A discussion on the estimation of apparent colour of water with the Forel-Ule index for medium-width sections of rivers with high hydrodynamic activity

Apparent colour has been used as an essential optical variable for surface waters. Expressing the apparent colour of waters is usually performed using the Forel-Ule Index (FUI) colour comparator scale, defined by 21 classes. The use of multispectral satellite images to extract the apparent colour of water expands the applicability of the FUI, as satellite images are characterized by mapping wide areas with a fixed revisit period, enabling monitoring of the evolution of the apparent colour of water. The use of satellite imagery has been explored for open ocean waters, mainly to characterize the optical properties of water. Less explored is their application to river environments, especially to sections where variations in geometry related to the emergence of islets and sandbanks, caused by water flow variations or high hydrodynamic activity. That is the case with river sections that are around 100 m wide, a dimension that is compatible with the spatial resolution of available multispectral imagery, e.g. Sentinel-2, meaning that the river section features can be mapped as continuous swaths of pixels. In this paper, we discuss the importance of the clear definition of land/river boundary geometry for the FUI classification of river sections, in order to prevent its application to the out-of -water area, a problem that may arise when fluvial hydrodynamic activity is high. We propose a solution based on a segmentation that uses a water index obtainable by multispectral band composition of Sentinel-2 images. An advantage of the process is the use of a unique image for both the water index segmentation and the FUI classification. The case study is the lower/medium section of the Tagus River in Portugal, susceptible to high levels of spatial changes in its water/land boundary geometry. Results show that an adequate delimitation of the water/land boundary is critical to prevent the consideration of land area in the process and bias in the FUI interpretation.