Improved surface water mapping using satellite remote sensing imagery based on optimization of the Otsu threshold and effective selection of remote-sensing water index

• A framework is proposed for surface water mapping in complex environments. • Two issues (fine-tuning threshold and selection index) in the optical remote-sensing water index are simultaneously solved. • The proposed method can improve the accuracy of the difference-based water indices after threshold optimization. Remote sensing monitoring is currently the main approach for global surface water dynamics. Several remote-sensing water indices, e.g. Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Automatic Water Extraction Index (AWEI), Water Index 2015 (WI2015), have been introduced in flood mapping, lake monitoring, and coastal extraction. When the threshold-based water indices were applied to images of different regions acquired at different times, it is challenging to determine an appropriate threshold to accurately discriminate water and non-water pixels in complex background. To overcome the limitation of most established water indices, considering fully the principles and advantages of different water indices, this study designed an automated surface water detection framework that optimizes the Otsu thresholds for water indices and picks the optimal water index based on a combination of 12 water indices. The method was tested on Landsat-9 images of Danjiangkou (DJK) Reservoir (mountain water bodies), Taihu Lake (plain water bodies), Lake Namtso (plateau lake in the Tibet Plateau), and Minjiangkou Estuary (estuary topography in Fujian Province), which shows that the overall accuracy (OA) is over 99 %, the kappa coefficient is in the range of 0.961 to 0.998, and F1 Score between 0.977 and 0.998, and it’s the OA is 2 % to 3 % higher than the results of the Otsu thresholds. Importantly, compared with the ratio-based water indices (e.g. NDWI and MNDWI), the proposed method can improve the accuracy of the difference-based water indices after threshold optimization. Meanwhile, we found that the optimal water indices are AWEIsh, WI2015, Multi-band Water Index (MBWI), and WI2015 for the DJK Reservoir, Taihu Lake, Lake Namtso, and Minjiangkou Estuary, respectively, which indicated that the optimal water index is variable in different regions. Our method not only improves the optimal threshold of water index, but also solves the problem of difficult acquisition of the optimal water indices for different scene images at a large spatial scale. The study implies great potential for the accurate monitoring of long-term and large-scale water bodies based on a variety of applications in Landsat data.