An official website of the United States government
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
Loop Current SSH Forecasting: A New Domain Partitioning Approach for a Machine Learning Model
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
- Award ID(s):
- 1828181
- PAR ID:
- 10312469
- Date Published:
- Journal Name:
- Forecasting
- Volume:
- 3
- Issue:
- 3
- ISSN:
- 2571-9394
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Nonparametric regression on complex domains has been a challenging task as most existing methods, such as ensemble models based on binary decision trees, are not designed to account for intrinsic geometries and domain boundaries. This article proposes a Bayesian additive regression spanning trees (BAST) model for nonparametric regression on manifolds, with an emphasis on complex constrained domains or irregularly shaped spaces embedded in Euclidean spaces. Our model is built upon a random spanning tree manifold partition model as each weak learner, which is capable of capturing any irregularly shaped spatially contiguous partitions while respecting intrinsic geometries and domain boundary constraints. Utilizing many nice properties of spanning tree structures, we design an efficient Bayesian inference algorithm. Equipped with a soft prediction scheme, BAST is demonstrated to significantly outperform other competing methods in simulation experiments and in an application to the chlorophyll data in Aral Sea, due to its strong local adaptivity to different levels of smoothness.more » « less
-
null (Ed.)Context. We present observations of ZTF20aatqesi (SN 2020faa). This Type II supernova (SN) displays a luminous light curve (LC) that started to rebrighten from an initial decline. We investigate this in relation to the famous SN iPTF14hls, which received a great deal of attention and multiple interpretations in the literature, but whose nature and source of energy still remain unknown. Aims. We demonstrate the great similarity between SN 2020faa and iPTF14hls during the first 6 months, and use this comparison to forecast the evolution of SN 2020faa and to reflect on the less well observed early evolution of iPTF14hls. Methods. We present and analyse our observational data, consisting mainly of optical LCs from the Zwicky Transient Facility in the gri bands and of a sequence of optical spectra. We construct colour curves and a bolometric lc, and we compare ejecta-velocity and black-body radius evolutions for the two supernovae (SNe) and for more typical Type II SNe. Results. The LCs show a great similarity with those of iPTF14hls over the first 6 months in luminosity, timescale, and colour. In addition, the spectral evolution of SN 2020faa is that of a Type II SN, although it probes earlier epochs than those available for iPTF14hls. Conclusions. The similar LC behaviour is suggestive of SN 2020faa being a new iPTF14hls. We present these observations now to advocate follow-up observations, since most of the more striking evolution of SN iPTF14hls came later, with LC undulations and a spectacular longevity. On the other hand, for SN 2020faa we have better constraints on the explosion epoch than we had for iPTF14hls, and we have been able to spectroscopically monitor it from earlier phases than was done for the more famous sibling.more » « less
-
Abstract Internal tides (ITs) play a critical role in ocean mixing, and have strong signatures in ocean observations. Here, global IT sea surface height (SSH) in nadir altimetry is compared with an ocean forecast model that assimilates de‐tided SSH from nadir altimetry. The forecast model removes IT SSH variance from nadir altimetry at skill levels comparable to those achieved with empirical analysis of nadir altimetry. Accurate removal of IT SSH is needed to fully reveal lower‐frequency mesoscale eddies and currents in altimeter data. Analysis windows of order 30–120 days, made possible by the frequent (hourly) outputs of the forecast model, remove more IT SSH variance than longer windows. Forecast models offer a promising new approach for global internal tide mapping and altimetry correction. Because they provide information on the full water column, forecast models can also help to improve understanding of the underlying dynamics of ITs.more » « less
-
Elsevier Publishing Company (Ed.)Loop Current Frontal Eddies (LCFEs) are known to intensify and assist in the Loop Current (LC) eddy shedding. In addition to interacting with the LC, these eddies also modify the circulation in the eastern Gulf of Mexico by attracting water and passive tracers such as chlorophyll, Mississippi freshwater, and pollutants to the LC-LCFE front. During the 2010 Deepwater Horizon oil spill, part of the oil was entrained not only in the LC-LCFE front but also inside an LCFE, where it remained for weeks. This study assesses the ability of the LCFEs to transport water and passive tracers without exchange with the exterior (i.e., Lagrangian coherence) using altimetry and a high-resolution model. The following open questions are answered: (1) How long can the LCFEs remain Lagrangian coherent at and below the surface? (2) What is the source of water for the formation of LCFEs? (3) Can the formation of Lagrangian coherent LCFEs attract shelf water? Strong frontal eddies leading to LC eddy shedding are investigated using a 1-km resolution model for the Gulf of Mexico and altimetry. The results show that LCFEs are composed of waters originating from the outer band of the LC front, the region north of the LC, and the western West Florida Shelf and Mississippi/Alabama/Florida shelf, and potentially drive cross-shelf exchange of particles, water properties, and nutrients. At depth (≈180 m), most LCFE water comes from the outer band of the LC front in the form of smaller frontal eddies. Once formed, LCFEs can transport water and passive tracers in their interior without exchange with the exterior for weeks: these eddies remained Lagrangian coherent for up to 25 days in the altimetry dataset and 18 days at the surface and 29 days at depth (≈180 m) in the simulation. LCFE can remain Lagrangian coherent up to a depth of ≈ 560 m. Additional analyses show that the LCFE involved in the Deepwater Horizon oil spill formed from water near the oil rig location, in agreement with previous studies. Temperature-salinity diagrams from a high-resolution model and aircraft expendable profilers show that LCFEs are composed of Gulf of Mexico water as opposed to LC water. Therefore, LCFE formation and propagation actively modify the surrounding circulation and affect the evolution of the flow and the transport of oil and other passive tracers in the Eastern Gulf of Mexico.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.3390/forecast3030036
