Deep Learning Based Ocean Current Feature Detection for Prediction

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a method for automated feature detection in ocean currents using CNNs and multi-sensor fusion, valuable for improving prediction models in marine renewable energy.

To enable the efficient integration of renewable energy sources into the power grid, accurately forecasting power production from the resources remains crucial. Ensuring electricity production from ocean current-based systems encompasses accurately predicting ocean current speeds at installations. Critical steps in this process entail identifying and tracking oceanographic features that impact flow speeds utilizing available data. Building on previous studies that quantified ocean current flows through direct measurements and numerical models, the present research advances the field by focusing on refined feature extraction techniques, and the impact of these features on flow velocity. Prior work has examined detection tools for ocean eddies and current boundaries, as well as extreme event mapping at proposed energy production sites like the Florida Current (FC). For instance, [1] descriptively portrays a comparative analysis amongst disparate forms of oceanographic satellite data, such as surface temperature, height, chlorophyll-a, and salinity, for discerningoceanic features. Additionally, an evaluation of sea surface temperature perturbations with concurrent edge detection during times of aberrant flow speeds measured by in situ instrumentation is delineated in [2]. Such developments have established a foundation for enhanced analysis of ocean dynamics and the identification of persistent flow structures. The proposed methodology integrates high-resolution sea surface temperature (SST), and sea surface chlorophyll-a (SSCa), and high-frequency (HF) radar data into a comprehensive data fusion tool designed to identify and track key oceanographic featureswithin the FC. Unlike geostrophic surface current velocity profiles procured from seasurface height measurements, HF radar offers enhanced spatial resolution capturing small-scale eddies and detailed current boundaries where the geostrophic assumption becomes impractical, revealing finer-scale dynamics that are often overlooked. SST intermittencies elucidate current boundaries and detect submesoscale eddies, displaying distinctive thermal patterns. Analogously, SSCa imaging supplies complementary insights into both biological and physical interactions influenced by eddy activity and meandering shifts associated with a current system. Together, these diverse techniques contribute to a clearer understanding of how significant flow features can be identified and impact ocean current velocities at potential ocean current electricity production sites. Convolutional neural networks (CNNs) are employed to automate the detection and extraction of these features. This deep learning component streamlines the analysis process and facilitates a comprehensive, quantitative assessment of ocean behavior. Hence, enhancements in feature recognition accuracy prevail with a corresponding robustness in classifying complex current behavior. Future work envisions integrating the developed data fusion tool into an advanced deep learning ocean current prediction model pertaining to the FC. This integration will extend the application of CNNs from feature extraction to full-scale predictive analytics, harnessing the power of fused multi-sensor data to model the intricate, nonlinear interactions that govern ocean dynamics. The resulting framework seeks to improve forecast accuracy for ocean current based electricity production, which will benefit renewable energy generation strategies.

• Zenodo https://zenodo.org/records/20533860first seen 2026-06-04 04:26:53