Human interference with the land surface is occurring at unprecedented frequency and intensity, posing substantial sustainability challenges. Continuously monitoring the land surface and promptly detecting land-cover changes are crucial for understanding human–environment interactions and balancing social development with natural resource management. While many Landsat time series–based change detection methods have been developed to capture fine-scale human-induced changes with temporal accuracy at the monthly scale or finer, most rely solely on independent pixel-level temporal modeling and thus underuse the added value of spatial information. This research introduces a novel quantile regression-based change detection method that integrates spatial context via empirical cumulative distribution functions and quantiles into a pixel-based change detection framework. We applied the proposed method to central Worcester, Massachusetts, an area that experienced intensive human activities in the early 2000s, using Landsat analysis-ready data surface reflectance data. Compared to state-of-the-art pixel-based methods, the proposed method achieved satisfactory detection accuracy with a F1 score of 88% and correctly detected 16 of 25 changes. It also demonstrated the lowest temporal root mean square error of 10.9 days, indicating its capability for early detection. Additionally, the method showed superior performance without requiring an extra cloud filter, highlighting its robustness to outliers. The source code for implementing the proposed method is publicly available at https://github.com/liangxy-geog/Quantile-Regression-based-Change-Detection-using-Landsat-Analysis-Ready-Data.
Liang et al. (Thu,) studied this question.