Ensembles of multiple spectral water indices for improving surface water classification

Mapping surface water distribution and its dynamics over various environments with robust methods is essential for managing resources supporting water-related policy design. Thresholding Single Water Index image (TSWI) threshold a common way of using index (WI) mapping it easy to use could obtain acceptable accuracies in many applications. As more WIs are available each has distinct merits, the real-world application TSWI, however, often face two practical concerns: (1) selection an appropriate WI (2) determination given WI. These issues problematic users who rely either on trial-and-error procedures that time-consuming or their personal preferences somewhat subjective. To better deal these concerns, alternative suggested here by transforming current paradigm into simple but ensemble approach called Collaborative Decision-making Indices (CDWI). A total 145 subsite images (900 × 900 m) from 22 Landsat-8 OLI scenes covering water-land around world were used assess performance TSWI CDWI. Five benchmark adopted five CDWI method: Normalized Difference (NDWI), Modified NDWI (MNDWI), Automated Extraction without considering (AWEI0) (AWEI1) shadows, state-of-the-art 2015 (WI2015). Two aspects analyzed: comparing (indicated both F1-scores Youden’s Index) accuracy sensitivities threshold. The results demonstrate produced higher than other most cases. Particularly, cases percentage) methods, i.e. 67% (CDWI) vs. 15% (TSWINDWI), 54% 22% (TSWIMNDWI), 42% 12% (TSWIAWEI0), 57% 17% (TSWIAWEI1), 34% (TSWIWI2015). Moreover, F1-score less sensitive change thresholds compared methods. important benefits make water. uncertainty method was thoroughly discussed general guidance (or look-up-table) determining parameters also suggested. underlying framework be readily generalizable applicable satellite images, such as Landsat TM/ETM+, MODIS, Sentinel-2 images.