Measuring polycentric urban development with multi-temporal Sentinel-1 SAR imagery: A case study in Shanghai, China

Studies on polycentric urban development (PUD) have been done by using LandScan and point of interest data. So far, the quality of PUD products has been limited by the spatio-temporal resolution of these datasets. Using Sentinel-1 SAR missions with a 10 m pixel spacing and a weekly repeat cycle, this study, for the first time, explores the potential of using a time series of Sentinel-1 SAR images for measuring urban polycentricity. In particular, we develop a variance-based filtering method to mitigate speckle noise. We propose the Kittler and Illingworth with a Modified Model (KI-MM) method to accurately identify PUD-related changes. We focus on the mean distance of new-born patches and the mean patch area for a PUD-related change analysis. These allow us to associate SAR tailor-made output with PUDs. The proposed methods were implemented on Google Earth Engine platform using 304 Sentinel-1 SAR images on the city of Shanghai, China, acquired between 2015 and 2018. We have tested 80 distribution models and found that the Laplace distribution is the best model for the KI-MM method. Our results show 2526, 2409, 4232 new-born patches for 2015–2016, 2016–2017, 2017-2018, with areas equal to 18, 20, 36 km2, respectively. Using cross-validation with Sentinel-2 (optical) reference, we found that the matching rates and F1-score between detected changes and reference for the years 2015–2016, 2016–2017, 2017-2018 were equal to 89.79% and 91.67%, 100% and 100%, 88.64% and 94.06%, respectively. We conclude that Sentinel-1 SAR images are suited to PUD applications at an intra-city scale with a high spatio-temporal resolution.