Detailed detection and extraction of estuarine tidal channels with multispectral and full-polarised SAR remote sensing

Abstract

Estuarine tidal channels are active geomorphic units in tidal flats. However, accurate information on the spatiotemporal changes in tidal channel systems remains scarce. The width of the tidal channels may vary from several kilometres to tens of centimetres. Monitoring tidal channel evolution is complicated because of periodic tidal scouring, anthropogenic activities and sea level rise. In this study, we propose a synergetic classification method to detect and extract morphological information of estuarine tidal channels with a spatial resolution of up to 3 m by fusing PlanetScope multispectral data with C-band GaoFen-3 fully polarised Synthetic Aperture Radar (SAR) data and machine learning algorithms. Considering the Yellow River Estuary as an example, the spectral features, vegetation and water index, polarisation and texture features derived from the multispectral and SAR images were selected as input data for classifiers according to feature importance ranking. Comparison to the maximum likelihood, and support vector machine classifiers, the synergetic classification with random forest showed the best performance, with an overall accuracy of 99.6%. Based on these results, the total number of tidal channels in the Yellow River Estuary reached 872, with a total length of 348.8 km. The spatiotemporal changes in the central axis over the last 4 years (2019–2022) suggest that the evolution of tidal channels was mainly controlled by ocean dynamics and anthropogenic activities. This method provides a cost-effective alternative to accurately map tidal channel systems in global estuarine and coastal zones and helps to quantitatively describe their morphological evolution, stability and drivers.