Project 1: Socio-Spatial Dynamics of Water Sustainability. The overarching goal of the project is to determine how interactions among social, hydrological, and biological spatiotemporal dynamics shape the sustainability of human-freshwater systems under incentive-based conservation. In pursuit of that overarching goal, we 1) characterize conservation organizations’ willingness to offer financial/technological incentives or engage in coalition building among water users, 2) characterize water users’ willingness to participate in voluntary incentive-based and coalition-based water-conservation programs, 3) use dynamic game theory models to explore conflict and cooperation among local coalitions of water users, 4) model ecosystem services as a function of human water usage, conservation incentives and surface-groundwater interactions, and 5) integrate the results of tasks 1-4 to understand the dynamics of integrated human-freshwater systems under incentive-based conservation, and derive strategies and guidelines for navigating conservation dilemmas that occur in water-limited human-freshwater systems worldwide. The proposed project is piloted on the Red River of the south-central United States. 

​

Project 2: Infection Transmission and Prevention in Metropolises with Heterogeneous and Dynamic Populations. Due to dense social contacts, metropolises are considered as epicenters of contagious infectious diseases. Developing models for infection transmission and mitigation in metropolises is a challenging problem because of spatiotemporal variations in their population structures. Employing the fact that intra-urban mobility is mainly shaped by the transportation infrastructure and reflected in the traffic flux information, we develop a novel multi-scale reaction-diffusion process to model infection transmission in a metropolis. To mitigate the infection, the impact of three non-pharmaceutical preventive interventions (movement restrictions, social distancing, and proactive testing/screening) is analytically investigated on the reaction-diffusion process. Finally, three mathematical models and their solution approaches are developed to optimize the implementation of the interventions with the least economic burden. The performance of the optimization models is compared with two simpler implementation schemes (uniform and pro rata implementations of interventions) using the information of Sioux Falls metropolitan area in the U.S. 

​

Project 3: Mass Casualty Management in Disaster Scene. The spatial dispersion of casualties and temporal variations of emergency resources are the two critical factors that complicate the casualty management operation in large-scale mass casualty incidents. These factors are investigated in this research to develop an optimal casualty treatment strategy for a medical station that provides first aid assistance for severely injured casualties on the site. This study shows that the optimal treatment strategy to save the highest number of casualties at a medical station with nonstationary arrival and treatment rates is dynamic with time-dependent resource allocation and casualty prioritization strategies. The dynamic treatment strategy is compared with static treatment strategies used in practice. Findings show that the expected performance of the dynamic strategy is always better than the static approaches regardless of the disaster severity level. 

 

Project 4: Agent-based Learning to Utilize Local Data for Activity Recognition. The Department of Homeland Security (DHS) obtains and retrieves a high amount of video data from various surveillance systems (e.g., CCTV cameras, vehicle-mounted cameras), and videos captured by citizens via home CCTV systems or vehicle cameras. A video of a single incident can be captured by multiple cameras of a surveillance system. Each camera may have a partial scene of an event that other cameras do not observe. In such a case, reconstruction of the overall incident using local computational resources, and discovering valuable insight from a vast stream of video data is a challenge in terms of computation and memory usage. Besides, due to privacy, many citizens do not prefer to share their personal or premises data with a third party or an external entity. In this project, we propose to conduct the computation at the edge, and to locally identify a potentially risky situation and notify the authorities. This significantly reduces computational cost and time. This project develops agent-based modeling by applying FL technique on distributed agents residing within the citizens’ personal and premises. In our proposed approach, an agent of a home/vehicle surveillance system locally stores all captured videos from cameras, carry-out on-device model training, and performs prediction by analyzing the captured data. However, we cannot achieve satisfactory prediction accuracy if the agent’s model only trained with its local data. Therefore, we propose a distributed machine learning-based FL strategy to ascertain any violation within a personal or premises. The major advantage of FL is that an agent can recognize an unseen and risky situation by learning from another home surveillance system’s agent observation. Moreover, if an agent has partial information of an event and another agent has the rest of the portion, they can reconstruct the whole event through the FL procedure. It is to be noted that none of the agents needs to share their raw data with other agents or even with the server. Alternatively, every agent can share their model parameters with the server (i.e., each individual citizen records and analyzes the data locally and shares the results with a central entity instead of sharing the raw video data). Then, the server broadcasts the aggregated model parameters to all the available agents.  

​

Project 5: Real-Time Road Network Restoration Using Unmanned Aerial Vehicles (UAVs). The goal of the research is to study two fundamental barriers of efficiency in road network restorations, namely, the lack of complete damage/debris information and the lack of coordination among the restoration operations. We developed an integrative decision-support framework with a model-based data diffusion component for online coordination of three restoration-interdependent operations in the disaster response phase: “damage assessment”, “road recovery” and “relief distribution”. The model developed for the damage assessment operation controls the damage/debris data diffusion speed in the integrative framework (via UAVs). This data is instantly shared with an online model developed to prioritize the recovery process for blocked roads. Road prioritization is done in a way to make the highest acceleration in the relief distribution operation. The integrative framework is tested on the road network of the Miami-Dade and Broward counties. 

​

Project 6: Behaviorally Enriched Learning Mechanism for Road Network Restoration After Disasters. In this project, we study emergency post-disaster restoration of road networks. Timely restoration of disrupted (e.g., damaged or blocked) roads plays a critical role in the response operations after disasters and helps communities turn back to their normal operations soon. Scarcity of restoration resources, uncertainty of recovery times, and gradual adaptations of travelers to new travel paths are the major factors that highly complicate road network restoration operations. We address these challenges by developing a Behaviorally-enriched Reinforcement Learning Mechanism (BRLM). Considering gradual adaptation of travelers, BRLM optimizes scheduling and resource allocation decisions in the restoration process in a way to make the highest acceleration in post-disaster traffic movements. The performance of BRLM is tested on the road network of Sioux Falls in South Dakota for several tornado scenarios that are generated based on the historical report of National Oceanic and Atmospheric Administration.

​

Project 7: A Coupled RLs Paradigm for Simultaneous Restoration of Interdependent Critical Infrastructures. The proper functioning of any society heavily depends on its interdependent critical infrastructures (CIs), where the functionality of components in one CI relies on the services provided by others. The restoration of CIs following a disaster poses several challenges, including the heterogeneity of each CI's operating context, operational interdependencies, decentralized decision-making, and the stochastic nature of post-disaster restoration. Addressing these challenges requires innovative modeling and solution approaches that bridge the gap between integrative and distinct decision-making, facilitating synergistic planning for interdependent CIs. In this project, we present a novel mechanism for generating cooperative restoration policies in a decentralized context. This approach is designed to handle the complexities of post-disaster operations, such as diverse and stochastic recovery times, varying resource availability, and the need for multiple recovery teams to expedite critical component restoration. To evaluate its performance, we focused on restoring road and power CIs in Sioux Falls, South Dakota, disrupted by a tornado. The results demonstrate the mechanism's effectiveness, showcasing a 20 percent improvement in the restoration of CI services during post-disaster response periods across various tornado scenarios.   

​