Quantifying urban areas with multi-source data based on percolation theory

TL;DR

This paper introduces a percolation theory-based method to quantify urban areas using multi-source data, demonstrating its effectiveness across 28 countries and validating results with Landsat data, thus advancing urban system analysis.

Contribution

The study develops a novel percolation-based approach for urban area delineation from diverse open-source datasets, enhancing robustness and data fusion capabilities.

Findings

Method captures similar urban characteristics across datasets

Zipf's law applies to most countries' urban areas

Validation shows high accuracy with Landsat data (mean κ=0.78)

Abstract

Quantifying urban areas is crucial for addressing associated urban issues such as environmental and sustainable problems. Remote sensing data, especially the nighttime light images, have been widely used to delineate urbanized areas across the world. Meanwhile, some emerging urban data, such as volunteered geographical information (e.g., OpenStreetMap) and social sensing data (e.g., mobile phone and social media), have also shown great potential in revealing urban boundaries and dynamics. However, consistent and robust methods to quantify urban areas from these multi-source data have remained elusive. Here, we propose a percolation-based method to extract urban areas from these multi-source urban data. We derive the optimal urban/non-urban threshold by considering the critical nature of urban systems with the support of the percolation theory. Furthermore, we apply the method with three open-source datasets - population, road, and nighttime light - to 28 countries. We show that the proposed method captures the similar urban characteristics in terms of urban areas from multi-source data, and Zipf's law holds well in most countries. The accuracy of the derived urban areas by different datasets has been validated with the Landsat-based reference data in 10 cities, and the accuracy can be further improved through data fusion ($\kappa=0.69-0.85$, mean $\kappa=0.78$). Our study not only provides an efficient method to quantify urban areas with open-source data, but also deepens the understanding of urban systems and sheds some light on multi-source data fusion in geographical fields.