Spatially non‐stationary relationships between sea fog and marine meteorological factors with a satellite‐based geographically weighted regression model

Abstract

Sea fog is a common hazardous marine weather phenomenon driven by combined meteorological factors. Traditional monitoring and modeling methods face significant challenges in accurately representing their complex relationships. Hence, this study introduces a geographically weighted regression (GWR) model that investigates the spatially non‐stationary relationships between sea fog occurrence and marine meteorological factors (mean sea level pressure [MSL], relative humidity [RH], air–sea temperature difference [TD], and wind speed [WSD]) in the Yellow and Bohai Seas. Two key features distinguish the methodology: (1) the integration of multisource data, including Himawari‐9 satellite‐based sea fog detection and the fifth‐generation atmospheric reanalysis, achieving a more accurate and comprehensive representation of the spatially non‐stationary relationship; and (2) the standardization of these datasets into monthly averages and hexagonal grid cells, improving spatial consistency and the reliability of local parameter estimation within the GWR framework. Results show that sea fog occurrence and marine meteorological factors exhibit significant spatial clustering, based on separate spatial autocorrelation analyses. The GWR model performed significantly better and had an R ² value of 0.536 higher than the ordinary least squares (OLS) model. Additionally, for peak sea fog (March–June), optimal formation ranges are: MSL 1013.73–1021.74 hPa, RH 76.48%–86.38%, TD 0.18–0.84°C, WSD 5.44–6.35 m·s ⁻¹ . The GWR's local regression coefficients revealed that MSL and RH primarily regulate sea fog occurrence. Notably, the dominant factors and their influence magnitudes vary substantially across sea areas, regions, and months. Thus, monitoring these high‐impact factors dynamically across regions enhances the effectiveness of sea fog early warning systems.