Digital Blue Foam Case Studies
Better Urban Walkability
Designing walkable pedestrian districts with Machine Learning
by Rutvik Deshpande and Cesar Cheng
Walkability as Quality of Life
Cities around the world are looking at ways to improve the quality of life of citizens, reduce their carbon footprint, and rejuvenate their urban core. By prioritizing walkability — a measure of how friendly an area is to walk — it is possible to address these issues through intelligent urban design.
There are a number of problems associated with neighborhoods where vehicular traffic is more prevalent than walkable streets. In particular, negative impact on people’s health and lives due to traffic, noise and pollution. Moreover, the amount of land area taken up by cars in the form of surface parking, On-site and On-street parking can be considerable and oftentimes un-necessary — for many cities, street parking spaces can cost tens of thousands to construct and maintain 1.
To rediscover economic, environmental, and social benefits, many cities have the potential to reclaim, repurpose, and redesign existing parking areas. Through intelligent redesign of these lots, it is possible to enhance the pedestrian experience, while reducing traffic, and ultimately carbon emissions 2.
Park-once-and-walk districts
The design of walkable neighborhoods requires the consideration of many aspects of urban mobility. This includes the design of more efficient parking solutions. Our study is based on supporting the concept of the 15 minute walkable city 3, through the intelligent planning of park-once-and-walk districts and walkable parking 5.
The idea behind this strategy is to make public parking areas available to everyone; to make local parking pools in order to discourage on-site parking as much as possible.
The idea behind this strategy is to make public parking areas available to everyone; to make local parking pools in order to discourage on-site parking as much as possible.
This strategy is ideal for transit-oriented developments and urban revitalization efforts where walkable urbanism is a priority.
Getting started
Our study focuses on a car-oriented neighborhood in Panama City with potential for redevelopment.
Within the study area, there are many large areas dedicated to car infrastructure. Here, the pedestrian witnesses all the associated problems of car-dependent mobility systems: long commutes, congestion, noise, pollution and little pedestrian walkways.
To reduce the amount of parking present in this neighborhood, we can redistribute parking capacity in a way that facilitates a park-once-and-walk experience . The goal of the study is to devise a strategy to reduce excessive on-site parking. Presently, people drive and park directly at their destination, resulting in the distribution of numerous small lots. Instead, we propose to select and consolidate parking to a few strategically located parking sites positioned to encourage people to leave their car and walk through the neighborhood to their final destination.
Using available urban data, we looked at land-use distribution in the neighborhood to identify the areas that are attracting the most visits, these are: commercial buildings, public institutions and health care centers.
We looked at what areas are being served by public transport, and finally we identified the available lots where we could relocate parking areas.
Once we identified the areas where parking could be relocated, we used our pre-trained Machine Learning model to help us understand parking capacity and distribution scenarios and ultimately decide the best location site for a parking facility within the neighborhood.
ML Workflow
In this study, we used a new workflow which allows designers to translate their expertise into ML models which can then be shared online. The workflow has three step basic steps:
- Step 1, Generate Data: using Grasshopper3D, a visual programming environment commonly used by architects, to create a training data-set for machine learning data-set of thousands of parking layouts;
- Step 2, Train and Predict: using python to create a Machine Learning model to predict parking capacities
- Step 3, Visualize and Compare, online using Digital Blue Foam, a new online design tool. This workflow accelerates the process of site analysis and design generation providing us with a quick and powerful tool to evaluate urban solutions for our cities.
Step 1, Generate Data
We use Grasshopper 3D to generate our training data set.
The advantage of Grasshopper 3D is that it uses a visual interface and it is well adopted in architecture already. This makes it easier for non-coders to get involved.
On an average the professionals would spend close to 60–80% of the time cleaning and wrangling the data. Preprocessing would take up a lot of effort and of course time, if we used some available dataset. Since we are generating data ourselves, we can control inputs, and we could eliminate the pre-processing step.
We wrote a Grasshopper3D script which generated the number of parking spots for a given set of parameters of a polygon :
- Area of the polygon
- Perimeter of the polygon
From Grasshopper 3D, we can create a CSV (Comma Separated Values) file that can be imported into python for model training in step 2.
Step 2, Train and Predict
The next step is to train our Machine Learning model.
A Machine Learning model is a file that has been trained to recognize hidden relationships between features in a dataset. At this step, input features (Predictor variables) are mapped to the variables we are predicting (Target variable). These learned patterns are then applied to testing data’s input features to predict the required outcomes which are the labels.
In this project we compared the results of few of the popular and explainable machine learning regression algorithms :
- Linear Regression
- K-Nearest Neighbor Regression
- Random Forest Regression
We further used Stacking 4, which is a method to combine the strength of different estimators.
All the estimators, we used so far; Linear Regression, K-Nearest Neighbor Regression, and Random Forest Regression are individually fit on our training dataset, while a final estimator is trained using the stacked predictions of these base estimators, which will combine the strengths of the different regressors.
Assessing The Results
The metric we used to evaluate is R² value (proportion of the variance in the dependent variable that is predictable from the independent variable), which has the best possible score of 1.0 . For initial phase we got the R² value for:
- Linear Regression : 0.98
- K-Nearest Neighbor Regression 0.992
- Random Forest Regression : 0.994
The Stacked Regression resulted in a high R² value of 0.997.
Step 3, Visualize and Compare
The goal of the study is to improve the pedestrian quality of the neighborhood. Our approach involves relocating On-Street parking and removing unnecessary On-Site parking from the area of intervention.
With this in mind, we create a proposal for a walking district. In this proposal, streets that were previously occupied by cars are replaced by pedestrian friendly greenways. Furthermore, the majority of On-Site parking in the area is replaced for public spaces.
We used the Machine Learning(ML) developed in step 2, to estimate the parking capacity of 60 possible lots available for parking relocation. Next, we selected lots located within 5 min walking distance of the area where future redevelopment will take place. From these inputs, we can use Digital Blue Foam we can quickly visualize and compare different scenarios through the use of generative design.
We simulate a variety of possible urban development scenarios. For each, we calculate parking capacity and average walking distance for new developments in the study area.
Conclusion
In this article, we outlined a set of data-driven tools and methods that can be applied to design park-and-walk-districts for better urban walkability. Using designer-generated parking layout data we were able to create a Machine Learning model to predict parking capacity. This model is implemented within Digital Blue Foam’s open map data to explore potential design scenarios to redistribute on-site and on-street parking.
Building multi-modal mobility strategies for the redevelopment of cities usually involves comparing multiple scenarios. To move away from traditional planning strategies we need to start by understanding how people move within the city, we can build strategies that focus on enhancing pedestrian mobility and use of public transport without ignoring that cars are still one element of the system that needs to be addressed. Park-and-walk-districts, is a strategy that can be included in the planning agenda for many cities across the world that need to understand how to better distribute land-use in order to create better neighborhoods for their citizens. At Digital Blue Foam, we are building tools to make it easier for urban planners, developers and city agencies perform rapid urban studies to make informed design decisions for the future of their cities.
N O T E S
- What is walkability? https://www.bluezones.com/2020/05/what-is-walkability-and-why-it-matters-for-health-resiliency-happiness-and-sustainability/
- The price of parking https://medium.com/r?url=https%3A%2F%2Fcityobservatory.org%2Fthe-price-of-parking%2F
- https://www.bloomberg.com/news/features/2020-11-12/paris-s-15-minute-city-could-be-coming-to-an-urban-area-near-you
- https://www.reinventingparking.org/2018/01/walkable-parking-how-to-create-park.html
- https://towardsdatascience.com/just-keep-stacking-implement-stacking-regression-in-python-using-mlxtend-3250ff327ee5
About the Authors
Rutvik Deshpande is an Associate Data Engineer at Digital Blue Foam. He aims to use AI in Architecture, with his work focusing on data-driven design workflows in both urban and architectural scales. He also has interests in the use of data science, machine learning, and computational design in the built environment. He is an active Kaggler too.
Cesar Cheng is product developer at Digital Blue Foam. He is an architect and urban designer specialized in computational design, urban data analysis and material research. He is a graduate from the Em-Tech program at the Architectural Association. His work focuses on the digital transformation of the AEC industries with particular interest in computational design, artificial intelligence, spatial data analytics and material research for applications in digital solutions for the built environment.
About Digital Blue Foam
At Digital Blue Foam we believe buildings and cities must reduce their dependence on fossil fuels in all aspects of the process design. This includes materials, construction processes, and operations. Presently, we are researching new ways to reduce energy use , and ultimately carbon emissions, and boost occupant comfort by leveraging tried and tested passive design principles in our design software.
