At the end of June, I attended the Computational Urban Planning and Management Conference (CUPUM 2025) in London. It was my first time visiting the UK, and the experience was very inspiring and eye-opening. Researchers from around the world presented studies on a wide range of topics in urban planning—such as urban greenery, mobility, climate modeling, participatory planning, and cultural heritage. What tied all the research together is the use of computational tools and data-driven methods. I was especially interested in the methodologies discussed, ranging from machine learning approaches to cutting-edge applications of large language models (LLMs). There was a lot to take in and reflect on, but in this blog, I’ll focus on the immediate connections I see between these methods and their potential applications in recreation monitoring and planning. Broadly, these technologies can support recreation in two main areas: (1) recreation monitoring and (2) recreation planning and management.
Links to Recreation monitoring
When it comes to recreation monitoring, the choice of methods and data sources is always a key topic of discussion. In many cases, the data we use depends on the data that is available, especially in this era of big data. Over the past decade, the emergence of mobile big data, such as social media, has introduced more diverse sources for understanding the movements and behaviors of recreationists. This data can offer multiple perspectives and different types of information.
However, despite the abundance of data, we still rely on traditional monitoring methods, which are often more accurate and representative. This raises an important question: how can we calibrate and integrate multi-source mobile big data into recreation monitoring in a meaningful and reliable way?
This also makes me wonder whether new AI-based methods might offer better solutions for data calibration, or whether traditional regression techniques might still prove to be more reliable. Additionally, as mobile applications quickly rise and fall in popularity, we need to ask: can our proposed methods adapt to these constantly evolving data sources with different characteristics?
Another challenge is how to translate complex AI models into user-friendly workflows or tools that can be applied in recreation management. There’s still a long way to go in unlocking the full potential of these technologies and validating their effectiveness. Compared to fields like traffic monitoring, recreation involves more spontaneous and random behavior, making it harder to track and validate.
Given the vast amount and variety of available data related to recreation, I believe it is worthwhile to explore data science approaches. These approaches can help reveal correlations between different types of data, such as visitor data, environmental conditions, and socio-demographic factors, across both space and time. This will be essential for building a systematic and multi-dimensional framework for recreation monitoring.
For example, machine learning models like gradient-boosted trees can be used to estimate visitor numbers in unmonitored areas by learning from data in areas with known visitor counts, a process we might call spatial generalization. Furthermore, Additionally, graph neural networks (GNNs) can model the spatial relationships between nearby locations, such as roads, parks, sub-regions within parks, or individual counting stations. This reflects the idea that visitor activity in one area may influence or correlate with that in surrounding areas. One of the most exciting sessions at the conference focused on large language models (LLMs), which I found particularly intriguing. Many presentations showcased the use of multimodal LLMs (MLLMs) in a wide range of applications, including neighborhood quality assessment, detecting social interactions, simulating complex mobility patterns, and modeling urban climate. MLLMs integrate multiple types of data, such as images and text. In the context of recreation studies, these models could potentially be used to estimate the distribution and characteristics of various visitor groups by combining video, text, image, and location data from various social media data sources.
Links to Recreation Planning and Management
Monitoring provides valuable insights that can inform recreation planning and management, such as evaluating the usage of recreational facilities or identifying potential threats to ecosystems. However, monitoring is often just one component within a broader planning framework, which is often intertwined with other systems such as nature restoration and green infrastructure planning. This makes the full integration of monitoring data into actual planning more difficult.
Beyond monitoring, digital tools offer new possibilities for supporting planning processes. These tools can help with data analysis, scenario simulation, and decision support. At the conference, I was particularly inspired by a presentation on adaptive pedestrian mobility simulation. The researchers used an AI agent-based framework to model pedestrian flow, incorporating adaptive agent logic and real-time behavioral parameters. Most notably, their method was validated using survey data and the received dynamic feedback based on spatial and temporal variations in demand. Agent-based modeling (ABM) has been applied in recreation flow simulation for a long time. Advancing traditional ABM with AI agents and fine-grained environmental data may potentially improve simulation accuracy.
This presentation made me reflect on the differences between ABM and statistical approaches, especially regarding their suitability for recreation planning. In the poster I presented at the conference, I used machine learning methods to assess changes in recreational use in response to a new national park plan. This analysis focused on correlations between environmental factors and visitation data, highlighting group behavior in specific contexts. In contrast, ABM is a bottom-up approach that simulates the interactions between individuals and their environment based on general behavioral rules. It would be interesting to further investigate the differences between these two approaches and their respective strengths in assessing recreational planning strategies.
In addition, engaging users in the planning process and improving communication is crucial for effective recreation planning and management. For example, chatbots powered by large language models (LLMs) offer an interactive way to share information with users while also collecting their opinions and preferences. Moreover, virtual and augmented reality (VR/AR) technologies can allow users to explore simulated nature experiences, while also tracking their movement during these interactions.
The CUPUM conference was an exciting journey for me. The intensive presentations gave me a glimpse into how fast the field is evolving and how much potential it holds. I also realized how much I still need to and want to learn. It was my first time in the UK. Meeting new peers while reconnecting with old friends made the experience even more special. Looking ahead, I hope to further explore some of the methods I came across and see how they might help address the challenges in my PhD research. There’s still a long way to go, but I’m feeling more motivated now.
