Landscape and Land Cover Factors Influence the Presence of Aedes and Anopheles Larvae (original) (raw)
Abstract
The objective of this study was to test for associations between land cover data and the presence of mosquito larvae of the genera Aedes Meigen and Anopheles Meigen in northern Thailand at the landscape scale. These associations were compared with associations between larval habitat variables and the presence of mosquito larvae at a finer spatial scale. Collection data for the larvae of one Aedes species and three species-groups of Anopheles, all of which are involved in pathogen transmission, were used. A variety of northern Thai landscapes were included, such as upland villages, lowland villages and peri-urban areas. Logistic regression was used to evaluate associations. Generally, land cover and landscape variables explained the presence of larvae as well as did larval habitat variables. Results were best for species/species-groups with specific habitat requirements. Land cover variables act as proxies for the types of habitat available and their attributes. Good knowledge of the habitat requirements of the immature stages of mosquitoes is necessary for interpreting the effects of land cover.
In view of often scarce financial resources, more efficient methods to allocate budgets for disease prevention are useful. Most countries where vector-borne diseases are widespread urgently need such budget-efficient methods. This is the case for dengue fever and malaria, two important mosquito-borne diseases prevalent in tropical areas, including Thailand. In Thailand, although mortality and morbidity have dramatically decreased since the 1950s, malaria remains a concern in border areas, and in the south, where epidemic events were recorded in the 1990s, following the disruption of control programs (Ministry of Public Health 2003). Dengue is also a major concern, with recurrent epidemics, and incidence of infection of ≈6% (Endy et al. 2002, Vanwambeke et al. 2006). Vector control is currently the main tool for dengue limitation, and it is still a tool for malaria control (TDR 2002). Because life cycle variables of vectors largely depend on the environment, the biology and behavior of the vectors have to be considered in designing effective intervention (Carter et al. 2000), especially regarding source reduction by elimination of aquatic habitats, or mapping of areas at risk. There is a need to focus vector control on areas that are most at risk and to recommend social and environmental changes that will have the greatest impact on vector abundance (Nagao et al. 2003). Here, field-collected variables describing larval habitats are compared with remotely sensed landscape variables for predicting the presence of mosquito larvae, to test whether the latter variables are good proxies for the former variables.
Most adult female mosquitoes lay their eggs directly on the surface of water, but some oviposit in areas that become inundated with water. Oviposition sites include artificial or natural containers and surface water. Artificial containers include artificial vessels used for water storage or left in the open where they collect rainwater. Natural containers include any plant part that collects water. Surface water includes lakes, rivers, ponds, marshes, swamps, and smaller sites such as puddles, hoof- or footprints, and wheel tracks. Irrigated fields also constitute an important source of surface water for some species, including some important vector species (Service and Townson 2002). Knowledge of larval distribution and larval habitats has potential value for implementing source reduction measures in vector control (Muir 1988). Here, we refer to larval habitats as the specific places where mosquito larvae are found. In some cases, it is an entire habitat (e.g., an artificial container), but sometimes it refers to a particular area of a habitat, i.e., a microhabitat, for example a stream margin.
Remote sensing and geographical information systems (GIS) provide efficient tools for locating environments capable of maintaining vector populations, provided that landscape elements critical to vector survival are known and can be detected by remote sensing (Beck et al. 2000). With mosquitoes, the dependence of immature stages on aquatic habitats has allowed remote sensing techniques to be exploited (Hay et al. 1997), mostly where these habitats are related to surrounding landscape elements. Landscape elements also can be related to the geographical range of vector species, which restricts the spatial distribution of vector-borne diseases (Kitron 1998). These elements include land cover, which is determined by attributes of land surface and immediate subsurface, diagnosed by a set of categorical or continuous attributes per spatial unit (Lambin et al. 2003). Remote sensing provides land cover data on large areas with little requirement for fieldwork in regions where up-to-date information on land cover or disease prevalence and incidence is often lacking.
Several studies have linked mosquito populations to remotely sensed land cover characteristics. Wood et al. (1992) studied the production of Anopheles freeborni Aitken in rice (Oryza spp.) fields in California by using remotely sensed growth characteristics of rice and GIS-derived distances to feeding sources. High production was related to early canopy development and proximity to bloodmeal sources. In a study of malaria transmission risk in Chiapas, Mexico, Beck et al. (1994) related the abundance of Anopheles albimanus Wiedemann adults in 40 villages to the surrounding landscape. Abundance was related to the quantity of transitional swamp and unmanaged pasture, which provided larval habitats. Roberts et al. (1996) successfully predicted the density of Anopheles pseudopunctipennis Theobald in Belize based on remotely sensed environmental data. Thomas and Lindsay (2000) investigated the link between abundance of Anopheles gambiae Giles adults and surrounding land cover in Gambian villages. Proximity of a village to "pooled sediments" was correlated with high adult mosquito abundance. Larval sampling supported this link. Jacob et al. (2005) tried to visually discriminate larval habitats occupied by members of the An. gambiae complex in urban areas in Kenya on multispectral images of 5-m resolution, but results were unsatisfactory. Moloney et al. (1998) used aerial photography to identify houses at high risk of Aedes aegypti (L.) (=Stegomyia aegypti of Reinert et al. 2004) in Australia but did not obtain accurate results. The latter two studies, however, tried to link habitat directly to remotely sensed radiance data, whereas the other studies linked mosquito abundance to land cover information.
In this study, we investigated the link between larval habitats and characteristics of the landscape, including land cover. Larval habitats are the target variable, but they occur at a spatial scale too fine to be detected by remote sensing. Land cover is an auxiliary variable that is easy to map on a broader scale by remote sensing, and its link with the presence and characteristics of larval habitats needs to be tested quantitatively.
Mosquitoes Studied.
Aedes albopictus Skuse (=Stegomyia albopicta of Reinert et al. 2004) and members of the Anopheles minimus group (sensu Harbach 1994; superseded by the An. funestus group of Garros et al. 2005), the An. maculatus group, and the An. barbirostris group were studied. Ae. aegypti was present in only 19/790 habitats surveyed and has been excluded from the analysis. Ae. albopictus has a controversial role in dengue transmission, but it is considered a vector in Asia in places where Ae. aegypti is absent, or a maintenance vector in places where Ae. aegypti is present (Gratz 2004). Ae. albopictus is found mainly in rural and peri-urban areas and breeds in artificial and natural containers.
Species of the An. minimus group are considered to be vectors of malaria throughout Southeast Asia, but the vector status of individual species varies from area to area, according to variations in behavior (Trung et al. 2005). The potential of species of the An. maculatus group to be involved in malaria transmission also has been noted (Trung et al. 2005). An. barbirostris was found to be a probable vector in Sa Kaeo Province in eastern Thailand based on population density, biting habits, and absence of other main vectors (Limrat et al. 2001).
Anopheles minimus Theobald (this nominal species includes two cryptic species in Thailand) is generally found among vegetation along the margins of partially shaded, small- to moderate-sized streams, as found in mountain foothills (Tagaki et al. 1995). Overgaard et al. (2002) suggested that factors such as stream sinuosity or pollution with chemicals also might be important factors. Larvae of the An. maculatus group (An. maculatus, An. sawadwongporni Rattanarithikul, and Green) are found in stream margins, streams pools, rice fields, swamps (Sithiprasasna et al. 2003), and temporary habitats (Rattanarithikul et al. 1995). Members of the An. barbirostris group (An. barbirostris, Anopheles barbumbrosus Strickland, and Chowdury, and Anopheles campestris Reid) occur in ground pools, rice fields, stream pools, swamps (Sithiprasasna et al. 2003), stream margins, and seepage (Harbach et al. 1987).
Data Collection and Methods.
Collection of Larvae.
Larvae were collected in and around seven villages in northern Thailand (Fig. 1). The study area includes a variety of landscapes, ranging from the suburbs of Chiang Mai to irrigated valleys surrounded by orchards and mountain villages with nonirrigated subsistence agriculture supplemented with a few irrigated fields. Altitude varies between 250 and 600 m, except for one village at ≈1,000 m. Each village and its surroundings (up to 5 km from the village center) were surveyed eight times between May 2003 and April 2004. Data were collected for each of 790 larval habitats, including the type of larval habitat, its size, presence of floating, emergent or submerged vegetation and type of vegetation, presence of floating, emergent or submerged debris and type of debris, presence of algae, turbidity, shadow, and immediately surrounding land cover characteristics, including presence of deciduous, evergreen or fruit trees, bamboo, and type of crops or fallow. Habitats were classified as permanent or temporary based on the likelihood of drying out in the absence of rain. The presence of livestock (such as cows and pigs) also was recorded. The date of collection was recorded. The location of each collection was georeferenced using a global positioning system (GPS).
Fig. 1
Locations of study villages in northern Thailand.
Larval habitats were located by walking transects through five aggregated land cover types detected by remote sensing in the area around each focal village (see below). More than 900 individual transects were conducted, each within the confines of a single land cover class. Transects were ≈200 m long and well within the land cover type (>30 m from the edge). Starting points were selected from the land cover maps to ensure that all classes of land cover surrounding the villages were surveyed, but bearing was random. Transects were not repeated, thereby avoiding repeated sampling of stable larval habitats. All larval habitats encountered in the field were investigated, except in land cover types where larval habitats were very abundant, such as irrigated fields, where a random representative sample was taken without regard for presence or absence of larvae. Although some larval habitat types are clustered spatially, within-transect spatial autocorrelation is minimized by the low number of larval habitats encountered in a transect (average of 2.3 habitats per transect), and between-transect spatial autocorrelation is very low due to the scattered distribution of transects around villages, in seven villages.
More than 200 samples were taken in each of these classes: forest and fallow, irrigated fields and villages. Orchards and nonirrigated fields, and peri-urban areas had fewer larval habitats and fewer than 50 samples were taken in these two classes. Habitats were searched for larvae by emptying the water from containers or by dipping with a white pan in larger water bodies. A sample of larvae was preserved in ethanol for identification. Morphological identification to species of Aedes or to species-group of Anopheles was conducted by R.E.H. Species within groups were identified by molecular methods (to be reported elsewhere).
Data analysis was carried out at the group level for Anopheles mosquitoes, which has practical advantages, such as easier and more reliable identification. Moreover, species-level information indicates that the An. minimus group was mostly represented by An. minimus s.s. (=species A; Harbach et al. 2006) and some Anopheles aconitus Dönitz; the An. maculatus group mostly by Anopheles maculatus s.s. and some Anopheles pseudowillmori Theobald and An. sawadwongporni; and the An. barbirostris group by An. barbirostris and two closely related but unidentified genetic forms. Except in the latter case, for which the mix of species is possibly more balanced, this should prevent bias related to varying behavior between species of the same group. Larvae of the An. hyrcanus group also were collected but almost entirely in irrigated rice fields (>90%). Therefore, land cover is clearly a predictor for the occurrence of larvae of this group but, due to high spatial autocorrelation between the observations, they could not be analyzed further.
Presence or absence of larvae is considered separately by species or species-group and does not take into account competition effects. Intraspecific (Dye 1984) and interspecific effects such as competition for food and chemical or physical interference (Lounibos et al. 2003, Juliano and Lounibos 2005) have been documented. In this study, however, most habitats surveyed had larvae of only one species or species-group.
Land Cover Data.
Land cover data were derived from Landsat Enhanced Thematic Mapper (LETM) imagery acquired on 5 March 2000. The Landsat imagery used has a spatial resolution of 30 by 30 m, which is appropriate for exploring the relationships between land cover and disease vectors provided the link between larval habitat and land cover is statistically significant (Beck et al. 2000). The panchromatic LETM data were not used. Image subsets of 12 by 12 km corresponding to the study villages were geometrically referenced using the 1:50,000 topographic map of Thailand (Royal Thai Survey Department 1992). Maximum likelihood classification (Richards 1993) was used to create land cover maps that included 10 classes: 1) mixed deciduous forest; 2) dry deciduous forest; 3) bushes or sparse forest; 4) irrigated wet areas (with a crop cover in March); 5) irrigated dry areas (with no crop cover in March); 6) orchards and house gardens (tree cover >60%); 7) upland fields and young orchards (tree cover <60%); 8) sparsely vegetated areas related to various human activities (no building or agriculture, e.g., wasteland, grassy areas); 9) densely built-up areas (peri-urban housing); and 10) water, shadows, and burned areas.
Accuracy of the classification was assessed by randomly selecting an equivalent number of pixels in each land cover class that were classified manually, independently from the classification result, by using visual interpretation, aerial pictures, and field observations. The estimated Kappa statistic (Congalton 1991) ranged between 0.76 and 0.84. Villages were first included in the orchard class, due to the presence of tall fruit trees around houses, or with bare soil. Two classes were therefore created using the housing information from the 1:50,000 topographic maps and field-collected village maps (van Benthem et al. 2005): village zones with dense vegetation and village zones with sparse vegetation. This detailed map was used. Topographic shadows and burned areas were not taken into account due to their temporary nature.
The field surveys, which spanned a whole year, encompassed seasonal variation that the remotesensing "snapshot" was not able to describe. Even though the landscape is varied and heterogeneous, most landscape units encountered (human settlements, fields, and orchards) are larger than the Landsat pixel, or clumped in large areas. This resolution obviously does not allow the detection of larval habitats. However, it appropriately describes the landscape, allowing the investigation of the relationships between landscape-level characteristics and the presence of larval habitats and larvae. This scale also allows linking land use with human exposure to vector-borne diseases (van Benthem et al. 2005, Vanwambeke et al. 2006). Seasonality of irrigated fields is partly included with the difference made between fields cultivated in the dry season and those only cultivated in the wet season.
Using a GIS, larval habitats were overlaid on the land cover map. This allowed us to derive landscape variables for each collection site. The proportion of each land cover class in a 200-m circular buffer around the collection points was calculated as well as a land cover diversity index for the same buffer. This buffer size was selected to capture two effects: the dependence of a larval habitat on the land cover patch where it is located (e.g., forest versus scattered trees in a field); and influences of the vegetation (leave shedding, production of vegetation debris, shadowing, and so on) and livestock in a landscape patch of a given size on larval habitats. This buffer size is slightly smaller than the median area of landscape patches where larval collections were made (equivalent to a radius of 223 m). It allows one to capture the diversity and the structure of locality at a sufficiently small scale. The shape index of the patch in which each collection site was located was calculated. The shape index compares patches of the landscape with a standard shape (McGarigal and Marks 1994) and indicates the complexity of a patch. The higher the index value, the more contact the patch has with its surroundings. The modified Simpson index was used for assessing diversity. Diversity describes the richness and evenness of a landscape: the number of different land cover types present and the area covered by each type. The absolute value of the index is not particularly meaningful and has to be used as a relative measure to compare landscapes (McGarigal and Marks 1994). The influence of the proximity of a land cover type beyond 200 m was not tested. Although discrepancy was low, to avoid potential positional errors related to the use of a GPS, we used the land cover at the habitat that was recorded when larvae were collected. This also partly avoids differences in land cover due to land cover change that could have happened between 2000 and 2003. Field visits confirmed that this was very limited, and transects were done in unchanged areas.
Statistical Methods.
Presence or absence of mosquito larvae in a habitat was used as the binary dependent variable, separately per species/species-group. The dependent variables were used in logistic bivariate regressions with larval habitat and land cover variables as independent variables. Two multiple logistic regressions were thus fitted, for each species/species-group. The first model used the habitat type and habitat characteristics (from field survey data), including season. The dry season is from November to April and the wet season is from May to October. The second model used the land cover data (from remotely sensed data), including season. Continuous variables were categorized using quantiles; the number of quantiles was chosen to have detailed information with sufficient frequencies in each class. All variables significant at the 0.15 level in the bivariate analysis were introduced in the multiple logistic regressions. Variables that were not significant at the 0.05 level were deleted from the multiple model, one at a time. Confounding and interaction effects were tested and significant effects were integrated in the model (P < 0.05). Colinearity effects also were tested. The models were compared using the pseudo-_R_2 (_ρ_2) measure. Values between 0.2 and 0.4 indicate a good model fit (Wrigley 1985). Statistical analyses were done using the SAS software (SAS Institute, Cary, NC). Percentage of correctly predicted records of larvae presence was calculated at the 0.1 probability threshold, given the low occurrence of presence (Greene 2003).
Results
Description of Larval Habitats
Ae. albopictus larvae were found in 90 collection sites, mostly in villages (70%), and nonirrigated fields (including orchards) (14%). Artificial containers made up 87% of collections (Table 1), and natural containers made up 11% of collections. Two collections were made in surface-water habitats. About half of the Ae. albopictus collections were made in partially shaded habitats, with fruit trees close enough to influence the larval habitat by shade or debris. Most of the larvae were in temporary, stagnant water. They were found near animals in 80% of the cases. Larvae were found in habitats with a moderate or abundant quantity of leaves (60%), stems (80%), or excrement (66%). Ninety-five percent (95%) of the Ae. albopictus larvae were collected in the wet season.
Table 1
Collections of Ae. albopictus larvae in artificial containers
Table 1
Collections of Ae. albopictus larvae in artificial containers
Larvae of the An. minimus group were collected from 93 sites, mostly in stream margins (71% of collections) and stream pools (Table 2). They were mostly present in permanent bodies of water with slow or moderate movement. Seventy-three percent were found in partial shade and 20% in habitats with no shade. Fruit trees were rarely found around (i.e., close enough to contribute shade and debris) sites harboring larvae, but evergreen trees were often present. Deciduous trees were much less frequent. Larvae were often found in habitats with emergent vegetation, submerged leaves, stems, and excrement. Water was clear in 93% of the habitats. Ninety percent of larvae were collected in two villages with highly fragmented landscapes. Seventy-eight percent were collected in the dry season (November–March).
Table 2
Collections of Anopheles in surface-water habitats
Table 2
Collections of Anopheles in surface-water habitats
Larvae of the An. maculatus group were collected 42 times, mostly in forests (69% of collections), often in stream pools (52%) (Table 2). Most of these habitats had temporary, stagnant water, with partial or no shade. Evergreen trees were present around 76% of the collection sites harboring larvae; deciduous trees were much less frequent. Aquatic vegetation was present in 76% of the habitats, leaves in 62%, stems in 69%, and excrement in 81%. The water was clear in 67% of the habitats. Sixty-nine percent of the larvae were found in the dry season.
Larvae of the An. barbirostris group were found in all villages, in a total of 38 collection sites (Table 2). Larvae were found in a variety of mostly temporary surface-water habitats, including ditches, rice paddies, ponds, stream margins, and stream pools, and once in an artificial container. Larvae were found mainly in clear water where emergent vegetation was present without algae. They were found principally in the dry season.
Larval Habitats and Land Cover
Fifteen percent of all artificial containers were found in orchards and these were mostly tires. Only 9% of artificial containers were found in peri-urban housing areas, possibly because of the availability of modern facilities and of well-kept gardens and surroundings. However, the peri-urban areas surveyed were mostly housing compounds where high walls enclose homes and people are often away during the day. Collections were therefore biased toward villages. Seventy-six percent of artificial containers were found in villages. Most of the records of artificial containers were made in the wet season, when they were filled by rain. Twelve percent of plant containers were found in orchards and 18% in villages. All degrees of shade (at the time of collection) were found in orchards. Most collection sites in orchards had temporary, stagnant water. In villages, half of the collection sites were in partial shade and 37% had no shade. Sites were mostly partially shaded in peri-urban housing areas. Most collection sites in peri-urban housing areas were temporary and stagnant, with no aquatic vegetation.
Stream margins and stream pools constituted 74% of the surface-water larval habitats sampled in forest (63% when fallow and forest are merged). Ground pools, rock pools, and hoof- or footprints totaled ≈16% of the surface-water habitats sampled. Most larval habitats sampled in fallow areas were exposed to sunlight most of the day. In forests, most habitats were partially shaded. Most larval habitats sampled in fallow and forest were temporary. Seventy-five percent in fallow had stagnant water, but 36% had slow or moderate current in forest. Seventy percent of habitats in fallow had aquatic vegetation, but only half of those in forest contained vegetation. Ditches constituted ≈40% of surface-water habitats found in villages and stream margins 23%. Stream pools, ground pools, and ponds also were found in villages. Half of the collection sites in villages were temporary, stagnant, and partially shaded. Twenty-eight percent of larval habitats found in villages had aquatic vegetation.
Some 86% of surface-water habitats found in irrigated fields were rice paddy. Most larval habitats in irrigated fields had no shade and 98% were temporary and stagnant. Ditches were found in each land cover type, but more than half of them occurred in peri-urban housing areas.
Logistic Regression Modeling of Presence or Absence of Larvae
Ae. albopictus.Larval Habitat Model. The chance of finding Ae. albopictus larvae in an artificial container was significantly higher than in natural containers, the Odds Ratio (OR) being the highest for klong jars and pots (Table 3). The presence of banana (Musa spp.) or longan (Dimocarpus longan Lour.) trees close enough to influence the habitat by their shadow or debris increased the chance of finding Ae. albopictus larvae. The presence of animals also significantly increased the probability of finding Ae. albopictus larvae. Collection sites with partial and no shade had significantly lower probabilities of harboring Ae. albopictus larvae. The location of collection sites in grassy fallow was also positive and significant after adjusting for other variables. Finally, season was a highly significant variable: Ae. albopictus larvae were 19 times more likely to be found in the wet season than in the dry season. The model had a very high _ρ_2 of 0.61 (89% of presence records correctly predicted at the 0.1 probability threshold).
Table 3
Logistic regressions for Ae. albopictus larvae: larval habitat and land cover
Table 3
Logistic regressions for Ae. albopictus larvae: larval habitat and land cover
Land Cover Model. The probability of finding Ae. albopictus larvae in orchards, peri-urban settlements, or villages was higher than in other land covers (Table 3). The highest OR was for collection sites in villages. The probability was higher for collection sites with a higher proportion of orchards. Intermediate fragmentation levels have a positive influence on the presence of Ae. albopictus larvae. In this case again, habitats had a much higher probability of harboring Ae. albopictus in the wet season. The model had a very high _ρ_2 of 0.52 (90% of presence records correctly predicted).
Considering that artificial containers are very important for Ae. albopictus and that these habitats are filled with water mainly in the wet season, models were recomputed to include only collections made in the wet season. The larval habitat model was the same except that the presence of longan trees was no longer significant, and the land cover model was identical (except for the absence of the season variable). These models had _ρ_2 values of 0.48 and 0.49, respectively.
Anopheles minimus Group.
Larvae of the An. minimus group were collected in surface-water habitats and some larval habitat variables only apply to those habitats (Table 4). A reduced database including only surface-water habitats was therefore created and used in the logistic regression analysis.
Table 4
Logistic regressions for larvae of the An. minimus group: larval habitat and land cover
Table 4
Logistic regressions for larvae of the An. minimus group: larval habitat and land cover
Larval Habitat Model. Once adjusted for other variables, the presence of floating or emergent plants increased the probability of finding larvae of the An. minimus group four times (Table 4). The presence of algae and excrement decreased the probability of finding larvae. Turbid and temporary water also had a lower chance of harboring larvae. Stream margins and stream pools were the larval habitats most likely to harbor larvae. An interaction term between the presence of emergent plants and the presence of algae was significant. This indicates that, when both emergent plants and algae are present, emergent plants favor the presence of larvae even more than when algae are absent, and, inversely, that the effect of the presence of algae is less unfavorable when emergent plants are present. The odds ratio (OR) for the presence of emergent plants was 3.60 when algae were absent and 157 when algae were present. This model had a _ρ_2 of 0.55 (93% of presence records correctly predicted).
Land Cover Model. Increasing proportions of mixed and dry deciduous forest both increased the probability of finding larvae, whereas the proportion of irrigated field decreased it (Table 4). The odds of finding larvae of the An. minimus group were significantly higher for habitats located in orchards and villages than in forests. Sites located in the most diverse landscapes, as indicated by the modified Simpson index, had a fivefold higher chance of harboring larvae of the An. minimus group. The probability of finding larvae was significantly lower in the wet season. An interaction term between season and proportion of mixed deciduous forest was significant, indicating that the chance of finding larvae of the An. minimus group in areas with mixed deciduous forest was much greater in the dry season. The model has a very high _ρ_2 of 0.51 (92% of presence records correctly predicted).
Anopheles maculatus Group.
Larvae of the An. maculatus group also were studied in a reduced database including only surface-water habitats (Table 5).
Table 5
Logistic regressions for larvae of the An. maculatus group: larval habitat and land cover
Table 5
Logistic regressions for larvae of the An. maculatus group: larval habitat and land cover
Larval Habitat Model. Larvae of the An. maculatus group were more likely to be found in stream margins, rice paddies, and ground pools than in stream pools (Table 5). Scarce floating plants reduced the chance of finding larvae, whereas the presence of fruit trees, evergreen trees, and animals increased it. Location of a collection site in grassy fallow or in irrigated fields with garlic increased the probability of finding larvae of the An. maculatus group. The model has a _ρ_2 of 0.34 (81% of presence records correctly predicted).
Land Cover Model. The probability of finding larvae of the An. maculatus group was higher in collection sites with bush/sparse forest, and with villages with vegetation within a 200-m radius (Table 5). Higher proportions of orchards and higher diversity of the landscape (modified Simpson index) decreased the probability of finding larvae of the An. maculatus group. This model had a _ρ_2 of 0.36 (90% of presence records correctly predicted).
Anopheles barbirostris Group.
Larval Habitat Model.
This model had few significant variables and a low _ρ_2 of 0.12 (53% of presence records correctly predicted). Larvae of the An. barbirostris group had a significantly higher probability of being found in ditches, with scarce or moderate amounts of floating debris, and in habitats where no monocotyledons were found (Table 6). The presence of longan trees also increased the probability of finding larvae of the An. barbirostris group in a collection site.
Table 6
Logistic regressions for larvae of the An. barbirostris group: larval habitat and land cover
Table 6
Logistic regressions for larvae of the An. barbirostris group: larval habitat and land cover
Land Cover Model. The land cover model also yielded poor results with a _ρ_2 of 0.14 (53% of presence records correctly predicted). The presence of dry/deciduous forest within a 200-m radius around the habitat increased the probability of finding larvae of the An. barbirostris group, as did the presence of over 35% of the land cover type village with vegetation (Table 6). Collection sites surrounded by intermediate proportions of orchards had a significantly lower chance of harboring larvae. Habitats located in areas with a large diversity index had a smaller chance of harboring them.
Discussion
The analysis of occurrence of various larval habitats and their characteristics in different land cover types indicated that larval habitats and land cover are strongly associated. The most obvious case is irrigated field habitats, which are defined by their land cover. But other significant associations were found, such as the occurrence of stream habitats in forests. The presence of evergreen vegetation usually indicates higher moisture content of the soil (Schmidt-Vogt 1999) and possibly the presence of a stream, and it was a significant predictor of the presence of larvae of the An. minimus group, which are most often found in stream habitats. Evergreen vegetation can be easily distinguished from deciduous vegetation by remote sensing in the dry season. Larvae of the An. minimus group were also found in villages, which are the second most important source of stream margins and pools. Other studies also found associations between landscape features and mosquito habitats, e.g., Rejmankova et al. (1998) associated herbaceous wetlands with An. albimanus larvae in Belize. Beck et al. (1994) associated transitional swamp and unmanaged pastures with An. albimanus in Chiapas, Mexico. Artificial containers were found in very high densities in areas related to human activities, but great variation was observed between types of human settlements. This might partially reflect sampling constraints but houses in the suburbs of Chiang Mai probably had tidier surroundings. Factors such as housing quality and socioeconomic status cannot be easily detected by remote sensing. This limits our ability to generalize these observations to other areas. Direct detection of larval habitats in housing areas on remotely sensed images or even aerial photographs also proves challenging, as indicated by Moloney et al. (1998) and Jacob et al. (2005). The surrounding land cover heavily influences the characteristics of larval habitat, such as shade. Emergent and floating plants were mostly found in habitats in irrigated fields, or in forests. Remotely sensed information is most useful if landscape elements that are critical to survival of the vectors can be identified (Beck et al. 2000), not only due to spatial autocorrelation of landscape attributes but also due to biologically meaningful associations (Kitron et al. 1996).
Regression models using larval habitat or land cover variables showed comparable ability to predict the presence or absence of larvae. The presence of fruit trees increased the probability of finding Ae. albopictus, which also was indicated in the land cover model by the significance of the proportion of orchards around collection sites and the higher probability of finding Ae. albopictus in orchards than in other land covers (except villages). The impact of shade is represented in the land cover model by the proportion of orchards. It also might be related to villages and peri-urban housing. Lowland villages have many old trees, which is not the case for recent housing compounds. Other factors were only represented in one of the two models, e.g., grassy fallow. In the land cover data, fallows were included with forest, whereas there can be many types of vegetation on fallow land. This emphasizes the point that an accurate knowledge of mosquito ecology should drive land cover classification of remotely sensed images. The land cover model for the An. maculatus group is partly consistent with the microhabitat model that considers the presence of fruit or wild trees. This also was described in the land cover model through the presence of bush and sparse forest, orchards, and villages with vegetation. The presence of dry irrigated fields increases the likelihood of finding larvae of the An. barbirostris group in a habitat, and this is consistent with the larval habitat model where the presence of monocotyledons decreased the presence of these larvae. Dry irrigated fields are a seasonal land cover that describes irrigated fields when they are not in use. The influence of the proportion of orchard was less clear.
Species-groups with more plastic habitat requirements made the modeling more difficult, as was noticed for the An. barbirostris group, and to a lesser extent the An. maculatus group. Larvae of the latter group inhabit a variety of habitats that can be found in several land cover types. The models for the An. maculatus group nonetheless gave satisfactory results, indicating that the variables appropriately describe the habitat preferences of species that comprise this group. Members of the An. barbirostris group are even more flexible in terms of larval habitat than members of the An. maculatus group, because they use a variety of habitat types with various characteristics. Other explanations could be that none of the factors included in this study captured the characteristics of suitable habitats for species of the An. barbirostris group or that the group was represented by a balanced mix of species with different requirements.
Variables describing the landscape structure were also significant in several land cover models. The Shape index was included in the model for Ae. albopictus and the modified Simpson index of diversity was significant in the models for the An. minimus and An. maculatus groups. Members of the An. minimus group are well suited for forest fringe near human activities and settlements where the landscape is the most diverse, whereas members of the An. maculatus group were found in more homogeneous landscapes. The results for the An. minimus group differ from what Overgaard et al. (2003) found for adult Anopheles mosquitoes in northern Thailand. Results of that study showed contradictory effects of various fragmentation measures, and adult mosquito diversity might relate differently to fragmentation than the larval habitats of individual species. The effects of landscape fragmentation on insects are poorly understood but are thought to influence abundance, diversity, and the interaction between species (Didham et al. 1996). The present results, however, indicate potential additional input that land cover analysis can bring to studies on vector ecology.
The larval habitat models mostly agreed with other studies of Anopheles species-groups (Harbach et al. 1987. Rattanarithikul et al. 1994, Rattanarithikul et al. 1995, Sithiprasasna et al. 2003), except for the presence of An. minimus s.l. in irrigated fields, (Rattanarithikul et al. 1995, Oo et al. 2004), which was not found in our survey. As in other studies, members of the An. minimus group were mostly found in the dry season. Habitat factors might be conditioned by season: stream margins might be more hospitable in the dry season due to slower stream flow. This effect was described in the land cover model by the interaction between mixed deciduous forest and season.
The link between land cover and microhabitat could be used in control programs by yielding information on potential larval habitats more efficiently than by field survey. Specific characteristics of habitats can be used to minimize manipulations of the environment (Coosemans and Mouchet 1990). Elimination of larval habitats needs to focus on the most productive targets to reach thresholds beyond which transmission cannot take place (Focks and Chadee 1997, Focks et al. 2000). Recent studies of An. gambiae in Kenya indicate that pupal productivity varied between larval habitat types (Mutuku et al. 2006) and by land cover type (Munga et al. 2006). Abundance and survival of larvae would therefore be a valuable addition to the data used here.
Larvae of Ae. albopictus were found in orchards as well as in villages. Orchards have increased substantially in the region and more information on the role of Ae. albopictus in dengue transmission would be useful in assessing the potential health impact of this increase. Sulaiman and Jeffrey (1986) found Ae. albopictus biting and developing in a rubber estate in Malaysia. Also, several studies indicate that forest-dwelling species such as members of the Anopheles dirus Peyton & Harrison or An. minimus complexes could adapt well to tree crops as a result of destruction of their usual habitat (Rosenberg et al. 1990, Singhasivanon et al. 1999). This study shows no evidence of members of Anopheles species-groups occurring in orchards. Absence of suitable surface-water habitats and use of pesticides could explain why mosquitoes do not colonize such areas (Gingrich et al. 1990, Overgaard et al. 2002). However, Ae. albopictus larvae were found in orchards.
In conclusion, the land cover and larval habitat variables that proved to be useful in predicting the presence of mosquito larvae in potential habitats were similar for most species and species-groups studied. Land cover is a good proxy for several important larval habitat variables but is also, in itself, one of the characteristics that influences the suitability of habitats, through the landscape structure. These attributes could influence the suitability of the habitat depending on whether fragmented transition zones between land cover types are appropriate. However, great care is needed when interpreting the role of land cover. The link between habitat and land cover characteristics is not always straightforward. For this reason, associating land cover with mosquito habitat preferences is helped considerably by knowledge of mosquito ecology.
The results show very good potential for land cover to predict mosquito presence and absence. This could be of great use in designing effective control programs. The direct applicability of these results is, however, constrained by two factors. First, regional variation in mosquito behavior, including selection of larval habitats, could possibly modify the relationships presented here. Second, the link between presence of larval habitats and land cover is likely to vary between places and possibly through time as well. This is most important for artificial containers. The association between natural land cover types and habitats is probably more apparent within a region with similar vegetation and topographic characteristics. We did not include the productivity of larval habitats, even though this would be of importance in studying disease transmission. The results, however, indicate that land cover is a good predictor of larval presence, especially for species or species-groups with specific habitat preferences.
Acknowledgements
This study was supported by European Union Grant QLRT-1999-31787, provided within the Quality of Life and Management of Living Resources Programme (1998–2002).
References Cited
.
2000
.
Remote sensing and human health: new sensors and new opportunities
.
Emerg. Infect. Dis.
6
:
217
–
227
.
et al. .
1994
.
Remote sensing as a landscape epidemiologic tool to identify villages at high risk for malaria transmission
.
Am. J. Trop. Med. Hyg.
51
:
271
–
280
.
.
2000
.
Spatial targeting of intervention against malaria
.
Bull. World Health Organ.
78
:
1401
–
1411
.
1991
.
A review of assessing the accuracy of classification of remotely sensed data
.
Rem. Sens. Environ.
37
:
35
–
46
.
.
1990
.
Consequences of rural development on vectors and their control
.
Ann. Soc. Belg. Med. Trop.
70
:
5
–
23
.
.
1996
.
Insects in fragmented forests: a functional approach
.
Trends Ecol. Evol.
11
:
255
–
260
.
1984
.
Competition amongst larval Aedes aegypti: the role of interference
.
Ecol. Entomol.
9
:
355
–
357
.
.
2002
.
Epidemiology of inapparent and symptomatic acute dengue virus infection: a prospective study of primary school children in Kamphaeng Phet, Thailand
.
Am. J. Epidemiol.
156
:
40
–
51
.
.
2000
.
Transmission thresholds for dengue in terms of Aedes aegypti pupae per person with discussion of their utility in source reduction efforts
.
Am. J. Trop. Med. Hyg.
62
:
11
–
18
.
.
1997
.
Pupal survey: an epidemiologically significant surveillance method for Aedes aegypti: an example using data from Trinidad
.
Am. J. Trop. Med. Hyg.
56
:
159
–
167
.
.
2005
.
Morphological assessment and molecular phylogenetics of the funestus and minimus groups of Anopheles (Cellia)
.
J. Med. Entomol.
42
:
522
–
536
.
.
1990
.
Hyperendemic malaria in a Thai village: dependence of year-round transmission on focal and seasonally circumscribed mosquito (Diptera: Culicidae) habitats
.
J. Med. Entomol.
27
:
1016
–
1026
.
2004
.
Critical review of the vector status of Aedes albopictus
.
Med. Vet. Entomol.
18
:
215
–
227
.
2003
.
Econometric analysis, 5th ed
.
Prentice Hall
,
Englewood Cliffs, NJ
.
1994
.
Review of the internal classification of the genus Anopheles (Diptera: Culicidae): the foundation for comparative systematics and phylogenetic research
.
Bull. Entomol. Res.
84
:
331
–
342
.
.
1987
.
Some entomological observation on malaria transmission in a remote village in northwestern Thailand
.
J. Am. Mosq. Control Assoc.
3
:
296
–
301
.
.
2006
.
Anopheles (Cellia) minimus Theobald (Diptera: Culicidae): neotype designation, characterization, and systematics
.
Proc. Entomol. Soc. Wash.
108
:
198
–
209
.
.
1997
.
The impact of remote sensing on the study and control of invertebrate intermediate hosts and vectors for diseases
.
Int. J. Rem. Sens.
18
:
2899
–
2930
.
.
2005
.
Capabilities of multisprectral thermal data for identification of Anopheles gambiae mosquito larval habitats in African urban environments
.
Int. J. Rem. Sens.
26
:
523
–
534
.
.
2005
.
Ecology of invasive mosquitoes: effects on resident species and on human health
.
Ecol. Lett.
8
:
558
–
574
.
1998
.
Landscape ecology and epidemiology of vector-borne diseases: tools for spatial analysis
.
J. Med. Entomol.
35
:
435
–
445
.
.
1996
.
Spatial analysis of the distribution of tsetse flies in the Lambwe Valley, Kenya, using Landsat TM satellite imagery and GIS
.
J. Anim. Ecol.
65
:
371
–
380
.
.
2003
.
Dynamics of land-use and land-cover change in tropical regions
.
Annu. Rev. Environ. Res.
28
:
205
–
241
.
.
2001
.
Anopheles barbirsotris/campestris as a probable vector of malaria in Aranyaprathet, SA Kaeo Province
.
Southeast Asian J. Trop. Med. Public Health
32
:
739
–
744
.
.
2003
.
Interactions with native mosquito larvae regulate the production of Aedes albopictus from bromeliads in Florida
.
Ecol. Entomol.
28
:
551
–
558
.
.
1994
.
Fragstats: spatial pattern analysis program for quantifying landscape structure, version 2.0
.
Forest Science Department, Oregon State University
,
Corvallis, OR
.
.
1998
.
Domestic Aedes aegypti breeding site surveillance: limitations of remote sensing as a predictive surveillance tool
.
Am. J. Trop. Med. Hyg.
59
:
261
–
264
.
1988
.
Anopheline mosquitos [sic]: vector reproduction, life-cycle and biotope, pp. 431–451
. In [eds.],
Principles and practice of malariology
.
Churchill Livingstone
,
Edinburgh, United Kingdom
.
.
2006
.
Association between land cover and habitat productivity of malaria vectors in western Kenyan highlands, Am
.
J. Trop. Med. Hyg.
74
:
69
–
75
.
.
2006
.
Pupal habitat productivity of Anopheles Gambiae complex mosquitoes in a rural village in western Kenya
.
Am. J. Trop. Med. Hyg.
74
:
54
–
61
.
.
2003
.
Climatic and social risk factors for Aedes infestation in rural Thailand
.
Trop. Med. Int. Health
8
:
650
–
659
.
.
2004
.
Review of the anopheline mosquitoes of Myanmar
.
J. Vector Ecol.
29
:
21
–
40
.
.
2003
.
Effect of landscape structure on anopheline mosquito density and diversity in northern Thailand: implications for malaria transmission and control
.
Landsc. Ecol.
18
:
605
–
619
.
.
2002
.
Characteristics of Anopheles minimus (Diptera: Culicidae) larval habitats in northern Thailand
.
Environ. Entomol.
31
:
134
–
141
.
.
1995
.
Larval habitats of malaria vectors and other Anopheles mosquitoes around a transmission focus in northwestern Thailand
.
J. Am. Mosq. Control Assoc.
11
:
428
–
433
.
.
1994
.
Dry season distribution of mosquito larvae in the bed of the Mekong river, northeastern Thailand
.
J. Am. Mosq. Control Assoc.
10
:
197
–
201
.
.
2004
.
Phylogeny and classification of Aedini (Diptera: Culicidae), based on morphological characters of all life stages
.
Zool. J. Linn. Soc.
142
:
289
–
368
.
.
1998
.
Characterization and detection of Anopheles vestitipennis and Anopheles punctimacula (Diptera: Culicidae) larval habitats in Belize with field survey and SPOT satellite imagery
.
J. Vector Ecol.
23
:
74
–
88
.
1993
.
Remote sensing digital image analysis: an introduction
.
Springer
,
Berlin, Germany
.
.
1996
.
Predictions of malaria vector distribution in Belize based on multispectral satellite data
.
Am. J. Trop. Med. Hyg.
54
:
304
–
308
.
.
1990
.
Highly efficient dry season transmission of malaria in Thailand
.
Trans. R. Soc. Trop. Med. Hyg.
84
:
22
–
28
.
1999
.
Swidden farming and fallow vegetation in Northern Thailand
.
Franz Steiner Verlag
,
Stuttgart, Germany
.
.
2002
.
The Anopheles vector, pp. 59–84
. In [eds.],
Essential malariology
, 4th ed.
Arnold
,
London, United Kingdom
.
.
1999
.
Malaria in tree crop plantations in south-eastern and western provinces of Thailand
.
Southeast Asian J. Trop. Med. Public Health
30
:
399
–
404
.
.
2003
.
Use of GIS-based spatial modeling approach to characterize the spatial patterns of malaria mosquito vector breeding habitats in northwestern Thailand
.
Southeast Asian J. Trop. Med. Public Health
34
:
517
–
528
.
.
1986
.
The ecology of Aedes albopictus (Skuse) (Diptera: Culicidae) in a rubber estate in Malaysia
.
Bull. Entomol. Res.
76
:
533
–
557
.
.
1995
.
Seasonal density and malaria vectors competence of Anopheles minimus and other anophelines at a shallow valley in north Thailand
.
Jpn. J. Trop. Med. Hyg.
23
:
177
–
182
.
.
2000
.
Local-scale variation in malaria infection amongst rural Gambian children estimated by satellite remote sensing
.
Trans. R. Soc. Trop. Med. Hyg.
94
:
159
–
163
.
.
2005
.
Behavioural heterogeneity of Anopheles species in ecologically different localities in Southeast Asia: a challenge for vector control
.
Trop. Med. Int. Health
10
:
251
–
262
.
.
2005
.
Spatial patterns of and risk factors for seropositivity for dengue infection
.
Am. J. Trop. Med. Hyg.
72
:
201
–
208
.
.
1992
.
Estimating high mosquito-producing rice fields using spectral and spatial data
.
Int. J. Rem. Sens.
13
:
2813
–
2826
.
1985
.
Categorical data analysis for geographers and environmental scientists
.
Longman
,
London, United Kingdom
.
© 2007 Entomological Society of America
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/), which permits non-commercial reuse, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com