GIScience & Remote Sensing ( IF 6.7 ) Pub Date : 2021-07-30 , DOI: 10.1080/15481603.2021.1948275 Jinzhu Wang 1, 2 , Michalis Hadjikakou 1 , Brett A. Bryan 1, 2
ABSTRACT
Accurate, long time-series, high-resolution mapping of built-up land dynamics is essential for understanding urbanization and its environmental impacts. Despite advances in remote sensing and classification algorithms, built-up land mapping which only uses spectral data and derived indices remains prone to uncertainty. We mapped the extent of built-up land in the North China Plain, one of China’s most important agricultural regions, from 1990 to 2019 at three-yearly intervals and 30 m spatial resolution. We applied Discrete Fourier Transformation to dense time-stack Landsat data to create Fourier predictors to reduce mapping uncertainty. As a result, we improved the overall accuracy of built-up land mapping by 8% compared to using spectral data and derived indices. In addition, a temporal correction algorithm applied to remove misclassified pixels further improved mapping accuracy to a consistently high level (>94%) over the time periods. A cross-product comparison showed that our maps achieved the highest accuracies across all years. The built-up land area in the North China Plain increased from 37,941 km2 in 1990–1992 to 131,578 km2 in 2017–2019. Consistent, high-accuracy, long time-series built-up land mapping provides a reliable basis for formulating policy and planning in one of the most rapidly urbanizing regions on this planet.