s and 1980s. Past investigations have looked at identifying causes of both the annual rainfall cycle and the larger decadal-scale patterns. Connections have been made with global sea surface temperatures and circulation patterns, as just discussed, along with other ocean and land surface processes including soil moisture. Hagos and Cook (2008) explain the decreasing Sahelian rainfall in the 1980s through increased sea surface temperatures in the Indian Ocean. Warming produced a region of subsidence over the Sahel blocking monsoonadvected moisture from the Atlantic. Continued increases in Indian Ocean SSTs have shifted this zone westward over the Atlantic leading to an increase though still depressed rainfall over the Sahel. Haarsma (2005) found a strong link between rainfall and mean sea level pressure over the Sahara, the summer location of the WAHL. Increases in surface air temperatures suggest a deepened heat low and increased rainfall over the Sahel. Eltahir and Gong (1996) investigated the sources of precipitation in West Africa and found the tropical Atlantic contributes 23%, central Africa contributes 17%, and precipitation recycling within West African contributes 27%. The Gulf of Guinea source is controlled by the southwest monsoon flow, the central African source is controlled by westerly flows generated by monsoon circulation, and the precipitation recycling is controlled by land surface properties.Using a Lagrangian approach, Nieto et al. (2006) tracked the sources of moisture for the Sahel and found that, in summer, precipitation recycling over the Sahel was the most important, pointing out the importance of soil moisture in sustaining a strongWAM. Coinciding with the West African monsoon region is the African meningitis belt (Lapeyssonnie 1963) extending through the semiarid region south of the Sahara. Several studies have indicated a strong link between atmospheric moisture, in the form of relative humidity or specific humidity, and meningococcal meningitis susceptibility. Molesworth et al. (2003) classified districts by their seasonal specific humidity profiles found that this classification along with land cover were the best predictors in a meningitis epidemic risk model. This relationship appears robust, as studies by Besancenot et al. (1997) in Benin and Yaka et al. (2008) in Niger and Burkina Faso reached similar conclusions. This link between relative humidity and its predictive capability of meningitis risk is corroborated in preliminary analysis (Pandya et al. 2014), shown in Fig. 1. This figure indicates an inverse relationship between relative humidity and meningitis risk. The probability of exceedance is based on the mean relative humidity for the preceding 4 weeks at a 2-week lag. The red dashed line indicates the inherent background risk of a meningitis epidemic independent of relative humidity. Humidity in the region and more importantly the timing of humidity increase and decrease are controlled by the WAM system. Thus, understanding monsoon dynamics to better predict monsoon onset and retreat in the context of increasing and decreasing relative humidity would allow better prediction ofmeningitis epidemic risk (Pandya et al. 2014). Prior to the development of a conjugate vaccine for serogroup A meningococcal meningitis, the primary method of treating all epidemics of meningococcal meningitis relied on the distribution of a polysaccharide vaccine to regions at risk for epidemics. The motivation was to contain the disease before it spread to surrounding districts. For districts at alert level, 5 cases in 100 000 received the vaccine if surrounding districts had already reached the epidemic level of 10 cases in 100 000. If a district reached the alert level without neighboring a district at the epidemic level, the decision to allocate vaccine was based on vaccine supply and time to the end of meningitis season. This allocation, managed by the International Coordinating Group (ICG) on Vaccine Provision, is still used to manage epidemics of other meningococcal meningitis serogroups, particularly W-135 (Pandya et al. 2014). While prior research efforts largely focused on monsoon seasonal rainfall and its variability, this study is motivated by the need to provide better tools to help mitigate and manage the meningitis risk. To this end, here we propose to investigate the interannual variability and predictability of relative humidity with its association to the large-scale geophysical drivers, during the onset and retreat phase of the monsoon season, which coincides with the retreat and onset season of meningitis risk. Note that monsoon onset as defined by Sultan and Janicot (2003) is split into two phases. The first phase is a ''preonset,'' which is identified as the date the ITF reaches 158N (with the ITF being the zero mean zonal wind component at 925 hPa). This phase represents the start of the rainy season in the region and the mean date of occurrence is 14 May, with a 9.8-day standard deviation. Monsoon onset is defined by an abrupt transition or ''jump'' of the ITCZ from 58 to 108N corresponding with increases in rainfall and a deepening of the heat low. The mean date of occurrence is 25 June, with a 9-day standard deviation. The deepening of the WAHL occurs on a mean date of 20 June, 5 days before the mean date of monsoon jump (Lavaysse et al. 2009). The paper is organized as follows: The study region and datasets used are first described followed by the methods. Weather station data for the time period 1973- 2012 were used along with climate reanalysis data to identify potential predictors of relative humidity behavior. Identified predictors were used to develop predictive models of relative humidity. Results from climate diagnostics and predictability are then presented, followed by results from predictive models and concluding with summary and discussion of the results. An understanding of the interseasonal variability of relative humidity could provide better prediction of the end of themeningitis season, allowing formore informed decisions while allocating vaccine and health-care resources. This study aims to investigate this interseasonal variability. 2. Methods a. Data and study region The study region encompasses countries falling within both the meningitis belt and WAM region and includes Mali, Burkina Faso, Togo, Benin, Chad, and Cameroon, as shown in Fig. 2. Daily calculated relative humidity data from theGlobalHistoricalClimatology Network (GHCN) were obtained through the National Oceanic and Atmospheric Administration (NOAA) Climate Data Online (CDO) portal (http://www.ncdc.noaa.gov/cdo-web/). The World Meteorological Organization (WMO) maintains the GHCN network, constructed from data collected by national meteorological services. These data were obtained for 32 stations (Fig. 2) within the study region with at least 90% coverage over 1973-2012. Investigations of large-scale climate variability used the gridded National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis data (Kalnay et al. 1996) and gridded Kaplan sea surface temperature reconstructions (Kaplan et al. 1998). b. Methodology Monsoon dynamics are examined through variations in relative humidity in two periods nominally defined here as monsoon onset, 15 May-30 June, and monsoon retreat, 15 September-15 October. These periods are similar to those selected by the African Monsoon Multidisciplinary Analysis for their special observing period (Redelsperger et al. 2006). For each period, mean relative humidity is computed for each station and year providing a 40-yr time series. A K-means cluster analysis (Scott and Knott 1974) is performed separately for each period to identify the spatial variability and coherence of relative humidity. In this, locations are grouped in homogeneous clusters such that within-cluster variability is minimum and betweencluster variability is maximum. A cluster mean relative humidity is then computed by averaging the relative humidity across stations in each cluster, to produce representative time series for each spatial region (cluster). The cluster relative humidity for each period are then correlated with a suite of contemporaneous global circulation fields including surface temperature; mean sea level pressure; zonal and meridional winds at 925, 600, and 200 hPa (925U and 925V, 600U and 600V, and 200U and 200V, respectively); and global Kaplan SST. The sea level pressure and average winds were selected to investigate links to atmospheric circulation including the African easterly jet and tropical easterly jet. Surface temperatures and SSTs investigate large-scale features such as the WAHL, ENSO, and Atlantic equatorial patterns. The resulting spatial correlation maps are used to identify the large-scale ocean and atmospheric mechanisms that drive the variability of relative humidity in the study region. Composite maps of selected fields corresponding to high and low relative humidity years are also produced to understand the physical links to extremes in relative humidity. To explore the predictability of relative humidity in the region, lagged correlation maps are produced with circulation fields and SST, where cluster relative humidity is correlated with large-scale fields from preceding time periods. For example, the onset period relative humidity is correlated with large-scale fields from the preceding January, February, March, and April. Regions of high correlation values are used to develop potential predictors by spatially averaging over this region. The predictors are then used in a generalized linear modeling (GLM) framework to develop predictive models at different lead times. This approach was selected as it is flexible and general. In GLM (McCullagh and Nelder 1989), the response or the dependent variable Y can be assumed to be a realization from any distribution in the exponential family with a set of parameters. A smooth and invertible link function transforms the conditional expectation of Y to the set of predictors. ... (1) where G(...) is the link function, X is the set of predictors or independent variables, E(Y) is the expected value of the response variable, and « is the error. In a linearmodel, the function G(...) is identity. Depending on the assumed distribution ofY, there are appropriate link functions (see McCullagh and Nelder 1989). The model parameters b are estimated using an iterated weighted least squares method that maximizes the likelihood function as opposed to an ordinary least squares method in linear modeling. Here we fit a linear regression model: that is, normality of variable Y and the identity link function. Models are fitted with different combination of predictors, and for each the Akaike information criteria (AIC) are calculated. The best model is selected as the one that minimizes the AIC, where AIC is calculated as follows: ... (2) where L is the logarithm of the likelihood function of the model with the predictor subset under consideration and k, the number of parameters to be estimated in thismodel, serves as a penalty. The AIC penalizes models with higher numbers of predictors, thus favoring parsimony. For the selected best model, two performance metrics are computed. A fitted R2 explains the variance captured by the model and a cross-validated (CV) R2 explains, where an observation is dropped, the model fitted using the rest of the observations, and the dropped pointed is predicted. This indicates the variance explained by the model in a predictive mode. To further assess the predictive capability of the models, we performed cross validation on our model results. This is performed by leaving 10% of observations out at random, which are then predicted using a model fitted on the rest of the data. A final measure of model fit, root-mean-square error (RMSE), is computed using the same drop-10% method as above. This is repeated 1000 times and the median RMSE is selected, providing a robust assessment of the predictive skill. This method has been widely used in application to western U.S. streamflow forecasting (Grantz et al. 2005; Regonda et al. 2006; Bracken et al. 2010). To evaluate the effectiveness of these forecasts for public health applications, the climatological risk of the meningitis epidemic is compared to the forecast risk using the relationship shown in Fig. 1. This is done by calculating the risk of epidemic using the 25th and 75th percentiles of the mean cluster humidity. The forecast risk is then calculated from the 25th and 75th percentiles of the mean forecast relative humidity calculated from the 25th and 75th percentiles of the model error using ... where «IQR is the interquartile range of the model error at 95% confidence. 3. Results and discussion a. Spatial variability A K-means cluster analysis was performed on the average relative humidity at all the locations for the onset, peak, and retreat periods. The data are grouped into several clusters and for each the within-cluster variance is computed. The number of clusters where this variance drops offand stabilizes is the optimal number selected. Figure 3 shows the within-cluster variance versus number of clusters for the three periods. It can be seen that the variance drops offaround three clusters in all the periods, indicating that higher number of clusters is unlikely to result in distinct homogeneous clusters. The spatial locations of the clusters in the three periods are shown in Fig. 4. In all the periods, the three clusters exhibit a clear north-south spatial pattern with stations grouped along east-west. The north-south stratification is consistent with movement of the tropical rain belt through the monsoon season and the background strong climatological humidity gradient in the region. The stations in the three clusters for the onset and retreat periods are almost identical with slight differences only at the cluster boundaries (Fig. 4). This similarity can be attributed to the onset northward and retreat southward movement of the rain belt being fairly uniform. Slight differences between the onset rain belt transition, with its jump from ;108 to 158N in June, and the retreat rain belt transition, with its smooth transition, explain the slight variations. During the peak season, the rain belt is within 108-158N latitudes: hence the diffusion of cluster boundaries. Several of the stations found in the northern cluster during onset and retreat are found in the middle cluster during the peak monsoon period. Only the three most northern stations (Tombouctou, Mali; Gao, Mali; and Agadez, Niger), all located close or within the Sahara, remain in the northern cluster in all three periods. The relative humidity is then averaged over the stations in each cluster to obtain cluster indices for the three periods and time series of their standardized values are shown in Fig. 5. Although the actual relative humidity varies between each cluster, the figure shows that they all have similar temporal variability (not shown). All three periods exhibit an upward trend from ;1988 through 2012, which corresponds to the upward trend found in the Sahel rainfall index (Janowiak 1988). The Sahel drought is also clearly visible in Fig. 4 during the peak period, with a strong dip in relative humidity in the mid-1980s. b. Links to large-scale climate To understand the drivers of year-to-year variability of the relative humidity, each cluster relative humidity was correlated with large-scale climate fields of the concurrent period. In particular, we selected five variables to correlate: surface temperature, mean sea level pressure, different pressure-level zonal and meridional winds, and global SST. The sea level pressure and pressure-level winds were selected to capture the links to atmospheric circulation such as the African easterly jet and tropical easterly jet, while the surface temperatures and SSTs are for large-scale oceanic features such as ENSO, Atlantic equatorial patterns, etc. Figure 6 shows the correlations with each of the variables' fields to the three clusters' mean relative humidity during the monsoon onset period, with correlations greater than 0.35 (less than 20.35) significant at the 95% confidence interval. In the onset period, all three clusters have a positive correlation with the Sahara desert surface temperature and a negative correlation with the Guinea Coast surface temperature [panel (i) of Figs. 6a-c]. This negative correlation extends to the Gulf of Guinea and to South Atlantic sea surface temperatures. This dipole pattern of positive correlation over land coupled with negative correlation to the south and over the ocean is indicative of a strong land-ocean thermal gradient, a key component of the monsoon. The correlation pattern with sea level pressure [panel (ii) of Figs. 6a-c] shows negative correlations where the pressure is low: over the warmer land of the Sahara (the approximate location of West African heat low). Positive correlations exist to the south and along the Guinea Coast, where the pressure is high [panel (i) of Figs. 6a-c]. With SSTs, a dipole pattern is apparent in the tropical Atlantic Ocean, with positive correlations in the northern tropical Atlantic and negative to the south [panel (iv) of Figs. 6a-c]. This dipole is well known as influencing the rainfall over northeastern Brazil and West Africa (Nobre and Srukla 1996). In the Pacific there is a weak positive correlation in the central and eastern tropical Pacific and a weak negative correlation in the west, reminiscent of the ENSO pattern. Overall, the correlation patterns are remarkably similar for all three clusters [panel (iv) of Figs. 6a-c]. Monsoon retreat period correlations are shown in Fig. 7. For all clusters, the land temperature correlations show positive correlation over western Africa similar to that in the onset period (Fig. 6) but a negative correlation to the south that is much weaker. The correlations with sea level pressure (SLP) also mirror this [panel (iv) of Figs. 6a-c]. The correlation pattern with 600-hPa zonal winds shows a stronger negative correlation over the region of the African easterly jet, which is also the location of the ITCZ [panel (iii) of Figs. 7a-c]. This correlation is stronger and more coherent than its counterpart in the monsoon onset period [panel (iii) of Figs. 6a-c], explained by an established land-ocean thermal gradient and soilmoisture- driven land surface processes (Cook 1999). The correlation with SST is much weaker than that observed during onset. This is consistent with the fact that, during the end of the monsoon season, the tropical Atlantic SST gradient is much diminished as the ITCZ is in the Northern Hemisphere and on its way south and the gradient is established in fall. Composite analysis was performed to investigate the large-scale climate features responsible for relative humidity extremes. For this, we selected years with high and low relative humidity, outside of one standard deviation away from the mean, for a given period, and maps of climate variables averaged over these years are produced. We show representative composite maps for the south cluster for the onset (Fig. 8) and retreat periods (Fig. 9). Composite maps of surface temperature for low years for the onset period (Fig. 8) shows a cooler land and warmer ocean, indicative of a weaker land-ocean temperature gradient. Stronger than normal Harmattan winds exist, which is consistent with the temperature pattern. During high years, the patterns are reversed, although the warming over land is a bit stronger than during low years and also the wind pattern is weaker than that of the low years. The asymmetry in the relationship during low and high years indicates nonlinearity in the relationship and the difference maps in the same figure show this. These patterns are similar during retreat (Fig. 9) and also for other clusters (figures not shown). c. Predictability As mentioned in section 1, relative humidity during monsoon onset and retreat periods is important for the retreat and onset of the African meningitis season, respectively. Therefore, the ability to predict the relative humidity during these periods is of specific interest. To this end, predictors and forecasting models were developed to understand the predictability and the potential long-lead skill. To identify predictors, each cluster relative humidity was correlated with lagged large-scale climate variables. We selected three lead times to issue forecasts for the monsoon onset (nominally 15 May-30 June): 1 March, 1 April, and 1 May (giving lead times of 75, 45, and 15 days, respectively). Figure 10 shows the correlation between the onset south cluster relative humidity with January climate variables. Onset predictors were identified using January, February, March, and April climate variables. At each lead time, different predictors were retained in eachmonth (TableA1 lists all predictors identified). Note that the correlation patterns with surface temperature, sea level pressure, and winds are similar to the correlation patterns seen during the concurrent period (Fig. 6). This indicates that the largescale patterns are persistent and thus lend potential predictability. Ward (1998) produced hindcasts of precipitation over West Africa using sea surface temperatures as predictors and found the models could explain around 30% of the variability. The boxes indicate regions of high absolute correlation. Correlations with April climate variables (Fig. 11) also show similar patterns. Regions with high absolute correlation, approximately 0.4 or above in these maps, are identified, and the corresponding climate variables averaged over these regions provide potential predictors [correlations greater than 0.35 (less than 20.35) are significant at the 95% confidence interval]. Figures 12 and 13 show the correlation plots for June andAugust used to select retreat predictors.Retreat predictors were identified using May, June, July, and August climate variables. As with the onset, different predictors were retained in each month (Table A2 lists the predictors identified). The red boxes in Figs. 10-13 show the regions used to generate these predictors. The best model (or set of best predictors) based on AIC selects 2-3 predictors for almost all the lead times, except for a few cases during withdrawal where it selects 4 predictors. The predictor sets and model skill scores can be found in Table 1 (onset period) and Table 2 (retreat period). Furthermore, predictor sets were remarkably consistent between the south and middle clusters, with notable exceptions being the 1 April south cluster model containing February North Atlantic SST and the 1 May south cluster model containing April Guinea Coast 200-hPa meridional winds. Guinea Coast 600-hPa meridional winds were found in the north cluster replacing South Atlantic mean SLP (MSLP) in the 1 March and 1 April models. Plots of observed, estimated, and cross-validated estimates of south onset cluster relative humidity are shown in Fig. 14. It can be seen that the model estimates and predicts the values fairly well at all the lead times. It is interesting to note that the models exhibit good skill at all lead times, especially at 75-day lead time, as can be seen by the cross-validated R2 in Tables 1 and 2. Models at shorter lead times still retain longer lead predictors, attributed to disruption of the teleconnections with SST and pressure centers as the monsoon is established in the south. This ''predictability barrier'' has been found with Indian Ocean monsoon rainfall (Rajagopalan andMolnar 2013) and East African rainfall (Nicholson 2013). Future investigation incorporating additional land processes could provide better understanding of the WAM predictability barrier. Forecasts of retreat period relative humidity also show good skill at all lead times, including long leads (Fig. 15). The retreat period model predictor sets are more varied. Land surface-atmosphere interactions are responsible for the majority of the predictors, with Guinea Coast 925-hPa meridional winds, central Africa 600- and 200-hPa winds, and central Africa surface temperature showing up in the predictor sets. There are no other common predictors found between the models. Model and cross-validated estimates capture the observed variability well, with median RMSE values at a 75-day lag ranging between 1.38% for the south cluster and 3.01% for the north cluster. The utility of these forecasts is best illustrated by the narrowing of the epidemic risk range as compared to climatological variability in relative humidity for each cluster. The risk of epidemic for the north onset cluster ranges between 1.6% and 3.0% for climatology. The risk of epidemic from the RH forecasts is relatively consistent for all lead times and ranges between 2.1% and 2.2%, a great reduction in the range. The narrowing of the risk range is seen in all clusters for both onset and retreat but is most pronounced for the north clusters, where relative humidity is lowest. 4. Summary Skillful seasonal forecasts of relative humidity in this region have been developed to aid in the development of meningitis forecasting. These forecasts can provide a more informed decision-making process to aid in vaccine resource allocation. Relative humidity forecasts were developed by exploiting the persistence of largescale climate features including sea surface temperature, sea level pressure, and winds. Spatially coherent clusters were identified through the K-means method and exhibit a strong north-south dependence. Composite time series created by averaging all stations within each cluster were then modeled using large-scale climate variables as predictors. These models were created to produce forecasts of relative humidity at lead times of 75, 45, and 15 days. The model results indicate skill in predicting relative humidity from these climate variables. The predictions provide the mean relative humidity for the period of interest, which can inform an earlier or later end (or start) to the meningitis season. The best parameter sets for the models indicate that the strengths of the WAHL and South Atlantic anticyclone are the largest controls on relative humidity during monsoon onset. Secondary controls include the Gulf of Guinea SSTs, which influence the local MSLP and winds and modify the crossequator pressure gradient responsible for monsoon flow. Another secondary influence is the strength of the North Atlantic anticyclone, which along with the WAHL directs the strength of the hot, dry Harmattan winds from the northeast. These secondary influences appear to be responsible for much of the intercluster variability. Combined they have a large influence on the location of the intertropical front (ITF), the boundary between monsoon and Harmattan flows. During monsoon retreat, the strength of the South Atlantic anticyclone remains important, but surface-atmospheric interactions become the secondary controls. There is a lot of variability surrounding the predictors identified for each model; given the complexity of these interactions, it becomes more difficult. This knowledge offers the International Coordinating Group of the World Health Organization a potentially better forecast of the start and end of the meningitis season, which can provide a more informed decisionmaking process while allocating vaccine in the region. A more precise range epidemic risk with these forecasts provides a better understanding of when epidemic risk has fallen below a designated threshold marking the end of the meningitis season and negating the need for additional vaccination. A companion paper (D. Broman et al. 2013, unpublished manuscript) extends this modeling to meningitis incidence. Acknowledgments. This work was partially supported by a generous grant from the Google.org foundation. We thank our three anonymous reviewers and Dr. Kerry H. Cook, Editor, Journal of Climate, for their work in improving this manuscript. REFERENCES Besancenot, J. P., M. Boko, and P. C. Oke, 1997: Weather conditions and cerebrospinal meningitis in Benin (Gulf of Guinea, West Africa). Eur. J. Epidemiol., 13, 807-815, doi:10.1023/ A:1007365919013. Bracken, C., B. Rajagopalan, and J. Prairie, 2010: A multisite seasonal ensemble streamflow forecasting technique. Water Resour. Res., 46, W03532, doi:10.1029/2009WR007965. Caniaux, G., H. Giordani, J.-L. Redelsperger, F. Guichard, E. Key, and M. Wade, 2011: Coupling between the Atlantic cold tongue and the West African monsoon in boreal spring and summer. J. Geophys. Res., 116, C04003, doi:10.1029/ 2010JC006570. Cook, K. H., 1999: Generation of the African easterly jet and its role in determining West African precipitation. J. Climate, 12, 1165-1185, doi:10.1175/1520-0442(1999)012,1165: GOTAEJ.2.0.CO;2. Drobinski, P., B. Sultan, and S. Janicot, 2005: Role of the Hoggar massif in the West African monsoon onset. Geophys. Res. Lett., 32, 481-484, doi:10.1029/2004GL020710. Eltahir, E., and C. Gong, 1996: Dynamics of wet and dry years in West Africa. J. Climate, 9, 1030-1042, doi:10.1175/ 1520-0442(1996)009,1030:DOWADY.2.0.CO;2. Fontaine, B., and S. Louvet, 2006: Sudan-Sahel rainfall onset: Definition of an objective index, types of years, and experimental hindcasts. J. Geophys. Res., 111, D20103, doi:10.1029/ 2005JD007019. Grantz, K., B. Rajagopalan, M. Clark, and E. Zagona, 2005: A technique for incorporating large-scale climate information in basin-scale ensemble streamflow forecasts. Water Resour. Res., 41, W10410, doi:10.1029/2004WR003467. Haarsma, R. J., 2005: Sahel rainfall variability and response to greenhouse warming. Geophys. Res. Lett., 32, L17702, doi:10.1029/2005GL023232. Hagos, S. M., and K. H. Cook, 2008: Ocean warming and latetwentieth- century Sahel drought and recovery. J. Climate, 21, 3797-3814, doi:10.1175/2008JCLI2055.1. Jackson, B., S. E. Nicholson, and D. Klotter, 2009: Mesoscale convective systems over western equatorial Africa and their relationship to large-scale circulation. Mon. Wea. Rev., 137, 1272-1294, doi:10.1175/2008MWR2525.1. Janowiak, J., 1988: An investigation of the interannual rainfall variability in Africa. J. Climate, 1, 240-255, doi:10.1175/ 1520-0442(1988)001,0240:AIOIRV.2.0.CO;2. Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437-471, doi:10.1175/1520-0477(1996)077,0437:TNYRP.2.0.CO;2. Kaplan, A., M. A. Cane, A. C. Clement, M. B. Blumenthal, B. Rajagopalan, and C. Bottoraley, 1998: Analyses of global sea surface temperature 1856-1991. J. Geophys. Res., 103, 18 567- 18 589, doi:10.1029/97JC01736. Lapeyssonnie, L., 1963: La méningite cérébrospinale en Afrique. Bull. WHO, 28 (Suppl.), 1-114. Lavaysse, C., C. Flamant, S. Janicot, D. J. Parker, J.-P. Lafore, B. Sultan, and J. Pelon, 2009: Seasonal evolution of the West African heat low: A climatological perspective. Climate Dyn., 33, 313-330, doi:10.1007/s00382-009-0553-4. Lough, J. M., 1986: Tropical Atlantic sea surface temperatures and rainfall variations in Subsaharan Africa. Mon. Wea. Rev., 114, 561-570, doi:10.1175/1520-0493(1986)114,0561: TASSTA.2.0.CO;2. McCullagh, P., and J. A. Nelder, 1989: Generalized Linear Models. 2nd ed. Chapman & Hall/CRC, 511 pp. Mohr, K. I., and C. D. Thorncroft, 2006: Intense convective systems in West Africa and their relationship to theAfrican easterly jet. Quart. J. Roy. Meteor. Soc., 132, 163-176, doi:10.1256/qj.05.55. Molesworth, A. M., L. E. Cuevas, S. J. Connor, A. P. Morse, and M. C. Thomson, 2003: Environmental risk and meningitis epidemics in Africa. Emerging Infect. Dis., 9, 1287-1293, doi:10.3201/eid0910.030182. Nicholson, S. E., 2009: A revised picture of the structure of the ''monsoon'' and land ITCZ over West Africa. Climate Dyn., 32, 1155-1171, doi:10.1007/s00382-008-0514-3. -, 2013: The West African Sahel: A review of recent studies on the rainfall regime and its interannual variability. ISRN Meteor., 453521, doi:10.1155/2013/453521. Nieto, R., L. Gimeno, and R. M. Trigo, 2006: A Lagrangian identification of major sources of Sahel moisture. Geophys. Res. Lett., 33, L18707, doi:10.1029/2006GL027232. Nobre, P., and J. Srukla, 1996: Variations of sea surface temperature, wind stress, and rainfall over the tropical Atlantic and South America. J. Climate, 9, 2464-2479, doi:10.1175/ 1520-0442(1996)009,2464:VOSSTW.2.0.CO;2. Pandya, R., and Coauthors, 2014: Using weather forecasts to help manage meningitis in the West African Sahel. Bull. Amer. Meteor. Soc., in press. Rajagopalan, B., and P. Molnar, 2013: Signatures of Tibetan Plateau heating on Indian summer monsoon rainfall variability. J.Geophys. Res. Atmos., 118, 1170-1178, doi:10.1002/ jgrd.50124. Ramel, R., H. Gallée, and C. Messager, 2006: On the northward shiftof the West African monsoon. Climate Dyn., 26, 429-440, doi:10.1007/s00382-005-0093-5. Redelsperger, J.-L., C. D. Thorncroft, A. Diedhiou, T. Lebel, D. J. Parker, and J. Polcher, 2006: African Monsoon Multidisciplinary Analysis: An international research project and field campaign. Bull. Amer. Meteor. Soc., 87, 1739-1746, doi:10.1175/BAMS-87-12-1739. Regonda, S. K., B. Rajagopalan, M. Clark, and E. Zagona, 2006: A multimodel ensemble forecast framework: Application to spring seasonal flows in the Gunnison River basin. Water Resour. Res., 42, W09404, doi:10.1029/2005WR004653. Scott, A. A. J., and M. Knott, 1974: A cluster analysis method for grouping means in the analysis of variance. Biometrics, 30, 507-512, doi:10.2307/2529204. Seefeldt, M. W., T. M. Hopson, and T. T. Warner, 2012: A characterization of the variation in relative humidity across West Africa during the dry season. J. Appl. Meteor. Climatol., 51, 2077-2089, doi:10.1175/JAMC-D-11-0196.1. Sultan, B., and S. Janicot, 2003: The West African monsoon dynamics. Part II: The ''preonset'' and ''onset'' of the summer monsoon. J. Climate, 16, 3407-3427, doi:10.1175/ 1520-0442(2003)016,3407:TWAMDP.2.0.CO;2. Thorncroft, C. D., and M. Blackburn, 1999: Maintenance of the African easterly jet. Quart. J. Roy. Meteor. Soc., 125, 763-786, doi:10.1002/qj.49712555502. Vizy, E. K., and K. H. Cook, 2002: Development and application of a mesoscale climate model for the tropics: Influence of sea surface temperature anomalies on the West African monsoon. J. Geophys. Res., 107, doi:10.1029/2001JD000686. Ward, M. N., 1998: Diagnosis and short-lead time prediction of summer rainfall in tropical North Africa at interannual and multidecadal timescales. J. Climate, 11, 3167-3191, doi:10.1175/ 1520-0442(1998)011,3167:DASLTP.2.0.CO;2. Yaka, P., B. Sultan, H. Broutin, S. Janicot, S. Philippon, and N. Fourquet, 2008:Relationships between climate and year-to-year variability in meningitis outbreaks: A case study in Burkina Faso and Niger. Int. J. Health Geogr., 7, 34, doi:10.1186/1476-072X-7-34. DANIEL BROMAN Civil, Environmental, and Architectural Engineering, University of Colorado, Boulder, Colorado BALAJI RAJAGOPALAN Civil, Environmental, and Architectural Engineering, and CIRES, University of Colorado, Boulder, Colorado THOMAS HOPSON National Center for Atmospheric Research, Boulder, Colorado (Manuscript received 15 July 2013, in final form 27 March 2014) Corresponding author address: Thomas Hopson, Hydrologic Applications Program, National Center for Atmospheric Research, P.O. Box 3000, Boulder, CO 80307-3000. E-mail: [email protected] APPENDIX Model Climate Predictors The entire suite of climate predictors considered for inclusion in the onset and retreat relative humidity models are found in Tables A1 and A2, respectively. (c) 2014 American Meteorological Society |