This study has revealed that drainage canals, hoof prints, tire tracks, rice paddies and watering taps, either leaking or broken with a constant flow of water, were important mosquito breeding habitats. These habitats were either man-made or associated with human activities. By contrast, natural water bodies such as erosion pits and habitats along the river fringe, were poorly represented among the available mosquito habitats, even in the rainy season. In summary, with availability of water (rainfall), our results indicate that anopheline larval abundance varied depending on the habitat suitability, habitat stability, vegetation covering the habitat and rainfall intensity.
Natural water sources such as pools and erosion pits relied on rainfall for water supply and in the absence of rain the habitats dried out and hence ceased to contribute to the local vector population. In contrast, habitats along the river fringe were more important for breeding during the dry season when the water levels reduced, stagnant pools of water suitable for mosquito breeding were created. In addition, when it rained heavily, flooding would occasionally occur especially in habitats along the river fringe, erosion pits and within drainage canals rendering the habitats unsuitable for mosquito breeding. This was facilitated more by the topography of the area. For Fort Ternan due to the V shaped valleys water drained into Kipchorian river down hill. In contrast, Lunyerere with its basin shaped valleys all the drainage canals were flooded while for Nyalenda being more less flat, the whole area flooded.
Results in Fort Ternan and Nyalenda show that temporary habitats were more preferred breeding habitats for both anopheline and culicine (data not included) mosquitoes. Conversely, in Lunyerere anopheline larvae occupied temporary and permanent habitats equally. These results are similar to reports by Fillinger et al  who found that semi-permanent and permanent habitats were suitable for proliferation of both anophelines and culicines. Temporary habitats due to their transient nature they held water for a shorter period of time whereas permanent habitats held water for a long period of time and after the rains, these habitats were more preferred by the culicines as observed in Fort Ternan and Nyalenda (S.S.I., personal observation). Of the highland villages studied, though at similar altitudes, more vector species were recorded from Lunyerere and in higher abundance than in Fort Ternan. In Lunyerere the mosquito breeding habitats were mainly in the broad, flat, valley floor, which until recently was a natural swamp forest. Lunyerere experienced underground seepage of water that ensured the presence of abundant, clean sunlit shallow bodies of water that were preferred mosquito breeding habitats all year round. By contrast, the Fort Ternan valley is narrow with steep slopes, causing a rapid drainage of water unless human activities create or construct water-retaining structures. In addition, Paaijmans et al  showed that the mean daily water temperatures (in small, medium and large pools) in these sites were generally higher than the mean air temperature, which eventually led to rapid development times. Nyalenda was a natural drainage area, which in the last decade has been turned into an agricultural area linked to the urban growth of Kisumu city, creating many variable water bodies suitable for mosquito breeding.
This study revealed that habitats that had grass growing in them had more anopheline mosquitoes than habitats with other vegetation types and open habitats. A previous study by Fillinger and others  found similar results. These findings were apparent in the two highland village sites and suggest that grass protected mosquito larvae from being swept or flushed away by running water  and/or predation [30, 31]. Grass could have also been convenient in offering newly emerged adult and gravid mosquitoes a shaded resting site. The role of grass as a resting site was not investigated and more research is needed on this topic. Habitats containing papyrus reeds were more important for anopheline larval breeding in Nyalenda. Natural and undisturbed papyrus swamps were found unsuitable for anopheline breeding [32, 21]. However, once the natural state of Papyrus spp. is disturbed, as was the case in Nyalenda and Lunyerere, mosquito breeding can occur due to incursion of other vegetation. Consequently, this may lead to an increase in mosquito-human contact, eventually leading to an increase in malaria transmission.
Anopheles arabiensis, An. gambiae and An. coustani were present in all three study sites. Anopheles arabiensis comprised 71% of An. gambiae s.l. collected from Fort Ternan. This is the highest proportion of this species that has ever been recorded in the highlands of western Kenya. Previous studies in the highlands of western Kenya report the presence of An. gambiae, but not An. arabiensis, a major vector species present in the Lake Victoria basin [33–35]. The presence of An. arabiensis in Fort Ternan and Lunyerere can be attributed to the slow changes in land use, such as deforestation that lead to changes in micro-climatic conditions favouring the survival of this species. The air temperatures used in the present study may have been lower than the corresponding water temperatures in the sampled habitats, a factor that seems to favour the existence of An. arabiensis in high altitude areas. In their study Paaijmans et al , have shown clearly that the use of air temperature rather than water temperature will tend to underestimate the population growth rates of mosquito populations. Anopheles funestus was only found in the highland villages, while An. christyi was only present in Fort Ternan, in line with previous studies by Koenraadt and others . The vector status of An. christyi has not been resolved, although Garnham  suggested that the species might occasionally be implicated as a malaria vector. Anopheles coustani was present in all the study sites and this species has recently been reported as a possible malaria vector in East Africa . In Nyalenda, a site where on many occasions the water was polluted with human waste and other debris (S.S.I, personal observation), An. arabiensis was more abundant than any other anopheline species. Anopheles species seem to have adapted to breeding in environments rich in organic waste, which was also noted in urban Dar es Salaam [10, 11] contrary to the findings of Robert et al . Anopheles gambiae was not found in Nyalenda, which is in line with recent changes of the distribution of this species in the Lake Victoria basin, where it seems to be replaced by An. arabiensis, especially in agricultural areas .
These results indicate that whenever it rained in Nyalenda, anopheline larvae were reduced with heavy rains leading to extreme reductions in larval abundance. It is possible that heavy rains flushed away and/or killed the larvae. These results are supported by experiments by Paaijmans et al , who reported that precipitation flushed, ejected and killed An. gambiae larvae in a semi-natural environment. Contrary to these results, in Fort Ternan when it rained heavily, more young larvae were recorded with a time lag of fourteen days, while late larvae were recorded in twenty-one days. At a slightly longer development period when compared to Nyalenda, which had no time lag between rain and the presence of larvae. In these highlands, the rains seem to have provided new habitats suitable for oviposition and larval development.
The current study was done under natural conditions and external factors could have played an important role in the colonization and growth of mosquito larvae in the respective habitats. Factors such as water turbidity, nutrient content of the water, cannibalism, predation of immature stages, parasitism, pathogens, competition, water temperature and plant odours that could have either repelled or attracted female mosquitoes during oviposition [39–45], were not controlled, which could have played a role in the results obtained. Studies on the ecology of larval anopheline mosquitoes are methodologically challenging and the current study had several limitations. Quantitative sampling was problematic because the standard dipping technique adapted for this study was unsuitable in small habitats and consequently the density of mosquito larvae and pupae may have been underestimated [25, 46]. In this study, productivity of a habitat is expressed as the density/abundance of larvae [5, 6, 39–41] and not as the number of pupae or adult mosquitoes that emerged from a habitat [46–48]. Although larvae were recorded in many habitats, not all larvae might have successfully developed into adults. However, because spatial models have shown the importance of non-productive habitats as part of transmission foci in facilitation of mosquito movement in search of a blood meal and aquatic habitats [49, 50], it is inevitable to focus anopheline larval control both in productive habitats and non-productive habitats.
The study has shown that mosquito larval abundance depends on the suitability, stability and productivity of the larval habitat and rainfall intensity. Suitability of any given habitat may be influenced by the vegetation type within the larval habitat and the land use in the area under study. Most of the mosquito larval habitats in the studied areas were a result of human activities. Lunyerere (highland) and Nyalenda (western lowlands) represent reclaimed swampy areas turned into farmland. This land use change seems to have contributed significantly to the suitability of habitats for anopheline mosquito breeding. By contrast, in Fort Ternan (highland), mosquito breeding was found not to be directly related to crop farming but rather to livestock holding, as most anopheline larvae were collected from cattle hoof prints. With the increasing human population in the highlands, enhanced human activities including deforestation, farming and livestock rearing are likely to create more vector habitats [15, 17, 18, 30, 46, 51].
This was a baseline study that provided insights in the development of integrated larval source management strategies suitable for the larval habitats present in the respective study sites. An integrated larval source management approach such as water and environmental management measures, biological control options like predatory fish/larvae are highly feasible , in addition to the use of microbial larvicides, especially in temporary habitats and in areas where water cannot be drained away. Considering the current study sites, source reduction through drainage and habitat manipulation would be suitable for Lunyerere and Fort Ternan. Whereas the application of microbial bio-larvicides such as Bacillus thuringiensis var israelensis (Bti) would be a better option for Nyalenda, because the community relies on the same water for domestic purposes . Due to the transient nature of temporary habitats, microbial bio-larvicides are most suitable irrespective of the area under study . Such mosquito vector control programmes should work closely with land and home-owners, as the majority of the mosquito breeding habitats were man made.