The COVID-19 pandemic's impact is significantly shaped by the timing and stringency of non-medical countermeasures. The absence of widespread vaccination and effective treatments makes these interventions crucial. Droplet transmission necessitates social distancing and lockdowns to mitigate infection spread. The exponential growth of COVID-19 necessitates precise timing and implementation of these measures, as their effectiveness varies greatly depending on their nature and the specific context. The United States serves as an example where delayed responses resulted in a massive surge of cases. Developing and transitional countries face additional challenges due to limited economic resilience, healthcare capacity, and testing capabilities, often exacerbated by factors like a young population, prevalent infectious diseases (e.g., tuberculosis, HIV), and unequal distribution of wealth and resources. South Africa, with its significant case numbers and recent drop in new cases, provides a critical case study. While the government eased lockdown measures due to economic and social concerns, the reliability of available data, particularly regarding unreported cases, remains questionable. This study uses agent-based modeling to simulate infection dynamics under varying lockdown scenarios to assess appropriate countermeasures and the actual epidemiological situation in South Africa. The Nelson Mandela Bay Municipality (NMBM) serves as a representative urban area in the Global South due to its socioeconomic disparities and high levels of segregation. The study integrates transportation science, behavioral economics, and epidemiological models. A novel aspect is the use of the MATSim framework to simulate a synthetic population's movements and activities, complemented by the Episim framework to track infection events and disease progression based on parameters derived from epidemiological findings. This approach helps overcome the challenge of limited micro-level data on social interactions.

The study draws upon existing literature on the impact of non-medical interventions on COVID-19 transmission dynamics. References are made to studies examining the influence of various countermeasures (lockdowns, social distancing) on infection rates, highlighting the importance of timing and stringency. It also addresses the economic trade-offs associated with effective countermeasures, particularly in developing countries with limited resilience. The specific challenges faced by South Africa, including factors like a young population, existing infectious diseases, and socioeconomic disparities, are contextualized with relevant research. Agent-based modeling (ABM) is presented as a suitable approach to capture spatial heterogeneity and stochasticity in human behavior, contrasting it with traditional mathematical epidemiological models.

The study employs a two-step approach using agent-based simulations. First, MATSim (version 12.0-2019w48-SNAPSHOT) conducts transportation simulations, creating realistic activity profiles for a synthetic population within a virtual network representing NMBM’s streets and locations (data derived from OpenStreetMap and the 2011 Census and 2004 travel survey data). This synthetic population consists of 114,346 agents (a 10% sample) across 32,597 households. Key characteristics like age distribution, household size, income, and transport modes (cars, minibus taxis, walking) are defined. Minibus taxis are modelled using a Demand Responsive Transport (DRT) framework to reflect their crucial role in NMBM's public transportation system. Second, Episim, a framework that builds upon MATSim output, conducts epidemic simulations. Episim identifies potential infection events based on co-location of agents in specific containers (work, home, leisure, etc.). The disease progression follows an extended SEIQR model, incorporating exposed, infectious, quarantined, seriously sick, and critical states. Parameters such as incubation period, asymptomatic infection probability, and recovery time are defined based on existing epidemiological data. Infection events are probabilistically determined, considering infectivity parameters for each activity type (Table 3). The model is calibrated by adjusting a parameter (θ) to match simulated infections to reported case numbers for NMBM, intentionally underestimating real infections to account for unreported cases. A conservative approach is taken by setting θ to 1.5 × 10⁻⁷. The simulations are initialized with ten randomly selected infected agents in early May 2020. Multiple simulations (100) are run to account for stochasticity. Six policy scenarios, reflecting different lockdown levels (Level 1 to Level 5 based on the South African government’s risk-adjusted strategy), are simulated. These scenarios incorporate activity parameters (Table 3) reflecting the restrictions imposed at each level. Simulations are run with lockdown periods of 30, 60, 90 days, and continuously until the end of the year, to assess the impact of various lockdown durations.

Simulations show that the shape of infection curves is similar across scenarios until August 26th, after which divergence occurs based on lockdown stringency. Lax and short lockdowns result in a tenfold increase in active cases within 30 days. Maintaining stricter Level 3 measures leads to a decline. Extending strict lockdowns to 60, 90 days, or indefinitely significantly reduces active case numbers. Extended strict measures (Level 4 and 5) can even eradicate the virus by the end of 2020. However, ending lockdowns, even with low case numbers, invariably leads to a return to exponential growth. Most scenarios indicate an autumn 2020 peak, where infected and recovered individuals comprise about half to two-thirds of the population, aligning with herd immunity estimates. The location of infections varies by lockdown level: home infections dominate across all scenarios (~86% to 96%), with the proportion increasing with stricter measures due to reduced out-of-home activity. The percentage of infections in minibus taxis and educational settings also decreases with tighter restrictions. The simulations suggest that a combination of intermediate measures could consolidate active cases, although the actual level required for containment might be more stringent than the simulations suggest, given the intentional underestimation of case numbers.

The findings highlight the crucial role of strict and sustained lockdowns in controlling the COVID-19 epidemic. Lenient measures prove insufficient to contain the virus's spread, leading to overwhelming healthcare resources. The results emphasize the need for a cautious approach to lifting restrictions, as any easing can trigger renewed exponential growth. The study demonstrates the utility of agent-based transportation simulation in modeling complex scenarios with limited micro-level data. This approach captures realistic human behavior and allows for integrating various data sources and theoretical assumptions. The conservative approach, focusing on underestimating case numbers, might lead to a delayed peak in simulations compared to the actual events. The discrepancy between the simulated autumn peak and the observed July/August peak in South Africa could be due to the significant number of unreported cases, potentially up to 98%. Other factors not included in the model, such as environmental effects, behavioral changes, or viral mutations, could also influence the observed dynamics.

The study underscores the epidemiological effectiveness of prolonged, stringent lockdowns in mitigating COVID-19 spread. Although the simulations suggest some intermediate measures might suffice to consolidate case numbers, the intentional underestimation of infections warrants caution. While not definitively confirming the peak in infections as declared by President Ramaphosa, the results, interpreted with awareness of potential limitations, indirectly lend support to this assumption. Agent-based transportation simulations, as demonstrated here, provide valuable tools for understanding complex epidemiological scenarios and informing policy decisions.

The study's findings are subject to several limitations. The calibration relies on assumptions regarding the proportion of unreported cases and adherence to government regulations, which may lead to underestimation of infection dynamics. The initial conditions of the simulation (10 randomly infected agents) introduce a short-term distortion in the early stages of the simulated epidemics. The use of a synthetic population and simplified representation of human behavior could also affect the accuracy of the results. Furthermore, factors not explicitly modeled, such as environmental influences, changing behaviors, or viral mutations, may influence the real-world dynamics.