Projected Impacts of Climate Change on Environmental Suitability for Malaria Transmission in West Africa

Background: Climate change is expected to affect the distribution of environmental suitability for malaria transmission by altering temperature and rainfall patterns; however, the local and global impacts of climate change on malaria transmission are uncertain. Objective: We assessed the effect of climate change on malaria transmission in West Africa. Methods: We coupled a detailed mechanistic hydrology and entomology model with climate projections from general circulation models (GCMs) to predict changes in vectorial capacity, an indication of the risk of human malaria infections, resulting from changes in the availability of mosquito breeding sites and temperature-dependent development rates. Because there is strong disagreement in climate predictions from different GCMs, we focused on the GCM projections that produced the best and worst conditions for malaria transmission in each zone of the study area. Results: Simulation-based estimates suggest that in the desert fringes of the Sahara, vectorial capacity would increase under the worst-case scenario, but not enough to sustain transmission. In the transitional zone of the Sahel, climate change is predicted to decrease vectorial capacity. In the wetter regions to the south, our estimates suggest an increase in vectorial capacity under all scenarios. However, because malaria is already highly endemic among human populations in these regions, we expect that changes in malaria incidence would be small. Conclusion: Our findings highlight the importance of rainfall in shaping the impact of climate change on malaria transmission in future climates. Even under the GCM predictions most conducive to malaria transmission, we do not expect to see a significant increase in malaria prevalence in this region. Citation: Yamana TK, Eltahir EA. 2013. Projected impacts of climate change on environmental suitability for malaria transmission in West Africa. Environ Health Perspect 121:1179–1186; http://dx.doi.org/10.1289/ehp.1206174


Table of Contents
where T is the mean water temperature over the time interval (1 hour); p 25 oc is the development rate at 25°C giVen no temperature inactivation of critical enzyme; M-f"A is the enthalpy of activation of the catalyzed reaction; !!J!L and !!J!H are the enthalpy changes due inactivation of the enzyme from low and high temperature, respectively; To sL and To sH are the water temperatures at which 50% of the enzyme is inactivated by low and high temperature respectively; and R = 1.987cal/mol is the universal gas constant.

Methods: Alternate EIP formulation
In our analysis, we follow the majority of current malaria models (Craig et al. 1999;Ermert et al. 2012;Guerra et al. 2008) in using the well-established Detinova (1962) et al. 1995). However a contradictory study found that parasites were able to develop in wild strain mosquitoes at temperature of 30 o C and 32 o C, though with decreased survival (Okech et al. 2004).
Here, we consider the alternate non-linear curve proposed by Paaijmans (2009). The timescales for the temperature dependent processes for malaria transmission are shown in Figure S3. At temperatures above 30 o C, there is a significant difference between the two EIP curves; the Detinova curve continues to decrease while the Paaijmans curve increases rapidly. At temperatures above 34 o C, the Detinova curve gives an EIP of just over 6 days, while the Paaijmans curve indicates that transmission is blocked.

Methods: Disaggregation of CRU data into hourly rainfall time series
The environmental processes simulated by HYDREMATS require meteorological inputs at high temporal resolution. The mean baseline rainfall from CRU required disaggregation into hourly resolution before being used as an input. The spatio-temporal dissagregation of rainfall for hydrological applications is a well researched problem, and is often done using various statistical models parameterized by assumed or observed characteristics of finer scale rainfall events (for example Bo et al. 1994;Mackay et al. 2001;Margulis 2001;Segond et al. 2007). Here, we take advantage of high resolution satellite observations of rainfall from the Climate Prediction Center Morphing Technique (CMORPH) data set, which gives ~8km resolution rainfall data every 30 minutes (Joyce et al. 2004). After applying the bias-correction described in Yamana and Eltahir (2011), we use the hourly rainfall observations from CMORPH data at each village to disaggregate baseline CRU rainfall into realistic storm events. The result is an hourly rainfall time series with mean annual rainfall equal to long-term observations from CRU, and patterns of hourly rainfall observations from CMORPH.

ERA Interim data
Temperature, wind speed, wind direction, and radiation data were taken from the ERA Interim data set (Dee et al. 2011) for the grid cell containing each village being simulated; we assume uniform conditions within the 0.75 degree ERA grid cell. ERA Interim data were adjusted for HYDREMATS as follows. Wind speed was brought from 10 m to 2 m by assuming a logarithmic profile. Wind and radiation data were linearly extrapolated from the 3-hour resolution provided by ERA to the 1-hour resolution required by HYDREMATS.

8
A comparison of the diurnal temperature cycles given by ERA Interim and ground observations at three locations across ecoclimate zones in our study region (Banizoumbou, Niger; Agoufou, Mali; and Djougou, Benin) indicated that while the reanalysis data gave good estimates of the daily mean temperature, the diurnal range was underestimated. The regression coefficients below were calculated using the diurnal temperature ranges of the three ground stations over one year and applied to daily ERA temperature ranges: The maximum and minimum daily temperatures were then computed as the daily temperature mean from ERA plus or minus the corrected range divided by two. The hourly temperature was calculated assuming a sinusoidal curve during daylight hours and an exponential decrease between sunset and sunrise, as described in Paaijmans et al.(2009)

Vegetation and soil properties
The dominant vegetation type at each location was obtained from the University of Maryland Land Cover Classification (Hansen et al. 2000). Soil properties were taken from the Harmonized World Soil Database (FAO and ISRIC). A thin layer of low-permeability soil is included in the model to account for soil crusting that occurs throughout West Africa under cultivated conditions (Morin 1993).

Results: Alternate EIP formulation
Simulations using the alternate EIP (Paaijmans 2009) generally had lower D, and thus lower VC, throughout West Africa, with the effect being most pronounced in the hottest regions (Zones 1 and 2). While the magnitudes of the values are lowered, the overall observations in terms of seasonality, difference between zones, and differences between current and future climates remain largely unchanged.
When we average the changes over the seven-year simulation period, shown in Figure S4, the results are again very similar to our original findings. The main differences are in the dry-hot scenarios in Zones 4 and 5; while the warming leads to slight increases in vectorial capacity using Detinova equation for EIP, there is a slight decrease in VC using the Paaijmans equation.