[Paper Review] Land Use Detection & Identification using Geo-tagged Tweets.

Saeed Khan, Shahzamal|arXiv (Cornell University)|Jan 9, 2021

Human Mobility and Location-Based Analysis14 references5 citations

TL;DR

This paper proposes a supervised learning method that uses spatio-temporal patterns in geo-tagged Twitter data to identify and predict urban land use types—such as business, residential, recreational, and educational zones—in Brisbane, Melbourne, and Sydney. It achieves high overlap (up to 67.2%) between predicted clusters and official zoning maps, demonstrating that social media activity can effectively reflect real-world land use patterns with minimal data preprocessing.

ABSTRACT

Geo-tagged tweets can potentially help with sensing the interaction of people with their surrounding environment. Based on this hypothesis, this paper makes use of geotagged tweets in order to ascertain various land uses with a broader goal to help with urban/city planning. The proposed method utilises supervised learning to reveal spatial land use within cities with the help of Twitter activity signatures. Specifically, the technique involves using tweets from three cities of Australia namely Brisbane, Melbourne and Sydney. Analytical results are checked against the zoning data provided by respective city councils and a good match is observed between the predicted land use and existing land zoning by the city councils. We show that geo-tagged tweets contain features that can be useful for land use identification.

Motivation & Objective

• To develop a scalable, data-driven method for identifying urban land use types using real-time social media activity.
• To assess the feasibility of using geo-tagged Twitter data as a proxy for official land zoning and urban planning.
• To evaluate the spatial accuracy of land use clusters derived from Twitter activity against city council zoning maps.
• To explore the potential of social media data as a low-cost, up-to-date alternative to traditional survey-based land use assessment.
• To determine the extent to which temporal and spatial tweet patterns correlate with known land use categories.

Proposed method

• The method applies supervised learning to extract spatial and temporal patterns from geo-tagged tweets across three Australian cities: Brisbane, Melbourne, and Sydney.
• Twitter data is clustered using a spatial clustering algorithm that groups tweets based on location and time, forming distinct activity signatures per zone.
• The clustering process is designed to preserve underlying spatial features without filtering or discarding raw data points.
• The predicted clusters are compared to official land use zoning maps from city councils using polygon overlap analysis to measure spatial similarity.
• The technique relies on temporal behavior—such as peak activity times and tweet volume—to distinguish between land use types like residential, commercial, and recreational.
• The method is scalable and efficient, capable of processing large volumes of Twitter data without data reduction or loss of spatial resolution.

Experimental results

Research questions

• RQ1 To what extent do spatio-temporal patterns in geo-tagged Twitter data align with official land use zoning classifications?
• RQ2 Can supervised learning models trained on Twitter activity accurately predict known land use types such as residential, commercial, and recreational zones?
• RQ3 How does the overlap between predicted tweet clusters and official zoning polygons vary across different land use categories?
• RQ4 What is the impact of cluster size and zone area on the accuracy of overlap measurements between predicted and official land use zones?
• RQ5 Can social media data serve as a viable, real-time alternative to traditional survey-based land use assessment methods?

Key findings

• The proposed method achieved a maximum overlap of 67.2% between predicted clusters and official recreation zones in Melbourne, indicating strong spatial alignment.
• For business zones, the overlap ranged from 52.4% in Melbourne to 62.5% in Sydney, demonstrating consistent performance across cities.
• Residential land use clusters showed 50–56% overlap with official zoning, indicating reliable detection of residential activity patterns.
• Education zones were poorly predicted in Melbourne (0% overlap), but achieved 53.5% overlap in Sydney, suggesting regional variability in data representation.
• Recreational zones had the highest average overlap (52.6%–67.2%), indicating that Twitter activity patterns are particularly effective for identifying public and leisure spaces.
• The method demonstrated scalability and robustness by processing raw, unfiltered Twitter data without data loss, maintaining spatial fidelity across all clusters.