Limits to Adaptive Capacity of Airports using Aerial Imagery and Machine Learning

Abstract

Due to exponential aviation demand in the last two decades, and conversely due to COVID19 and environmental pressures, airports have seen pressure to undergo unprecedented levels of transformation which includes upscaling, downscaling, and reconfiguration. For instance, the new Beijing Daxing airport designed to handle over 100 million passengers, the complete relocation of Istanbul airport to a new site away from the congested city center, the Jewel in Changi Airport combining airport functions into mixed-use spaces that mirror central business districts, and Anchorage Airport’s development as a logistics hub due to reduced East Asian passenger flights and recent Russian aerospace closing. These are just a sample of the range and extent of recent global changes that have and will impact airports and thus cities land use. Responding to this global change, our study investigated the ‘adaptive capacity’ of airports in a novel data-driven way.

We explored the spatial relationship between the world’s top 500 cities and their airports using machine learning techniques to perform automatic land use classification on aerial imagery, sourced from the Sentinel2 satellite data set, using a bespoke trained U-Net based image segmentation approach. These images were analytically compared at scale to derive relationships between airport-city land usage pairs, comparing airport network metrics including centrality, size, against the complex land usage, level of development and ability to expand or contract. Based on this unique dataset, we explored the future adaptive capacity of airports, as a result of spatial pressures from the city, airport demand driven by changes in local population, and passenger footfall. Identifying the cites that are the least and most able to adapt to future changes.

Keywords: Aviation, Aerotropolis, Aerial Analytics, Machine Learning, Urban Planning, Adaptive