Case Study: Southeast Michigan Council of Governments (SEMCOG)

What was the organization using before UrbanSim?

Southeast Michigan Council of Governments (SEMCOG) has been one of UrbanSim’s longest customers, utilizing the model since approximately 2001. Before transitioning to the UrbanSim model for SEMCOG’s 2035 forecast, they used DRAM/EMPAL (later updated and rebranded as Metropolis), as well as a few custom tools. These gravity-based models used forecast districts, regions, and larger zones, rather than smaller units of geography to simulate scenarios. Rather than using Traffic Analysis Zones (TAZs), SEMCOG became interested in forecasting at smaller geographies, summarizing results at larger geographies as needed. They initially moved towards this approach by generating nested geographies in DRAM/EMPAL, but they soon reached the technical limits of the models. The model system was stretched to 2,800 zones, but SEMCOG modelers found the results to be unsatisfying. This approach required developing a post-processing zonal allocation program (ZAP) to process DRAM/IMPAL’s output. They were “pushing the technology of the time” to its limits in this approach.

How did your work change after you started using UrbanSim?

SEMCOG had three motivations for transitioning to the UrbanSim model. First off, SEMCOG wanted to work at smaller grid- or parcel-based geographies and to have more flexibility in aggregating the results. They were also interested in having a discrete-choice, behavioral model that could analyze consumer choice more accurately. Thirdly, SEMCOG had data which could not fully be utilized without adopting the UrbanSim model, but once the data was cleaned it could easily be fed into the UrbanSim model.

The first land use forecast generated by the UrbanSim model was the 2035 forecast for Southeast Michigan. The small area forecast was created using five-acre grid cells, accomplishing the agency’s goals of forecasting at a small geographic scale. For subsequent forecasts after the 2035 forecast, SEMCOG switched to UrbanSim’s parcel-based model. In SEMCOG’s 2045 forecast, they used 1.8 million land parcels to simulate location patterns of future population and jobs and to model residential and nonresidential development as demands shift over time. The key benefit of using the parcel-based system was that all previous forecasts had used arbitrarily defined geographies. Utilizing UrbanSim’s parcel-based model allowed the team to more accurately simulate reality, engage in meaningful discussion with community members, skip time consuming post-processing steps, and reduce opportunities for error.

After the transition to UrbanSim’s parcel-based model, forecasts were more readily adopted by SEMCOG’s membership. As they’ve introduced more variables and segmentation, the team has continued to receive positive feedback, in turn giving more credibility to the forecast and streamlining the adoption process. “The last forecast was unanimously adopted... The parcel-based models, in particular, have been much easier to process and control than the grid-based ones.” There are a few driving dynamics which have made adoption challenging for SEMCOG’s seven counties. For example the region’s outermost municipalities are growth resistant, while the team is modeling decline in the region’s urban core and growth in the suburbs in between. They reflected this variability by creating models that were differentiated by county.

“Moving to a parcel-based system has expanded the level of contact we have with communities and their willingness or interest to participate in the process, especially in terms of providing information... Since we’ve moved to the parcel-model and really placed an emphasis on getting communities to tell us what is going to happen at a micro-level,... over 120 communities have agreed to meet with us and talk about what is going to happen in
the future.” The team has used parcel maps during these conversations, enabling participants to show precisely where they expect growth patterns to occur. Getting input from approximately half of all of the communities SEMCOG represents is an astonishing improvement from the 20 or so communities which were involved in the modeling processes back in the 1990s. Simply put, presenting information in parcels allows the modelers and community members to speak in less abstract, more tangible geographies they both understand.

What’s next?

With each iteration of the forecast, SEMCOG has improved its process and outcomes. One key takeaway is the recognition that models are imperfect due to their exceedingly challenging goal of simulating human behavior. UrbanSim has allowed the SEMCOG team to model scenarios based upon many more inputs than what was previously possible, therefore adding sophistication and contextual nuance to the model. As SEMCOG’s modelers develop more expertise, as tools evolve, and as they become more familiar with the data, they’ve gotten more confident with the model, employing it to solve critical problems SEMCOG is facing. The model will continue to evolve, capture complexities, and provide more nuanced insights with each future iteration.