Sensitivity analysis of Lagrangian tracking models OpenDrift and Parcels: An analysis on OpenMetBuoy-v2021 drifter in-situ data and Lagrangian simulations in the Agulhas Current System.

Abstract

Research on the drift at the surface of the ocean with Lagrangian particle tracking tools has numerous academical and practical purposes including the faith of pollutants in the ocean as marine plastics (van Sebille et al. 2015), search and rescue missions (Breivik et al. 2013), the spread of living organisms (Röhrs et al. 2014), oil spill evolution and clean-up (Mellor 2016) and hydrodynamic connectivity (van Sebille et al. 2010). To acquire a thorough understanding of the limitations and optimal utilization of Lagrangian tracking tools a sensitivity analysis of Lagrangian tracking models will be performed by comparison of six observed OpenMetBuoy-v2021 drifter trajectories to simulated trajectories in the highly dynamical Agulhas Current System by MCSD, Liu-Weisberg skillscore andWillmott skillscore. Three different aspects will be analysed: (1) two different offline Lagrangian tracking tools, OpenDrift and Parcels, (2) three different Eulerian ocean current products, HYCOM, Mercator and Globcurrent, and (3) the addition of wind and/or wave parameterizations. Additionally, the importance of the use of additional wind and/or wave velocity field(s) forcing in Lagrangian trajectory simulations will be evaluated by region and certain ocean and atmospheric conditions.

First, using the same time stepping scheme and linear interpolation methods, the different Lagrangian simulators OpenDrift and Parcels, perform identical. Second, the Globcurrent product shows the highest mean skill of the three ocean models, although it underestimates the speed for strong ocean currents due to its spatial resolution. The HYCOM and Mercator simulations show, respectively, 40% and 15% lower skill than the Globcurrent simulations. Finally, the Stokes drift will be incorporated in the Lagrangian trajectory simulations as a simple addition to the Eulerian ocean current field and for the wind, a factor of the 10m-wind speed will be used, the wind drift factor (WDF). Different simulations with WDF’s from 0-5% will be conducted. The addition of the Stokes drift and wind parameterizations, resulted in the best Lagrangian simulation performance compared to the observed drifter trajectories in skillscore, separation distance and speed (23% Liu-Weisberg skillscore; 16% MCSD).

This study shows that the addition of a separate Stokes drift field as forcing to the Lagrangian simulation shows a significant increase in performance, especially for high Stokes drift regimes (>0.15 m/s). Although, the simple addition of the Stokes drift field is still a parameterization and wave-current interactions are not taken into account. The addition of the WDF paramterization solely shows contrary results in MCSD and skill. The WDF parameterization seems to cover for many processes in the ocean surface that are not directly related to the wind speed. For a more physically based incorporation of direct wind drag on Lagrangian trajectory modelling in the future, these other processes need to be resolved first. Overall, the performance of the Eulerian ocean model seems to be the most important aspect in the skill of the Lagrangian trajectory simulations studied in this research, as where there is dislocation of the main ocean circulation show the lowest skill (Agulhas Retroflection and Eddies).

A final remark should be made regarding the limited scope of the drifter dataset used in this research. To statistically strengthen the drawn conclusions from this research, incorporating additional drifter trajectories favourable.