GIS-Based Model for Mapping Malaria Risk under Climate Change Case Study: Burundi

Malaria is one of the largest problems threatening global public health that is expected to increase in the future under climate change due to associated warming and wetter conditions. This will exacerbate disease burden in Burundi as one of sub-Saharan African countries, where 2 million cases of malaria were reported in 2015. This highlights the need for developing a methodology for mapping malaria risk under climate change and delineating those regions that may potentially experience malaria epidemics in the future. Malaria transmission and distribution are generally determined by a wide range of climatic, topographic and socioeconomic factors. The paper in hand is intended to map malaria risk in Burundi under climate change up to 2050. For this purpose, a GIS-based model was developed for mapping malaria as a function of various climatic and topographic determinants of malaria. The developed GIS-model was used in mapping malaria risk under current climatic conditions. Thereafter, the produced risk map was validated compared to malaria morbidity data in Burundi at health district level. Finally, the GIS-model was applied to map malaria risk in the future under RCPs 2.6 and 8.5 scenarios up to 2050. It was found that about 34.6% and 44% of Burundi land surface will be highly vulnerable to malaria risk by 2050 under RCPs 2.6 and 8.5 scenario, respectively. Also, it was noted that such highly vulnerable areas are distributed mainly in northern parts of the country. The suggested GIS-based model for mapping malaria risk under climate change can contribute largely to more informed decision-making and policy making process in terms of planning for intervention and control malaria risk. This in turn can support reducing disease burden and improving resilience to climate change.


Introduction
Malaria represents one of the largest problems threatening global public health [1] with total cases exceeding 212 million worldwide in 2015, of which 90% were in Africa. Moreover, Malaria caused 429 thousand deaths globally in 2015 [2]. In this respect, it was estimate that Malaria was the seventh leading cause of death in low-income countries by 2015, leading to 34.5 causes of deaths per 100,000 of population on average in these countries [3]. As one of vector-borne disease systems, malaria is caused by Plasmodium parasites that transmitted by Anopheles mosquito. Malaria transmission and distribution are generally determined by a wide range of climatic, topographic and socioeconomic factors [4] [5]. While both climatic and topographic factors determine the suitable habitats for Plasmodium and Anopheles, socioeconomic factors are more relevant to adaptive capacities of individuals or communities and consequently determine their vulnerability to malaria.
A wide range of research work was undertaken to recognize the various factors underlying malaria risk in different countries [6]. The contribution of climate in malaria transmission was already recognized [7] and the association between epidemics of malaria and climate was evidenced in different regions worldwide in particular in sub-Saharan African countries [1] [8] [9], Latin America [10] and Asia [11]. Generally, malaria incidence increases with warm and wet conditions, where higher temperature leads to increasing malaria incidence as it contributes to shortening extrinsic incubation period of Anopheles mosquito [8] [12]. Generally, it was suggested that the optimum temperature for Anopheles mosquito and parasite development ranges between 20˚C and 30˚C and they are subjected to thermal death if temperature rises above 40˚C or drops below 16˚C [13]. Yet, it is worth mentioning that the impact of temperature on malaria cannot be fully explained only by mean temperature [14]. Actually, diurnal temperature range has significant impacts on malaria transmission, where more diurnal temperature range in relatively cold conditions, with mean temperature below 21˚C, may increase malaria transmission as it accelerates mosquito development. Meanwhile, more diurnal temperature range in warmer conditions, with mean temperature above 21˚C, may lead usually to decrease malaria transmission as it may slow-down mosquito development [14] [15]. Also, malaria incidence strongly associated with wet conditions, higher levels of rainfall [8] or relative humidity [4] that usually support Anopheles mosquito to complete their life cycle [10]. In this respect, 80 mm per month of rainfall for at least five months was suggested as a threshold limit that is sufficient to malaria transmission [4].
Moreover, malaria incidence is significantly affected by some topographic factors such as altitude, slope and land cover. In this respect, the negative relationship between the risk of malaria infection and altitude was well-recognized.
For instance, it was argued that malaria morbidity was significantly less prevalent in the highlands that experience relatively cold conditions compared to M. A. Hassaan et al. Journal of Geoscience and Environment Protection lowlands [16]. Such low temperature in high altitude limits parasite development and consequently restricts the spread of malaria there [12]. Similarly, malaria risk is negatively associated with slope, where breeding sites of mosquitos are more likely to be in gentle slopes and planes [17]. Usually, katabatic wind in steep slopes and a relatively cold temperature during nights restrict dispersion of mosquitos into higher altitudes [17] [18]. Also, malaria risk increases in some patterns of land cover that are dominated by wetlands, farmlands and regularly flooded land. Such type of land covers provides suitable sites for Anopheles mosquitos to complete their life cycle [10] [12] [19].
As a result of sensitivity of malaria transmission to climate and topography, malaria was defined as a highly climate sensitive disease [10] and an environmental related disease [20]. Such sensitivity, also, implies that the great potentials for mapping malaria risk through various climatic and topographic variables that determine malaria transmission. Recently, mapping malaria has been promoted by increasing data availability, improved insight into environmental covariates of malaria risk, advances in analytical techniques [21] and availability of affordable Geographical Information Systems (GIS) software [4]. The application of GIS was not restricted only to mapping malaria incidence but extended also to include assessing the relationship between malaria and underlying factors, developing malaria early warning systems [13] and planning for malaria control programs [22]. This was motivated by the nature of data on malaria incidence and underlying climatic, topographic and socioeconomic factors are spatial variables that can be modeled and manipulated through GIS due the great capabilities of such systems in terms of handling and analyzing spatial data. In this respect, it is worth mentioning that the successful application of GIS in mapping malaria and planning for control programs depends largely on availability of accurate and updated data on malaria cases [22].
Malaria is one of the main health issues in Burundi in terms of morbidity, mortality and associated sever economic costs [8]. In 2015, Burundi had 2 million cases of malaria, which led to about 5600 deaths [2]. Due to its significant magnitude, malaria in Burundi was considered repeatedly.  [23]. It was projected that climate change will have significant impacts on expanding spatial extent of the areas vulnerable to malaria risk [24] [25] as a result of expected warming and increasing humidity [26]. However, none of the previous research work considered the potential impacts of climate change on malaria incidence in Burundi. This, in turn, emphasizes the need for developing a methodology for mapping malaria risk and delineating those regions that may potentially experience malaria epidemics under climate change in the future.
The paper in hand is intended to map malaria risk in Burundi under climate change up to 2050. For this purpose, a GIS-based model will be developed for mapping malaria through its climatic and topographic determinants. The spatial

Case Study
Burundi is one of the sub-Saharan African countries that is located between 2˚20' and 4˚27' South and between 28˚50' and 30˚53' East [8]. Covering a total area of 27,834 km 2 , of which 8% are water surface occupied by Lake Tanganyika; the second largest freshwater lake in the world [27]. The country is one of the most populated countries in East Africa with estimated total population size of 9.8 million inhabitants in 2015. This population primarily lives in rural environments, where urban population represents only about 15% of the total population [28]. The country is administratively subdivided into 18 provinces ( Figure   1), which are subdivided into 46 health districts [29]. Burundi has a mountainous surface with altitude ranging between 775 meters and 2670 meters above mean sea level [8]. The country is dominated by high plateau in the center with an average altitude of 1500 meter above mean sea level [27] ( Figure 2). Also, Burundi territory has some plains including Imbo in the West, the Kumoso depression in the East [30]. The varied altitude has considerable impact on temperature, where the highest mountain areas have an average annual temperature of 16˚C, while the central plateau has cool temperature with annual average temperature of 20˚C. Meanwhile, the relatively low areas near Lake Tanganyika has warmer climate with an annual temperature of 23˚C on average [8].
Most parts of Burundi receive considerable amount of rainfall exceeding 1300 mm/year. Yet, the rainfall is generally irregular in space and time. Spatially, the north-west parts of Burundi receive heavy rains compared to other parts of the country [8]. Temporally, Burundi climate is characterized by four seasons, a short rainy season (from October to December), a short dry season (January to February), a long rainy season (March to May) and a long dry season (from June to September [30].

Data
Mapping malaria risk as a function of climatic and topographic conditions requires as a prerequisite identifying those determinants that would be considered.
In this respect, a list of seven determinants was developed including: average annual temperature, rainfall, relative humidity, land cover, altitude, terrain slope and distance to wetlands and regularly flooded land. Data on topographic determinants including altitude and land cover was acquired in raster format from DIVA-GIS, which provide a wide range of data in vector and raster formats at country level. Additionally, the base map of Burundi including administrative units at different levels was acquired in vector format [31].
Meanwhile, data on temperature, rainfall and relative humidity as key climatic variables determining malaria risk was obtained from The National Center for Atmospheric Research (NCAR). The obtained data, which was in shapefile format, included historical data for the period 1986-2005 and downscaled data future data under RCP8.5 scenario for the period 2040-2060 [32]. Using ArcGIS (V. 9.3) the acquired data was processed and integrated into a geodatabase for Burundi. In this respect, point feature classes of climatic determinants were interpolated to create raster surfaces representing the spatial variation of these determinants within Burundi territory. Also, slope was derived from altitude raster surface.
Moreover, data on malaria incidence was essential for validation purposes.
Accordingly, previous research work and available data sources were surveyed to acquire data on malaria incidence in Burundi. In this respect, data on malaria cases reported by health facilities in 2010 were obtained from the Ministry of Public Health in the service of Epidemiology and health statistics. The district collects health information regularly from health facilities and compiles the data on a monthly basis. The completeness of data in the districts was about 97% during the study period [33].

Methodology
Using acquired data on various climatic and topographic determinants, malaria risk was mapped according a methodology of four main steps (Figure 3).

1) Evaluating determinants weights
This step involved identifying weight for different climatic and topographic determinants as they don't equally contribute to malaria incidence. For this purpose, Analytic Hierarchy Process (AHP) was applied. AHP is simply a way to assign relative weights for different factors through applying pairwise comparisons [34]. According to pairwise comparisons each factor is evaluated according to an evaluation scale ranging from 1 to 5 (Table 1)   The two factors contribute equally to malaria risk (Equal relative importance) 2 The considered factor contributes more slightly to malaria risk compared to the other factor (slight relative importance) 3 The considered factor contributes moderately to malaria risk compared to the other factor (Moderate relative importance) 4 The considered factor contributes strongly to malaria risk compared to the other factor (Strong relative importance) 5 The considered factor contributes very strongly to malaria risk compared to the other factor (Very strong relative importance) in the produced reciprocal matrix was divided by the sum of its column. Finally, the weight of each factor was calculated through averaging across the rows [34] [35].
After estimating weights for all considered determinants, the consistency of the estimated weights was assessed using Consistency Ratio (CR), which was about 9.3% indicating satisfactory consistency level of the estimated weights as it didn't exceed 10% level.

2) Mapping malaria risk under current climatic conditions
This step involved applying Weighted Overlay analysis to combine various climatic and topographic determinants of malaria risk. According to Weighted Overlay analysis the raster surfaces of different considered malaria determinants were reclassified into a common preference scale of different malaria risk levels.
Then, each of reclassified raster surfaces is multiplied by its assigned weight, identified in the previous step, then added together ( Table 2). As a result, a new raster surface was generated representing different levels of malaria risk at the present according to the considered current climatic and topographic determinants.

3) Assessing accuracy of produced malaria risk map
This step was intended to evaluate the accuracy of the produced malaria risk map compared to morbidity data and identify to what extent the produced malaria risk reflects spatial pattern of malaria incidence in Burundi.

Results and Discussion
Mapping malaria risk means delineating areas that are vulnerable to host breeding sites of malaria vectors and parasites. Such delineation, which was carried out as a function of a number of climatic and topographic determinants of malaria incidence, involved combining these determinants with different weights. As a result, a risk map was produced, in which various parts of Burundi were classified into three levels of risk; low, moderate and high malaria risk.
Generally, the produced malaria risk map, based on current climatic condi-   district were overlaid interest, then, the area of high-risk zone and the no of malaria cases were summarized. It was found that the high-risk zone captured 65% of the total cases of malaria that are reported at district level in Burundi in 2010 (Table 3)   Such expected changes in climate variables under different scenarios, particularly, temperature and rainfall, is thought that, will have significant impact on malaria incidence in Burundi in the future. This is especially in eastern parts that are expected to experience higher levels of warming and the northern parts that are expected to experience wetter conditions. This was emphasized by the produced malaria risk maps under RCPs 2.6 ( Figure 6) and 8.5 (Figure 7)    In this respect, it was found that 26% and 35% of total area of currently moderate-risk zone is expected to become high-risk zone under RCP2 2.6 and 8.5 scenarios, respectively. Meanwhile, 11% and 12% of total area of currently low-risk zone is expected to become moderate-risk zone under the same two scenarios, respectively (

Conclusions
Mapping malaria risk under expected climate change in the future can contribute largely to improving resilience of Burundi to climate change associated risks and their health implications. In this respect, GIS can play a crucial role in mapping malaria risk under different scenarios of climate change and identify hot sports of malaria incidence that will need to intervention and control measure.
The developed GIS-based methodology for mapping malaria under climate change revealed that considerable parts of Burundi, particularly northern parts, will be highly vulnerable to malaria risk in the future under different scenarios of climate change.
The accuracy of mapping process of malaria risk under climate change is determined by availability of data on malaria determinants currently and in the future. Also, to improve the accuracy of mapping process, there is a need to apply an integrated analysis considering, in addition to topographic and climatic determinants, socioeconomic determinants of malaria such as income, population density, access to health services and effectiveness of malaria control measures, etc. Yet, this entails projecting these variables in the future as a prerequisite of such integrated analysis.