Abstract
Reconstructions of Lagrangian drift, for example for objects lost at sea, are often uncertain due to unresolved physical phenomena within the data. Uncertainty is usually overcome by introducing stochasticity into the drift, but this approach requires specific assumptions for modelling uncertainty. We remove this constraint by presenting a purely data-driven framework for modelling probabilistic drift in flexible environments. Using ocean circulation model simulations, we generate probabilistic trajectories of object location by simulating uncertainty in the initial object position. We train an emulator of probabilistic drift over one day given perfectly known velocities and observe good agreement with numerical simulations. Several loss functions are tested. Then, we strain our framework by training models where the input information is imperfect. On these harder scenarios, we observe reasonable predictions although the effects of data drift become noticeable when evaluating the models against unseen flow scenarios.
Similar content being viewed by others
Data Availability
Code Availability
Notes
Due to the limited number of one-day snapshots in a year, the same flow scenarios are used for both training and evaluation for the year 2018. However, trajectories are sampled across different locations (see Section 3.2) which prevents the model from relying on memory.
Although U and V are gridded on curvilinear coordinates, the limited region that we consider makes it reasonable to neglect projection errors and to associate each grid cell to a Cartesian pixel.
If this results in particles to lie outside of the ocean’s domain then we discard them, and hence the actual number of particles may be less than \(N_P\).
The training split for the 2016 dataset is not used in this study.
References
Gill AE (1982) Atmosphere-ocean dynamics. Academic Press 30
Van Sebille E, Griffies SM, Abernathey R et al (2018) Lagrangian ocean analysis: Fundamentals and practices. Ocean Model 121:49–75
Koszalka IM, Haine TW, Magaldi MG (2013) Fates and travel times of Denmark Strait overflow water in the Irminger Basin. J Phys Oceanogr 43(12):2611–2628
Visser AW (2008) Lagrangian modelling of plankton motion: From deceptively simple random walks to fokker-planck and back again. J Mar Syst 70(3–4):287–299
LaCasce J (2008) Statistics from lagrangian observations. Prog Oceanogr 77(1):1–29
Han M Sane S Johnson CR (2021) Exploratory Lagrangian-based particle tracing using deep learning. arXiv:2110.08338
Grossi MD, Kubat M, Özgökmen TM (2020) Predicting particle trajectories in oceanic flows using artificial neural networks. Ocean Model 56(101):707
Nam YW, Cho HY, Kim DY et al (2020) An improvement on estimated drifter tracking through machine learning and evolutionary search. Applied Sciences 10(22):8123
Aksamit NO, Sapsis T, Haller G (2020) Machine-learning mesoscale and submesoscale surface dynamics from lagrangian ocean drifter trajectories. J Phys Oceanogr 50(5):1179–1196
Zhuang J Kochkov D Bar-Sinai Y et al (2021) Learned discretizations for passive scalar advection in a two-dimensional turbulent flow. Phys Rev Fluids 6(6):064,605
Ourmières Y, Zakardjian B, Béranger K et al (2011) Assessment of a NEMO-based downscaling experiment for the North-Western Mediterranean region: impacts on the Northern Current and comparison with ADCP data and altimetry products. Ocean Model 39(34):386–404
Gurvan M, Bourdallé-Badie R Chanut J et al (2017) NEMO ocean engine. https://doi.org/10.5281/zenodo.1464816
Guihou K, Marmain J, Ourmieres Y et al (2013) A case study of the mesoscale dynamics in the North-Western Mediterranean Sea: a combined data-model approach. Ocean Dyn 63(7):793–808
Mansui J, Molcard A, Ourmières Y (2015) Modelling the transport and accumulation of floating marine debris in the Mediterranean basin. Mar Pollut Bull 91(1):249–257
Delandmeter P Sebille Ev (2019) The Parcels v2.0 Lagrangian framework: new field interpolation schemes. Geosci Model Dev 12(8):3571–3584
Chen LC Papandreou G Schroff F et al (2017) Rethinking atrous convolution for semantic image segmentation. arXiv:1706.05587
He K Gkioxari G Dollár P et al (2017) Mask R-CNN. In: International conference on computer vision, pp 2961–2969
Ronneberger O Fischer P Brox T (2015) U-Net: Convolutional networks for biomedical image segmentation. In: Medical image computing and computer-assisted intervention, pp 234–241
Ioffe S Szegedy C (2015) Batch normalization: Accelerating deep network training by reducing internal covariate shift. In: International conference on machine learning, PMLR, pp 448–456
He K Zhang X Ren S et al (2016) Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 770–778
Hayou S Clerico E He B et al (2021) Stable resnet. In: International conference on artificial intelligence and statistics, PMLR, pp 1324–1332
Carvalho M Le Saux B Trouvé-Peloux P et al (2018) On regression losses for deep depth estimation. In: 2018 25th IEEE international conference on image processing (ICIP), IEEE, pp 2915–2919
Loshchilov I Hutter F (2017) Decoupled weight decay regularization. arXiv:1711.05101
Loshchilov I Hutter F (2016) Sgdr: stochastic gradient descent with warm restarts. arXiv:1608.03983
Ma J Yarats D (2021) On the adequacy of untuned warmup for adaptive optimization. In: Proceedings of the AAAI conference on artificial intelligence, pp 8828–8836
Micikevicius P Narang S Alben J et al (2017) Mixed precision training. arXiv:1710.03740
Jenkins J Paiement A Ourmières Y et al (2022) A dnn framework for learning lagrangian drift with uncertainty. arXiv:2204.05891
Funding
This work has been supported by Ocean Next, Datlas, and ANRT by means of a PhD CIFRE grant attributed to Joseph Jenkins. It has been co-granted by the FEDER MARITTIMO GIAS (Geolocalisation by AI for maritime Security). Funding was received from Chaire Intelligence Artificielle ADSIL ANR-20-CHIA-0014 and ANR-18-CE40-0014 SMILES. The NEMO calculations were performed using GENCI-IDRIS resources, grant A0110101707.
Author information
Authors and Affiliations
Contributions
Conceptualization: Y.O., A.P., J.J., J.S. and H.G.; Methodology: J.J., A.P., Y.O., J.S., and C.U.; Data curation: J.J. and Y.O.; Investigation: J.J.; Software & validation: J.J.; Visualization: J.J. and A.P.; Writing - original draft: J.J. and A.P.; Writing - review & editing: A.P., Y.O. and J.S.; Supervision: A.P., Y.O. and H.G.; Project administration: H.G., J.V., and C.U.; Funding acquisition: H.G., J.V., C.U. and Y.O.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Jenkins, J., Paiement, A., Ourmières, Y. et al. A DNN Framework for Learning Lagrangian Drift With Uncertainty. Appl Intell 53, 23729–23739 (2023). https://doi.org/10.1007/s10489-023-04625-1
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10489-023-04625-1