Better Methods for Studying Ocean Currents

2023-05-19

A new machine learning model has made more accurate predictions about ocean currents, which could help track plastic pollution and oil spills and aid in search and rescue efforts. Computer scientists at MIT, in collaboration with oceanographers, have developed a machine learning model that combines knowledge of fluid dynamics to more accurately predict the speed of ocean currents. The image below shows the drift trajectory of buoys overlaid on the surface ocean currents in the Gulf of Mexico. The red dot represents the position of the buoys on March 9, 2016, with the tail extending up to 14 days. To study ocean currents, scientists release buoys with GPS tags into the ocean and record their speeds to reconstruct the currents that transport them. These buoy data are also used to identify "divergence points," areas where water rises or sinks below the surface. By accurately predicting ocean currents and precisely locating divergence points, scientists can make more accurate weather predictions, estimate the spread of oil spills, or measure energy transfer in the ocean. A new research report claims that a new model combining machine learning makes more accurate predictions than traditional models. A multidisciplinary research team consisting of computer scientists and oceanographers from MIT found that the standard statistical models typically used for buoy data are difficult to accurately reconstruct ocean currents or identify divergence points because they make unrealistic assumptions about the behavior of water. The researchers developed a new model that incorporates knowledge of fluid mechanics to better reflect the physical phenomena in ocean currents. They demonstrated that their approach only requires a small additional computational cost but can predict ocean currents and identify divergence points more accurately than traditional models. This new model can help oceanographers make more accurate estimates from buoy data, enabling them to more effectively monitor the transport of biomass (such as seaweed), carbon, plastics, oil, and nutrients in the ocean. This information is also crucial for understanding and tracking climate change. "Our method captures the physical assumptions more appropriately and accurately. In this case, we have a lot of physical principles that we understand. We provide the model with some information so that it can focus on learning what is important to us, such as what the currents are away from the buoys or what these divergence points are and where they occur," said Tamara Broderick, senior author, associate professor in the Department of Electrical Engineering and Computer Science at MIT, and member of the Laboratory for Information and Decision Systems and the Institute for Data, Systems, and Society. In-depth analysis of the data Oceanographers use buoy velocity data to predict ocean currents and identify areas where water rises to the surface or sinks deeper, known as "divergence points." To estimate ocean currents and identify divergence points, oceanographers use a machine learning technique called Gaussian processes, which can make predictions even with sparse data. To work well in this scenario, Gaussian processes must make assumptions about the data to generate predictions. A standard approach to applying Gaussian processes to ocean data is to assume that the latitude and longitude components of the ocean currents are independent. However, this assumption is not accurate from a physical standpoint. Existing models, for example, assume that the divergence and vorticity (rotational motion of the fluid) of ocean currents operate at the same amplitude and length scale. Broderick said oceanographers know this is not true. Previous models also assumed the reference frame is important, meaning that the behavior of the fluid in the latitude and longitude directions would differ. "We thought we could address these issues with a model that incorporates physics," she said. They built a new model using the Helmholtz decomposition method to accurately represent the principles of fluid mechanics. This approach models ocean currents by decomposing them into vorticity components (capturing rotational motion) and divergence components (capturing the rise or sink of water). In this way, they provided the model with some basic physical knowledge to make more accurate predictions. This new model uses the same data as the old model. While their approach may involve more computational cost, the researchers demonstrated that the additional cost is relatively small. Floating performance They used synthetic and real ocean buoy data to evaluate the new model. Since synthetic data is fabricated by researchers, they can compare the model's predictions with ground truth ocean currents and divergence points. However, the assumptions involved in the simulation may not reflect reality, so the researchers also tested their model using data captured by buoys released in the Gulf of Mexico. In each case, their approach outperformed the standard Gaussian process and another machine learning method that used neural networks in predicting ocean currents and identifying divergence points. For example, in a simulation that included vortices adjacent to the ocean currents, the new method correctly predicted no divergence, while the previous Gaussian process method and neural network method both predicted divergence with very high confidence. Broderick added that the technique is also good at identifying vortices from a small set of buoys. Now that they have demonstrated the effectiveness of using the Helmholtz decomposition method, the researchers hope to incorporate a temporal element into their model, as ocean currents can change over time and space. Additionally, they want to better capture the influence of noise on the data, such as wind that can affect buoy velocity at times. Separating this noise from the data can make their method more accurate. "Our hope is to extract useful information from these velocity fields observed by buoys, infer actual divergence points and vortices, and make predictions outside of the buoys. We think our new technique will be very helpful for this," she said. "The authors cleverly integrate known behaviors from fluid mechanics into a flexible model to simulate ocean currents," said Massimiliano Russo, a biostatistician at Brigham and Women's Hospital and lecturer at Harvard Medical School, who was not involved in the work. "The resulting approach retains the flexibility of nonlinear modeling of ocean currents but can also capture phenomena such as vortices and related ocean currents that would only be noticed when fluid mechanics is integrated into the model in a thoughtful and scientifically sound way. This is a nice example showing that flexible models can be substantially improved through careful and scientifically motivated specifications." This research was supported in part by the Office of Naval Research, the National Science Foundation CAREER Award, and the Rosenstiel School of Marine and Atmospheric Science at the University of Miami.

RECENT AI NEWS

RECENT AI TOOLS