High-resolution mapping of surface water dynamics using restricted random forest: A case study in the Yellow River Basin

Numerous datasets revealing locations and alterations of water bodies have been produced from field investigations and remote sensing imagery. However, measuring surface water changes with high resolution remains a challenge. Here, a high-precision random forest (RF) model constrained by the annual maximum remote sensing indices was developed. Validation result from visual inspection shows that the accuracy of the model has reached 99.46%. Based on the improved RF model, monthly surface water variations in the Yellow River Basin over the past 10 years were quantified at 30-meter resolution using Landsat 8 and Sentinel 2 satellite images. The variations of water bodies including when water was presented, where occurrence changed and what form changes took in terms of seasonality and persistence were obtained. It is found that between 2014 and 2023, there are evident variations of permanent water bodies including formation and disappearance of surface permanent water bodies in the Yellow River Basin. Further research can be conducted on the intricate impact of climate and human activity on water bodies using the high-resolution surface water dataset provided.