Do Fish Bite Before a Front? Barometric Pressure and 390 Days of Taiwan Catch Records

"Fish go on a feeding frenzy before a front and shut down after it" is angling lore second only to the moon. Unlike the moon, which left no trace in our data, this question got a power analysis first: we confirmed that 390 pressure-tagged days could detect roughly a 9% effect per 3 hPa of fall before deciding to write it up. There is a signal. It just does not look like the lore.

Finding: There is a signal, but it is small and it flips with the water: freshwater catch per angler rises 13% on sharp pressure-drop days, marine falls 8%. Anglers go out 7% more when pressure falls. The pre-front frenzy of lore does not exist.

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

With only 27 sharp-drop days, the first question is what effect size the data could detect. The simulation keeps observed daily effort and multiplies catchability on drop days by R, then runs the same test:

The binary design reaches 80% power at ×1.3; the continuous design (slope per hPa over all 390 days) reaches it at about ×1.09 per 3 hPa of fall. The lore implies far more than that, so the study was worth running. Hypotheses, fixed in advance:

• H1 Continuous: the more pressure falls over 24 hours, the higher catch per angler. → holds, small (+1.9% per 3 hPa)
• H2 Sharp-drop days: days with change ≤ −3 hPa beat other days. → holds in freshwater (+13%), reverses in marine (-8%)
• H3 Absolute low pressure: lower pressure means better fishing. → rejected (p = 0.58)
• H5 After the front: rise days (≥ +3) fall below other days. → right direction, not significant (-3.0%, p = 0.22)
• H0: after day-of-week and month, pressure change is unrelated to catch. → rejected, but the effect is far smaller than lore

Binning the 390 days by 24-hour change, after removing day-of-week and month effects: catch per angler falls monotonically from 1.047 on sharp drops to 0.975 on sharp rises. The gradient matches lore exactly; its whole range is seven points. Continuous slope: +0.64% per hPa of fall (p = 0.027), so a 3 hPa drop is worth about +1.9%.

Freshwater: catch per angler on drop days +13.2% (p < 0.001, n = 27), continuous +3.6% per 3 hPa (p = 0.004). Removing the 27 drop days one at a time keeps the effect between +10% and +15%; no single day carries it. Marine: catch per angler on drop days -7.8% (p = 0.034), catches -18.6%. Opposite sign.

Anglers move first. Per 3 hPa of fall, logging-angler counts rise +6.7% (p = 0.002); freshwater +7.5%, marine +5.7%. Anglers believe the lore and act on it, including at sea, where drop-day results actually get worse. This behavioural signal (7%) is larger than the fish signal (2%).

The freshwater drop-day effect holds under every check. The overall continuous effect does not survive subsampling: restrict to EXIF timestamps, or loosen the definition of a drop, and it goes non-significant. That is what a real but small effect looks like: halve the sample and it disappears. Fish length is unrelated to pressure, so this is not "bigger fish on a falling barometer", only slightly more of them.

Measured

• All waters, catch per angler: +0.64% per hPa of fall (p 0.027); five bins fall monotonically from 1.047 to 0.975.
• Freshwater drop days +13.2% (p < 0.001, jackknife +10% to +15%); marine -7.8% (p 0.034).
• Angler counts +6.7% per 3 hPa of fall (p 0.002).
• Absolute pressure and fish length: no effect.

Inferred, to be read separately

• "Marine reverses because the sea gets rough" is a reasonable inference, not a measurement; wave-height coverage is only a third of records, too thin to test.
• The mechanism can only be a weather proxy: 1 hPa ≈ 1.02 cm of water, and fish cannot sense a 3 cm equivalent depth change. Cloud, light, wind or surface disturbance may be doing the work; this study does not separate them.
• The 390 days are days people fished and logged pressure; days too stormy to fish are absent. Extreme-weather fishing cannot be answered here.
• Pressure is the value attached to each catch at logging, averaged by day, not a continuous station record; daily means erase intraday change.
• Catch per angler is conditional on at least one catch. If falling pressure turns more anglers from zero to one, the measure is pushed down, not up; the freshwater effect appears against that direction.

Two rounds of web search found no verifiable peer-reviewed study that tests barometric change against catch per unit effort; every result was an angling blog or a tackle vendor, and none is cited here. Claims that circulate online, such as a Gulf of Mexico yellowfin tuna correlation or a twelve-month experiment finding no difference, could not be traced to a primary source and are marked unverified. The peer-reviewed base for this question is far thinner than for the lunar cycle; if you know a traceable original study, please write to us.

What is certain is the physics: standard atmospheric pressure of 1013 hPa corresponds to about 10.3 m of water column, so 1 hPa ≈ 1.02 cm. Direct swim-bladder sensing of barometric change fails by orders of magnitude.

1. Pressure. Each catch carries the Central Weather Administration pressure at logging; values of 980–1035 hPa are kept (zeros and outliers dropped); daily mean; change = today minus yesterday; both days need at least three readings, leaving 390 days.
2. Outcomes. log(catches + 1), log(anglers + 1), log(catch per angler), for all waters and for freshwater and marine separately.
3. Model. Residualise on day-of-week (6) and month (20) fixed effects, then regress on the negative change (continuous) or on drop-day and rise-day indicators (binary); p is the share of 3,000 circular-shift permutations (2,000 for binary) with |β| at or above the observed value.
4. Power. Pre-registered simulation: keep observed effort, multiply drop-day catchability by R or per-hPa catchability by r, draw Poisson catches, run the same test, record the detection rate.
5. Robustness. Rain-share control, EXIF subset, −2 hPa threshold, jackknife, Bonferroni, fish length, absolute pressure.

Read-only analysis; no production data modified. Daily aggregate series for replication are available on request.

The aggregates, figures and methods here are free to read and cite; please retain the data period, sample size 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. Separating the weather proxy into its parts needs hourly station cloud, wind, light and pressure alongside catch data with trip duration. If you work at the intersection of meteorology and fishing behaviour, we can provide de-identified daily series split by water type. Write to support@dudefishing.net.