A method to model the current speed and direction within different layers of the water column using ADCP data

Abstract

Water on Earth is constantly moving, both vertically and horizontally, primarily due to tidal forces and weather conditions. The water column, from the surface to the seabed, is not uniform in terms of density, temperature, or composition. These differences cause the water to form distinct layers, each of which may move at different speeds and in different directions. One key factor influencing this variation is the Coriolis effect, which creates a spiral pattern in the flow of water layers, known as the Ekman spiral. The movement of water plays a vital role in transporting natural and human-related materials, such as plankton, sediments, and pollutants from sources like shipwrecks. Traditionally, water movement is often considered as a single, uniform flow. However, this assumption overlooks important vertical differences in current behaviour. This research addresses that gap by investigating a method to model the speed and direction of currents in multiple layers of the water column to provide more detailed insight into how water actually moves at different depths, which can support better environmental monitoring, ecological understanding, and economic decision-making.

To achieve the main objective, the research was structured around four sub-objectives. First, potential predictors influencing water currents, such as wind, bathymetry, and tidal forces, were identified through a desk research. Second, the vertical structure of the water column was analysed using ADCP data collected along two vessel tracks at the entrance of the IJmuiden harbour. By applying hue histogram analysis to ADCP visualizations, the water column was divided into three depth layers: 0–10 metres, 10–20 metres, and 20–30 metres. For each layer, average current speed and direction were extracted and organised into regression matrices.

Environmental predictors were prepared for use in modelling: bathymetric data from Rijkswaterstaat and EMODnet were harmonised into a single dataset, and tidal predictions from NLTides were interpolated into raster layers using Inverse Distance Weighting (IDW). Although wind data from KNMI was initially considered, it was excluded due to insufficient spatial and temporal variability in the available dataset. Two regression approaches, Ordinary Least Squares (OLS) and Random Forest, were tested by predicting the current speed for layer 1 using track 1 as training dat. Their outcome were compared, using R² , MAE, MAPE and RMSE. Random Forest was selected and then used to generate raster outputs predicting current speed and direction for the first two layers of the water column. To enhance understanding of spatial and vertical dynamics, difference layers were created comparing model predictions with tidal forecasts and between the two modelled layers.

The modelled current speed closely matched observed and tidal values in both upper layers, while current direction showed more deviation, particularly in Layer 2. Visual comparisons indicated greater differences in direction near the harbour’s oval-shaped depression. Difference layers between predicted and tidal values, as well as between the two modelled layers, highlighted zones of vertical shear and spatial variability. Validation using independent data from Track 2 confirmed high accuracy in speed prediction but poor performance for direction.

Overall, the model proved effective for predicting current speed but requires refinement for direction modelling.