Intertidal Bathymetry Extraction with Multispectral Images: A Logistic Regression Approach

Abstract

In this study, a methodology to estimate the intertidal bathymetry from multispectral remote sensing images is presented. The technique is based on the temporal variability of the water and the intertidal zone reflectance and their correlation with the tidal height. The water spectral behavior is characterized by high absorption at the infrared (IR) band or radiation with higher wavelengths. Due to tidal cycles, pixels on the intertidal zone have higher temporal variability on the near IR spectral reflectance. The variability of IR reflectivity in time is modeled through a sigmoid function of three parameters, where the inflection parameter corresponds to the pixel elevation. The methodology was tested at the Tagus river estuary in Lisbon, Portugal, and at the Bijagós archipelago, in the West African nation of Guinea-Bissau. Multispectral images from Sentinel-2 satellites were used, after atmospheric corrections from ACOLITE processor and the derived bathymetric model validated with in situ data. The presented method does not require additional depth data for calibration, and the output can generate intertidal digital elevation models at 10 m spatial resolution, without any manual editing by the operator. The results show a standard deviation of 0.34 m at the Tagus tidal zone, with −0.50 m bias, performing better than the Stumpf ratio transform algorithm, also applied to the test areas to derive intertidal bathymetry. This methodology can be used to update intertidal elevation models with clear benefits to monitoring of intertidal dynamics, morphodynamic modeling, and cartographic update.

Using Sentinel-2 Images to Estimate Topography, Tidal-Stage Lags and Exposure Periods over Large Intertidal Areas

“…In this study, we propose a set of procedures to address the issues outlined above and further develop the quality of DEM estimation of large intertidal flats using images sensed by Sentinel-2 Multi Spectral Instrument (S2A&B, hereafter). Using a time-series of images of the Bijagós Archipelago off the coast of Guinea-Bissau, West Africa, we (i) propose a four-parameter alternative for the pixel-based logistic algorithm proposed by the authors of [27] to produce a DEM of the intertidal area, and (ii) implement a precalibration procedure for all bands of the images in the time-series to reduce temporal variability at the pixel level. Furthermore, we (iii) propose a novel method to estimate cotidal lines in the entire extent of the Sentinel-2 scene and (iv) produce a final DEM of the area, after correcting for differences in tidal-stage across the region.…”

“…The method developed by authors in [27] is computationally intensive, since it requires a set of operations carried out at each pixel of the image. To avoid unnecessary calculations, the process starts by identifying the zone of interest, comprising only pixels in the intertidal area.…”

“…Recently, the authors of [26] proposed a variation to the waterline method, consisting of a pixel-based technique using the backscatter of a time-series of SAR Sentinel-1 images to generate a DEM of intertidal areas. The technique was later generalized for multispectral images [27]. This technique explores the relationship between the near infrared (NIR) reflectance and pixel elevation, which can be described by a logistic regression, relating a set of values of the water height at the time of image acquisition with the corresponding NIR values of each pixel.…”

“…In this relationship, NIR reflectance values at a given point will suddenly shift from high values, when the pixel is exposed, to low values, when the pixel is covered by water (which strongly absorbs IR radiation). The water height at this shift point is assumed to represent the height of that particular pixel [26,27].…”

“…However, in remote locations the number of available gauges is often very limited and consequently tidal models are unable to capture spatial differences in the stage of the tide over large extents. In such situations, one often assumes that the water height measured at one reference point is the same across the entire region, a simplification that can lead to significant errors in the estimates of elevation values when tide-stage differences are significant [27]. An alternative approach is to use tidal prediction models, such as the Oregon State University Tidal Prediction Software (OTPS) [28], which had been used for continental scale time-series tide modelling [25,29].…”