skip to main content


Title: 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.  more » « less
Award ID(s):
1828181
NSF-PAR ID:
10312466
Author(s) / Creator(s):
; ; ; ;
Date Published:
Journal Name:
Frontiers in Artificial Intelligence
Volume:
4
ISSN:
2624-8212
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Despite the large efforts made by the ocean modeling community, such as the GODAE (Global Ocean Data Assimilation Experiment), which started in 1997 and was renamed as OceanPredict in 2019, the prediction of ocean currents has remained a challenge until the present day—particularly in ocean regions that are characterized by rapid changes in their circulation due to changes in atmospheric forcing or due to the release of available potential energy through the development of instabilities. Ocean numerical models’ useful forecast window is no longer than two days over a given area with the best initialization possible. Predictions quickly diverge from the observational field throughout the water and become unreliable, despite the fact that they can simulate the observed dynamics through other variables such as temperature, salinity and sea surface height. Numerical methods such as harmonic analysis are used to predict both short- and long-term tidal currents with significant accuracy. However, they are limited to the areas where the tide was measured. In this study, a new approach to ocean current prediction based on deep learning is proposed. This method is evaluated on the measured energetic currents of the Gulf of Mexico circulation dominated by the Loop Current (LC) at multiple spatial and temporal scales. The approach taken herein consists of dividing the velocity tensor into planes perpendicular to each of the three Cartesian coordinate system directions. A Long Short-Term Memory Recurrent Neural Network, which is best suited to handling long-term dependencies in the data, was thus used to predict the evolution of the velocity field in each plane, along each of the three directions. The predicted tensors, made of the planes perpendicular to each Cartesian direction, revealed that the model’s prediction skills were best for the flow field in the planes perpendicular to the direction of prediction. Furthermore, the fusion of all three predicted tensors significantly increased the overall skills of the flow prediction over the individual model’s predictions. The useful forecast period of this new model was greater than 4 days with a root mean square error less than 0.05 cm·s−1 and a correlation coefficient of 0.6. 
    more » « less
  2. Abstract We assess to what extent seven state-of-the-art dynamical prediction systems can retrospectively predict winter sea surface temperature (SST) in the subpolar North Atlantic and the Nordic seas in the period 1970–2005. We focus on the region where warm water flows poleward (i.e., the Atlantic water pathway to the Arctic) and on interannual-to-decadal time scales. Observational studies demonstrate predictability several years in advance in this region, but we find that SST skill is low with significant skill only at a lead time of 1–2 years. To better understand why the prediction systems have predictive skill or lack thereof, we assess the skill of the systems to reproduce a spatiotemporal SST pattern based on observations. The physical mechanism underlying this pattern is a propagation of oceanic anomalies from low to high latitudes along the major currents, the North Atlantic Current and the Norwegian Atlantic Current. We find that the prediction systems have difficulties in reproducing this pattern. To identify whether the misrepresentation is due to incorrect model physics, we assess the respective uninitialized historical simulations. These simulations also tend to misrepresent the spatiotemporal SST pattern, indicating that the physical mechanism is not properly simulated. However, the representation of the pattern is slightly degraded in the predictions compared to historical runs, which could be a result of initialization shocks and forecast drift effects. Ways to enhance predictions could include improved initialization and better simulation of poleward circulation of anomalies. This might require model resolutions in which flow over complex bathymetry and the physics of mesoscale ocean eddies and their interactions with the atmosphere are resolved. Significance Statement In this study, we find that dynamical prediction systems and their respective climate models struggle to realistically represent ocean surface temperature variability in the eastern subpolar North Atlantic and Nordic seas on interannual-to-decadal time scales. In previous studies, ocean advection is proposed as a key mechanism in propagating temperature anomalies along the Atlantic water pathway toward the Arctic Ocean. Our analysis suggests that the predicted temperature anomalies are not properly circulated to the north; this is a result of model errors that seems to be exacerbated by the effect of initialization shocks and forecast drift. Better climate predictions in the study region will thus require improving the initialization step, as well as enhancing process representation in the climate models. 
    more » « less
  3. A divide-and-conquer (DAC) machine learning approach was first proposed by Wang et al. to forecast the sea surface height (SSH) of the Loop Current System (LCS) in the Gulf of Mexico. In this DAC approach, the forecast domain was divided into non-overlapping partitions, each of which had their own prediction model. The full domain SSH prediction was recovered by interpolating the SSH across each partition boundaries. Although the original DAC model was able to predict the LCS evolution and eddy shedding more than two months and three months in advance, respectively, growing errors at the partition boundaries negatively affected the model forecasting skills. In the study herein, a new partitioning method, which consists of overlapping partitions is presented. The region of interest is divided into 50%-overlapping partitions. At each prediction step, the SSH value at each point is computed from overlapping partitions, which significantly reduces the occurrence of unrealistic SSH features at partition boundaries. This new approach led to a significant improvement of the overall model performance both in terms of features prediction such as the location of the LC eddy SSH contours but also in terms of event prediction, such as the LC ring separation. We observed an approximate 12% decrease in error over a 10-week prediction, and also show that this method can approximate the location and shedding of eddy Cameron better than the original DAC method. 
    more » « less
  4. Abstract

    The ocean circulation is modulated by meandering currents and eddies. Forecasting their evolution is a key target of operational models, but their forecast skill remains limited. We propose a machine learning approach that improves the output of an ocean circulation model by learning and predicting its systematic biases. This method can be applied a priori to any region, and is tested in the Gulf of Mexico, where the Loop Current (LC) and the large anticyclonic eddies that detach from it are major forecasting targets. The LC dynamics are recurrent and lie on a low‐dimensional dynamical attractor. Building upon the information gained analyzing this low dimensional attractor, we improve the representation of sea surface anomalies in model outputs through information from satellite altimeter data using a Sequence‐to‐Sequence model, which is a special class of Recurrent Neural Network. Building upon the HYCOM‐NCODA analysis system, we deliver a correction to the forecast at the observation resolution. For at least 15 days the proposed method learns to forecast the systematic bias in the HYCOM‐NCODA, outperforming persistence, and improving the forecast. This data‐driven approach is fast and can be implemented as an added step to any dynamical hindcasting or forecasting model. It offers an interesting avenue for further developing hybrid modeling tools. In these tools, fundamental physical conservations are preserved through the integration of partial differential equations which obey them. In addition, the method highlights specific deficiencies of the hindcast system that deserve further investigation in the future.

     
    more » « less
  5. Abstract

    This study evaluates the performance of deep learning approach in the prediction of the ionospheric total electron content (TEC) during magnetically quiet periods. Two deep learning techniques, long short‐term memory (LSTM) and convolutional LSTM (ConvLSTM), are employed to predict TEC values 24 hr ahead in the vicinity of the Korean Peninsula (26.5°–40°N, 121°–134.5°E). The LSTM method predicts TEC at a single point based on time series of data at that point, whereas the ConvLSTM method simultaneously predicts TEC values at multiple points using spatiotemporal distribution of TEC. Both the LSTM and ConvLSTM models are trained using the complete regional TEC maps reconstructed by applying the Deep Convolutional Generative Adversarial Network–Poisson Blending (DCGAN‐PB) method to observed TEC data. The training period spans from 2002 to 2018, and the model performance is evaluated using 2019 data. Our results show that the ConvLSTM method outperforms the LSTM method, generating more reliable TEC maps with smaller root mean square errors when compared to the ground truth (DCGAN‐PB TEC maps). This outcome indicates that deep learning models can improve the prediction accuracy of TEC at a specific point by taking into account spatial information of TEC. We conclude that ConvLSTM is a reliable and efficient approach for the prompt ionospheric prediction.

     
    more » « less