Dengue is a vector-borne virus transmitted by the bites of infected mosquitoes, mainly Aedes aegypti (Hotez et al. 2008; Kyle and Harris 2008; WHO 2009). It is an acute febrile disease causing 50–100 million cases per year and for which more than 2.5 billion people are at risk (WHO 2009). Dengue incidence increased 30-fold in the last 50 years (CDC 2009), coinciding with rapid urbanization. Given the close link of the vector to urban areas (WHO 2009), dengue incidence is expected to rise with increasing urban population, especially in developing countries where urban growth outpaces government’s capacity to provide infrastructure (Barrera et al. 1993, 1995; Troyo et al. 2009).

Individuals who acquire dengue develop immunity to the infecting serotype, but subsequent infection with a different serotype can lead to the more severe forms of dengue, including dengue hemorrhagic fever and dengue shock syndrome (Kendall et al. 2009). Due to this complication of multiple viral serotypes and immunological responses to sequential coinfection, there is currently no vaccine for dengue (CDC 2009); however, a tetravalent vaccine is currently in clinical trials (Morrison et al. 2010). With or without a vaccine, best practices for dengue prevention are integrated vector control methods including reduction of vector-producing containers and community education and participation (Gubler 2006). Sustaining community participation strategies in inter-epidemic years is difficult, though integral to the success of these programs (Lloyd 2003). By identifying areas of high dengue transmission risk, resources can be focused to the areas of greatest need (Wilson 2002).

Aedes aegypti lives in close association with humans and preferentially bites human hosts (Harrington et al. 2001; Kyle and Harris 2008; CDC 2009). Given its key role on the emergence of epidemic dengue in urban areas, vector control methods have focused on controlling Ae. aegypti by elimination of domestic and peridomestic water filled artificial containers that produce mosquitoes. However, the efficacy of vector control methods may be limited by the presence of alternate invasive or native container breeding species. In Puerto Rico, Ae. mediovittatus may be implicated in the transmission of dengue based on the laboratory observations that it feeds on humans, has greater vector competence and rates of vertical transmission of dengue viruses than Ae. aegypti (Gubler et al. 1985; Freier and Rosen 1988). Aedes mediovittatus is a native tree hole mosquito and is closely associated with arboreal vegetation (Cox et al. 2007). It has been proposed that Ae. mediovittatus can act as a reservoir in the maintenance of dengue viruses during inter-epidemic periods in rural areas with low human population density (Gubler et al. 1985). Areas where Ae. aegypti and Ae. mediovittatus co-occur may be at higher risk for dengue persistence between epidemics. The rationale is that: (1) during epidemics, high virus amplification by Ae. aegypti would increase the likelihood of Ae. mediovittatus becoming infected; (2) dengue viruses would persist in these areas during inter-epidemic periods due to the high rate of vertical transmission observed in Ae. mediovittatus in the laboratory; and (3) dengue transmission by Ae. mediovittatus to humans and Ae. aegypti areas of overlap could catalyze another epidemic. While targeting areas of high Ae. aegypti production is important during epidemic periods, habitats of high Ae. mediovittatus production may be better targets during inter-epidemic periods. Our aim in this study is to develop a predictive model of areas where both Aedes species overlap to conduct intensive sampling of Ae. mediovittatusand assess infection with DENV. Moreover, identifying the areas where these two species overlap is critical to focus interventions to curb future dengue epidemics.

The spatial overlap of adult Ae. mediovittatus, Ae. aegypti, dengue viruses, and humans in Puerto Rico has not been determined (Cox et al. 2007). In Puerto Rico, immature Ae. aegypti have been shown to occur close to houses (99%) while Ae. mediovittatus were more likely to occur far from houses (Moore 1983). In San Juan, Puerto Rico, immature Ae. aegypti and Ae. mediovittatus were found to segregate across an urban gradient (Cox et al. 2007) with Ae. aegypti found in areas with higher housing density (Smith et al. 2009). However, Ae. mediovittatus was found in containers within San Juan, particularly in areas of lower density housing and forested areas (Cox et al. 2007; Smith et al. 2009) and at the peak of the rainy season when most dengue viral disease transmission occurs (Smith et al. 2009). Ae. mediovittatus requires high humidity to survive as an adult (Cox et al. 2007) which may explain why it prefers vegetated areas.

Research utilizing remote sensing for the study of dengue has been limited (Eisen and Lozano-Fuentes 2009). There are four proposed explanations for the limited use of mapping, including remote sensing, for the study of dengue risk: (1) traps for adult Ae. aegypti had not been efficacious and emphasis has been placed on sampling immatures (Eisen and Lozano-Fuentes 2009); (2) dengue researchers have focused on the importance of small containers to determine the risk of dengue, precluding the utility of remote sensing which has not been shown to be able to discern containers at the fine scale needed (Moloney et al. 1998; Eisen and Lozano-Fuentes 2009); (3) environmental characteristics such as the normalized difference vegetation index (NDVI) may only be relevant in locations where containers are rain filled and trees prevent evapotranspiration rather than filled and maintained by humans (Eisen and Lozano-Fuentes 2009); and (4) the risk of dengue outbreaks is influenced not only by the abundance of mosquitoes but also the serotype specific herd immunity of the population (Eisen and Lozano-Fuentes 2009).

These concerns have resulted in a focus on epidemiological approaches rather than entomological or ecological. There are, however, exceptions. Remote sensing has been used to map landscape features such as tidiness and shade levels of backyards as predictors of suitable Ae. aegypti breeding habitats (Tun-Lin et al. 1995; Moloney et al. 1998) and to map environmental conditions, temperature, and precipitation, using NDVI as a proxy for precipitation (Estallo et al. 2008). Remote sensing has also been used to map dengue transmission risk by mapping land use change in Thailand (Vanwambeke et al. 2007) and landscape structure in Hawaii (Vanwambeke et al. 2011). Still there are few studies using remote sensing at high spatial resolution to assess Ae. aegypti habitat in urban areas (Troyo et al. 2009).

Here we used high resolution remote sensing to map elements of urban structure, tree cover and built-up areas, to predict Ae. aegypti within urban areas as has been done in previous studies (Troyo et al. 2008; Fuller et al. 2010). The goal of this research is to determine environmental features associated with high abundance of Ae. aegypti and Ae. mediovittatus at different spatial scales and identify and map the environmental conditions that result in high levels of co-occurrence. Since immature mosquitoes have been shown to segregate along an urbanization gradient (Cox et al. 2007), we hypothesize that Ae. aegypti will be more closely associated with urban areas while Ae. mediovittatus will be more likely to be found in forested areas. We used novel BG-sentinel traps previously shown to be efficacious at collecting adult Aedes spp. mosquitoes (Kroeckel et al. 2006; Maciel de Freitas et al. 2006; Williams et al. 2006) and data from a newly launched satellite of high spatial and spectral resolution to characterize features of the urban environment predictive of habitat segregation of the two Aedes species. Mapping areas of co-occurrence will identify areas that will be used to refine a trapping scheme aimed at determining the role of Ae. mediovittatus in the transmission dynamics of dengue in Puerto Rico.

We obtained 1,260 mosquito collections from 235 trapping sites, with an average of 5.4 collections per trap site (range 2–6). Over the entire sampling period we collected 7,057 males and 8,787 females (0.8♂:1♀) of Ae. aegypti and 92 males and 1,090 females (0.08♂:1♀) of Ae. mediovittatus. Average numbers of Ae. aegypti adults per trap per day were higher than those of Ae. mediovittatus in most study sites, with the exception of Mulas (Table 1). The two localities with the highest densities of Ae. aegypti per trap, such as Lamboglia and Quebrada Arriba also showed the highest densities of Ae. mediovittatus (Table 1). The percentages of traps with both species (co-occurrence) also were the highest in Lamboglia and Quebrada Arriba (Table 1). Co-occurrence was minimal in the two localities with the lowest densities (Mamey and Mulas). These areas received the lowest levels of rainfall that were observed during the field study (Table 1). Other mosquito species collected in BG traps during the study were Aedes taeniorhynchus, Ae. tortilis, Culex bahamensis, Cx. quinquefasciatus, Cx.antil-lummagnorum, and Toxorhynchites portoricensis.

Satellite image classification accuracy ranged from 80 to 99% with associated kappa coefficients of 0.73 to 0.9 (Fig. 1). For all three outcomes (presence of high numbers of Ae. aegypti compared to low numbers, presence of Ae. mediovittatus compared to absence, and co-occurrence of high numbers of Ae. aegypti and presence of Ae. mediovittatus), we found that mixed scale models had the best predictive values, indicated by lowest QIC scores (Table 2). For Ae. aegypti, the model with the best fit included the covariates number of tree patches within 50 m, housing density within 25 m, rainfall, and temperature and was statistically significant (χ² = 273, df = 5, P < 0.001). For Ae. mediovittatus, the final model included number of tree patches within 100 m, urban patch area within 50 m, rainfall, and temperature and was statistically significant (χ² = 62, df = 5, P < 0.001), The overall model predicting co-occurrence (statistically significant (χ² = 137, df = 5, P < 0.001)) included the covariates housing density within 25 m, the number of tree patches within 100 m, and rainfall.

We used Receiver Operating Characteristic (ROC) to plot the true positive rate (sensitivity) as a function of one minus the false positive rate (specificity) (STATA SE 10 2007). The model predicting Ae. aegypti results in an AUC value of 0.83, suggesting that a random sample of positive observations has a test value larger than that of negative observations 83% of the time (mediovittatus 64% of the time; co-occurrence 75% of the time). Moran’s I statistic based on the regression residuals was not significant in five of the seven study sites, and significant for two, indicating some residual spatial structure not explained by the covariates specified.

To create an ecological risk map of co-occurrence across the study area we used the logistic regression model predicting co-occurrence. To create the raster grid for house density we calculated the kernel density at 25 m throughout the entire study area. To extract values of the number of tree patches over the entire study area we overlaid a 50 m × 50 m grid and calculated the number of tree patches in each grid cell. The map depicting risk of cooccurrence illustrates the heterogeneity across the landscape within the inhabited areas of Patillas, Puerto Rico (Fig. 2). The communities Lamboglia and Providencia show the greatest risk for co-occurrence of Ae. aegypti and Ae. mediovittatus. These are areas of higher urban density but also have a high number of tree patches.

In this study, we identified density of urban and vegetation structure as important habitat features for Ae. aegypti and Ae. mediovittatus abundance and for their co-occurrence. The analysis of high resolution remotely sensed data enabled us to look at the scale at which the landscape metrics had the most influence on predicting mosquito distribution. Ae. aegypti is positively predicted by urban density but also by the number of tree patches. The scale at which these covariates predict Ae. aegypti distribution are localized to the immediate (25 m) urban environment and to an intermediate scale (50 m) for the influence of trees. Ae. mediovittatus is negatively predicted by the large urban areas at an intermediate scale (50 m) and positively predicted by the number of tree patches at a larger spatial scale of 100 m. The models suggest that Ae. aegypti may be more influenced by local environmental features, while Ae. mediovittatus may be more influenced by landscape structure at a coarser scale. The use of a model considering multiple scales simultaneously indicates the importance of landscape structure and composition at different spatial scales. However, the three scales analyzed are all relatively local. Further research should incorporate the influence of landscape structure and composition at a larger geographical scale.

The models showed significant positive associations of each mosquito species with accumulated rainfall (2–3 weeks before sampling) that must have resulted from an increased number of water-filled containers. Temperature was positively associated with high Ae. aegypti and negatively associated with presence of Ae. mediovittatus. This difference is likely associated with the main types of terrestrial habitats occupied by these species. For example, Ae. mediovittatus is more prevalent in areas with tall vegetation where air temperature should be lower than in built-up areas where Ae. aegypti is more abundant. Rainfall and temperature are important drivers of the population dynamics of Ae. aegypti (Moore et al. 1978) and, arguably more so, for Ae. mediovittatus (Gubler et al. 2001; Smith et al. 2009) as well as dengue in Puerto Rico (Johansson et al. 2009).

Overall the results of the model support previous findings about the habitat preferences of immature Ae. aegypti and Ae. mediovittatus, with Ae. aegypti preferring containers located in high density urban areas, Ae. mediovittatus preferring lower density urban areas and both Aedes species overlapping in urban areas with high canopy cover (Cox et al. 2007). We determined that the same variables are associated with co-occurrence of adults. Previous studies on the influence of landscape structure on Aedes spp. distribution and abundance, habitat preferences, and community structure are limited (Rey et al., 2006). Heterogeneities of urban/built up structure have been linked to Ae. aegypti distribution (Troyo et al. 2009). Recent research on Ae. albopictus in Hawaii specifically investigated vegetation and urban structure using landscape metrics (Vanwambeke et al. 2011). Vanwambeke et al. (2011) incorporated residential density data, recreational areas, mosquito abundance, and environmental factors to identify areas with abundant mosquitoes and where humans actively visit to pinpoint areas of high transmission risk. They found that the areas with highest abundance of Ae. albopictus were in a mixture of urban/built up and vegetated areas (Vanwambeke et al. 2011). This result is remarkably similar to our findings for Ae. mediovittatus. Ae. mediovittatus has been likened to Ae. albopictus, ‘‘the albopictus of the Caribbean’’, due to similar habitat requirements (Gubler et al. 1985). Understanding the habitat preferences of Ae. aegypti and Ae. mediovittatus should guide research of habitat preferences and community structure of Ae. aegypti and Ae. albopictus.

The biotic interactions between larvae of Ae. aegypti and Ae. mediovittatus have not yet been investigated. Ae. aegypti and Ae. albopictus overlap during immature development in containers and when natural (leaf litter) resources are limited Ae. albopictus outcompetes Ae. aegypti (Barrera 1996). Research on land use change in Thailand found Ae. aegypti was limited to artificial containers in settled areas while Ae. albopictus was in artificial and natural containers in both settled and agricultural areas (orchards) (Vanwambeke et al. 2007). Ae. aegypti adaptation to urban life and increased dengue transmission in densely populated areas are cited as the main reasons for the emergence of dengue as a public health problem in the last 50 years (WHO 2009). Rapid urbanization that outpaces governmental capacity to maintain housing standards and provide reliable access to drinking water, trash collection, and excreta disposal are main drivers for dengue fever emergence (Barrera et al. 1993; Barrera et al. 1995; Troyo et al. 2009). Ae. aegypti is well adapted to crowded tropical cities, but has also been found in other urban habitats (Cox et al. 2007). Troyo et al. (Troyo et al. 2009) developed a model including tree cover and built-up areas and found that moderately built up residential areas with moderate tree cover were likely to have high numbers of habitats suitable for Ae. aegypti larvae and that an important portion of habitats containing mosquito larvae were not near houses. A study in Brazil found that Ae. aegypti oviposition was related to the urbanization gradient, indicating preference for certain resources at the rural urban interface (Carbajo et al. 2006). And, in Australia an increase in shade in backyards was linked to increased risk of immature Ae. aegypti (Tun-Lin et al. 1995). Ae. aegypti exploits a wide range of containers as developmental habitats from discarded rain-filled containers to anthropogenic water storage containers (Barrera et al. 2006; Kendall et al. 2009). Previous observations in Puerto Rico suggest that most containers with water suitable for mosquito development are outdoors (Moore et al. 1978) and in unattended rain-filled containers (Barrera et al. 2006). Containers filled with water and organic matter or under trees were found to produce the largest adult Ae. aegypti, which has faster development and better survival rates (Tun-Lin et al. 2000). Tree canopies may provide benefits to larvae by reducing evapotranspiration and adding nutritious leaf detritus in rain filled containers (Barrera et al. 2006). Urban areas with high canopy cover may therefore be important targets for Aedes spp. vector control efforts.

The metrics developed in this study to map areas of cooccurrence will be employed to refine a trapping schedule aimed at determining the role of Ae. mediovittatus in the transmission dynamics of dengue in Puerto Rico. We speculate that in areas of co-occurrence Ae. mediovittatus will be more likely to carry dengue then in areas where it does not overlap with Ae. aegypti. However, this study is limited by the exclusion of human movement and human interaction with the environment as mechanisms of dengue transmission. Human movement is thought to be an important driver of dengue transmission and targeting key individuals and areas frequented by large numbers of people may improve disease prevention (Stoddard et al. 2009). Because Ae. mediovittatus prefers more forested areas it is possible that human movement between urban and rural areas facilitates transmission of dengue. If Ae. mediovittatus is incriminated as a competent dengue vector in the wild, future research should investigate where humans and Ae. mediovittatus come into contact.

Mapping areas of co-occurrence also has relevance to urban planning. The habitat of co-occurrence with a mix of urban and tree patches could be described as low-density development patterns that cut into forested areas. If Ae. mediovittatus is incriminated as a reservoir for dengue during inter-epidemic periods then areas of co-occurrence may be at higher risk for dengue emergence. If Ae. mediovittatus is involved in dengue transmission then it should be the object of vector control wherever it overlaps with humans.

The results also suggest that in rural areas like Patillas the intermixing of urban areas with tree patches positively predicts Ae. aegypti and reducing this type of development pattern may be important for reducing dengue transmission. The landscape of Puerto Rico is undergoing rapid transformation due to both urban expansion and forest recovery; Puerto Rico underwent conversion from an agricultural to an industrial economy 60 years ago (Martinuzzi et al. 2007). A total of 11% of the island is urban surfaces and a large portion of this development occurs in low-density patterns. Like other mountainous islands, urban areas are concentrated to the coasts and valleys, however, in Puerto Rico much of the low-density development occurs at the forest–urban interface (Martinuzzi et al. 2007). Land use change mediated by humans is linked to many vector-borne diseases (Patz et al. 2004) through the creation of new habitats, the expanding range of vectors, and changing mosquito community structure (Vanwambeke et al. 2007). Vanwambeke et al. (2007) challenge the theory that intact forest ecosystems regulate pathogens and disease, and suggest that some ecosystem services are actually ‘‘disservices’’ that create new nodes of disease transmission. In the case of Patillas, Puerto Rico the intermixing of urban with tree patches may result in such an ecosystem ‘‘disservice.’’ Puerto Rico is a ‘‘window to the future’’ for Caribbean islands because many Caribbean islands are only now making the transition from an agricultural economy (Martinuzzi et al. 2007). Informed urban planning that provides public infrastructure and minimizes sprawl through zoning initiatives such as urban growth boundaries may significantly reduce dengue transmission in Puerto Rico and provide a model for other Caribbean islands.

In conclusion, this research highlights the importance of landscape structure and composition for predicting Aedes mosquitoes and risk of dengue transmission. The use of high resolution remote sensing is key to determine the structure and composition of the landscape at a local scale. The use of high resolution or potentially medium-resolution remote sensing could be used to look at larger scale landscape patterns that influence the distribution of Ae. aegypti and Ae. mediovittatus. Further research should aim to clarify the occurrence and persistence of dengue in Ae. mediovittatus around areas of co-occurrence with Ae. aegypti. A landscape ecology approach contributes to our understanding of vector biology by adding the importance of landscape pattern to what has been previously elucidated about land cover types predictive of mosquitoes (e.g., NDVI). The importance of landscape pattern contributes to our understanding of vector control. Localized vector control efforts may fail if they do not consider the importance of urban development patterns. Future research should directly address how urban development patterns influence dengue epidemiology.