Nutrition

Malnutrition, resulting from both acute food shortages and chronic or seasonal undernourishment, is a significant health burden for many countries in Africa. Sub-Saharan Africa has the highest prevalence of hunger in the world, with undernourishment rates of approximately 25% (Food and Agriculture Organization of the United Nations [FAO], 2015). An estimated 220 million people are estimated to be calorie deficient. The problem is particularly acute in children: the percentage of children under 5 who are stunted exceeds 30% in sub-Saharan Africa, and Eastern and Southern Africa and West and Central Africa are the only two major developing regions in the world in which the number of stunted children increased between 1990 and 2014 (UNICEF, 2015). Stunting is a powerful indicator of health outcomes in general. Fifty-three percent of infectious disease-related deaths in developing countries are associated with stunting, and severe stunting causes a fourfold increase in a child’s risk of death (Black et al., 2008; Caulfield, de Onis, Blossner, & Black, 2004). Rates of low-birthweight infants, underweight children, wasting, and childhood deaths attributable directly or indirectly to undernutrition are all among the highest in the world (UNICEF, 2015).

For these reasons, food insecurity is perhaps the most pressing health issue in Africa. Its causes are a matter of considerable research and political debate. In a proximal sense, rates of food insecurity are tied directly to poverty (Food and Agriculture Organization of the United Nations [FAO], 2011). It has been demonstrated time and again that undernourishment is most prevalent in poorer households, and that over- and undernutrition frequently coexist at the same time within a single community. Lack of education also contributes to malnutrition, as ignorance of basic information on nutrition, sanitation, and disease prevention within a household has been shown to contribute significantly to negative nutritional outcomes (Food and Agriculture Organization of the United Nations [FAO], 2011). Poverty and lack of education, however, are a product of larger social and institutional drivers, including physical isolation due to lack of infrastructure, ineffective or corrupt governance, and sociocultural attitudes that can limit economic or educational opportunities (Bain et al., 2013).

Links between climate and food security, then, are complex and heavily mediated by local, national, and international social, political, and economic factors. Nevertheless, climate does have the potential to affect food security through multiple pathways. First, food availability is tied to the performance of rainfall-dependent crops and the potential for crop and livestock loss under extreme events. Food access is influenced both by local climate variability—which will impact the price of food in local markets and farmers’ ability to pay for it—and by global food price spikes that are associated with climate shocks in remote food basket regions (Lagi, Bar-Yam, Bertrand, & Bar-Yam, 2011; Wiggins, Keats, & Compton, 2010). Food utilization is also affected by climate, since the risk of spoilage and the availability of biomass energy for cooking are both climate sensitive, and the stability of food supply can be affected by climate variability via its impacts on availability, access, and utilization ((Ray, Gerber, MacDonald, & West, 2015).

Of these links, food availability is by far the most extensively studied. Broadly speaking, wet periods lead to higher crop yields and greater national-scale food availability in most countries of sub-Saharan Africa (Buhaug, Benaminsen, Sjaastad, & Theisen, 2015). But this macroscale association does not apply everywhere. A number of location-specific studies have failed to find a correlation between rainfall totals and crop yield and have noted more nuanced relationships in which length of growing season, timing of the onset of rains, or the pattern of wet or dry conditions within the cropping season are more important predictors of yield (Adejuwon, 2006). Temperature can also be an important predictor of yields (Lobell, Schlenker, & Costa-Roberts, 2011), with higher temperatures in sub-Saharan Africa expected to lead to reduced yields of many major crops (Sultan et al., 2013). In the observed record, however, variability in rainfall has been both larger and more significant to crop performance than interannual variability in temperature. Ray et al. (2015) find significant sensitivity of maize yields to ENSO-related rainfall variability over much of Africa, but they note that complex interactions between rainfall and temperature influence yields in many regions. While most studies of climate and yield variability are performed at large spatial scale, local rainfall variability can be a critical determinant of nutritional outcomes in subsistence-based communities (Grace, Davenport, Funk, & Lerner, 2012). This is relevant to climate change projections, as studies that fail to consider changes in variability or fail to account for highly local variability could underestimate the risk that climate change poses to food availability (Thornton, Ericksen, Herrero, & Challinor, 2014).

Though less studied in the climate context, inadequate access to food and markets is a major risk factor for food security that can be influenced by climate conditions. Transport costs in sub-Saharan Africa tend to be high (Haile, 2005), and general limitations in transportation and communication infrastructure can reduce market integration (Brown et al., 2012). Climate can influence market isolation through its impacts on transportation, including extreme events that damage infrastructure and regular seasonal wet and dry periods that affect river transport and the functioning of low-quality roads. Economic isolation resulting from lack of market integration is, in turn, associated with greater food insecurity (Burgess & Donaldson, 2010). This signal is communicated through food prices, which can differ widely between markets within a single country. Isolated markets exhibit low price correlation with integrated markets and global commodity indices (Davenport & Funk, 2015). This could, in principle, insulate isolated rural communities from global price shocks, but it frequently has a negative impact, as local prices spike under low-production or high-demand conditions, harming food access for poorer members of the community. Meanwhile, urban poor populations face access constraints when global commodity prices spike (Cohen & Garrett, 2010; Crush & Fayne, 2010). Volatility in global food prices is still poorly understood, but climate shocks to major food-producing regions play a significant role, either through directly impacting food supply or through triggering speculative bubbles in food prices (Lagi et al., 2011). The combined effects of a local climate shock with food price shocks can be especially detrimental to food security; countries in sub-Saharan Africa are particularly vulnerable to this combination on account of low purchasing power and a net dependence on food imports (Felix & Romuald, 2009).

The influence of climate variability on food utilization is relatively understudied, though it is understood that climate is relevant to the availability of biomass energy and clean water for cooking. Postharvest food loss during storage is another factor that affects both food access (via market availability) and utilization (when stored in the home). Postharvest loss is a leading constraint on food security in Africa, as poor storage techniques and infrastructure can lead to large food loss (Godfray et al., 2010; Parfitt, Barthel, & Macnaughton, 2010); for example, smallholder farmers in Africa lose an estimated 14 to 36% of their maize grain postharvest (Tefera, 2012). Food storage is sensitive to ambient temperature, humidity, and rainfall, as well as to extreme events that can disrupt postharvest processes. This points to the potential for climate variability and change to affect rates of postharvest loss, and to the need for adaptation strategies that target improved storage systems for African farmers (Stathers, Lamboll, & Mvumi, 2013). Food utilization could also be impacted by changes in crop choice in response to climate change and, potentially, by projected increases in diarrheal disease leading to reduced ability to absorb nutrients (Felix & Romuald, 2009).

The importance of climate variability to food security in Africa has motivated the creation of numerous climate outlooks and food security forecasts (Patt, Ogallo, & Hellmuth, 2007). The Famine Early Warning System Network (FEWS NET) is perhaps the most established and influential of these programs (Brown, 2008; Funk & Verdin, 2010; Verdin, Funk, Senay, & Choularton, 2005). Recognizing the multiple pathways to food insecurity, FEWS NET issues food security outlooks and acute warnings for 29 countries in Africa through a combination of seasonal weather forecasts, crop stress modeling, food price monitoring and projection, satellite-based rainfall and vegetation anomaly analysis, extreme weather alerts, and ENSO tracking. This provides a suite of risk indicators that capture climate influence on crop production and prices, allowing for decision-relevant food insecurity forecasts (Brown, Funk, Galu, & Choularton, 2007). Applying these forecasts to effective interventions, however, remains a challenge in many countries (Funk, 2011).

There is significant concern that climate change will exacerbate food insecurity over much of Africa, though with significant regional variability (Adejuwon, 2006; Adhikari, Nejadhashemi, & Woznicki, 2015; Berg, de Noblet-Ducoudré, Sultan, Lengaigne, & Guimberteau, 2013; Schlenker & Lobell, 2010; Sultan et al., 2013). This concern is based largely on projected climate stress on crop production. Climate change is expected to have significant negative impacts on grain production in sub-Saharan Africa, including potentially dramatic loss of productivity in wheat (Nelson et al., 2009) and maize (Schlenker & Lobell, 2010) in the absence of significant adaptation activities. Production of certain key export crops could also fall, affecting income and food access (Adhikari et al., 2015), and rising temperatures and altered drought patterns could have impacts on livestock (Niang et al., 2014). Investment in climate smart agriculture does have the potential to offset some of these risks (Brown & Funk, 2008; Challinor, Wheeler, Garforth, Craufurd, & Kassam, 2007). Climate change may also alter the range of pests that impact both crops and livestock systems, but there is substantial uncertainty in these projections (Niang et al., 2014).

The impact of these risks on food production, coupled with other factors influencing access and utilization, could have significant impacts on nutrition. Averaged across sub-Saharan Africa, climate change is projected to cause higher food prices, lower food affordability, a decrease in calorie intake, and increased malnourishment (Ringler, Zhu, Cai, Koo, & Wang, 2010). Childhood anemia, undernutrition, and stunting are projected to increase (Jankowska, Lopez-Carr, Funk, Husak, & Chafe, 2012); indeed, some studies indicate that climate change will cause a greater number of children to become undernourished in Africa than in any other region of the world (Nelson et al., 2009; Phalkey, Aranda-Jan, Marx, Hofle, & Sauerborn, 2015). This climate stress could be large enough to offset and possibly outweigh expected reductions in stunting due to economic development (Lloyd, Kovats, & Chalabi, 2011). There is also the potential for a gender imbalance in the impact of climate change on undernutrition, as girls may disproportionately suffer from reduced food intake under conditions of scarcity (Bain et al., 2013).

Conflict and Migration

The vast majority of research on African food security and nutrition under climate change has assumed relatively stable social conditions. Researchers attempt to account for factors such as economic development, changing fertility rates, and urbanization when possible, but the general assumption is that the pathways through which climate affects nutrition are stationary. A complementary line of analysis is concerned with the impact that climate change could have on fundamental social and economic stability. The idea that climate change could lead to increased conflict and large-scale migration is much discussed in research and policy circles (Buhaug et al., 2015; Salehyan, 2014), and weather-driven shocks to food production are frequently invoked as a stress that could provoke these phenomena (Homer-Dixon, 1991; Koubi, Bernauer, Kalbhenn, & Spilker, 2012) (Figure 3). Projections of potential climate–scarcity–conflict dynamics are most often studied in the developing world, and many studies have focused on Africa (Brown, Hammill, & McLeman, 2007; Hendrix & Glaser, 2007).

Open in new tab

Figure 3. Hypothesized climate–scarcity–conflict feedbacks. Flowchart is derived from similar figures in Brown et al., 2007; Buhaug et al., 2015; Homer-Dixon, 1991).

Empirical evidence, however, is limited and controversial. Reduced food access due to an increase in global food prices, for example, has been cited as a cause for widespread food riots in Africa in 2007–2008 and as a trigger for the “Arab Spring” unrest in North Africa in 2011 (Berazneva & Lee, 2013; Johnstone & Mazo, 2011; Sternberg, 2012), suggesting that food insecurity can lead to significant political instability. The strength of the link between climate and the global food commodity price spikes, however, is debated, as is the question of whether the food security of undernourished people was the actual reason for “food riots” in either 2007–2008 or 2011 (Sneyd, Legwegoh, & Fraser, 2013). Local climate variability leading to reduced food production in both farming and pastoral systems has also been identified as a driver of violent conflict in Africa, including the extended conflict and suspected genocide in Darfur that began in 2003 and the less widely reported but significant conflicts in Kenya, Ghana, Nigeria, and elsewhere. Again, the complex nature of climate–scarcity–conflict dynamics makes it difficult to draw systematic conclusions. In Darfur, for example, most analysis points to a combination of political and cultural triggers for conflict (Kevane & Gray, 2008; O’Fahey, 2006; Salih, Mohamed Abdel Rahim Mohamed, 2005), though a relatively direct climate trigger does appear to be plausible in other situations (Scheffran, Ide, & Schilling, 2014). A systematic analysis of climate variability, food prices, and violent conflict across sub-Saharan Africa found significant relationships between climate and food production but failed to find a robust link between this relationship and incidences of violent conflict (Buhaug et al., 2015). The study’s authors note, however, that their analysis did not consider the possibility that climate-mediated impacts on food prices influence conflict via international food prices, localized subnational disturbances, or chronic impacts of food production on development.

Large-scale migration can be both a response to experienced food insecurity and a cause for future food insecurity. As with conflict, relationships between climate, food security, migration and, ultimately, health outcomes are difficult to untangle, but the potential for climate shocks and long-term climate change to trigger migration has received significant attention (Black, Bennett, Thomas, & Beddington, 2011; Brown et al., 2007; McLeman & Smit, 2006; Reuveny, 2007). In Africa, there is evidence that patterns of voluntary and forced migration are influenced by climate stress. This includes within-region rural-to-urban migration and large-scale migration between regions (Barrios, Bertinelli, & Strobl, 2006; Greiner & Sakdapolrak, 2013; McMichael, Barnett, & McMichael, 2012). The drought in the Horn of Africa in 2011, in particular, highlights the complicated interplay between food security and political dynamics in the face of a climate shock. It is also a climate pattern that could become more frequent under climate change (Funk et al., 2008).

Voluntary migration can be viewed as an adaption to changing climate (Tacoli, 2009), though experience shows that both forced and voluntary migration in response to climate stress has a negative impact on food security and other health outcomes in vulnerable populations (McMichael et al., 2012; Toole, 2005). An applied research effort by CARE has sought to distinguish between climate-affected populations that use migration to flourish, those that migrate to maintain their level of well-being, those that migrate but suffer (“erosive migration”), and those that are trapped because they lack the capacity to migrate (Warner et al., 2012). In all cases, there is potential risk to food security, but risks are greatest for erosive migration and trapped populations.

Vector-Borne Disease

Malaria, on account of its enormous and widespread health burden, has received more attention than any other climate-sensitive disease. Multiple species of Anopheles mosquitoes serve as vectors across the continent, with An. gambiae s.s., An. arabiensis, and An. funestus responsible for the majority of transmission (Sinka et al., 2010). Notably, the relatively deadly Plasmodium falciparum malaria parasite is the most prevalent form of malaria in sub-Saharan Africa (Figure 4).

Open in new tab

Figure 4. Mean rate of observed clinical Plasmodium falciparium malaria cases per person per year for the year 2015 (Bhatt et al., 2015), overlain on a map of population centers derived from the Gridded Population of the World v4 database, 2010 count data (Center for International Earth Science Information Network—CIESIN—Columbia University, 2016).

When describing risk profiles, epidemiologists typically distinguish between endemic and epidemic malaria regions. The definition of these categories is more complicated than is generally acknowledged (Hay, Smith, & Snow, 2008) and is the subject of some debate. In general terms, however, one can understand that endemic areas are characterized by “stable” transmission rates, with transmission occurring year-round (holoendemic) or seasonally with high regularity and prevalence rates (hyperendemic). Rates of lower endemicity have unstable transmission that can be highly sensitive to climate variability and other disturbance. Regions with unstable transmission are variously defined as having low endemicity (mesoendemic to hypoendemic categories) or as being epidemic zones in which outbreaks occur at irregular intervals. In these regions, acquired immunity is low, so malaria outbreaks can be severe and include significant mortality. From a climate perspective, epidemic malaria has been the focus of early warning systems, while the distribution of endemic malaria may be sensitive to long-term climate variability and change.

Considerable efforts have been made to develop climate-informed malaria early warning systems (MEWS) for epidemic zones of Africa, based on the observation that malaria vectors and the malaria parasite are sensitive to rainfall and temperature conditions (Craig, Snow, & Le Sueur, 1999), with a lag of weeks to months (Bomblies, Duchemin, & Eltahir, 2009; Teklehaimanot, Lipsitch, Teklehaimanot, & Schwartz, 2004). These MEWS-oriented studies tend to focus on single regions within Africa, and they rely on different datasets for both prediction and evaluation, which makes it difficult to draw broad conclusions across studies (Mabaso & Ndlovu, 2012). Nevertheless, the majority of studies have found some relationship between rainfall variability and malaria epidemics. In general, malaria epidemics are more likely to occur after periods of unusually heavy rainfall. The most likely explanation for this is that heavy rainfall in areas that are typically water limited expands the number of Anopheles breeding sites, particularly as waters recede and leave stagnant pools and puddles behind. This increases vector capacity and the probability of encounters between humans and vectors (Grover-Kopec et al., 2005; Patz et al., 2002). The association is not universal, however, as excess rainfall in some regions may have the effect of washing out breeding sites, breaking the association between rainfall and malaria risk (Molineaux, Wernsdorfer, & McGregor, 1988). This points to the need for ecosystem-specific (Githeko, Ogallo, Lemnge, Okia, & Ototo, 2014) and spatially specific (Arab, Jackson, & Kongoli, 2014) studies to support locally customized MEWS.

Beyond rainfall, studies have shown that temperature variability can also impact malaria transmission rates, with warm temperatures leading to malaria outbreaks in regions where low temperatures typically limit vector development, Plasmodium growth rates, or biting frequency (Lindsay & Martens, 1998; Molineaux et al., 1988). Other studies have made use of satellite-derived vegetation estimates—most commonly a normalized difference vegetation index (NDVI), a nondimensional parameter that is a proxy for vegetation coverage or health (Ceccato et al., 2007; Gomez-Elipe, Otero, Van Herp, & Aguirre-Jaime, 2007). These vegetation studies have sometimes found promising statistical associations, though the mechanism of association could be that vegetation and malaria are responding in parallel to a common rainfall forcing, rather than that malaria risk is actually mediated by vegetation.

Taking a broader climate perspective and seeking to extend the time horizon of MEWS, a number of studies have attempted to predict malaria epidemics as a function of large-scale climate modes such as ENSO. Since ENSO is a major driver of rainfall variability in parts of Africa, it can be employed as a predictor of rainfall-mediated epidemics at several months’ lead (Hashizume, Terao, & Minakawa, 2009; Lindsay, Bødker, Malima, Msangeni, & Kisinza, 2000; Mabaso, Kleinschmidt, Sharp, & Smith, 2007). Other seasonal prediction studies have applied dynamically based seasonal forecast systems to predict rainfall and temperature anomalies directly and have used those predictions to drive malaria models (Ceccato et al., 2007; Hoshen & Morse, 2004; Jones & Morse, 2010; Tompkins & Di Giuseppe, 2015). At a research level, many of these approaches to MEWS have shown promise, but to date few operational systems are in use.

From a climate change perspective, a number of studies have projected changes in the range of epidemic or endemic malaria across Africa. In the East African Highlands, mean warming is expected to lead to increased malaria risk (Ermert, Fink, & Paeth, 2013; Siraj et al., 2014). Diurnal and day-to-day temperature variability can also be quite important (Blanford et al., 2013; Paaijmans et al., 2014), however, suggesting that climate-based malaria risk projections must account for changes in variability and transient temperature extremes. There is also a possibility that warming will push the upper end of the malaria transmission optimum; Ryan et al. (2015) find that warming would be expected to increase the suitable area for malaria in Africa but to decrease and shift the most suitable areas for year-round transmission as temperatures in some currently holoendemic areas rise above known physiological optima. Projected changes in rainfall are highly uncertain for many parts of Africa, but in areas with projected decreases in precipitation—including the Sahel—combined drying and warming trends could lead to reduced malaria risk, or at least to a neutral climate change impact on transmission rates (Ermert et al., 2013; Yamana, Bomblies, & Eltahir, 2016).

Beyond malaria, extensive work has been done on early warning and climate change analysis for a number of other vector-borne infectious diseases in Africa. Rift Valley Fever (RVF), a potentially deadly vector-borne disease found in Eastern, Southern, and Western Africa, can be transmitted by several genera of mosquito and by other insect vectors. Work on RVF has shown that outbreaks are almost always preceded by prolonged periods of excessive rainfall in savannah ecosystems (Anyamba, Linthicum, Mahoney, Tucker, & Kelley, 2002; Davies, Linthicum, & James, 1985; Linthicum, Britch, & Anyamba, 2016), indicating that there is potential for early warning based on rainfall, or vegetation, or large-scale sea-surface temperature monitoring. Indeed, successful warning systems have been developed on the basis of these variables, with demonstrated predictive associations between ENSO and RVF outbreaks (Anyamba et al., 2009, 2012) and an operational system in place based on NDVI anomalies. The proposed mechanism underlying NDVI-based warning is that enhanced vegetation promotes survival of Aedes mosquitoes (Linthicum, Bailey, Davies, & Tucker, 1987; Tucker, Hielkema, & Roffey, 1985) and potentially other insect vectors as well. The details of this mechanism, however, are not fully described and could vary by insect species. Related work on RVF in South Africa has suggested that EWS can be developed using rainfall and soil moisture monitoring (Williams, Malherbe, Weepener, Majiwa, & Swanepoel, 2016), due to a similar proposed ecological mechanism. Bubonic plague, a tickborne zoonotic disease that is a health threat in East and Southern Africa, has also been found to show time-lagged relationships with rainfall variability (Moore et al., 2012), with possible applicability to EWS. In this case, the ecological mechanism could be related either to the population dynamics of the rodents that serve as hosts for the bacteria or to direct climate influence on tick survival and reproduction.

Waterborne Disease

Cholera, a waterborne disease transmitted by the Vibrio cholerae bacterium, is another climate-sensitive infectious disease that places a significant health burden on Africa. In 2014, Africa accounted for 55% of global suspected cholera cases reported to the WHO Global Health Observatory. In addition, the case fatality ratio for cholera in Africa is approximately 2%, which is double the threshold used by the WHO when assessing effective case management systems (Mengel, Delrieu, Heyerdahl, & Gessner, 2014). The actual number of cholera cases and deaths is difficult to judge on account of underreporting, but even the reported numbers suggest that the disease causes tens of thousands of deaths in Africa each year. In some regions, including the East African lakes and some coastal locations, the disease is present year round with regular seasonal peaks, while in others it appears in sporadic epidemics. One notable characteristic of cholera in Africa is that roughly three quarters of all reported cases occurred in inland regions (Rebaudet, Sudre, Faucher, & Piarroux, 2013b), with the largest number of cases found in the East African Great Lakes region and around Lake Chad. The disease burden is also significant in some coastal areas (Rebaudet, Sudre, Faucher, & Piarroux, 2013a).

Climate variability can affect V. cholerae through its influence on aquatic conditions, which are relevant to bacteria and to the plants and animals with which they associate. Climate—particularly heavy precipitation—also influences human behavior and systems relevant to cholera transmission. Wet conditions lead to outbreaks for multiple reasons, including the potential for the mixing of human waste with drinking water supplies during flooding. In many regions of Africa, cholera transmission is known to peak during the rainy season (Bompangue et al., 2009; Colombo, Francisco, Ferreira, Rubino, & Cappuccinelli, 1993; de Magny, Guégan, Petit, & Cazelles, 2007; Schaetti et al., 2009), and extreme precipitation events are often associated with epidemic outbreaks (de Magny et al., 2012; Guevart et al., 2006; Sasaki, Suzuki, Fujino, Kimura, & Cheelo, 2009). Dry conditions can also enhance transmission by limiting access to water for sanitation or to safe drinking water sources (Lawoyin, Ogunbodede, Olumide, & Onadeko, 1999; Tauxe, Holmberg, Dodin, Wells, & Blake, 1988). Warmer air temperatures have been associated with increased cholera risk in southeastern Africa, due to the relationship between air temperature and water temperature (Paz, 2009; Trærup, Ortiz, & Markandya, 2011).

These complex climate responses, combined with limited understanding of cholera ecology and reporting limitations, have made it difficult to establish a unifying framework for climate-based EWS or for projections of climate change impacts on cholera. Furthermore, cholera epidemics are ultimately a product of failed human infrastructure that allows for fecal–oral transmission of the disease. As such, the most damaging outbreaks often occur in cities or large refugee camps where climate might play some role as a trigger or stressor, but where epidemic dynamics are driven by nonclimate factors. Nevertheless, the relationship between climate variability and cholera outbreaks is robust and potentially predictive. El Niño conditions have been associated with increased cholera at continental scale (Griffith, Kelly-Hope, & Miller, 2006), with strong associations found in specific regions where El Niño leads to high rainfall and flooding (Nkoko et al., 2011; Olago et al., 2007). In general, El Niño events appear to shift the cholera burden to continental East Africa, primarily owing to shifts in rainfall and associated ecological parameters, but the details of the pattern are complex. A relationship between cholera episodes and Indian Ocean variability has also been found in parts of West Africa (de Magny et al., 2007), while outbreaks in Southern Africa have been associated with both large-scale (Paz, 2009) and local (Mendelsohn & Dawson, 2008) sea-surface temperatures (SSTs). Large-scale SST anomalies influence cholera risk through climate teleconnections, while local SST variability can alter coastal ecological systems, influencing V. cholerae concentrations relevant to coastal populations, as well as local weather.

The future of cholera in Africa most likely depends more on sanitation and health infrastructure than it does on climate change. Nevertheless, projections for wetter conditions in the East African Great Lakes region and for increased variability and frequency of hydrological extremes over much of the continent (Niang et al., 2014) do suggest the potential for more frequent climate triggers of cholera outbreaks.

Other Infectious Disease

Meningococcal meningitis is another climate-sensitive infectious disease that poses a significant health risk in Africa. The 21-country African “meningitis belt” stretching from Senegal to Ethoipia has the highest attack and fatality rates for bacterial meningitis in the world (Jusot et al., 2016). Infection causes inflammation of the membranes that cover the brain and spinal cord and can cause hearing loss, brain damage, and death. The fatality rate, when untreated, is 50% (WHO, 2015). Within the meningitis belt, disease risk follows a regular seasonal cycle, with transmission highest during the dry season. There is also significant interannual variability, with severe outbreaks occurring every 7 to 14 years (WHO, 2015). Variability appears to be associated with both temperature and dust load. Meningitis is transmitted from person to person through exchange of droplets of respiratory or throat secretions. Transmission may be elevated under dry and dusty conditions because damage to the nasopharyngeal mucosa increases susceptibility to infection; because heat and inhaled dust promote expression of bacterial virulence factors; and/or because of mediating social dynamics that are sensitive to weather conditions (Ceccato et al., 2014; Jusot et al., 2016; Pandya et al., 2015; WHO, 2015).

The climate influence on meningitis transmission in Africa has been reported for some time (Cheesbrough, Morse, & Green, 1995; Lapeyssonnie, 1963). A number of studies and initiatives have attempted to apply this knowledge to predictive modeling and EWS (Agier et al., 2013; Thomson et al., 2013). Studies have consistently found that elevated temperatures and dust load are associated with increased transmission and that these variables have predictive potential (Abdussalam, Monaghan, Dukić et al., 2014; Agier et al., 2012; Ceccato et al., 2014; Jusot et al., 2016; Oluwole, 2015). Relative humidity is also an important predictor (Abdussalam, Monaghan, Dukić et al., 2014; Pandya et al., 2015), and humidity-based models can skillfully predict the end of seasonal outbreaks (Pandya et al., 2015). Predicting the end of an outbreak is quite useful, as it can inform reactive vaccination campaigns. Interannual variability in meningitis cases has also been tied to large-scale climate variability, including ENSO and the Pacific Decadal Oscillation (Oluwole, 2015), due to the influence that these large-scale climate patterns have on temperature, moisture conditions, and winds. Warming under climate change may extend the transmission season and increase the total number of cases in the meningitis belt (Abdussalam, Monaghan, Steinhoff et al., 2014).

For a number of other high-profile diseases, climate is a proposed but uncertain driver of risk. Ebola, for example, has been associated with climate variability: outbreaks have been found to occur when wet years follow a period of several dry years (Tucker et al., 2002), possibly because an increase in the population of reservoir species. Transmission to humans is then thought to be more likely at the conclusion of the wet season, when there is a dramatic transition from wet to dry conditions that causes animals to crowd together around available resources (Pinzon et al., 2004). But the significance of this association is unknown, and the potential to use climate information to predict or project risk has not yet been realized.

A number of arboviruses transmitted by Aedes mosquitoes are present in Africa and are of global concern, including dengue, yellow fever, zika, and chikungunya. Within Africa, however, the health burden of severe dengue cases is poorly characterized (Jaenisch et al., 2014). This could well be due to underreporting, but it is also possible that disease ecology or human genetics have acted to limit its impact (Sierra, Kouri, & Guzmán, 2007). Like dengue, chikungunya and zika are transmitted by Aedes aegypti and Aedes albopictus mosquitoes. Both originated in Africa and are of great concern owing to their rapid spread into other parts of the world. Some projections suggest that Aedes aegypti and Aedes albopictus ranges in Africa could expand under climate change (Campbell et al., 2015), and urbanization is a risk factor for any disease transmitted by the urban-dwelling Aedes mosquitoes. However, before one can draw conclusions about potential changes due to climate change, more must be learned about the present climate sensitivity of dengue, zika, and chikungunya in Africa.

This selective review of climate impact on infectious disease in Africa is not intended to be comprehensive. Rather, it provides an overview of observed and hypothesized climate links for representative vector-borne, waterborne, and directly transmitted diseases. The links are complex, on account of interacting ecological factors and disease dynamics, and they can be obscure owing to lack of relevant ecological data and underreporting or biased reporting of human cases. Nevertheless, robust climate associations with disease risk have been demonstrated for many diseases at the research level, and a number of EWS have been established. The ability to translate these EWS into decision-relevant information systems is an area of active work (Ceccato et al., 2014; Pandya et al., 2015; Thomson et al., 2014). Projections of future disease burden under climate change can be made on the basis of known disease sensitivities. Such projections can inform long-term planning for emerging health risks, evaluated relative to current conditions. At the same time, the rapid pace of social and economic change in most of Africa suggests that changing human conditions will be the largest driver of changes in infectious disease risk in coming decades.

Floods

The difficulty of distinguishing climatic from social factors clearly applies in the case of floods. Floods can be responsible for more than 80% of annual natural disaster fatalities in Africa (Hoedjes et al., 2014), and the number of reported fatalities due to flooding has increased dramatically in Africa over the past half century, from fewer than 2,000 in the period 1950–1969 to over 14,000 in the period 1990–2009 (Di Baldassarre et al., 2010; Guha-Sapir, Below, & Hoyois, 2015). But numerous studies have shown that it is difficult to attribute variability or trends in the number of flood deaths to climate relative to other human factors, including demographic change, land-use change, and river management (Bates, Kundzewicz, Wu, & Palutikof, 2008; Blöschl & Montanari, 2010; Kundzewicz et al., 2005). In an analysis of flood trends in river basins across Africa, Di Baldassarre et al. (2010) found no significant trend in flood magnitude in Africa and concluded that the large trend in health impacts is dominated by changes in vulnerability patterns, most notably unplanned urbanization that has placed growing populations into informal settlements in floodprone cities such as Lusaka (Zambia), Dakar (Senegal), Alexandria (Egypt), and Ouagadougou (Burkina Faso).

At meteorological to interannual timescales, however, flood occurrence and the potential for health impacts are strongly related to rainfall variability. Studies have shown evidence of an ENSO signal on flood disasters across Africa (Li, Chai, Yang, & Li, 2016), and flood EWS ranging from daily to seasonal scale are an area of significant interest. Numerous examples of flood warning systems have been developed at national and river basin scale in Africa (Gumbricht, Wolski, Frost, & McCarthy, 2004; Haile, Tefera, & Rientjes, 2016; Thiemig et al., 2010; Trambauer et al., 2015). There is also interest in systems that could be generalized to continental scale and that could be used in countries that lack the hydrometeorological forecasting capacity to maintain their own flood systems (Jubach & Tokar, 2016). These systems employ satellite data and model-based forecasts to issue warnings ranging from flash flood timescales (Hoedjes et al., 2014) to medium-range hydrological forecasts (Thielen-del Pozo et al., 2015; Thiemig, Bisselink, Pappenberger, & Thielen, 2015), to seasonal flood risk outlooks (Braman et al., 2013). As with any climate services effort, the forecast is only one component of an effective information system; enabling policy environments and community engagement critical to the success of the system (Hellmuth, Moorhead, Thomson, & Williams, 2007; Jubach & Tokar, 2016).

Projections of flood casualties under changing climate conditions are highly uncertain. Floods themselves are difficult to project since they are influenced by land and water management as well as by meteorology (Field, 2012). Accounting for social and economic factors that drive vulnerability is also fraught with uncertainty. The one point in which there is some confidence is that increases in precipitation extremes would be expected to result in increased flood risk. Climate projections indicate that high-intensity rainfall will become more frequent over much of Africa, indicating that there is a potential for an increase in floods, all else being equal.

Heat

Temperature extremes—both heat waves and cold snaps—are also known to impact human health. With few exceptions, however, studies of the health burden of extreme temperatures have focused on the midlatitudes. Only in the face of climate change and a number of high-profile heat wave events in tropical countries has attention turned to the potential for temperature extremes, particularly high-temperature extremes, to pose a health risk in Africa. Health risks associated with high temperatures, including impacts on blood viscosity and cardiac output, are highest for the elderly and people who suffer from existing health conditions, and they can also impact children (Smith et al., 2014). The limited number of studies that have investigated the health impacts of temperature extremes in Africa have found that both high temperatures (Azongo, Awine, Wak, Binka, & Oduro, 2012; Diboulo et al., 2012; Egondi et al., 2012) and cold temperatures (Egondi et al., 2012; Egondi, Kyobutungi, & Rocklöv, 2015; Mrema, Shamte, Selemani, & Masanja, 2012) can have significant impacts on mortality and morbidity in Africa. In a systematic review of temperature impacts on mortality in sub-Saharan Africa, Amegah et al. (2016) found, with moderate confidence, that high temperatures cause an increase in all-cause mortality. However, they note that the literature is thin and that it is difficult to assess confidence in specific health impacts on asthma, respiratory disease, undernutrition, and most infectious diseases. A broader review of temperature impacts on health across the tropics came to similar conclusions (Burkart et al., 2014).

The risks of heat exposure may be greatest for populations living in urban informal settlements, as urban heat island effects can cause cities to be several degrees hotter than surrounding areas, and those living in informal settlements tend not to have access to climate-controlled environments or to rapid medical care (Egondi et al., 2012; Scovronick, Lloyd, & Kovats, 2015). Unusually cold temperatures can also pose a health risk to these populations (Egondi et al., 2015). Climate change projections point to an increase in the number of days with high Apparent Temperature (a humidity and wind speed-adjusted heat measure) across much of Africa, particularly in the East African Highlands, with potential consequences for health (Garland et al., 2015).

Dust

Finally, climate extremes and trends have the potential to impact dust emission and transport. Dust emission is sensitive to soil moisture and vegetation status, which are, in turn, influenced by rainfall and temperature, as well as by wind speed. Higher atmospheric dust load has been associated with relatively direct health impacts, such as transport accidents and respiratory illnesses, and with more highly mediated processes, such as allergies and disease risk (Goudie, 2009). There is significant interannual and decadal scale variability in dust load in dust-affected regions of Africa, including the Sahel and portions of Southern Africa, but long-term trends are either unclear or not fully explained (Goudie, 2014). Dust exposure has been linked to lung cancer in Northern Africa (Giannadaki, Pozzer, & Lelieveld, 2014) as well as to the meningitis outbreaks, as described in the preceding section on infectious diseases (see Ecologically mediated: Infectious Disease). African-sourced dust is also known to impact asthma, respiratory complaint, and cardiovascular disease in southern European countries (Karanasiou et al., 2012), suggesting that it likely has an impact on health in Africa as well, though targeted studies are still required (De Longueville, Hountondji, Henry, & Ozer, 2010; De Longueville, Ozer, Doumbia, & Henry, 2013).

Application of Early Warning Systems

The idea that observations and predictions of climate can be applied to anticipate disease is ancient (Sargent, 1982). The aspiration to predict or control floods and other climate disasters that bring physically mediated harm is perhaps even more ancient: Utnapishtim, the hero of the flood in the Epic of Gilgamesh, is granted eternal life after receiving early warning of the flood (albeit with help from the gods) and responding appropriately; Yu’s successful control of the catastrophic Great Flood of Gun-Yu in Chinese mythology leads to the establishment of the Xia dynasty. The importance of famine early warning has roots in the remarkably effective predictions (and governance!) found in the biblical story of Joseph and Pharaoh, which spared Egypt from famine and made it a place of refuge across the region.

In the modern era, conceptual disease early warning systems based on climate variability have been in existence for close to a century, including work in the 1920s by C. A. Gill on malaria in India and Leonard Rogers on multiple diseases in India and beyond (Gill, 1923; Rogers, 1923, 1925, 1926). These early studies recognized clear patterns between climate variability and disease incidence, and they also appreciated the need to consider climate variables in the context of other social and environmental factors: Gill’s malaria risk model, for example, included economic and epidemiological observations in addition to rainfall. Over the past decade, numerous disease early warning systems for Africa have been proposed, including several that disseminate risk estimates on a real-time basis (Thomson et al., 2014). Several of these systems are described in the infectious disease section of this review (see Ecologically mediated: Infectious Disease). Skillful systems for predicting crop yields have also existed for many decades, including work in India in the 1940s and 1950s that was endorsed by the Food and Agriculture Organization of the United Nations as a model for agricultural forecasts and subsequent statistically based objective yield forecast work by the United States Department of Agriculture in the 1960s and 1970s (Abreu & Riberas, 2008) that was transferred to a number of other countries. The effort to apply crop yield projections to food security outlooks gained momentum with the establishment of FEWS NET in the 1980s. Meanwhile, storm, flood, and heat-wave prediction has traditionally been handled by meteorological service agencies. Mature and operational systems have existed in developed countries and in high-profile cities and river basins throughout the world for some time.

Despite this long history, operational use of health early warning systems—particularly for highly mediated health risks like hunger and infectious disease—has been relatively limited across the world, and the problem is particularly acute in Africa. There are a number of reasons for this situation. In the case of disease early warning, implementation of skillful and effective systems depends on the presence of reliable data for both predictor variables (climate and mediating factors) and disease incidence. The lack of reliable, consistent, and available data records in most African countries, particularly for human health outcomes, makes it difficult to develop, implement, or assess proposed disease prediction systems. The WHO has also noted that most conceptual EWS have been developed for specific case studies, with limited budget and in an academic rather than an operational context. This makes them ill equipped for widespread operational use, and it also means that there is a shortage of established metrics for evaluating the accuracy and utility of proposed warning systems.

Institutional stovepipes are another recognized challenge, as meteorological services and health ministries have little history of collaboration. Health officials lack expertise in climate data and probabilistic forecasts and usually have no experience applying them to operational activities. Providers of climate information, meanwhile, have little exposure to decision making on health matters and may provide products that are not well suited for health risk monitoring or prediction. Efforts such as the establishment of the joint World Meteorological Organization–World Health Organization (WMO–WHO) Joint Office for Climate and Health and the broader initiative for a Global Framework for Climate Services (GFCS) are important steps to overcome these institutional barriers.

Ultimately, however, even a seamless and skillful forecasting system must be tied to a capacity for effective intervention. In the case of epidemic disease in many parts of Africa, communication systems, effective and realizable risk reduction techniques, and the ability to distribute vaccines and treatments in a timely manner must all be improved alongside the development of climate-informed early warning systems.

Similarly, in the case of food security, FEWS NET and other systems successfully predicted several damaging climate-driven food crises months in advance. This includes the 2010–2011 drought crisis in the Horn of Africa and the 2015 drought affecting Ethiopia. It is difficult to quantify the realized benefit that these predictions provided. Nevertheless, there is consensus that skillful early warning is not being translated to effective early action in the way that it should be (Bailey, 2012; Funk, 2011). A critical factor in this failure is the presence of perverse incentives: no single actor or institution is accountable for preventing a crisis, whereas every decision maker is held accountable for multiple competing priorities other than crisis prevention. Given that the prediction of a food crisis always comes with some uncertainty, the tendency is to delay action until it is absolutely required (Bailey, 2012).

This tendency is institutionalized by the fact that it is extremely difficult for humanitarian organizations to raise funds or devote resources on the basis of forecasts, since no emergency is yet underway. Crisis response institutions also tend to be risk averse to the possibility of “acting in vain”—allocating resources and warnings to a crisis that never materializes (Coughlan de Perez et al., 2015a). This aversion stems from both concrete financial risks and the potential for negative repercussions from donors and partner institutions in the case of a false alarm. It applies to any form of prediction-based health warning, but it has been most damaging in terms of lost opportunity in the area of food insecurity in Africa. The problem is amplified by the fact that food security in many countries of Africa is still viewed as a responsibility of the international aid community. This further diffuses responsibility and means that operational actors—a combination of UN agencies, nongovernmental organizations (NGOs), and partner government agencies from donor and recipient nations—must coordinate decision making and efforts. It also means that the political priorities of donor countries can be decisive and that well-meaning but cash-limited NGOs might need a crisis to materialize in order to motivate donations (Bailey, 2012; Fink & Redaelli, 2011; Olsen, Carstensen, & Høyen, 2003).

Improved application of climate information to health early warning in Africa, then, requires both improvements in prediction systems and changes in the way that predictions are used to inform action. In the area of improved prediction, there is a need for (1) improved data collection and dissemination, particularly in the area of health outcomes; (2) more complete integration of human systems into climate-based warnings; and (3) rigorous testing for predictive capability and value of information in the decision context.

The need for improved data collection is widely recognized. Improved surveillance and monitoring systems for disease and undernutrition can be used to train better predictive models and to drive real-time risk prediction systems that use current incidence as one of the predictors of future burden (as is widely the case in epidemics modeling). Ancillary demographic information is also critical, as a disease case count is most meaningful if it is reported alongside estimates of population size, geographic distribution, and characteristics. The proliferation of cell phones and other communication technology has been pointed to as an opportunity area for collecting this information, though these systems need to be designed in a manner that encourages participation, ensures data quality, and does not depend on cutting-edge or data-intensive technologies that are presently inaccessible in many poor communities.

Integration of human systems into climate-based prediction is also important. While the incorporation of economic information was a feature of even the earliest disease prediction systems, and sophisticated systems like FEWS NET incorporate multiple forms of economic and demographic information as a matter of course, the majority of published disease prediction models for Africa still treat human social dynamics in a simplified manner, if they are addressed at all. It is more typical to see a correlation between rainfall or vegetation and case count, without consideration for livelihood patterns, access to health services, transportation networks, and other factors that can influence the magnitude of an impending epidemic and that could be relevant to preparedness and response efforts. Combining the best-available tools of climate analysis, including seasonal forecasts, remote sensing, and land data assimilation systems, with models of human settlement, mobility, and activity is both a substantial research challenge and an opportunity to better understand and predict health vulnerabilities.

Testing for the predictive skill and value of information represents its own set of challenges. The basic concept of evaluating models for out-of-sample predictive skill rather than just descriptive fit is straightforward from an analytical perspective, given sufficient data, even if it is not always practiced (Shortridge, Falconi, Zaitchik, & Guikema, 2015). It is critically important when proposing models for forecasts, since in-sample fit and predictive skill do not always correlate. Maximizing the value of information, however, requires collaborative design across disciplines. The value of information in an early warning system depends both on the accuracy, reliability, and time horizon of a prediction and on the information needs of decision makers. Seasonal climate forecasts, for example, are frequently disseminated in terms of terciles—above-average, average, or below-average rainfall. This is the primary approach used by the African Regional Climate Outlook Forums, even though it has long been recognized that the approach does not meet the needs of decision makers (Patt et al., 2007).

Collaborative design and evaluation of forecast products, communication of uncertainty in a meaningful way, and the use of “boundary organizations” specialized in spanning the worlds of climate information and decision making can be critical to the success of a forecast system (Buizer, Jacobs, & Cash, 2016). Working with decision makers to tie forecasts to specific actions is also critical. In this context, the use of “serious games” forecast-based decision-making exercises has proved successful in Africa and elsewhere when applied to health-relevant decision making in disaster preparedness and resource management (Suarez, Mendler de Suarez, Koelle, & Boykoff, 2014). This need for collaboration between climate information providers and decision makers is a pillar of the Climate Services movement (Hewitt, Mason, & Walland, 2012).

On the institutional side, there is a recognized need to establish decision-making processes that are capable of acting on forecast information. This includes changing the institutional cultures of accountability in a way that will encourage forecast-based decision making (Bailey, 2012), combined with a broader cultural change among donors, recipients, and operational institutions to be more accepting of false alarms. This second requirement is particularly challenging, as the negative impact of false alarms has been documented in a number of contexts (Dillon & Tinsley, 2008). Models of successful forecast-based action do exist, including the forecast-based financing approach for disaster preparedness (Coughlan de Perez et al., 2015a), use of climate-based food insecurity models within a weather index insurance scheme to speed delivery of famine relief (Chantarat, Barrett, Mude, & Turvey, 2007), and integrating climate forecast information into health programming at the local scale (Coughlan de Perez et al., 2015b). These models depend on effective and accepted financing mechanisms and outreach activities. Success stories, such as the Red Cross experience using seasonal forecasts to enhance flood preparedness in West Africa in 2008 (Braman et al., 2013), are causing these approaches to gain traction.

Climate Change Adaptation

Climate change is having significant impacts on health across the world (Smith et al., 2014). In Africa, projected impacts on crop production and food security (see Socially-Mediated: Food Security) are a cause for significant concern, and changing patterns of infectious disease (see Ecologically Mediated: Infectious Disease) and climate extremes (see Physically Mediated: Floods, heat, and dust) also require adaptive action (McMichael, 2013; Patz & Hatch, 2014). Public health, however, has received relatively little focus in National Adaptation Plan of Action (NAPA) reports developed by Least Developed Countries for the United Nations Framework Convention on Climate Change; these plans have tended to focus on agriculture and energy sectors, with less targeted attention to the health sector (Doumbia et al., 2014).

In general, health-oriented climate change adaptation plans are hindered by the large uncertainties present both in climate change projections for Africa and in quantitative understanding of the links between climate change and highly mediated health outcomes. It is not at all clear, for example, that ecologically or socially mediated health dynamics will remain the same in a changing climate or that statistical relationships found to be robust for interannual climate variability will apply over longer timescales (Patz, Campbell-Lendrum, Holloway, & Foley, 2005). These challenges apply across the world, but in Africa they are compounded by the fact that climate change is occurring in the presence of rapid economic development and demographic transformation. In this context, projecting the health impacts of future climate change requires coupled natural–human systems models that account for cross-sectoral interactions. Few such models exist for the health sector. Just as importantly, risk assessment and communication frameworks are required to apply the findings of these inherently uncertain projections to decisions on adaptation investment.

This is a daunting challenge, but there are a number of opportunities for research to meet the needs of application. The WHO has identified some of these opportunities in its Framework for Public Health Adaptation to Climate Change in the Africa Region (Regional Committee for Africa of the World Health Organization, 2011), which includes an eight-point adaptation action plan based on (1) risk and capacity assessment, (2) capacity building, (3) integrated health and environment surveillance, (4) social mobilization, (5) health-oriented environmental management, (6) scaling up existing public health activities, (7) strengthening cross-sectoral partnerships, and (8) enhanced research. The influence of this framework is still to be determined, and funds will be required for its implementation in most sub-Saharan African countries.

Spatial scale is another critical consideration. When evaluating health risks such as undernutrition, studies are required at the household scale, but these studies must be conducted in a manner that is scalable to large populations or, at a minimum, designed with an awareness of complementary studies and data-sharing standards that allow for meta-analysis across locally oriented studies (Phalkey et al., 2015). Placing health studies in ecological context is also important. Health statistics are most commonly aggregated to political units, but climate-related health burdens are, in many cases, better understood as a function of climate forcing imposed on natural and agricultural ecosystems. For this reason, ecosystem-based adaptation strategies offer an opportunity for translating research to adaptation action at scale (Doumbia et al., 2014).

The quantification and communication of risk under rapidly changing climatic and socioeconomic conditions is a grand challenge for all climate change adaptation strategies in developing countries. Tools such as Robust Decision Making (RDM) analysis have been applied successfully to characterize the robustness of infrastructure and adaptation strategies in the energy and environmental resources sectors (Lempert, 2011). RDM is founded on the principle that systems should be designed to be robust to uncertain future conditions rather than optimized for a specific existing or projected climate (Lempert, Groves, Popper, & Bankes, 2006). In the health sector in Africa, where there is deep uncertainty regarding the future dynamics of malnutrition, infectious disease, and ability to cope with climate extremes, frameworks like RDM offer an opportunity to identify risks that may be missed or understated in projections focused on mean climate projections rather than the range of potential outcomes.