MFogHub: Bridging Multi-Regional and Multi-Satellite Data for Global Marine Fog Detection and Forecasting

Abstract: Deep learning approaches for marine fog detection and forecasting have outperformed traditional methods, demonstrating significant scientific and practical importance. However, the limited availability of open-source datasets remains a major challenge. Existing datasets, often focused on a single region or satellite, restrict the ability to evaluate model performance across diverse conditions and hinder the exploration of intrinsic marine fog characteristics. To address these limitations, we introduce \textbf{MFogHub}, the first multi-regional and multi-satellite dataset to integrate annotated marine fog observations from 15 coastal fog-prone regions and six geostationary satellites, comprising over 68,000 high-resolution samples. By encompassing diverse regions and satellite perspectives, MFogHub facilitates rigorous evaluation of both detection and forecasting methods under varying conditions. Extensive experiments with 16 baseline models demonstrate that MFogHub can reveal generalization fluctuations due to regional and satellite discrepancy, while also serving as a valuable resource for the development of targeted and scalable fog prediction techniques. Through MFogHub, we aim to advance both the practical monitoring and scientific understanding of marine fog dynamics on a global scale. The dataset and code are at \href{https://github.com/kaka0910/MFogHub}{https://github.com/kaka0910/MFogHub}.

MFogHub: Bridging Multi-Regional and Multi-Satellite Data for Global Marine Fog Detection and Forecasting

The paper "MFogHub: Bridging Multi-Regional and Multi-Satellite Data for Global Marine Fog Detection and Forecasting" presents MFogHub, a comprehensive dataset designed to address challenges in detecting and forecasting marine fog using deep learning models. This dataset aims to overcome limitations in existing datasets, which often focus on single regions or satellites, thereby restricting the capability to evaluate model performance across diverse geophysical conditions.

The MFogHub dataset integrates annotated observations from 15 coastal regions prone to fog and six geostationary satellites, resulting in over 68,000 high-resolution samples. This amalgamation supports a robust analysis of detection and forecasting methods under varying conditions, surpassing traditional approaches that have leveraged singular datasets. The paper highlights that previous methods in marine fog research have been hampered by a reliance on isolated satellite and regional data, limiting the exploration of the intrinsic characteristics of marine fog dynamics.

The research utilizes 16 baseline models to demonstrate MFogHub's utility in evaluating the generalization capabilities across different regions and satellites. This dataset reveals substantial fluctuations due to regional and satellite discrepancies, which are pivotal for developing scalable and targeted fog prediction techniques. For instance, the study discloses how spatial distribution and satellite-specific variations lead to significant impacts on model generalization and accuracy, emphasizing MFogHub's role in prompting further exploration beyond local performance assessments.

Quantitative results reveal significant variations in model accuracy across the three evaluated regions using the CSI metric, highlighting how multi-region datasets can enhance model robustness compared to single-region data. Models such as DlinkViT and ABCNet exhibited reasonably consistent performances across diverse regions, demonstrating promise in satellite-based marine fog detection tasks.

The paper further explores the sensitivity of satellite spectral bands, particularly focusing on FY4A and H8/9 satellites, demonstrating how band selection influences detection performance. The use of true-color and natural-color synthesized images illustrates that natural-color images generally provide better distinction in fog detection tasks, reinforcing the advantage of using multi-spectral data.

The practical implications of this research are considerable, providing foundational knowledge and tools for advancing meteorological AI applications. The dataset's inclusion of diverse regional and satellite data is pivotal for improving model generalization, allowing researchers to refine detection algorithms and develop forecasting systems that better reflect real-world conditions. Theoretical implications suggest enhanced understanding of the dynamic marine fog phenomena, which could lead to novel methodologies in other environmental monitoring applications.

The future direction of this research includes optimizing the dataset for finer-grained region and spectral band analysis, providing a more nuanced understanding of fog dynamics. This will potentially manifest in developing more precise prediction models that can adapt to spatial and environmental variations, a critical aspect for advancements in meteorology using AI.

Overall, the MFogHub dataset represents a significant contribution to marine fog detection and forecasting, offering a platform for both practical and theoretical progression in this field. By facilitating a multi-regional and multi-satellite approach, MFogHub presents a pivotal step forward in leveraging deep learning for enhanced meteorological forecasting.