Study area
The study was conducted in SSA. This region accounts for 90 % of the global malaria burden, i.e., 174 million cases annually [31]. Currently, malaria transmission is generally stable in western and central Africa, unstable in much of eastern Africa and unstable or absent in southern Africa [32]. In SSA, intensive malaria control efforts since the beginning of the twenty-first century, have resulted in a reduction of malaria-associated mortality by 45 % and 1.6 million fewer malaria deaths occurred between 2001 and 2015 [33]. Encouraged by this progress, the World Health Organization (WHO) has set ambitious goals to reduce the global malaria burden by 90 % by 2030 when compared to 2015 [33].
Plasmodium falciparum is the most prevalent and most fatal malaria parasite in the region [31]. Anopheles gambiae and Anopheles funestus are the predominant malaria vector species in wet and humid areas, while Anopheles arabiensis is the most common vector in drier climates [34, 35]. The variation in intensity of malaria transmission across SSA is partly the result of the relative abundance of these species across different ecological settings [36].
Data source
The present study used four major data sources: (1) previously published global maps that show the future distribution of malaria under RCP 2.6 and 8.5 [27]; (2) databases of currently existing large dams; (3) projected population for the 2020s, 2050s and 2080s; and, (4) the Malaria Atlas Project (MAP) database of malaria incidence. These were used to estimate the population at risk around dams in areas of stable and unstable malaria transmission and to determine the possible impact of climate change on malaria transmission in the vicinity of reservoirs behind large dams.
Data on existing African dams
The distribution of existing dams in SSA was mapped using the geo-referenced locations of individual dams in the Food and Agriculture Organization (FAO) African dam database [37] and the International Rivers database [38]. Data on water storage capacity, dam height and reservoir surface area were obtained from the World Register of Dams [39] and the Global Reservoirs and Dams (GRanD) database [40]. Locations and parameters of additional dams were obtained from a number of journal articles, project reports and dissertations. Overall, geo-referenced locations and dam parameters were obtained for a total of 1268 existing dams (out of an estimated total of over 2000 [40]) in SSA (see Additional file 1).
Estimating reservoir perimeters
Data on reservoir perimeter are necessary to estimate the population at risk of malaria due to a dam. However, these data are not available for most dams, so reservoir perimeter was estimated using a method adapted from Keiser et al. [10]. The method estimated the perimeter based on dam and reservoir characteristics and an assumed rectangular shape. (For more details see Kibret et al. [11]).
Malaria data
The MAP database [41] assembled all existing malaria surveys to produce a global spatiotemporal malaria dataset at a 1 × 1-km spatial resolution. The data were interpolated within a Bayesian space–time geo-statistical framework to predict the average P. falciparum infection rate (PfIR) in the 2020s, 2050s and 2080s using biological models based on estimated average temperature increases in future years. These predictions were calibrated using the 2000–2010 malaria dataset from literature and the WHO [42]. This database was used to estimate the PfIR around dam-associated reservoirs in SSA for the 2020s, 2050s and 2080s in both RCP 2.6 and 8.5.
Non-linear relationships between malaria incidence and climatic predictor variables (i.e., temperature and rainfall) were constructed using a genetic programming method, to predict the spatial distributions of malaria under climate change scenarios. Spatiotemporal scan analysis was performed to identify and quantify the spatiotemporal clustering scales of malaria incidence to empirically estimate the relationship between malaria incidence and climatic factors using the output of the MAP Bayesian statistical model, which combines malaria survey data with climate predictors to provide a gridded malaria incidence for each of the future time intervals. Details of the spatiotemporal malaria incidence prediction procedures are documented in MAP database [41].
The classification described by Gething et al. [43] was used to characterize the epidemiological settings in which the dams were located. These were defined as:
-
stable transmission in areas with annual PfIR greater than 0.1 cases per 1000 population;
-
unstable transmission in areas with annual PfIR between 0 and 0.1 cases per 1000 population; and
-
no malaria in areas having zero annual PfIR.
Future climate scenarios
Caminade et al. [27] developed malaria maps using a multi-malaria model, which comprised five well-known, process-based, malaria models (i.e., LMM_RO, MIASMA, VECTRI, UMEA, MARA) driven by climate outputs from five global circulation models (GCMs) (i.e., HadGem2-ES, IPSL-CM5A-LR, MIROC-ESM-CHEM, GFDL-ESM2M, NorESM1-M). These all employed bias-corrected temperature and rainfall data from the coupled model intercomparison project phase 5 (CMIP5). A common metric (i.e., length of malaria transmission season) was used to compare all malaria models. Separate maps were produced for each of the four Intergovernmental Panel on Climate Change (IPCC) RCPs: (i.e., RCP 2.6, RCP 4.5, RCP 6.0, RCP 8.5), for three future time intervals (i.e., 2020s, 2050s, 2080s). RCPs are time and space-dependent trajectories of concentrations of greenhouse gases and pollutants resulting from human activities, including changes in land use. RCPs provide a quantitative description of concentrations of the climate change pollutants in the atmosphere over time, as well as their radiative forcing in 2100 (for example, RCP 6 achieves an overall impact of 6 W per sq. m by 2100). Radiative forcing, expressed as Watts per sq. m, is the additional energy taken up by the Earth system due to the enhanced greenhouse effect. The four RCPs illustrate the range of year-2100 radiative forcing values found in the literature, i.e., from 2.6 to 8.5 W per sq. m [44]. Details of the selection process and parametrization are documented in Caminade et al. [27].
Among the four RCPs, two (RCP 2.6 and RCP 8.5) were selected for the current study in order to show how malaria distribution changes between the lowest (i.e., RCP 2.6) and the highest (RCP 8.5). The maps for each RCP and time interval were overlaid separately on the malaria and dam distribution maps to develop future malaria distribution scenarios around dams in the 2020s, 2050s and 2080s.
Population data
1 × 1-km gridded projected future SSA annual population data were obtained from the International Institute for Applied Systems Analysis (IIASA) for RCP 2.6 and 8.5, for the three future time intervals [45]. In each RCP, the average population in the decade was determined and the data were imported to ArcGIS and overlaid on the dam-malaria-climate distribution maps. The population around dams (i.e., within 5-km radius) was then calculated separately for zones of stable, unstable and no malaria transmission for each time interval and climate scenario. The impact of a dam on malaria was assumed to be negligible beyond 5 km due to mosquitoes’ limited flight range [46]. Also, the population concentration around reservoir relative to non-reservoir communities was assumed constant between 2020s and 2080s.
Statistical analysis
Mapping dams and malaria in climate change scenarios
The locations of dams were overlaid on the malaria-climate scenario maps to develop separate maps for RCP 2.6 and RCP 8.5 in the 2020s, 2050s and 2080s using ArcGIS. For both RCPs, the number of dams in each malaria transmission stability zone (i.e., stable, unstable and no malaria) was counted for each future time interval.
Estimating population at risk around dams
Future populations were analysed separately for the two RCPs. It should be noted that future population projections vary between RCPs: the SSA population in the 2080s is estimated to be 1.55 billion and 1.65 billion in RCP 2.6 and 8.5, respectively. The populations at risk of malaria (<5 km) around existing dam-associated reservoirs in areas supporting stable and unstable transmission were estimated for the two RCPs for each of the three time intervals.
Estimating future malaria incidence around reservoirs
Using MAP spatiotemporal malaria database, annual PfIRs (hereafter called malaria incidence) for reservoir communities (i.e., located within 5 km of a reservoir shoreline) and non-reservoir communities (i.e., located 5–10 km from a reservoir-shoreline) were estimated for the two RCPs for each of the three time intervals. The annual malaria incidences in reservoir and non-reservoir communities were compared in stable and unstable areas for each RCP in each time interval. Previously documented data of annual malaria incidence for 2010 [11] were used as a baseline against which to measure the degree of future change in malaria incidence. Chi square tests were employed to compare the differences in malaria incidence between malaria stability zones and across time intervals. The same test was carried out to determine whether the difference in malaria incidence between reservoir and non-reservoir communities was significant within each malaria stability zone. Statistical analyses were done using statistical software, SPSS version 22 (SPSS Inc, Chicago, IL, USA). The level of significance was determined at the 95 % confidence interval (P = 0.05).
Estimating future malaria cases associated with dams
For each time interval, projected annual malaria incidence was multiplied by the population within 5 km and 5–10 km from reservoirs, to estimate the number of malaria cases in reservoir and non-reservoir communities, respectively, in areas of stable and unstable transmission. The number of annual malaria cases attributable to dams was estimated by calculating the difference in the number of annual malaria cases for communities less than 5 km and for communities 5–10 km from the reservoir, allowing for differences in population. The proportion (%) of cases attributable to dams relative to the total malaria cases (<5 km) was also calculated.
Identifying the increase in malaria cases attributable only to population increase
To distinguish the increased malaria due to population growth from that associated with climate change, the number of malaria cases due solely to population growth was determined. The 2010 malaria incidence rate was applied to population increases (i.e., the change in population between the 2010 and the average population in future time intervals of the 2020s, 2050s and 2080s for each RCP) to calculate the number of additional malaria cases due to population increases in each of the three periods for each RCP.
Identifying the increase in malaria cases attributable only to climate change
For each RCP, the number of cases in 2010 and the number of additional cases associated with population growth were subtracted from the total malaria cases around dams in future time intervals. The results were considered to be the number of cases attributable solely to climate change. The difference in socio-economic and other environmental factors was assumed to be minimal between reservoir and non-reservoir communities. Possible impacts of further malaria control interventions and other non-climate factors (e.g., socio-economic development) in the future were not captured.