Physics-Informed Tensor-Train ConvLSTM for Volumetric Velocity Forecasting of the Loop Current

According to the National Academies, a week long forecast of velocity, vertical structure, and duration of the Loop Current (LC) and its eddies at a given location is a critical step toward understanding their effects on the gulf ecosystems as well as toward anticipating and mitigating the outcomes of anthropogenic and natural disasters in the Gulf of Mexico (GoM). However, creating such a forecast has remained a challenging problem since LC behavior is dominated by dynamic processes across multiple time and spatial scales not resolved at once by conventional numerical models. In this paper, building on the foundation of spatiotemporal predictive learning in video prediction, we develop a physics informed deep learning based prediction model called—Physics-informed Tensor-train ConvLSTM (PITT-ConvLSTM)—for forecasting 3D geo-spatiotemporal sequences. Specifically, we propose (1) a novel 4D higher-order recurrent neural network with empirical orthogonal function analysis to capture the hidden uncorrelated patterns of each hierarchy, (2) a convolutional tensor-train decomposition to capture higher-order space-time correlations, and (3) a mechanism that incorporates prior physics from domain experts by informing the learning in latent space. The advantage of our proposed approach is clear: constrained by the law of physics, the prediction model simultaneously learns good representations for frame dependencies (both short-term and long-term high-level dependency) and inter-hierarchical relations within each time frame. Experiments on geo-spatiotemporal data collected from the GoM demonstrate that the PITT-ConvLSTM model can successfully forecast the volumetric velocity of the LC and its eddies for a period greater than 1 week.

[1]  G. Shapiro,et al.  A multi-envelope vertical coordinate system for numerical ocean modelling , 2018, Ocean Dynamics.

[2]  D. R. Watts,et al.  Loop Current Eddy formation and baroclinic instability , 2016 .

[3]  Yisong Yue,et al.  Long-term Forecasting using Tensor-Train RNNs , 2017, ArXiv.

[4]  Robert R. Leben,et al.  Loop Current Frontal Eddies Based on Satellite Remote-Sensing and Drifter Data , 2009 .

[5]  Tal Ezer,et al.  An exercise in forecasting loop current and eddy frontal positions in the Gulf of Mexico , 2005 .

[6]  G. Halliwell,et al.  Evaluation of vertical coordinate and vertical mixing algorithms in the HYbrid-Coordinate Ocean Model (HYCOM) , 2004 .

[7]  Paris Perdikaris,et al.  Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , 2019, J. Comput. Phys..

[8]  Bobby Bhattacharjee,et al.  Tensorial Neural Networks: Generalization of Neural Networks and Application to Model Compression , 2018, 1805.10352.

[9]  Kevin C. Dorff,et al.  The MicroArray Quality Control (MAQC)-II study of common practices for the development and validation of microarray-based predictive models , 2010, Nature Biotechnology.

[10]  Paris Perdikaris,et al.  Physics Informed Deep Learning (Part II): Data-driven Discovery of Nonlinear Partial Differential Equations , 2017, ArXiv.

[11]  Ivan Oseledets,et al.  Tensor-Train Decomposition , 2011, SIAM J. Sci. Comput..

[12]  Petros Koumoutsakos,et al.  ContextVP: Fully Context-Aware Video Prediction , 2017, ECCV.

[13]  Jan Kautz,et al.  Convolutional Tensor-Train LSTM for Spatio-temporal Learning , 2020, NeurIPS.

[14]  V. E. Tarasov.  Discrete Wave Equation with Infinite Differences , 2017 .

[15]  Fanghua Xu,et al.  Hindcasts and forecasts of Loop Current and eddies in the Gulf of Mexico using local ensemble transform Kalman filter and optimum-interpolation assimilation schemes , 2013 .

[16]  Tianrui Li,et al.  Predicting Citywide Crowd Flows Using Deep Spatio-Temporal Residual Networks , 2017, Artif. Intell..

[17]  V. Simoncini,et al.  A Guide to Empirical Orthogonal Functions for Climate Data Analysis , 2010 .

[18]  Maziar Raissi,et al.  Deep Hidden Physics Models: Deep Learning of Nonlinear Partial Differential Equations , 2018, J. Mach. Learn. Res..

[19]  Samy Bengio,et al.  Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks , 2015, NIPS.

[20]  I. Kevrekidis,et al.  Physics-informed machine learning , 2021, Nature Reviews Physics.

[21]  Philip S. Yu,et al.  PredRNN: Recurrent Neural Networks for Predictive Learning using Spatiotemporal LSTMs , 2017, NIPS.

[22]  Ruoying He,et al.  Predictability of the Loop Current Variation and Eddy Shedding Process in the Gulf of Mexico Using an Artificial Neural Network Approach , 2015 .

[23]  Hanqi Zhuang,et al.  Medium‐Term Forecasting of Loop Current Eddy Cameron and Eddy Darwin Formation in the Gulf of Mexico With a Divide‐and‐Conquer Machine Learning Approach , 2019, Journal of Geophysical Research: Oceans.

[24]  R. Zemel,et al.  Neural Relational Inference for Interacting Systems , 2018, ICML.

[25]  Hanqi Zhuang,et al.  Loop Current SSH Forecasting: A New Domain Partitioning Approach for a Machine Learning Model , 2021, Forecasting.

[26]  Yan Liu,et al.  Differentiable Physics-informed Graph Networks , 2019, ArXiv.

[27]  Geoffrey E. Hinton,et al.  Generating Text with Recurrent Neural Networks , 2011, ICML.

[28]  Hui Jiang,et al.  Higher Order Recurrent Neural Networks , 2016, ArXiv.

[29]  Patrick Gallinari,et al.  Deep learning for physical processes: incorporating prior scientific knowledge , 2017, ICLR.

[30]  Bobby Bhattacharjee,et al.  Tensorized Spectrum Preserving Compression for Neural Networks , 2018, ArXiv.

[31]  Colin Summerhayes,et al.  Deep Water – The Gulf Oil Disaster and the Future of Offshore Drilling , 2011 .

[32]  A. Wills,et al.  Physics-informed machine learning , 2021, Nature Reviews Physics.

[33]  Steven P. Anderson,et al.  Real Time Observing and Forecasting of Loop Currents in 2015 , 2016 .

[34]  Paris Perdikaris,et al.  Physics Informed Deep Learning (Part I): Data-driven Solutions of Nonlinear Partial Differential Equations , 2017, ArXiv.

[35]  Eero P. Simoncelli,et al.  Image quality assessment: from error visibility to structural similarity , 2004, IEEE Transactions on Image Processing.

[36]  Laurent Bertino,et al.  High-resolution ensemble forecasting for the Gulf of Mexico eddies and fronts , 2009 .

[37]  Ming Zhou,et al.  A Tensorized Transformer for Language Modeling , 2019, NeurIPS.

[38]  George E. Karniadakis,et al.  Hidden physics models: Machine learning of nonlinear partial differential equations , 2017, J. Comput. Phys..

[39]  X. Liang,et al.  Instabilities and Multiscale Interactions Underlying the Loop Current Eddy Shedding in the Gulf of Mexico , 2019, Journal of Physical Oceanography.

[40]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[41]  Dylan Zhuang,et al.  Deep Learning for Drug Discovery: A Study of Identifying High Efficacy Drug Compounds Using a Cascade Transfer Learning Approach , 2021, Applied Sciences.

[42]  D. R. Watts,et al.  Gulf of Mexico Loop Current path variability , 2016 .

[43]  Raia Hadsell,et al.  Graph networks as learnable physics engines for inference and control , 2018, ICML.

[44]  Volker Tresp,et al.  Tensor-Train Recurrent Neural Networks for Video Classification , 2017, ICML.

[45]  J. Sheinbaum,et al.  A Loop Current experiment: Field and remote measurements , 2016 .

[46]  Maqc Consortium.  The MicroArray Quality Control ( MAQC )-II study of common practices for the development and validation of microarray-based predictive models , 2012 .

[47]  Dit-Yan Yeung,et al.  Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting , 2015, NIPS.

[48]  Robust Deep Networks with Randomized Tensor Regression Layers , 2019 .