IBI Group Takes a Parametric Approach to Pandemic Resilience

Data-driven parametric modelling can provide information to decision-makers during uncertain times. The ability to quickly model a nearly unlimited number of scenarios, while using the processing power of the computer to optimize design and planning solutions, is key as we navigate the current public health crisis.

The COVID-19 pandemic has upended the operation of cities across the world. In most instances of disaster planning, historical events inform the response to future crises. While the world has endured viral outbreaks in the past, the global scale and the upheaval of every aspect of urban life due to COVID-19 is without precedent. Plans made prior to the pandemic by cities, businesses, educational facilities, and transit agencies must be completely reimagined to include pandemic resilience — and this needs to happen quickly.

Early on in the crisis, IBI Group’s computational team shifted its focus to pandemic resilience. Comprised of planners, architects and programmers, the team is equipped to develop custom workflows and parametric design tools for the design and operation of the built environment. Under regular circumstances, the team builds parametric models to assist in designing buildings, planning transit systems, and developing land use plans, among other aspects of urban planning and architecture. The parametric modelling process enables the efficient use of large datasets and the correlation of variables and can incorporate flexibility into the models.

The following is an overview of how parametric models can be built to support cities as they respond to COVID-19. By developing our own tools, and using publicly available health data, we have been able to investigate spatial and operational solutions, leaving the epidemiological solutions to the doctors and scientists.

In early April, we built a parametric model by looking holistically at the healthcare system in Los Angeles County and considered its preparedness to support an overwhelming number of new patients. Our model used the available geospatial data for every hospital in Los Angeles, including information about the number of available beds, staff and equipment. It also considered the locations of nearby high schools with gymnasiums in which temporary care facilities could be established. These data were associated with the daily infection counts across the county and organized by neighbourhood.

Leveraging the Insitute for Health Metric and Evaluation’s (IMHE) COVID-19 projection curve, our model anticipated potential infection counts by neighbourhood, associating a percentage of the infections with hospitals in a given radius. We could not assume every bed was available, as hospitals still had non-COVID-19 patients to treat. Thus the percentage of available beds was included as a variable to test. If any hospitals were to fill to capacity, the model would identify opportunities to send additional patients to nearby healthcare facilities. If all hospitals were filled, the model would find the most efficient high school gymnasiums to convert into temporary care facilities and alert us when these spaces should be converted. 

Finally, our model enabled us to test unlimited scenarios, replacing the IMHE curve with the best- and worst-case infection scenarios. It became immediately evident that if we considered the healthcare facilities holistically, even in the worst-case scenarios, the system had the capacity to provide care for all those that may need it, a fact that has proven to be true. This highlights the need for holistic thinking and wide-spread access to centralized databases and models in order to circumvent silos and make evidence-based decisions. Click here to watch a demonstration. 

The next phase of our modelling efforts focused on a strategic approach to returning children to school and parents to work. The push to return essential workers to their jobs has been complicated by critical safety concerns and the fact that many essential workers have school-aged children. With some schools still closed for in-class instruction, and others operating via hybrid models, some parents were unable to return to their jobs.

Using publicly available databases, our model created profiles of census blocks, including where essential workers live, the average family size and age of children. These data, combined with a district-wide analysis of school occupancy levels, enabled us to understand how to safely distribute children among available classrooms.

Our model focused on students aged 15 years and younger, with the assumption that older high school-aged students could study autonomously while their parents were at work. This strategy also made high school classrooms available for other uses, freeing up additional capacity across the school system. Ultimately, our model proved that at only 28 per cent of pre-COVID-19 occupancy, the children of essential workers could safely return to school and every essential worker in Los Angeles could return to work. Click here to watch a demonstration.

Observing physical distancing protocols in spaces as unique as offices, schools, hospitals, stadiums, and performing arts centres presented a challenge. Our evolutionary algorithms were designed to ensure physical distancing protocols could be abided within various highly trafficked spaces despite their idiosyncrasies. 

We built a distancing algorithm to inform our own return to office strategy, which was able to generate layouts and seating plans across our 60+ global offices in a matter of days. With physical distancing requirements as variables, enabling efficient re-optimization as they evolve, our algorithm can also consider personal space and major and minor circulation in seconds. When tested against traditional “manual” seating layouts, our evolutionary solver found an average of 4-5 per cent additional capacity.

Where the office model associated employees with a fixed-desk layout, the school model presented an opportunity to redistribute the desks. Given the easily repeatable module of a student’s desk, our evolutionary solver optimized the spatial layout within physical distancing parameters. Additionally, the model includes options for collaboration mode (clustered desks facing one another), or teaching mode (desks facing a teaching wall), along with a ‘teacher bubble’, giving the instructor room to move around safely without entering a student’s space. With a teaching wall defined, the algorithm can maximize desk layouts for entire schools in under one minute.

Districts are now treating the movement of students throughout the school day in different ways, and our model accounts for three options: static student pods with dynamic teachers, where students remain in the same space and the teachers move around throughout the day; dynamic student pods with static teachers, where students move in groups between classrooms while the teachers stay in the same space; and status quo, where teachers occupy assigned rooms with individual students who have unique schedules. Using the school’s daily schedules, the model simulates movement through the school, suggesting route optimizations as students and/or teachers move from classroom to classroom.

We are advising a large hospital in Toronto on its COVID-19 operations, and have organized our modelling efforts around answering three key questions:

1. How many patients can each clinic accommodate?
2. What is the safest and most efficient route to get patients to their examination or treatment rooms?
3. How many people can safely occupy waiting rooms?

Taken independently, each of these questions is relatively easy to model. However, the three questions need to inform each other recursively to find coalescence (e.g. clinic capacity needs to be reduced if hallways are overcrowded).

First, we used pre-COVID-19 patient visits to generate an average visit time for each doctor and clinic with a variable for enhanced cleaning time in between patients. Using these volumes, we built an evolutionary solver to associate each clinic with its own entrance and optimized route to minimize potential conflicts as patients flow throughout the facility. Finally, a distancing optimization was performed on all waiting areas to generate safe layouts to serve the required volumes of each clinic.

For venue models, the cluster size is greater than one. An analysis of historical ticket sales provides ratios for average group sizes by section within a specific venue. The various group sizes are pre-clustered with the potential to be efficiently recalibrated as new data become available. Additionally, each seat is associated with a parking area, restroom and a food and drink counter. Our model provides optimized routes from parking lot to seat and the ability to calculate the potential occupancy load of the restrooms and food service counters. It also has the functionality to export the seat clusters and associated service facility locations to the ticket sales platforms, allowing the customer to search for clusters serving their group size with advanced knowledge of parking, entry and service locations for their visit.

Overseen by Parametric Design Lead, Jason King, and Computational Design Lead, Evangelos Pantazis, IBI’s parametric approach to pandemic resilience has been embraced by communities across North America, and was recently profiled in Smart Cities Dive.