• 18 November 2026
    • DOI:10.20944/preprints202611.1925.v1
Abstract
Coastal monitoring and marine traffic management services can benefit from short-horizon estimates of gust exceedance, but precursor detection and probability forecasting require different evaluations. In this work, we analyse one year of observations from ten meteorological stations located in the Varna Bay and the area of the Varna–Beloslav Lake system. Exploratory pressure-fall and gust-rise rules are compared against 23 network-wise events. A pressure fall of 1 hPa over three hours detects 26% of events with a 98% false-alarm ratio; a network-median gust rise of 1.5 m/s over one hour detects all 23 events with a 96% false-alarm ratio. Separately, a pooled current-gust logistic model estimates station-level exceedance of 12 m/s within two hours. The reported later-period Brier scores are 0.01134 for this model and 0.01931 for a constant training-frequency reference, giving a Brier skill score of 0.413. Adding an analogous station-level preceding-hour change predictor gives a Brier score of 0.01151 and a skill difference of −0.009, with a reported 95% block-bootstrap interval of [−0.0185, −0.0006]. Thus, adding this predictor does not improve the evaluated logistic specification. The diagnostics motivate investigation of gust trend but do not establish its incremental predictive value. Full-record feature exploration, overlapping targets and limited seasonal coverage constrain interpretation of the probability comparison. The results support further validation of local gust probabilities rather than an operational warning claim.