Habitat characterization for malaria vector mosquito larvae in Gamo Gofa, Ethiopia.

(1)

Projektarbete 15 hp October 2014

Habitat characterization for malaria vector mosquito larvae in Gamo Gofa, Ethiopia

Rasmus Elleby Vilhelm Feltelius

(2)

(3)

Abstract

Malaria is a widespread parasitic disease in developing countries of the tropics and subtropics, infecting approximately 200 million people and causing over half a million deaths every year.

The disease is caused by the protozoan Plasmodium and is transferred to humans through infective bites from female mosquitoes of the genus Anopheles. In order to reduce malaria transmission, measures of larval control have been implemented throughout the tropics. This includes usage of larvicides, source reduction by environmental or physiochemical manipulation as well elimination of larval habitats.

The purpose of this study was to evaluate differences in occurrence and densities of anopheline larvae by investigating the environmental characteristics of their habitat. The study was conducted in the Gamo Gofa Zone, Great Rift Valley, Ethiopia where a total of 26 sampling sites were chosen for larval sampling. Each sampling site was characterized according to a protocol and sampled for water chemistry analysis. Environmental variables studied include water depth, habitat size, distance to nearest dwelling, land use within a 10 m and 100 m from the sampling site and number of domestic animals within a 100 m.

Physiochemical variables include water temperature, pH, electric conductivity (EC), total dissolved solids (TDS), dissolved oxygen (DO), turbidity and phosphate. Larval sampling was conducted on each site using a soup ladle dipper. The occurrence of anopheline larvae was statistically analysed using multiple logistic regressions, while using linear regression for analysing larval abundance at positive sites.

Larval sampling resulted in a total of 1245 mosquito larvae, 567 anopheline and 678 non- anopheline. Of the anopheline larvae, 118 were analysed morphologically by microscopy which resulted in 117 belonging to An. gambiae complex and one An. garnhami. Of the 26 sites investigated, 16 were positive for anopheline larvae. All sampled river fringes and flood pools were positive for anopheline larvae whereas none were found in irrigation channels.

Negative correlation for anopheline larval occurrence was obtained for both water depth and percentage of tall vegetation within 10 m radius of the sampling area. Anopheline larval abundance was only correlated, positively, with water temperature.

The study concludes that water depth, temperature and percentage of tall riparian vegetation are important factors to consider when designing a control program for anopheline larvae.

One should be aware of the fact that clearing riparian forest and other tall vegetation is likely to improve growing conditions for anopheline larvae. Furthermore, different habitat classes were either exclusively positive or negative for anopheline larvae, irrigation channels in the area not being suitable larval habitats during the time of measurements. The authors suggest that more studies are needed, preferably on a larger set of sampling sites and over a longer period.

Keywords: Anopheles, Ethiopia, larval habitats, water quality, environmental variables, Gamo Gofa zone.

(4)

Preface

This minor field study was carried out during the period January-March 2014 in the Gamo Gofa Zone in Southern Etiopia.

Supervisors for this study were Dr Richard Hopkins at the Department of Ecology, at the Swedish University of Agricultural Sciences (SLU) and Mr. Admasu Tassew Tsegaye, PhD Student in Water and Public Health at the Ethiopian Institute of Water Resources (EIWR).

A scholarship from the Swedish International Development Agency (Sida) was received through SLU in order to support the minor field study.

First of all, we would like to thank Dr. Richard Hopkins for all the support and

encouragement. Thank you for putting up with two engineering students who didn’t know the first thing about mosquitoes at the start of the project! Second, a big thanks goes to Mr.

Admasu Tassew Tsegayefor the invaluable help with all of the practical details regarding our stay in Ethiopia as well as academic input on the study. Third, we want to send our thanks to Mr. Abebe Bashe at Arba Minch University for his hard work and long days, both in the field and in the lab. We also want thank to Mr. Fekadu Massebo for letting us use the entomology lab at Arba Minch University and teaching us how to identify the larvae, Dr. Solomon

Gebreyohannis for hosting us at EIWR, Dr. Olle Terenius at SLU for help with ordering all of the instruments for the project and Monica Halling at SLU for support and helping us with applications. Last but not least, we would like to give big thanks to Professor Matt Low at SLU for his help with the statistical analysis of our data.

(5)

1. Introduction

1.1 Malaria in Ethiopia and the world

Malaria is regarded to be one of most important and widespread parasitic disease in developing countries of the tropics and subtropics (Mwandawiro and Okech, 2011; Okwa, 2012). Approximately 207 million people worldwide were infected by malaria in 2012, causing approximately 627 000 deaths (World Health Organisation (WHO), 2014). Sub- Saharan Africa is the most severely affected region, where children below five years old and pregnant women represent around 90 % of the total deaths (Mwandawiro and Okech, 2011;

Ovadje and Nriagu, 2011; WHO, 2014).

The number of confirmed cases of malaria in Ethiopia was in 2012 nearly 3.9 million (WHO, 2013). The transmission of the disease in the country is largely determined by factors such as altitude, temperature and rainfall. The majority of transmission of malaria occurs between September and December, following a period of rain between June and September. A smaller period of transmission is also occurring between April and May due to rain between February and March (WHO, 2014).

Ethiopia can, in regard to malaria, be divided into four epidemiological strata. The first stratum is the highland areas above 2500 meter which are considered to be free of malaria.

Highland fringe areas between 1500-2500 meters constitute the second strata and are frequently affected by the disease. The third and the fourth strata consists of lowland areas below 1500 meters where the malaria transmission is either seasonal or occurs all year round (WHO, 2014).

Malaria is caused by the protozoan Plasmodium, transferred to humans through bites from female Anopheles mosquitoes (Silver, 2008; Sathe and Tingare, 2010; Choumet, 2012). In order to produce eggs the female mosquito are in need of blood meals and it is therefore only the female mosquito that bites (Igweh, 2012). The protozoan Plasmodium has a development cycle that involves both the Anopheles mosquito and human host (Ovajde and Nriagu, 2011).

There are five plasmodium species that can be related to human malaria, P. falcifarum, P.

malariae, P. ovale and P.vivax and P knowlesi (WHO, 2012). The most frequently occurring species in Ethiopia are P falcifarium and P vivax (WHO, 2013).

1.2 Defining malaria vectors

Within the group of Anopheles mosquitoes, there are several sibling species that have more or less significance in regards to the transmission of malaria (Becker et al., 2003). The major anopheles species transmitting human malaria in Ethiopia consists of An. Arabiensis, An.

pharaoensis, An. nili and An. funestus (WHO, 2014). The capability of mosquitoes to function as disease vectors is characterized by variables such as life expectancy, density and competence. The latter variable includes behavioural, cellular, biochemical and environmental factors, which may in return have an impact on the connection between a vector, the pathogen transmitted by the vector and the vertebrate host being infected (Okwa et al., 2007).

Species of anopheline mosquitoes with a strong antropophilic behaviour, i.e. preferring blood meal from human, are the most potent vectors and therefore associated with stable transmission of the disease. However, most species of Anopheles do not feed exclusively on either humans or animals (Choumet, 2012). Studies have therefore suggested that keeping cattle in the vicinity of human dwellings may have a distractive effect on the feeding behaviour of Anopheles mosquitos. This is especially relevant for An. arabiensis, which have been showing a high tendency of resting and feeding outside (Hadis et al., 1997; Mahande et

(7)

al., 2007). By contrast, An. gambiae senso stricto (s.s) prefer to rest and feed inside, which in combination with a strong anthropophilic behaviour makes it a very potent vector of malaria transmission. It should also be mentioned that An. arabiensis have the ability to adapt to both available blood meal and whether the hosts are indoor and outdoor. When using indoor spraying as a measure of malaria control, An arabiensis are even able to adapt completely to resting outdoor.

It is important to take the biting cycle of the mosquito into consideration when discussing malaria control measures. A study made by Yohannes and Boelee (2012), revealed that 70%

of the biting activity of An. arabiensis occurred before 22.00 hours and thus before the time when people usually go to bed. An. pharaoensis have been shown to have a similar behaviour and suggests that a malaria control measure such as using insecticide-treated mosquito nets as malaria control measure is less effective (Kibret et al., 2010).

1.3 Early life cycle stages

The development of the female anopheles mosquito can be divided into an immature stage and a mature stage respectively, where the immature stage consist of egg, larva and pupa and are not associated with malaria (Ovadje and Nriagu, 2011). The majority of the eggs hatches into a first larval stage within 2-3 days and are defined as first instar larvae. Three larvae stages then follow over a period between 4-14 days, during which the larvae shed its skin and grow in size. The fourth instar stage is then followed by the non-feeding pupal stage, which lasts for about 1-4 days. After this time, the adult mosquito will emerge and mating and blood feeding will then follow a couple of days later (Sathe and Tingare, 2010).

Mosquito larvae usually have a non-selective behaviour in regards to feeding, but it is preferable if the particles they feed on are less than 50 micrometres. Some genera, for example Culex have predacious behaviour and even feed on other mosquito larvae. One easy way of visually separating anopheline larvae from other mosquito larvae is through the behaviour of feeding. Anopheline larvae rest in a horizontal position, feeding on microorganisms and dead organic material from the surface film. Other larvae species, such as Culex accomplish feeding through hanging on the surface filtering the water column below (Becker et al., 2003).

1.4 Larval control

Larval control, including the use of larvicides and source reduction by elimination of aquatic habitat, has long been used as a strategy for reducing malaria transmission throughout the tropics (Killeen et al., 2002). Several recent studies in Eastern African countries suggest that identifying and targeting larval habitats with larvicides is a viable strategy for managing malaria vectors (Fillinger and Lindsay 2006; Shililu et al., 2007; Fillinger et al., 2008). A study in Tigray, Ethiopia associated source reduction with a 49% decrease of An. arabiensis adults (Yohannes et al., 2005). As an alternative to larviciding, some studies suggest that anopheline larvae control can be managed through environmental manipulation (Imbahale et al., 2011) as well as altering physiochemical factors in the aquatic habitat (Mbuya, Kateyo and Lunyolo 2014).

Although malaria is a major health concern and anopheline larval control can be an important component of malaria control program, little is known about the ecology of Anopheles larvae.

Usually, the description of larval habitats has been given in more general terms such as permanent/temporary or natural/managed habitats. More studies, such as Munga et al. (2006), Fillinger et al. (2008), Kenea, Balkew and Gebre-Michael, (2011) and Mala and Irungu,

(8)

(2011), are however starting to focus on finding what factors are influencing the occurrence and abundance of anopheline larvae.

1.5 Environmental factors

The microclimate in the larval habitats can be investigated in many different ways.

Morphologically, the habitats can often relatively easily be characterized by size and water depth. Several studies have found depth to be a key factor affecting the occurrence of anopheline larvae, shallow sites being more favorable (Shililu et al., 2003; Fillinger et al., 2004; Kenea, Balkew and Gebre-Michael, 2011). A study by Fillinger et al. (2009) found a decrease in anopheline larval density in sites that were larger than 100 m in perimeter.

Vegetation within and around aquatic habitats and its effect on mosquito larvae have been the subject of several if recent studies. Emergent vegetation in the habitat was shown to have a negative correlation with presence of mosquito larvae in general (Greenway, Dale and Chapman, 2003), occurrence of An. gambiae (Gimnig et al., 2004; Munga et al., 2006) and An. arabiensis (Gimnig et al., 2004; Fillinger et al., 2009). Some species of Anopheles contrarily are positively correlated with aquatic vegetation, such as An. funestus (Gimnig et al., 2004) and An. pharoensis (Kenea, Balkew and Gebre-Michael, 2011), suggesting that different species within the genus have different habitat preferences. Other than the role that emergent plants may play in the aquatic ecosystem and food chain, the plants themselves contribute to shading. Canopy cover above the aquatic habitat has also been shown to be negatively correlated with the occurrence of anopheline larvae (Minakawa et al., 2005;

Munga et al., 2006; Fillinger et al., 2009).

Algal growth in the aquatic habitat doesn’t provide any shade but likely constitutes a food source for the anopheline larvae. Several studies in Africa have shown a positive correlation between algal growth and occurrence of anopheline larvae, supporting that theory (Gimnig et al., 2001, 2002; Fillinger et al., 2009; Kenea, Balkew and Gebre-Michael, 2011; Mala and Irungu, 2011).

1.6 Physiochemical factors

The physiochemical microclimate is an important aspect trying to characterize larval habitats.

Water temperature, as a first, is widely regarded to have a positive correlation with the densities of anopheline larvae (Shililu et al., 2003; Piyaratne et al., 2005; Munga et al., 2006;

Kenea, Balkew and Gebre-Michael, 2011). This is likely connected with anopheline larvae being more frequent in less shaded waters, which naturally should be warmer than those in the shade. DO has been shown to have a positive correlation with distribution and abundance of anopheline larvae (Mbuya, Kateyo and Lunyolo 2014). Other physiochemical variables positively correlated with anopheline larvae are concentration of nutrients phosphate (Rejmankova et al., 1991) and nitrate (Mala and Irungu, 2011).

Two things negatively correlated with anopheline larval densities are EC and pH, the former studied by both Fillinger et al. (2009) and Mbuya, Kateyo and Lunyolo (2014), the latter studied by Mala and Irungu (2011). For pH, it is most likely a case of an optimum rather than a steady increase in larval density with lower pH levels. The same thing goes for water temperature, with high or low enough values, the conditions in the aquatic environment will become too extreme to sustain any aquatic life.

One thing that is not consistent among studies is the relation between anopheline larvae and turbidity of the water. Shililu et al. (2003) and Kenea, Balkew and Gebre-Michael (2011) observed a negative correlation between turbidity and densitites of anopheline larvae. Mbuya, Kateyo and Lunyolo (2014) also observed a negative correlation with turbidity and larvae

(9)

from An. funestus and An. gambiae complex. However, Gimning et. al. (2001) as well as Mala and Irungu (2011) found a positive correlation between turbidity and both An. gambiae and An. arabiensis. Since several studies shows different results, more studies are needed to fully understand the impact turbidity has on larval habitat.

1.7 Larval habitat classification

Even though describing larval habitats in more general terms only tells a part of the tale, it may still give valuable some insight on the suitability of different habitats for mosquito larvae. According to Kenea, Balkew and Gebre-Michael (2011), the densities of total anopheline larvae and An. squamosus in natural habitats were higher but lower for An.

pharoensis. Habitat permanence is another factor that has been studied, classifying habitats as temporary, permanent or on a scale in between. Gimnig et al. (2001) showed that An.

gambiae and An. arabiensis were both associated with temporary habitats while An. funestus was associated with semipermanent bodies of water. Kenea, Balkew and Gebre-Michael (2011) showed a negative correlation between habitat permanence between total anophelines and An. Pharoensis, but a positive correlation with An. arabiensis, as opposed to Gimnig et al. (2001). Habitat classification has been shown to give good predictive power for some species of anopheline larvae, especially during dry season (Rejmankova et al., 1991).

The effect of irrigation systems on malaria transmission is a recurring question, weighing the benefits of a potential increase in food security against the possibility of creating new habitats for anopheline larvae. The matter have been subject for a number of studies with varying results: some studies suggest that irrigation increases malaria transmission in the area (Ghebreyesus et al., 1999; Koram et al., 2003; Keiser et al., 2005) and some showing unchanged conditions or even decreased transmission (Faye et al., 1995; Ijumba et al., 2002;

Assi et al., 2013). Another effect of agriculture, particularly growing maize has been shown by Ye-Ebiyo et al (2003). This study showed that the larval habitat’s proximity to flowering maize was positively correlated with development of anopheline larvae. However, the same study could not find any differences in initial larval densities. The role of agriculture and irrigation schemes as anopheline breeding sites needs further investigation to explain its effect on malaria transmission. It is likely that the suitability for anopheline larvae depends on both location and form of the irrigation system.

All these things and more are factors that may be worth considering when studying the ecology of anopheline larvae. Different species within Anopheles seem to have different preferences in larval habitats, as conditions for anopheline larvae change over seasons and regions. Clearly, more research is needed for a deeper understanding of what makes larval habitats suitable or not.

1.8 Aim of study

The aim of this experimental study is to explain differences in occurrence and densities of anopheline larvae by characterizing their habitats from an environmental and physiochemical standpoint. This could allow larval control to be more effectively targeted at specific sites which in its turn could reduce malaria transmission in the area.

(10)

2. Methods 2.1 Study area

The study was conducted in the Gamo Gofa Zone, Rift Valley, Ethiopia. A total of 26 sampling sites were chosen in the Arba Minch Zuria and Mirab Abaya Woredas. The sites are located in rural areas, most of them near farm and grazing lands, where malaria transmission is frequently occurring according to Fekadu Massebo¹ at Arba Minch University. All sites are located around 1100-1200 m above sea level and could be considered to have seasonal or all year round transmission of malaria.

Figure 1. Sampling sites in relation to the city if Arba Minch, Lake Abaya and Lake Chamo.

1Fekadu Massebo, Department of Biology, Arba Minch University, Ethiopia. Meeting 30 January 2014.

(11)

2.2 Selection and definition of sampling sites

Possible Anopheles larval habitats were selected as sampling sites both through communication with entomologists and by search in the field. Fekadu Massebo (Massebo, pers. comm.) advised to dismiss any fast flowing waters as sampling sites because of the unlikeliness of larval occurrence. For each of the sites, the size of the area needed to be defined. In some cases, such as with cut off drainage channels or small pools, the whole water body could be used as the defined area. In the case with rivers, a part of the river fringe was selected and defined as a sampling area. After the area was properly defined, the mean length and width was estimated visually and recorded. The location and altitude of each site was recorded with a Garmin Oregon 550 global positioning system device (GPS).

The sites were divided in to two main classes: natural and artificial habitat. Then, the sites were divided in to different subclasses (see Table 1).

Table 1. Main and sub classes for habitat types.

Artificial habitats Natural habitats Irrigation channels River fringes

Construction site Marsh

Artificial watering point Flood pool

Man-made pond Lakeside

Natural pond Rain pool

Every site was visited twice, with 10 ± 1 days in between visits (except for site O, P, Q, R, S and T that were revisited after 14 days). The time in between visits was selected so that any larvae present at the first sampling should have developed to adults by the time of the second sampling, making the samplings independent from each other in that sense.

To give the reader an idea about what the area of sampling looks like, pictures of six sites are shown in Figure 2-8.

(12)

Figure 2. Site K, classified as natural habitat, sub class “marsh”. Larvae were sampled in several water-filled cow tracks.

Figure 3. Close-up on water-filled cow track in site K.

(13)

Figure 4. Site F, classified as natural habitat, sub class “river fringes”. Sampling was conducted in the foreground where the water was less fast-flowing than in the main river channel.

Figure 5. Site L, classified as artificial habitat, sub class “construction site”. The construction of new bridges led to stagnant water around the concrete foundation on some places.

(14)

Figure 6. Site J, classified as artificial habitat, sub class “irrigation channel”. Sampling was conducted under the shade of a mango tree.

Figure 7. Site U, classified as artificial habitat, sub class “artificial watering point”. The site was full of blooming algae on the first visit. The second visit, the site was completely dry.

(15)

Figure 8. Site Q, classified as natural habitat, sub class “flood pool”. The pool was situated less than ten meters from a river and likely formed during higher discharge.

2.3 Environmental analysis

The distance from the sampling site to the nearest dwelling was estimated visually. This was done in order to investigate the correlation between larval occurrence, abundance and potential human blood meals. Locals were consulted whenever there were any uncertainties if houses were human dwellings or not. The land cover was estimated within both within 10 m and 100 m radius of the site as percentages within 5% of different categories (see Appendix 1). Aquatic vegetation within the sampling sites was estimated in the same way. The number of domestic animals in a 100 m radius of the sampling site was counted using binoculars. This was done to investigate the effect of livestock as potential blood meals on mosquito larval occurrence. If domestic animals were passing through the area during measurements, recordings were done when possible and added for a final number. Canopy and cloud cover at the time of sampling was also recorded.

2.4 Larval sampling

Sampling of mosquito larvae were conducted with a 293 ml dipper. A minimum of five dips were done on each site, distributed across the defined area at suitable places for anopheline larvae. For river fringes, this meant excluding the fast flowing middle part and sometimes even one of the banks. When the dipper could not be properly submerged in the water, stones were sometimes removed in order to do so. However, this was only done when removing stones was regarded to create little disturbance in the water. At each location where dipping was done, measurements of depth were done with a ruler in millimetres.

Mosquito larvae were identified as Anopheles or non-Anopheles genus directly in the dipper by resting/feeding behaviour and either early (1-2) or late (3-4) instar and recorded thereafter.

(16)

The anopheline larvae from every dip was then stored in Eppendorf tubes with 78% ethanol for Polymerase Chain Reaction (PCR) analysis.

2.5 Water chemistry analysis Temperature, pH, EC and TDS

Measurements of temperature, pH, EC and TDS were conducted in situ with a portable meter HI 991301 from HANNA instruments. When the water depth allowed submerging the meter, measurements were taken at three different locations within the defined area. When the water was too shallow, water from different locations were gathered in a cut off, transparent plastic bottle three times. In both cases, measurements were taken by holding the meter still in the water and recording pH, EC, TDS and temperature in the given order. Each pH, EC, TDS and temperature value was recorded when the instrument’s display showed that the readings were stabilized. The range of measurements are 0.0-60.0°C, 0.00-14.00, 0-20.00 mS/cm and 0- 10.00 parts-per-thousand for temperature, pH, EC and TDS respectively. Accuracies of measurements are ± 0.5°C and ± 0.01 and for temperature and pH. For EC and TDS, the accuracy of measurements is ± 2% F.S.

DO

DO was measured in situ with a portable meter HI 9146 from HANNA instruments. When the water depth allowed submerging the meter, measurements were taken on the same three different locations within the defined area as for pH, EC, TDS and temperature. The meter was submerged in the water by holding the cable and spun slowly in a small circle. This was to ensure that the oxygen-depleted membrane surface is constantly replenished. When the water was too shallow, same procedures as for pH etc. were conducted. In this case, the water container itself was spun slowly for water circulation. Each DO value was recorded when the instrument’s display showed that the readings were stabilized. The range of measurements is 0.00-45.00 ppm or 0.00-300.0%, the resolution 0.01 ppm or 0.1% and the accuracy 20°C ± 1.5% of full scale or ± 1 digit (whichever is greater).

Turbidity

Turbidity was measured with a portable meter HI 93703 from HANNA instruments. Water samples were collected in situ and brought back to the lab for analysis because of the inconvenience in keeping the cuvettes clean in the field. Due to lack of appropriate containers, only one water sample from each site was collected. To compensate for the reduced ability in representing the whole defined water area with just one sample, water were taken from at least three different locations and mixed in a 100 m plastic container. Water samples were taken before conducting any dipping or water chemical analysis so that the sampled water remained undisturbed. In the lab, turbidity measurements were taken on the same sample three times. The range of measurements is 0.00-50.00 FTU and 50 to 1000 FTU, the resolution 0.01 and 1 FTU and the accuracy ± 0.5 FTU or ± 5% of reading (whichever is greater).

Phosphate

The concentration of phosphate (PO4₄^3-) was measured in the lab with the photometer HI 83200 from HANNA instruments. The photometer uses a tungsten lamp with narrow band interference filter, with a wavelength of 610 nm. The water used for analysis was taken from the same samples used for turbidity analysis. If the turbidity or colour of the water was high, active carbon was added to it. In any case, the water was filtrated before added to a 10 ml cuvette. After the sample was used to zero the meter, one package of HI 93713-01 powder reagent was added and the cuvette was shaken until the reagent was completely solved.

(17)

Before reading the concentration of phosphate, a countdown of 3 minutes was initiated. One reading per site was conducted.

2.6 Larval identification

Morphological identification of the mosquito larvae was done using a microscope and a species identification key by Verrone (1962). The morphological identification was then used to be able to select the appropriate primers for PCR analysis, which in its turn was used to identify the species of individual specimens from the Anopheles gambiae complex. Without the initial morphological analysis, DNA sequence analysis is necessary to be performed on each specimen. PCR analysis is possible using only small amounts of any life cycle stage (Rafferty et al, 2002). Both morphological and PCR analysis was performed by Admasu Tassew Tsegaye, EIWR at the Department of Ecology at SLU.

2.7 Statistical analysis

Multiple logistic regression was performed in R (R Development Core Team 2014), assessing the correlation between environmental variables and presence of anopheline and non- anopheline larvae. This statistical method estimates the probabilities of possible outcomes of a single trial, in this case the probability of finding anopheline larvae. Presence/absence of larvae were defined as a binary (y=1 or y=0) dependent variable in a generalized linear model (GLM), e.g. a logistic regression model. The independent environmental variables in the model include temperature, turbidity, DO, depth, pH, EC, distance to nearest dwelling, PO₄, percentage of tall vegetation within respectively 10 m and 100 m radius, presence/absence of algae, artificial/natural site, number of domestic animals, presence/absence of shading canopy and presence/absence of clouds. Percentage of tall vegetation was defined as a sum of the following categories: shrubs, banana, mango, cassava, cotton, sugarcane, orchard and trees.

For sites where we sampled a second time, we treated these resamples as independent for the analyses because (1) many of the environmental factors had changed between samplings, (2) the timing of resampling occurred after any larvae that were present during the first sample had developed into adults, (3) preliminary analyses showed that results including or removing the resamples produced qualitatively similar results, and (4) treating resampling as independent made it possible to include some environmental factors in the GLM that couldn’t be fit in the mixed models (GLMM) because the mixed models wouldn’t converge. Stepwise backwards selection of variables (based on minimizing the Akaike Information Criterion (AIC)) was used for model selection to find the best model structure.

Microsoft Excel was used for t-tests (two-sample assuming unequal variances). Also, the software was used for scatter plots and linear regression fitting for mosquito abundance, analysing correlations between environmental variables and relative abundances of anopheline and non-anopheline larvae. Because of difficulties in fitting data with many zeros (based on the assumptions of a linear model with Gaussian residuals), we only fit models for larval abundance using site data with positive larval abundance. Therefore, any resulting correlation with larval abundance can only be interpreted relative to sites with positive larval counts. The significant variables were combined in a GLM and statistically analysed in R, looking at the model assumption to see if the linear regression should be rejected or not. The model assumptions are (1) the residuals are independent, (2) the residuals follow a Normal distribution and (3) the residuals have the same variance, independent of X, i.e. the residuals are homoscedastic.

(18)

3. Results

3.1 Larval identification

A total of 1245 mosquito larvae were counted during dipping throughout the study. In the field, 569 of those were identified as Anopheles and 676 as non-Anopheles. Among the collected anopheline larvae, 537 were observed for morphological identification. The remaining 32 were probably so small that they were either missed during collection or disintegrated into an unrecognizable state by the alcohol. Of the observed larvae, two were identified as non-anopheline meaning that they were misidentified in the field. Among the rest, 118 larvae were well preserved enough for further investigation where 117 (99.2%) were identified as members of the Anopheles gambiae complex and one as Anopheles garnhami (0.8%), leaving 378 larvae unidentified.

Samples from eight larvae of the An. gambiae complex were extracted for PCR analysis. Five larvae could be properly identified, all of them An. arabiensis.

3.2 Larval habitat

A total number of 26 sites were sampled for mosquito larvae (see

Table 2). Among these, 16 were positive for anopheline larvae. Of the 26 sites visited, eight were either dry or overflowed during the second visit and resampling was not conducted. This resulted in 18 sites with two samplings and eight sites with one sampling, giving a total of 44 samplings.

Table 2. Distribution of collected anopheline among different types of habitats.

Main class Sub class No. of sites Positive Negative Total anophelines

Artificial Irrigation channels 6 0 6 0

Artificial Construction site 3 3 0 123

Artificial Artificial watering point 2 0 2 0

Artificial Man-made pond 1 1 0 2

Natural River fringes 4 4 0 151

Natural Marsh 4 4 0 173

Natural Flood pool 3 3 0 108

Natural Lakeside 1 0 1 0

Natural Natural pond 1 1 0 10

Natural Rain pool 1 0 1 0

Sum: 26 16 10 567

3.3 Larval occurrence

Anopheline larvae were found in aquatic habitats with a high range of environmental and physiochemical variables (Table 3).

(19)

Table 3. Range of environmental and physiochemical variables in aquatic habitats where anopheline larvae were found. Values within parenthesis are the lowest/highest values found on sites without larvae.

Environmental and physiochemical variables Min Max

pH 7.23 (7.22) 9.56

EC (mS/cm) 0.52 (0.24) 11.29 (16.46)

Temperature (°C) 23.4 (20.7) 36.6

Depth (cm) 1.8 11.1 (15.2)

DO (ppm) 0.90 (0.14) 16.02 (22.41)

Turbidity (FTU) 1.28 733

Phosphate (mg/l) 0.07 12.4 (19.2)

After stepwise backwards selection, the model was left with depth, distance to nearest dwelling, percentage of tall vegetation within respectively 10 m and 100 m radius and number of domestic animals (see Table 4). For anopheline larvae, significant correlations were obtained for depth and percentage of tall vegetation within 10 m radius. No significant results were obtained for occurrence of non-anopheline larvae.

Table 4. Correlation coefficients between environmental variables and occurrence of anopheline larvae.

Environmental variables Estimate, n = 43 Std. Error P-value

(Intercept) 3.318 2.449 0.175

Depth -0.662 0.309 0.032*

Distance to nearest dwelling 0.005 0.003 0.110

Tall vegetation, 10 m radius -0.192 0.066 0.004^†

Tall vegetation, 100 m radius 0.065 0.033 0.051

Number of domestic animals 0.015 0.009 0.090

*Correlation significant at 0.05 level; ^†Correlation significant at 0.01 level

From the results, the probability of finding mosquitoes can be presented as a function of water depth (see Figure 9) and percentage of vegetation within a 10 m radius (see Figure 10).

(20)

Figure 9. Probability of finding mosquitoes as a function of water depth. The coloured areas represent standard error.

Figure 10. Probability of finding mosquitoes as a function of percentage of tall riparian vegetation within 10 m radius around the sampling site. The coloured areas represent standard error.

(21)

3.4 Larval abundance

Correlation between larval abundance, i.e. average number of larvae per dip, and

environmental variables resulted in a significant positive correlation between anopheline larvae and the variables water temperature and PO₄ (see ). For non-anopheline larvae, a significant positive correlation was obtained with turbidity.

Table 5). For non-anopheline larvae, a significant positive correlation was obtained with turbidity.

Table 5. Correlation coefficients between environmental and physiochemical variables and abundance of anopheline larvae.

To validate the result, the model assumptions for linear regression were analysed by

combining temperature and PO4, with anopheline larval abundance as response variable, in a GLM (see Table 6 and Figure 11). In Figure 11, the plot in top left corner shows the residuals against fitted values, displaying possible non-linearity, outliers and error variances. A normal quantile-quantile plot is displayed in the top right corner, describing the validity of

distributional assumption for the data set. The more linear pattern, the more normally

distributed the data is. In the lower left corner, a scale-location plot with the square root of the residuals against the fitted values show possible trends in the residuals. The residuals are plotted against the leverage in a plot displayed in the lower right corner, describing if the results are driven by some data points.

Table 6. GLM for temperature and PO₄ with mean anopheles per dip as response variable.

(Intercept) -12.7591 5.6897 0.0364*

Temperature 0.5439 0.1915 0.0101*

PO₄ 0.7266 0.2592 0.0110*

*Correlation significant at 0.05 level Environmental and

physiochemical variables Anopheline, n = 23 P-values Non anopheline, n = 14 P-values

Water temperature 0.503 0.033* 1.113 0.214

Water depth -0.814 0.057 1.698 0.055

Turbidity 0.007 0.220 0.047 0.028*

pH 0.551 0.745 -4.681 0.395

DO 0.242 0.283 -1.164 0.101

EC 0.139 0.761 -1.345 0.801

PO₄ 0.670 0.036* 0.623 0.548

Distance to nearest dwelling 0.000 0.961 -0.012 0.357

Tall vegetation, 10 m radius -0.093 0.209 -0.084 0.607

Tall vegetation, 100 m radius -0.057 0.251 -0.118 0.325

Number of domestic animals 0.000 0.995 0.058 0.387

*Correlation significant at 0.05 level

(22)

Figure 11. GLM for anopheline abundance as a function of water temperature and PO4. Top left corner: Plot of residuals against fitted values. Top right corner: A normal quantile-quantile plot. Lower left corner: a Scale- Location plot of the square root of the residuals against the fitted values Lower right corner: Residuals plotted against leverage.

When checking the model assumptions, the quantile-quantile plot showed a non-linear pattern suggesting that the data is not sufficiently normally distributed. Judging from the residuals versus leverage plot, it also seemed like there were influential observations driving the results.

Because of this, two points in the PO4-data set was considered as outliers and therefore removed. A GLM was then fitted to the new data set, where PO4 was no longer showing a significant correlation with anopheline larval abundance (see

(23)

Table 7). The GLM showed a less driven result and a more ideal quantile-quantile plot compared to when including the two outliers (see Figure 12).

(24)

Table 7. GLM for anopheline abundance as a function of water temperature and PO₄. Two outliers in the PO₄ data set have been removed.

(Intercept) -10.9734 4.1388 0.0164*

Temperature 0.5077 0.1360 0.00152^†

PO₄ -0.3240 0.3709 0.39389

Figure 12. GLM for anopheline abundance as a function of water temperature and PO₄. Top left corner: Plot of residuals against fitted values. Top right corner: A normal quantile-quantile plot. Lower left corner: a Scale- Location plot of the square root of the residuals against the fitted values Lower right corner: Residuals plotted against leverage. Two largest in the PO₄ data set have been removed.

The model assumptions for linear regression of non-anopheline larvae and turbidity were analysed in the same way as with anopheline abundance (see

(25)

Table 8 and Figure 13). The model assumptions seemed to be correct and no outliers were removed.

(26)

Table 8. GLM for non-anopheline abundance as a function of turbidity.

(Intercept) 1.51694 3.54463 0.6763

Turbidity 0.04743 0.01906 0.0285*

*Correlation significant at 0.05 level

Figure 13. GLM for non-anopheline abundance as a function of turbidity. Top left corner: Plot of residuals against fitted values. Top right corner: A normal quantile-quantile plot. Lower left corner: a Scale-Location plot of the square root of the residuals against the fitted values Lower right corner: Residuals plotted against leverage.

A comparison between river fringes and flood pools combined and irrigation channels for anopheline larval abundance and 11 environmental and physiochemical variables was conducted by performing a two-tailed t-test (see Table 9). The average larval abundance, pH and DO were significantly higher for river fringes and flood pools than irrigation channels.

(27)

Table 9. Two-tailed t-test for 12 variables, river fringes and flood pools as one sample and irrigation channels as another sample.

Variable

Mean river fringes and flood

pools, n = 12

Mean irrigation channels,

n = 10

df P-value (two-tail)

Anopheline larvae per dip 3.63 0.00 11 0,011*

Water temperature 29.8 27.8 20 0,346

Water depth 4.92 6.89 19 0,064

Turbidity 138 30,5 12 0,091

pH 8.32 7,71 17 0,004^†

DO 7.84 2,50 14 0,008^†

EC 1.29 1,07 19 0,297

PO4 0.591 0,422 14 0,195

Distance to nearest dwelling 243 248 19 0,952

Tall vegetation, 10 m radius 18,3 28,0 19 0,195

Tall vegetation, 100 m radius 59,2 41,0 10 0,196

Number of domestic animals 56,7 24,0 14 0,168

4. Discussion

The results of the study showed a decrease in the probability of finding anopheline larvae with increasing water depth and higher percentage of tall riparian vegetation. A negative correlation between depth and occurrence of anophelines matches the results of the studies by and Shililu et al. (2003), Fillinger et al. (2009) and Kenea, Balkew and Gebre-Michael (2011). Applied to habitat elimination as a mean of larval control, this suggests that effort should be focused on shallow waters. It is also possible that increasing the water depth as a mean of environmental manipulation could be used at sites where habitat elimination is not suitable. Artificial watering points for example, could be a potential habitat for anopheline larvae but are still of importance for keeping livestock. Making sure that the water depth is not too low could be a way of reducing the risk of anopheline larval occurrence and thus potential malaria transmission in the longer run. Water storage and channels should be made deep and with steep sides in order to minimise the amount of anopheline habitats. Bear in mind though that naturally occurring, shallow, potential larval habitats will almost always be present and the importance of those should be investigated in relation to what effect any environmental manipulation would result in.

Shading from canopy cover is mentioned to have a negative effect on occurrence of An.

gambiae larvae in the study by Munga et al. (2006). No similar correlation with canopy cover could be shown in this study, maybe due to the lack of an accurate way of measuring canopy shading or a lack of shaded sampling sites. However, it is possible that a high percentage of tall riparian vegetation could provide shading, affecting the occurrence of anopheline larvae.

The sharp decline in probability of anopheline larval occurrence at around 20% tall riparian vegetation should be considered when implementing larval control (see Figure 10). Targeting larval habitats with none or little tall riparian vegetation with larvicides, could mean a more effective malaria vector control program. Preserving riparian forest or other tall vegetation may be an effective way of keeping anopheline larvae from breeding in rivers or other waters.

(28)

The abundance of anopheline larvae showed to have a positive correlation with temperature, which goes hand in hand with previous studies larvae (Shililu et al., 2003; Piyaratne et al., 2005; Munga et al., 2006; Kenea, Balkew and Gebre-Michael, 2011). The importance of temperature with regards to larval control remains to be seen, since it is difficult to manipulate other than indirectly using vegetation or other means of shading. Higher water temperature could otherwise be considered as an indication of shallow water depth and lack of tall riparian vegetation.

The morphological identification showed that out of the 537 larvae identified as Anopheles in the field only two were misidentified, being members of a different genus (0.37%). The result shows that the classification of mosquito larvae into anopheline and non-anopheline directly in the dipper is reliable to a high degree. The result from the PCR analysis is in line with the study by Seyoum et al. (2004) in Ethiopia, which found An. arabiensis to be the dominant anopheline mosquito larvae. However, considering the small sample size used in the PCR analysis, it is difficult to draw any conclusion from the obtained result. PCR analysis performed on a larger sample is needed to better understand the composition of anopheline species in the Gamo Gofa zone.

Among the different subclasses of aquatic habitats, the sites within each class were either all positive or all negative for anopheline larvae. This could suggest that adult females are selective with oviposition, choosing only some types of aquatic habitats as breeding sites. It may also be because some aquatic habitats provide very poor conditions for anopheline larval growth. Further studies with a larger number sampling sites under a longer period of time are probably needed in order to better understand the differences of anopheline larval occurrence between different aquatic habitats.

Sampling in six different irrigation channels resulted in zero collected larvae which indicate that they are not suitable breeding sites for anopheline larvae, at least not during the sampling period. The reason behind this could be because of several factors. When comparing irrigation channels with river fringes and flood pools, the only significant differences were lower pH values and higher DO concentrations in the former. According to Fillinger et al. (2009), lower pH is negatively correlated with anopheline larvae density and according to Mbuya, Kateyo and Lunyolo (2014), DO is positively correlated with anopheline larval abundance. This suggests that obtained pH and DO levels are less suitable for anopheline larvae in sampled irrigation channels than river fringes and flood pools. However, no correlations between pH or DO and larval occurrence or abundance were found in the overall study. These physiochemical variables are therefore unlikely to solely explain the absence of anopheline larvae in irrigation channels.

It is more likely that varying discharge in rivers and irrigation channels due to precipitation affected larval occurrence and abundance during the sampling period. Some unusually early rain falls occurred during the sampling period and as a result, this study should not be considered to represent typical conditions during this time of year. Shililu et al. (2003) and Kenea, Balkew and Gebre-Michael (2011) both found water current to have a significant negative correlation with anopheline larval density. Due to a smaller capacity in irrigation channels compared to river channels, the effect of precipitation on water current is likely to be greater on irrigation channels.

Especially during dry season, when the river is occupying a small part of the floodway, a sudden peak in discharge will flood large parts of the flat river bed creating new possible larval habitats. In the case of irrigation channels, the ones sampled in this study were dug deep enough not to overflow due to the rain falls during measurements and thus less likely to

(29)

create new larval habitats. It is however important to consider that by redirecting water with irrigation schemes, the water balance for the whole irrigated area is changed. This may affect larval occurrence and abundance in habitats other than irrigation schemes themselves. When studying the correlation between irrigation and breeding anopheline mosquitoes, it might be better to do so in a larger scale, as done by Keiser et al. (2005) with malaria transmission.

When designing a study, one should consider comparing irrigated with non-irrigated areas rather than larval occurrence and abundance only in different habitats providing it is possible to sample a large area. If the aim of a study is to investigate anopheline larval ecology in agricultural lands, one should also consider measuring levels of pesticides in the water since it might have an effect on larval growth.

Even though algae have been suggested as a possible food source in the literature (Gimnig et al., 2001, 2002; Fillinger et al., 2009; Kenea, Balkew and Gebre-Michael, 2011; Mala and Irungu, 2011), no correlation between this parameter and occurrence was found. This might be because only the presence or absence of algae was used in the model and in order to further investigate the effect of algae a more qualitative mean of measurement is probably needed.

Neither could the study succeed in finding correlations between the variables DO, turbidity, pH and EC and occurrence or abundance of anopheline larvae, even though literature indicated that such a connection was possible. Increasing the data set with more samplings could improve the models and help detect trends that this study was not able to.

The visual estimation of land cover was probably more accurate for the 10 m than 100 m distance from the aquatic habitat. Usage of remote sensing could be one way of decreasing the error in the estimation of the land cover at 100 m or longer distances, since there will always be difficulties in overviewing such an area from the ground. Likewise, future studies should consider using a more accurate way of measurements than visual estimation of the distance to nearest dwelling in order to improve the quality of this variable. Measuring tape has for example been used by Kenea, Balkew and Gebre-Michael (2011) for distances less than 100 m. In the same study, longer distances were classified as 100-300 m or >300 m. Another alternative could be calculating the distance from GPS coordinates taken at the sampling site and nearest dwelling respectively.

The number of domestic animals could not be correlated with larval occurrence. Livestock are usually kept in herds and were often on the move during the day. Missing a herd of 200 cattle passing through just after sampling was finished have a big impact on the data, making the variable unpredictable. However, it is curious that the P-value for the positive correlation between the number of domestic animals and occurrence of anopheline larvae in the GLMM was so low. This might be because high numbers of domestic animals indicate open areas. In open areas, there is little tall vegetation obstructing the view when counting domestic animals and a higher percentage of grass where livestock are kept for grazing.

5. Conclusion

This study concludes that occurrence of anopheline larvae in sampled sites are negatively correlated with water depth and percentage of tall riparian vegetation. Anopheline larval abundance is positively correlated with water temperature. As a result, clearing riparian forest and other tall vegetation improves growing conditions for anopheline larvae. The majority of anopheline larvae belong to An. gambiae complex. Different habitat classes were either exclusively positive or negative for anopheline larvae, irrigation channels in the area not being suitable larval habitats during the time of measurements. When designing a control

(30)

program for malaria vector mosquito larvae, a higher efficiency could be obtained when focusing measures on a selected few habitat classes, at sites with shallow waters and little riparian vegetation. The authors suggest that more studies are needed, preferably on a larger set of sampling sites and during a longer period.

(31)

6. References

Assi, S. B., Henry, M. C., Rogier, C., Dossou-Yovo, J., Audibert, H., Mathonnat, J., Teucher, T. & Carnevale, P. (2013). Inland valley rice production systems and malaria infection and disease in the forest region of western Côte d'Ivoire. Malar J, 12, 233.

Becker N., Petrić D., Zgomba M., Boase C., Dahl C., Lane J., Kaiser A. (2003). Mosquitoes and Their Control. New York: Kluwer Academic/Plenum Publishers

Choumet, V. (2012). Proteomics of Anopheles gambiae. Malaria Parasites, 39-56.

Faye, O., Fontenille, D., Gaye, O., Sy, N., Molez, J. F., Konate, L., Hebrard, G., Herve, J. P., Trouillet, J., & Diallo, S. (1995, September). [Malaria and rice growing in the Senegal River delta (Senegal)]. In Annales de la Societe belge de medecine tropicale (Vol. 75, No. 3, pp.

179-189).

Fillinger, U., Sonye, G., Killeen, G. F., Knols, B. G., & Becker, N. (2004). The practical importance of permanent and semipermanent habitats for controlling aquatic stages of Anopheles gambiae sensu lato mosquitoes: operational observations from a rural town in western Kenya. Tropical Medicine & International Health, 9(12), 1274-1289.

Fillinger, U., & Lindsay, S. W. (2006). Suppression of exposure to malaria vectors by an order of magnitude using microbial larvicides in rural Kenya. Tropical Medicine &

International Health, 11(11), 1629-1642.

Fillinger, U., Kannady, K., William, G., Vanek, M. J., Dongus, S., Nyika, D., Geissbühler Y., Chaki P.P., Govella N.J., Mathenge E.M., Singer B.H., Mshinda H., Lindsay S.W., Tanner M., Mtasiwa D., de Castro M.C.& Killeen, G. F. (2008). A tool box for operational mosquito larval control: preliminary results and early lessons from the Urban Malaria Control

Programme in Dar es Salaam, Tanzania. Malaria Journal, 7(1), 20.

Fillinger, U., Sombroek, H., Majambere, S., van Loon, E., Takken, W., & Lindsay, S. W.

(2009). Identifying the most productive breeding sites for malaria mosquitoes in The Gambia.

Malaria journal, 8(1), 62.

Ghebreyesus, T. A., Haile, M., Witten, K. H., Getachew, A., Yohannes, A. M., Yohannes, M.,

& Byass, P. (1999). Incidence of malaria among children living near dams in northern Ethiopia: community based incidence survey. Bmj,319(7211), 663-666.

Gimnig, J. E., Ombok, M., Kamau, L., & Hawley, W. A. (2001). Characteristics of larval anopheline (Diptera: Culicidae) habitats in Western Kenya. Journal of Medical Entomology, 38(2), 282-288.

Gimnig, J. E., Ombok, M., Otieno, S., Kaufman, M. G., Vulule, J. M., & Walker, E. D.

(2002). Density-dependent development of Anopheles gambiae (Diptera: Culicidae) larvae in artificial habitats. Journal of Medical Entomology, 39(1), 162-172.

Greenway, M., Dale, P., & Chapman, H. (2003). An assessment of mosquito breeding and control in 4 surface flow wetlands in tropical-subtropical Australia. Water Sci Technol, 48(5), 249-256.

Habtewold, T., Prior, A., Torr, S. J., & Gibson, G. (2004). Could insecticide‐treated cattle reduce Afrotropical malaria transmission? Effects of deltamethrin‐treated Zebu on Anopheles

(32)

arabiensis behaviour and survival in Ethiopia. Medical and veterinary entomology, 18(4), 408-417.

Hadis, M., Lulu, M., Makonnen, Y., & Asfaw, T. (1997). Host choice by indoor-resting Anopheles arabiensis in Ethiopia. Transactions of the Royal Society of Tropical Medicine and Hygiene, 91(4), 376-378.

Igweh, J. C. (2012). Biology of malaria parasites. Malaria parasites, 17-36.

Ijumba, J. N., Shenton, F. C., Clarke, S. E., Mosha, F. W., & Lindsay, S. W. (2002). Irrigated crop production is associated with less malaria than traditional agricultural practices in Tanzania. Transactions of the Royal Society of Tropical Medicine and Hygiene, 96(5), 476- 480.

Imbahale, S. S., Mweresa, C. K., Takken, W., & Mukabana, W. R. (2011). Development of environmental tools for anopheline larval control. Parasit Vectors, 4(1), 130.

Keiser, J., de Castro, M. C., Maltese, M. F., Bos, R., Tanner, M., Singer, B. H., & Utzinger, J.

(2005). Effect of irrigation and large dams on the burden of malaria on a global and regional scale. The American journal of tropical medicine and hygiene, 72(4), 392-406.

Kenea, O., Balkew, M., & Gebre-Michael, T. (2011). Environmental factors associated with larval habitats of anopheline mosquitoes (Diptera: Culicidae) in irrigation and major drainage areas in the middle course of the Rift Valley, central Ethiopia. Journal of Vector Borne Diseases 48(2), 85-92.

Kibret, S., Alemu, Y., Boelee, E., Tekie, H., Alemu, D., & Petros, B. (2010). The impact of a small‐scale irrigation scheme on malaria transmission in Ziway area, Central Ethiopia.

Tropical Medicine & International Health, 15(1), 41-50

Killeen, G. F., Fillinger, U., Kiche, I., Gouagna, L. C., & Knols, B. G. (2002). Eradication of Anopheles gambiae from Brazil: lessons for malaria control in Africa?. The Lancet infectious diseases, 2(10), 618-627.

Koram, K. A., Owusu‐Agyei, S., Fryauff, D. J., Anto, F., Atuguba, F., Hodgson, A., &

Nkrumah, F. K. (2003). Seasonal profiles of malaria infection, anaemia, and bednet use among age groups and communities in northern Ghana.Tropical Medicine & International Health, 8(9), 793-802.

Mala, A. & Irungu, L. W. (2011). Factors influencing differential larval habitat productivity of Anophelesvgambiae complex mosquitoes in a western Kenyan village. Journal of Vector Borne Diseases 48(1), 52-57.

Mbuya, N. P., Kateyo, E., & Lunyolo, F. (2014). Assessment of Anopheles Larval Source Reduction Using Cow Dung: Environmental Perspective on Pro-poor Tool for Malaria Vector Control. International Journal of Innovation and Applied Studies 5(1), 30.

Minakawa, N., Munga, S., Atieli, F., Mushinzimana, E., Zhou, G., Githeko, A. K., & Yan, G.

(2005). Spatial distribution of anopheline larval habitats in Western Kenyan highlands: effects of land cover types and topography. The American journal of tropical medicine and

hygiene, 73(1), 157-165.

Munga, S., Minakawa, N., Zhou, G., Mushinzimana, E., Barrack, O. O. J., Githeko, A. K., &

Yan, G. (2006). Association between land cover and habitat productivity of malaria vectors in

Habitat characterization for malaria vector mosquito larvae in Gamo Gofa, Ethiopia.