Abstract:Numerical models have long been used to understand geoscientific phenomena, including tidal currents, crucial for renewable energy production and coastal engineering. However, their computational cost hinders generating data of varying resolutions. As an alternative, deep learning-based downscaling methods have gained traction due to their faster inference speeds. But most of them are limited to only inference fixed scale and overlook important characteristics of target geoscientific data. In this paper, we propose a novel downscaling framework for tidal current data, addressing its unique characteristics, which are dissimilar to images: heterogeneity and local dependency. Moreover, our framework can generate any arbitrary-scale output utilizing a continuous representation model. Our proposed framework demonstrates significantly improved flow velocity predictions by 93.21% (MSE) and 63.85% (MAE) compared to the Baseline model while achieving a remarkable 33.2% reduction in FLOPs.

What problem does this paper attempt to address?

The paper primarily proposes a new solution to the downscaling problem of tidal current data. Specifically, the paper addresses the following key issues: 1. **Background and Challenges**: Traditional numerical models, while capable of simulating complex geoscientific phenomena including tidal currents, are costly for generating data at different resolutions. Although deep learning-based methods offer fast inference speeds, most are limited to fixed-scale inference and overlook important characteristics of target geoscientific data. 2. **Uniqueness Analysis**: Tidal current data differ from image data and have the following characteristics: - The data contain different types of information (flow velocity and water level), which are highly correlated but have different features. - Each grid corresponds to a specific geographic area, unlike the translation invariance assumed in the image super-resolution field. 3. **Proposed Framework**: To address the above issues, the authors propose a novel downscaling framework with the following three key components: - **Auxiliary Training Module (ATM)**: Helps to quickly and stably train the feature extractor from scratch. - **Feature Map Segmentation (FMS)**: Segments the feature map according to predefined ratios to downscale flow velocity and water level separately. - **Positional Encoding (PE)**: Adds a positional encoding layer in the feature extractor to utilize geographic area clues. 4. **Experimental Results**: A series of experiments validate the effectiveness of the proposed method, including ablation studies, trade-offs under different feature map segmentation ratios, and qualitative results. The results show that the method not only improves prediction performance but also reduces computational costs and can produce stable high-resolution predictions at untrained scales. In summary, this paper aims to overcome the limitations of existing methods by proposing a novel downscaling framework that considers the characteristics of tidal current data, thereby improving prediction accuracy and efficiency.

Arbitrary-Scale Downscaling of Tidal Current Data Using Implicit Continuous Representation

Statistical Downscaling of Coastal Directional Wave Spectra Using Deep Learning

Physics-informed neural networks of the Saint-Venant equations for downscaling a large-scale river model

Super-Resolution works for coastal simulations

Deep learning for statistical downscaling of sea states

GSDNet: A deep learning model for downscaling the significant wave height based on NAFNet

Efficient downscaling of satellite oceanographic data with convolutional neural networks

Generative Diffusion Model-based Downscaling of Observed Sea Surface Height over Kuroshio Extension since 2000

Fourier Neural Operators for Arbitrary Resolution Climate Data Downscaling

Downscaling of wind fields on the east coast of India using deep convolutional neural networks and their applications in storm surge computations

Physics-Aware Downsampling with Deep Learning for Scalable Flood Modeling

Spatial Downscaling of Sea Surface Temperature Using Diffusion Model

Generative downscaling of PDE solvers with physics-guided diffusion models

A deep learning super-resolution model to speed up computations of coastal sea states

Deep learning-based spatial downscaling and its application for tropical cyclone detection in the western North Pacific

Reducing Spatial Discretization Error on Coarse CFD Simulations Using an OpenFOAM-Embedded Deep Learning Framework

A Transfer Learning-Enhanced Generative Adversarial Network for Downscaling Sea Surface Height through Heterogeneous Data Fusion

Debias Coarsely, Sample Conditionally: Statistical Downscaling through Optimal Transport and Probabilistic Diffusion Models

A Tidal Current Speed Forecasting Model based on Multiple Periodicity Learning

Learning computational upscaling models for a class of convection–diffusion equations

Deep learning modeling framework for multi-resolution streamflow generation