Massively parallel implicit equal-weights particle filter for ocean drift trajectory forecasting

Abstract

Forecasting ocean drift trajectories are important for many applications, including search and rescue operations, oil spill cleanup and iceberg risk mitigation. In an operational setting, forecasts of drift trajectories are produced based on computationally demanding forecasts of three-dimensional ocean currents. Herein, we investigate a complementary approach for shorter time scales by using a recent state-of-the-art implicit equal-weights particle filter applied to a simplified ocean model. To achieve this, we present a new algorithmic design for a data-assimilation system in which all components -including the model, model errors, and particle filter -take advantage of massively parallel compute architectures, such as graphical processing units. Faster computations can enable in-situ and ad-hoc model runs for emergency management, and larger ensembles for better uncertainty quantification. Using a challenging test case with near-realistic chaotic instabilities, we run data-assimilation experiments based on synthetic observations from drifting and moored buoys, and analyse the trajectory forecasts for the drifters. Our results show that even sparse drifter observations are sufficient to significantly improve short-term drift forecasts up to twelve hours. With equidistant moored buoys observing only 0.1% of the state space, the ensemble gives an accurate description of the true state after data assimilation followed by a high-quality probabilistic forecast. * Corresponding author: havard.heitlo.holm@sintef.no 1 arXiv:1910.01031v1 [stat.CO] 2 Oct 2019 which is an offline trajectory model. It reads the ocean current forecasts produced by the ocean circulation models, and uses these to predict drift trajectories. Although OpenDrift is computationally efficient, the ocean circulation models still require access to supercomputers. This paper explores the option of using a state-of-the-art particle filter method applied to a simplified ocean model for efficient drift trajectory forecasting. The aim is to build a data-assimilation system that can run efficiently on commodity-level desktop computers, and also be extendable to supercomputers. We achieve this by using a simplified ocean model and a data-assimilation method that both are able to take advantage of massively parallel accelerator hardware, such as the graphical processing unit (GPU). This work is not intended as a substitute of current operational systems, but as a complementary approach, in which the predicted currents may even be updated with in-situ observations, e.g., during ongoing search and rescue operations. Furthermore, by enabling research models to run on individual desktop and laptop computers, researchers are able to do more rapid prototyping. At the same time, this work will contribute to more efficient simulations also on supercomputers, since all algorithms may be extended to run on multiple GPUs and compute nodes.The paper is organized as follows: We start by reviewing related work relevant for Lagrangian data assimilation with accelerated pa...

Comparison of Ensemble-Based Data Assimilation Methods for Sparse Oceanographic Data

“…With β = 0, the implicit transform has a gap that leads to asymptotic bias (Skauvold et al, 2019), but this seems to be adjusted reasonably well by the second part having β > 0. The IEWPF has recently shown applicable and efficient for assimilating point-based observations into a simplified ocean model based on the shallow water equations (Holm et al, 2020). Herein, this method represents a state-of-the-art PF and is investigated more thoroughly.…”

“…Albeit highly informative about the ocean state at the buoy locations, they can be many kilometers apart, and spatio-temporal modeling is required to fill in the gaps between the sparse data. Buoy information is here used i) to constrain an advection-diffusion process for particle concentration (Foss et al, 2021), and ii) to constrain drift trajectories in an ocean model (Holm et al, 2020). Case i) represents a linear system in space-time, and we can study properties of new data assimilation approaches with the optimal Kalman filter (KF) solution.…”

“…While particle filters (PFs), see e.g. Van Leeuwen (2009) or Chopin and Papaspiliopoulos (2020), are less seen in high-dimensional real-world applications since they are prone to degenerate, Holm et al (2020) demonstrated a modern PF approach based on the implicit equal-weight particle filter (IEWPF) of Zhu et al (2016), that shows promising forecasting results for drift trajectories.…”

Data Assimilation for Ocean Drift Trajectories Using Massive Ensembles and GPUs

“…The scheme uses H as the water depth, η as the deviation from mean sea level, and hu and hv as the momentum along the abscissa and ordinate, respectively. We can perturb this ocean state using the approach in [12], in which we first generate a smooth random field, ∆ η, for each ensemble member, representing deviations of the ocean surface elevation. We continue by computing the momentum required to balance this perturbation, namely…”

A testbed for geomagnetic data assimilation

“…The Lorenz problems (Lorenz 1963(Lorenz , 1995, which are simple ordinary differential equations, are prominent examples of proxy models in NWP. Other examples include the Kuramoto-Sivashinsky equation (Kuramoto & Tsuzuki 1975;Sivashinsky 1977), quasi-geostrophic models (see, e.g., Evensen 1994) the primitive equations (see, e.g., Ades & van Leeuwen 2015) the shallow water equations (see, e.g., Holm et al 2020), and a barotropic vorticity model (see, e.g., Browne 2016). While these models may not accurately describe an atmospheric flow, they have been extremely useful for prototyping and testing new schemes for DA, which has in turn led to more accurate forecasts.…”