Changing the Unpredictable Nature of Internal Tides Through Deep Learning. Issue 8 (17th April 2023)

• Record Type: Journal Article
• Authors: Li, Bingtian; Wang, Yufei; Wei, Zexun; Pan, Haidong; Xu, Tengfei; Lv, Xianqing
• Is Part Of: Geophysical research letters. Volume 50:Issue 8(2023)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2023-04-17
• Subjects: Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2022GL102227
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: Nonstationary internal tides (ITs) are formed from their interactions with background currents. Harmonic analysis (HA), which cannot be used to estimate the incoherent component, is almost exclusively used method to predict ITs from observations. This remains ITs prediction challenge. In this study, we establish a deep learning framework to predict semidiurnal ITs. The model is established and trained with observed semidiurnal internal tidal currents that are 172 days long, and then ITs over the next 42 days are forecasted. The prediction accuracy is greatly improved using the deep learning framework. The magnitudes of errors using the deep learning framework are approximately 35% of those obtained using HA. Most temporal and spatial variations in baroclinic currents can successfully be forecasted using deep learning. In addition, the kinetic energy and incoherent components of ITs can be accurately predicted. Moreover, the relatively high adoptability of the established deep learning model is shown. Plain Language Summary: Internal tides (ITs) can be evidently influenced by their interactions with background currents and horizontal variations in the density field during propagation. ITs become nonstationary and uncorrelated to the spring neap forcing of surface tides. Harmonic analysis (HA), which is used to predict stationary ITs, cannot be used to forecast nonstationary ITs. Thus, IT forecasting is challenging. Many researchers have categorized ITs as unpredictable motions. The adoption of deep learning techniques that do not rely on any dynamic processes related to IT propagation and generation is one possible approach that can be used to improve the predictability of ITs. In this study, we establish a deep learning framework that can be used to significantly increase the predictability of ITs. Thus, the vertical displacement and temporal variation trends of ITs can be successfully forecasted. Baroclinic currents forecasted using deep learning are much closer to the observed currents than the currents obtained using HA. In addition to internal tidal currents, deep learning can also be used to predict the kinetic energy of ITs and the incoherent components, which have been extremely difficult to estimate in the past. Moreover, an established deep learning model can be used to predict ITs in nearby locations. Key Points: A deep learning approach can be used to overcome internal tide (IT) prediction challenges The baroclinic current, kinetic energy and incoherent components can be accurately forecasted using deep learning techniques The relatively high adoptability of the established deep learning model for the prediction of ITs in nearby locations is demonstrated