A novel approach for quantifying high-frequency urban land cover changes at the block level with scarce clear-sky Landsat observations

Abstract High-frequency and complex transitions are two important characteristics of urban land cover changes (ULCCs). However, the change magnitude and duration of ULCCs have rarely been discussed in the context of high-frequency ULCCs. Moreover, previous models for quantifying high-frequency changes commonly required numerous clear-sky observations. Such requirement is difficult to be satisfied in widespread cloud-prone cities. Considering these characteristics of ULCCs and data availability, we developed a novel approach that can comprehensively quantify ULCCs with relatively limited data. The approach successfully integrated two commonly used approaches in change detection, namely temporal segmentation and trajectory classification. In addition, sparse clear-sky observations were used in the time series of fractional land cover (TSFLC). Specifically, we first conducted classifications for each year, and then extracted the TSFLC from each block. Second, we used temporal segmentation to segment the TSFLC based on the local change trend. Third, we classified the segments of time series by trajectory classification, and recorded the change information. The proposed approach can provide comprehensive details for ULCCs including the frequency of changes during the observation time, the transition, the change time, and the change magnitude for each change. The approach quantifies multiple changes with a slightly higher accuracy (93.33%) than that of previous approaches for only one change (91.00%). The categories/ transitions of ULCC were also quantified, with an accuracy of 81.67%. Moreover, the approach effectively monitored ULCCs in a cloud-prone city and successfully captured and analyzed two common transitions, block construction and urban renewal. The present approach offers a new perspective for providing complete and accurate ULCC information with scarce clear-sky observations.