Do Anglers Stay Home During Typhoon Warnings? Effort on 21 Warning Days Across Nine Taiwan Typhoons

A typhoon warning is one of the few external events that is timed to the half hour, nationwide, and directly relevant to fishing. It lets us ask a question that is normally unanswerable: do official warnings actually change what anglers do, by how much, which kind of warning works, and what happens after the lift? We align the sea and land warning windows of all 9 typhoons warned by Taiwan's Central Weather Administration in 2025–2026 with the daily number of anglers logging marine and freshwater catches. Freshwater anglers face the same holidays and rain but not the sea state, so they are a natural control.

Finding: Land warnings cut marine angling to 0.40 of normal; sea-only warnings barely change behaviour. Nobody rushes back after the lift: marine participation on the first day after is still 0.73 of normal. Those who fish through a warning are disproportionately active regulars in the Kaohsiung and Tainan estuaries, facing waves 1.3 times the usual and three-metre seas six times as often.

This page describes what anglers actually do, not what they should do. A sea warning means the gale radius is about to enter coastal waters, and a lifted warning does not mean the sea has calmed. Follow the Central Weather Administration and Coast Guard.

Analysis date 2026-09-05 · Data 2025-01-01 to 2026-09-04 · 中文版

• H1 Land warning: marine participation falls on land-warning days, by more than freshwater. → supported: marine 0.40× (0.24–0.68, p < 0.001), freshwater 0.67× (p 0.115), marine/freshwater difference 0.60× (p < 0.001)
• H2 Sea warning: marine participation falls on sea-warning-only days. → not supported: 0.85× (0.58–1.24, p 0.360); only a drop of forty percent or more can be excluded
• H3 Rebound: marine participation on the first day after the lift exceeds normal. → rejected, opposite sign: 0.73× (p 0.103; difference from freshwater 0.77×, p 0.022)
• H4 Anticipation: marine participation already falls on the day before the first warning day. → not supported: 0.95× (p 0.736)

With Bonferroni correction across the four primary tests the threshold is p < 0.0125; only H1 passes. All p-values are two-sided, from 2,000 recomputations in which each typhoon's affected window is shifted independently onto typhoon-free dates.

The land-warning day is the only point in the chart that leaves the band: marine participation 0.40×, freshwater 0.67×. Freshwater falls too, because typhoon holidays and heavy rain keep people home, but marine falls far more; the 0.60× gap is the most robust number on this page (EXIF-timestamp subset 0.34×, catch counts 0.34×, same direction and size).

On sea-only days the marine point is 0.85 and freshwater 0.98, both intervals covering 1.0. By the CWA definition a sea warning means the gale radius may enter waters within 100 km inside 24 hours, which for shore anglers is when the swell arrives. The behaviour data say anglers treat a sea warning as "nothing yet".

The marine points for the first and second day after the lift sit below 1.0 (0.73, 0.83) while freshwater sits at 1.0. There is no catch-up rush. By the hour, marine angler-hours in the first 0–6 hours after the lift are 0.85 of normal, 6–12 hours 0.64, 12–24 hours 0.75, 24–48 hours 0.87; the first 24 hours combined 0.75 (p 0.120).

This table makes "which warning works" sharper. Across sea-only typhoons, marine angler-hours run at 0.87 of normal (freshwater 1.14). Within the same typhoon, the hours after the sea warning but before the land warning are 0.68; once the land warning is issued, 0.34; after the land warning is lifted but while the sea warning stays up, 0.48. Behaviour tracks the land warning, not the sea warning.

All 5 typhoons with a land warning have warning-day marine ratios at or below 0.65; three of the four sea-only typhoons are above 0.9. The 2025 user base was small (expectations of 5–20 anglers a day), so individual storms need their intervals, but the direction is consistent.

Three things stand out. First, these records are not back-filled: the EXIF-timestamp share (53%) is higher than baseline (38%), so people really were at the water taking photos during the warning. Second, they are regulars: median active days 2 against 1 for all marine anglers, with 28% in the most active tenth. Fishing through a warning is a familiar-spot, experienced-angler behaviour, not novices wandering in. Third, the sea really was worse: median wave height 1.3 times baseline, three-metre seas 6 times as often; during land warnings the median is 2.565 m with 19% at three metres or more. The locations cluster in the estuaries and harbours of Kaohsiung and Tainan, the relatively sheltered south-west, consistent with experienced anglers choosing their ground. If safety messaging is to work, its audience is the regulars at known spots, not the general public.

A typhoon holiday does not send people fishing. On the five verified 2025 days off (Danas 6–7 July, Podul 13 August, Fung-wong 11–12 November), freshwater participation was 0.68 of normal and marine 0.32. The extra free day was eaten by rain.

A rainstorm with no warning keeps people home just as well. Typhoon Fengshen in October 2025 never entered the warning area, but interacting with the north-east monsoon it dropped four days of heavy rain on the north and east. From 20 to 23 October marine participation was 0.48 of normal and freshwater 0.46, both falling together by about as much as a land-warning day. This shows that the freshwater 0.67 on land-warning days is rain, and the extra marine drop is the warning plus the sea.

Exploratory aside: on the 47 unaffected days when the maximum wave height attached to any catch reached 3 m or more, marine participation was 0.77 of expected (0.87 on the 55 days at 1 m or less). Wave height is only observed when someone logged, so this comparison is selected and serves only as a side note: without a warning, rough seas alone trim marine participation by a little over a tenth, far less than the sixty percent of a land warning.

Robustness: land-warning marine on the EXIF subset 0.34× (p < 0.001), by catch count 0.34× (p 0.001); sea-only EXIF 0.90× (p 0.496), catch count 0.81× (p 0.335). All warning days combined: marine 0.68× (p 0.030), marine/freshwater difference 0.79× (p 0.003).

Measured

• Land-warning marine anglers 0.40× (0.24–0.68, p < 0.001), freshwater 0.67×, difference 0.60× (p < 0.001).
• Sea-only marine 0.85× (0.58–1.24, p 0.360); minimum detectable ratio 0.68.
• Lift +1 day 0.73×, +2 days 0.83×; first 24 hours 0.75×.
• 272 marine catches by 189 anglers during warning hours; EXIF share, wave heights and counties as tabled.
• Five typhoon holidays: freshwater 0.68×; four Fengshen rain days: marine 0.48×, freshwater 0.46×.

Inferred, to be read separately

• Logging anglers proxy for anglers who went out. Blank trips are invisible; if those who went out during warnings caught less, the drop is overstated. The bias acts on marine and freshwater alike, so the difference is more reliable than either ratio.
• "Sea warnings do not change behaviour" rests on 12 sea-only days and excludes a drop of forty percent or more; a drop of ten to thirty percent cannot be excluded.
• Three of the four sea-only typhoons tracked through the Bashi Channel or the northern waters, with little effect on the south-west; the apparent ineffectiveness of sea warnings may partly reflect anglers judging the track rather than ignoring the warning. The data cannot separate the two.
• Why participation stays low after the lift (sea state, road and harbour closures, rain) cannot be separated here.
• Wave heights are nearest-station values attached to each catch, not measured at the fishing spot.
• This page does not estimate how dangerous fishing during a warning is; that needs incident data we do not have.

The closest peer-reviewed work is a study of weekly angling effort in coastal Louisiana, 2015–2021: tropical storms and hurricanes reduce effort, and effort recovers rapidly afterwards. Taiwan shows the same drop, but only under land warnings, and no rapid recovery: participation is still below normal two days after the lift. We found no study that tests angler behaviour against the exact issue and lift times of official warnings while separating sea from land warnings; that is the gap this page tries to fill.

• 中央氣象署颱風資料庫：有發警報颱風列表與各颱風概況（海上／陸上颱風警報發布與解除時間）。Central Weather Administration Typhoon Database, warned-typhoon list and per-typhoon overview. sourceSource of every warning window in this study; retrieved 2026-09-05.
• 中央氣象署：颱風警報（海上颱風警報＝預測七級風暴風範圍 24 小時內可能侵襲臺灣本島、澎湖、金門或馬祖 100 公里以內海域；陸上颱風警報＝18 小時內可能侵襲陸上）。CWA typhoon warning definitions. sourceOfficial definitions: a sea warning means the gale radius is about to reach coastal waters, not that nothing is happening yet.
• Heat, hurricanes, and health: Effects of natural disturbances on angling effort. PLOS ONE (2023), PMC10490919. Weekly angling-effort estimates, coastal Louisiana 2015–2021. sourcePeer-reviewed: tropical storms and hurricanes reduce angling effort, which then recovers rapidly. We see the same drop in Taiwan, but effort on the first two days after the lift is still below normal.
• Exploring spatio-temporal changes in coastal recreational fisheries and potential links to extreme weather events. (2024), PMC11161121. Eastern Australia. sourceExtreme weather can affect coastal recreational fisheries by altering angler behaviour and decisions; this study measures that decision directly as daily participation.

1. Warning windows. Sea warning issue and lift times for all 9 typhoons in the CWA typhoon database's warned-typhoon list for 2025–2026; land warning times from each storm's narrative field. Retrieved 2026-09-05.
2. Day roles. A day with ≥3 hours under a land warning between 06:00 and 18:00 is a land-warning day; otherwise ≥3 hours under a sea warning makes a sea-warning day. The day before the first warning day is the pre day; the lift day (if lifted before noon) or the next day is lift +1, then +2. 48 affected days in total.
3. Outcome. Distinct anglers logging a marine (coastal or estuary) catch per day; freshwater computed identically as control. Robustness: catch counts, EXIF-timestamp catches only.
4. Expectation. Mean over the same weekday at ±1, 2, 3, 4 weeks (eight days, excluding affected days). Ratio = sum observed / sum expected (plus 0.5 smoothing).
5. Inference. Each typhoon's affected window shifted independently by ±12–80 days to dates overlapping no affected day; the same statistic recomputed 2,000 times as the null; two-sided p; intervals exp(log ratio ± 1.96 × null SD). Bonferroni across the four primary hypotheses.
6. Hourly analysis. Warning phases and post-lift bins in angler-hours, expectation from the same hour on the same weekday ±1–4 weeks.
7. Power. Minimum detectable ratio and power at set multiples from the null SD under a normal approximation.

Read-only analysis; no production data modified. The day-level table (48 affected days, observed and expected) is available to researchers on request.

The aggregates, figures and methods here are free to read and cite; please retain the data period, number of typhoons and analysis date. Public pages are free to read and cite page by page. Raw records are not available for download, and bulk reuse licensing for aggregated data is still under review.

We are looking for collaborators. The design needs only official warning times and timestamped behaviour, so it transfers directly to coast guard rescue logs, harbour entry records, beach attendance or other angling communities' data. Anyone holding incident or attendance data could match "who fished during warnings" against "who got into trouble during warnings", which is the actual measure of risk. Write to support@dudefishing.net.