Parking Requirement Comparison based on Zoning Codes
Project Contexts & Goals
Parking has always been a key topic in U.S. planning and construction. The main stream of vehicle parking requirements is to set a strict minimum for vehicle parking spaces to ensure enough parking for potential drivers, both in urban and suburban areas, aligned with car-oriented development. The minimum standards are dramatically high compared with the real demand in many places, especially suburban areas, therefore resulted in low occupation and significant waste of land. In recent years, people more and more recognize the negative impacts of excessive parking requirements, including land use inefficiency, economic inefficiency, environmental problems related to impervious surfaces, encouragement for car dependency and discouragement for active transportation. For sustainable development, some cities in the U.S. have abolished minimum parking requirements. Other cities have decreased the minimum number and set maximum restrictions. Parking reform is spreading across the United States, with trends of loosening the standards or making the use-based and scale-based requirements more specific to guide efficient and sustainable parking. In this context, the differences between cities are growing and U.S. parking requirements system is becoming more and more complex. Although there are many slogans about parking reform, only by digging into the actual zoning documents of cities can we find the truth of U.S. parking reform and make detailed comparison and analysis.
Zoning documents are hard to read, with highly fragmented and detailed information, let alone compare and summarize zoning codes from different cities with different structures. In this project, I’m going to use AI to extract off-street vehicle parking spaces number requirements from zoning documents of 7 cities with largest populations in Michigan, then ask AI to compare and summarize the gathered information. The outcome can provide planners and public with a comprehensive and reliable summary of current parking requirements in Michigan. Another significance of this project is to try out a process of multiple zoning documents analysis in a unified schema.
Methods
Download Zoning Code Documents
I select 7 cities in Michigan which have populations above 100,000 in Michigan, according to the 2023 Population Estimates Program and 2023 American Community Survey, and download their zoning code PDF documents from Municode Library. https://library.municode.com/
The two things I care about in zoning code in this project:
off-street vehicle parking spaces number requirement for retail use
maximum standard for parking spaces number which can not be exceeded
Use LLM In-Context Learning to extract designated parking requirement information from different codes through unified prompts
I interact with OpenAI GPT-4o model through API to coduct the extraction tasks.
Extract Texts, Split Texts into Chunks, and Create Text Embeddings
Key Parameter:
chunk_size=3000 (for more complete and consistent context in each chunk)
Load Document, Search Relevant Chunk Based on the User's Query, Generate Response
System Message: You are an assistant extracting accurate and complete information from zoning code documents as the user requires.
Key Parameters:
temprature=0.1 (low temprature for focused and deterministic reponses, reduce randomness)
max_tokens=1000 (high max_tokens for complete and detailed information)
top_p=0.2 (low top_p for focused, predictable responses)
user_input_1: Extract the off-street vehicle parking spaces number requirement for retail or ratail stores or retail sales use in the table for Parking Requirements. Is this minimum or maximum requirement?
user_input_2: Is there maximum standard for parking spaces number which can not be exceeded?
Conduct comparison and analysis based on the gathering of extracted information from different cities’ zoning codes
System Message: You are an assistant making comparison, summary and analysis for zoning information of different cities. Your response should be readable.
user_input: Make a comparison chart for the off-street vehicle parking spaces number requirement for retail use and maximum standard for parking spaces number which can not be exceeded in these cities. Then write a paragraph for comparison and summary. Finally analyze the parking requirement status in Michigan based on this.
Manual audit to evaluate the accuracy and reliability of the extraction and analysis
Outcomes
Extraction
Detroit - Audit result: Accurate
Grand Rapids - Audit result: Fail to extract the maximum standard that 20% spaces exceeding minimum requirement need approval
Warren - Audit result: Accurate
Sterling Heights - Audit result: Fail to exract both two information
Ann Arbor - Audit result: Accurate
Lansing - Audit result: Accurate
Dearborn - Audit result: Accurate
Analysis
### Comparison Chart (modified by Web ChatGPT)
### Comparison and Summary (GPT-generated)
The parking requirements for retail use across these Michigan cities vary significantly, reflecting different urban planning priorities and zoning strategies. Detroit and Lansing have tiered minimum parking requirements based on the size of the retail space, with Detroit also distinguishing between general retail and shopping centers. Grand Rapids and Warren have simpler requirements, with Grand Rapids specifying different requirements for various zoning districts. Ann Arbor stands out by having a maximum parking requirement, emphasizing a cap on parking spaces to manage urban density and promote sustainable transportation. Lansing also imposes a form of maximum by requiring approval for exceeding the minimum by more than 20%. In contrast, cities like Dearborn and Warren focus solely on minimum requirements without specifying maximum limits.
### Analysis of Parking Requirement Status in Michigan (GPT-generated)
The parking requirements in Michigan cities reveal a diverse approach to urban planning and zoning. Cities like Ann Arbor and Lansing are more progressive in managing parking supply, potentially to curb urban sprawl and encourage alternative transportation modes. Ann Arbor's maximum parking standards are particularly noteworthy, as they aim to prevent excessive parking infrastructure that can lead to increased traffic congestion and environmental impacts. On the other hand, cities like Detroit and Grand Rapids emphasize minimum parking requirements, which may cater to the needs of car-dependent populations but could also contribute to over-parking and inefficient land use. The lack of maximum standards in most cities suggests a traditional approach to zoning, prioritizing accessibility and convenience for drivers over environmental and urban design considerations. Overall, Michigan's parking requirements reflect a balance between accommodating vehicular traffic and managing urban growth, with some cities beginning to adopt more sustainable practices.
Reflections
This project made me realize that AI does have great potential in helping us understand and extract complex text documents in the field of urban planning, but there are also many limitations. In particular, planning legal texts such as zoning codes are not only complex in format (containing a large number of tables, diagrams, and nested structures), but also highly specialized in content. It is difficult to accurately extract key information with existing large language models alone. To develop a truly reliable zoning chatbot, not only stronger information structure recognition capabilities are required, but also manual training with a planning professional background is required.
This project revolves around parking space requirements. I tried to use LLM to extract parking regulations for commercial purposes in zoning documents in different cities. In this process, I not only improved my prompt design capabilities, but also realized that AI is relatively accurate in extracting some direct statement information, but there are significant deviations for content with complex conditions and tabular expressions. This shows that the current LLM is more suitable as a "preliminary reading and classification tool", and truly reliable structured output still requires planners to assist in interpretation. It was also during this process that I came to understand more clearly that urban planners, due to their understanding of the structure of regulations and the logic behind them, can often raise more targeted questions and judge the accuracy of AI outputs. This is the "recognition" and "correction" that professional knowledge gives us when using AI.
Through this project, I not only deepened my understanding of the boundaries of AI capabilities, but also realized that human-machine collaboration is the key direction for the development of future planning technology tools: using AI for preliminary extraction and humans for high-level interpretation to jointly improve the readability of regulations and the efficiency of policy evaluation.