The transmission cycle of dengue is notably complex, involving two different replication cycles of DENV alternating in human and Aedes species mosquito hosts (Figure 1). These cycles are often controlled by fluctuations dominated by the random nature of population events (e.g., mosquito dynamics and human mobility) and the space-time variability of suitable environmental conditions. The development of DENV inside adult mosquitoes, which are ectothermic, is air temperature-dependent and expected to be affected by global warming. Current evidence suggests that future climate change, particularly with more frequent extreme weather events, may amplify dengue transmission [19, 20]. This is through increasing mosquito abundance, higher rates of both mosquito biting and survival, and a shorter DENV extrinsic incubation period—the duration between when a mosquito takes a viremic blood meal and when it becomes infectious.

Based on our previous work [21], it is possible to develop a stochastic lattice-based integrated dengue model to predict the dynamics of DENV transmission under climate change for the urban environment. The proposed model consists of two time-continuous space-discrete sub-models that consider the non-linear relationship between temperature and DENV development (including of all four serotypes) inside Aedes vectors to estimate the DENV extrinsic incubation period under global warming. It incorporates explicit coupling between entomological and viral processes, epidemiological statuses, and hydroclimatic conditions to capture the dynamics of dengue. The stochastic entomological model resolves the virus life cycle in aquatic (egg, larva, pupa) and adult stages of Aedes mosquitoes. The coupled epidemiology-virology model simulates stochastically the circulation of DENV in human hosts and adult Aedes vectors using the well-known susceptible-exposed-infected-recovered (SEIR) approach. These two sub-models can be iteratively coupled in discrete space domains through the equality constraint of adult vector populations. In addition, this model also incorporates a meta-population approach to describe spatial movements among discrete geographic domains in urban environments.

The proposed model moves beyond the current state-of-the-art modeling of VBDs. These explicitly incorporate the aquatic stage of the mosquito life cycle and the spatial dynamics of mosquitoes, the distribution of human population, atmospheric conditions, and hydrological conditions into an SEIR epidemiological model [21]. While such models have improved predictive ability for infection rates, they were not developed to gain insights into interventional approaches by coupling the life cycles of vector mosquitoes and arbovirus across the horizontal and vertical heterogeneity of the modern, planned urban environment. In order to design sustainable and healthy metropolitan communities, consideration should be given to the role of parks and detention ponds/lakes that support green spaces, the density of population that enables rapid spread of VBDs, and changing micro-habitats of the mosquito life cycle under climate change. Accordingly, the model may better inform local government policies for the good maintenance of public spaces that should be implemented in order to reduce the risk of mosquito breeding in sites that have been specifically designed for human recreational and leisure activities.

Ae. aegypti and Ae. albopictus, the main vectors of DENV transmission in urban environments, are anthropophilic mosquito species [22]. The preferred habitat for adult Ae. aegypti, in particular, can be described as a household setting. Typically, gravid adult females lay desiccation-resistant eggs in water-holding artificial containers (e.g., plant pots, buckets, cans, and lids) in and around the home, many of which are discarded and/or small, making source reduction of juvenile habitats in urban environments especially challenging. Mosquito larvae graze on aquatic microbial growth, which may vary considerably in quality and is driven largely by autochthonous inputs of terrestrial leaf litter detritus [22]. In areas where electricity and water supplies are unreliable, households may store water, increasing the likelihood of colonization by Aedes mosquitoes [23]. Due to variations in the reliability of these municipal services, the presence of artificial container habitats, and human population density, in the example of Vietnam the risk of exposure to DENV varies with human land-use gradients [24], socioeconomic gradients [25], and the built environment [26]. However, scant attention has been given to either the vertical built environment or the strategic implementation of stormwater infrastructure to minimize mosquito habitats in tropical urban environments.

The model developed in task 1 may be validated and refined using epidemiological data for disease incidence and field data for mosquito abundance and dengue prevalence. For instance, in Vietnam, the two largest cities Hanoi and Ho Chi Minh City could be selected, together with a third location, Cam Lam, which is under development into a major urban center. Hanoi to the north and especially Ho Chi Minh City to the south are often hotspots of dengue outbreaks [11]. These are conurbations of several million people, with high population densities and increasingly developed residential and business infrastructure, which also serve as major transportation hubs in Vietnam. The south-central coastal city of Cam Lam, on the other hand, is being developed into a smart city of more modest size. These locations have distinct hydroclimatic conditions that may influence mosquito habitats.

By searching for larval habitats in and around households and public green spaces, including water-holding containers and stormwater habitats, mosquito juvenile stages may be visually and/or dip-sampled and their presence/abundance quantified. For surveillance of adult mosquitoes, a combination of traditional ovitrap [27], BG-Sentinel trap [28], and gravid Aedes trap [29] can be deployed, as different trap types may vary in efficacy across environmental contexts. Temperature, humidity, and wind speed require measuring at sampling sites. As mosquito habitats depend on vegetation and the chemical composition of the water [30, 31], the types and coverage of plants in each area require documenting. Water samples need to be characterized chemically for ammonia, phosphate, nitrate, and pH using field sensor kits. Sampling frequency should be at least weekly and increased depending on weather conditions. In order to study the movement of mosquitoes horizontally and vertically in the urban built environment, an iridescent dye is applied to sterilized adult male Ae. aegypti, which are then released as part of a mark-recapture study [32]. Sterile male mosquitoes pose no health risk to humans since they are unable to contribute to mosquito reproduction and do not feed on blood. In initial studies, the number of sterilized mosquitoes released into the environment will be so few as to minimize the potential ecological and evolutionary impacts. Recapturing at different distances from release points enables modeling of the probability of dispersal distances both horizontally and vertically in the built environment. Past datasets, including mosquito abundance and dengue disease incidence (weekly and monthly) at sub-province levels, may also be obtained from a nation’s Ministry of Health to support task 1.

Existing survey-based studies that use aggregated counts over large spatial units are often inaccurate, either over- or under-estimating the effect of a containment measure. This is because such methods assume the instances of activities to be distributed uniformly across entire cities, provinces, or states [33]. Given the localized effectiveness of a single instance of one containment activity, there is a need for analyses at similarly small (sub-city) spatial resolutions. This requires collecting empirical evidence regarding the efficacy of highly localized containment and intervention activities that target vector mosquito species. Findings may facilitate the optimization of near real-time prediction-based control and prevention strategies in resource-limited, local-level urban settings. If the conceptual framework is successful, consideration should be given to challenges in scaling up effective interventions from a small trial into broader policy and practice. Key factors might include maintaining accuracy and required sources for validating the outcomes, as well as linking data from implementation monitoring to decision making throughout the process [34].

The following interventions can be considered: managing the spatiotemporal distribution of rainwater puddles in present and future climate; distracting the ability of mosquitoes to find hosts by disrupting CO₂ gradients that they use to seek them out; and considering stochasticity [e.g., strategies for reducing entrainment of dengue in a domain (neighborhood)]. The non-random deployment of each intervention activity in relation to dengue cases, as well as the deployment of organized containment activities in different epidemiological contexts, such as indoor residual spraying and fogging in the vicinity of a residential building, are accounted for by different modeling frameworks. These also incorporate individual behaviors regarding such activities as keeping plant pots and other containers in which residual water may collect on apartment balconies at different elevations above ground level to which mosquitoes will have variable access. Predictive results from these spatiotemporal analyses will reinforce existing knowledge from observational data on the efficacy of containment activities aimed at adult mosquitoes. By applying the model under varying climate conditions, this informs better building designs and ensures ‘future-proof’ urban planning that discourages and distracts mosquitoes from breeding in a peridomestic environment. Mosquito data should be collected as described in task 2.

A digital twin may be built by leveraging our newly developed computational fluid dynamics (CFD) model to test alternative urban designs. Since a mosquito cannot be treated as a passive scalar field, there is a two-way coupling between its flying behavior and the surrounding air flow. First, a local scale structure model [35, 36] should be constructed to characterize the aerodynamic behaviors of mosquito migration. The flying characteristics are extracted and fed into a large-scale meteorology-coupled CFD model. Second, the high-fidelity meteorology-coupled CFD simulations characterize the local microclimatic dynamics under different urban configurations. The digital twin is a large eddy simulation (LES)-based model to resolve the atmospheric boundary layer dynamics and meteorology of built environments (Figure 2). These CFD simulations provide detailed spatiotemporal microclimatic fields and reveal the effects of urban configuration on their evolution under both present-day and future climate conditions. Multiple representative urban types in different geographical regions of Vietnam may be used as test sites. The high-fidelity model is built on a finite element discretization that can resolve complex urban geometry and incorporate realistic boundary conditions [37, 38]. These LES simulations are conducted over a domain size of 1 km × 1 km (i.e., neighborhood scale) at a mesh resolution of 1 m, coupled with real-world climate-modeled meteorology. We have recently developed a physics-informed transfer learning framework leveraging satellite imagery, satellite-measured land surface temperature, forcing meteorology data, physics-informed neural networks, and transfer learning techniques to produce urban air temperatures [7, 39]. This set of high-resolution urban air temperature data allows validation of the LES simulations.

The high-fidelity modeled data is used to predict spatial (both horizontal and vertical) and temporal heterogeneity of mosquito habitats in the urban environment that affect human infection rate. In order to ensure the usability of the digital twin, which needs real-time sensing data and to be computationally efficient, there is a need to develop fast surrogate models using physics-informed machine learning [40, 41] by leveraging labeled data generated by the high-fidelity model.

Sensors for temperature, humidity, and wind speed should be installed in selected residential locations, which are studied in tasks 2 and 3. The digital twin provides simulation of dengue disease risk and intervention strategies in near real-time based on the models developed in task 1, and empirical data collected in tasks 2 and 3. For locations that are planned for future development (e.g., Cam Lam), the digital twin will allow near real-time simulation of alternative neighborhood designs to reduce dengue disease risk.