Reconstruction of long-term high-resolution lake variability: Algorithm improvement and applications in China

Temporal monitoring of inland water bodies using remote sensing images is often impeded by missing data caused by clouds and other adverse conditions. To date, various data recovery algorithms have been developed based on the water occurrence threshold (WOT), where the contaminated pixels are recovered by using longterm historical water distribution information. Here, we propose an improved algorithm, enhanced WOT (EWOT), which addresses the issue of mismatch between the water occurrence product and the actual historical water presence that has been neglected by previous WOT algorithms. The EWOT algorithm achieved an overall high accuracy (with a mean absolute percentage error (MAPE) = 5.1%) and prevailed against a representative WOT algorithm. The accuracy could be further reduced (MAPE = 1.6%) after the application of a novel quality control process. In addition, the temporal coverage of the high-quality surface water area time series was improved by an average of 26.2%, and the percent count and percent area of lakes with high-quality reconstructed data reached as high as 84.5% and 94.7%, respectively, facilitating the utilization of these data in further time series analysis. In general, the improvement was closely associated with the extent of the contamination before recovery. We evaluated the algorithm's ability to be implemented on a large scale in China, and the results generally were in line with previous studies. Nonetheless, our high-quality annual-based dataset presented a more comprehensive and continuous representation of the changes in lake area spanning from 2000 to 2019. The significance of improving the existing WOT algorithms is highlighted in this study, and the proposed method can be readily extended to lakes worldwide, thus providing a valuable data source to examine the causes and possible impacts of lake dynamics.

Automated summary

This study proposes an improved algorithm for monitoring inland water bodies using remote sensing images. The algorithm addresses the issue of missing data caused by clouds and other adverse conditions, and achieves high accuracy in recovering contaminated pixels. The algorithm improves the temporal coverage of the water area time series and provides valuable data for further analysis of lake dynamics.