2026 Pacific Saury Catch Trajectory and Season Outlook
How the Taiwan-origin season has developed, what the remaining path requires, and the assumptions and measured evidence behind the season outlook.
Taiwan-origin catch · 2026 season
WK18 · September 27–October 3, 2026 · assessment dated October 6
Key figures Pacific Saury methodologySeason outlook
Fishing through October 31 · end-of-October return assumed
- Low endModel’s lower estimate
- 20,073MT
- −7.8%vs WK17 estimate
- Previously 21,778 MT
- The model gives a 1-in-5 chance of finishing below this level.
- Base caseOur principal season estimate
- 26,277MT
- −9.2%vs WK17 estimate
- Previously 28,948 MT
- Catch still needs to accelerate before the fleet’s assumed return.
- Upper limitPhysical capacity ceiling
- 46,354MT
- −14.9%vs WK17 estimate
- Previously 54,492 MT
- A capacity limit, with no probability assigned to this endpoint.
Season estimates rounded to the nearest MT; exact figures appear below.
- Catch this week
- 1,226.9MT
- ↑ +115.8%vs revised WK17 catch
- Season catch reported
- 15,015.2MT
- 46.7%of the historical mean at WK18
- Catch rate required
- 2.18×
- the recent four-week catch per hull-day, to reach the base case
- Premium share
- 9.27%
- ↓ −0.41percentage points vs WK17
Different fishing-end assumption; separate from the base case.
Catch to date remains well below the matched historical mean.
Continuing recent catch per hull-day places the season close to the low end of the range.
The return date materially changes the outlook; the upper end remains a physical limit.
The assessment at a glance
At week 18, reported season catch stands at 15,015.2 MT. Our base case is 26,277.4 MT, with a low end of 20,072.6 MT and an upper limit of 46,354.4 MT. The low end means a one-in-five chance of finishing below it; the upper limit requires the ceiling explanation carried below.
We assume the fleet begins returning by the end of October. The base case ends fishing in week 22, October 25–31, and includes no catch after October 31. This is BX-TRADE’s field assessment. The sensitivity follows the model’s own end-of-season timing and gives 36,841.2 MT.
This assessment is dated October 6, 2026 and uses the WK18 scientific snapshot. It is a fixed edition. Later weekly publications provide new observations without silently replacing the figures or comparisons in this assessment.
E01 2026 is running at 46.7% of the panel mean. The fan says it does not catch up.
Begin with the catch already reported, then examine the remaining path. The return assumption is part of the estimate, rather than an observation that the fleet has already returned.
WK18 · September 27–October 3, 2026 · assessment dated October 6
The season so far
The historical comparison is 46.7% of the matched panel’s 32,124 MT mean through WK18. Seasons are aligned by fishing-season week so that the same stage of the season is compared. Differences in final totals do not show how much a season had already caught at week 18.
The comparison uses the reporting series identified by the exhibit. Annual catch statistics and the sum of weekly in-season reports are different source series and can differ. A longer annual chart offers context; it is not a substitute for the issue’s matched weekly panel.
The latest issued outlook changed by a reduction of 2,670.2 MT (9.2%) from 28,947.6 MT, for the recorded reasons the new catch report and updated source vintage. Reporting vessels fell from 80 to 77. Separately, the capacity limit fell 8,138 MT: +1,223 MT from the cumulative advance, −1,526 MT from the lower reporting count, and −7,835 MT from fewer calendar days. These are capacity arithmetic, not a causal decomposition of the base-case revision. The explanation separates revised source data, the arrival of a new catch report and a change in the assessment. A revised observation does not rewrite the record of an earlier forecast.
E15 Catch since 2014: down by roughly three quarters. Fleet down by roughly a sixth: 91 vessels to 75.
Check the series basis, week and issue vintage. Historical weekly lines remain visible as approved plates, without exporting protected point values to the browser.
WK18 · September 27–October 3, 2026 · assessment dated October 6
Fleet participation and catch rate
At the reporting stage, 77 vessels supplied catch reports and 80 roster hulls were observed in the fishing box. The complete position window is September 27–October 2, captured October 6; six of seven WK18 days. The complete weekly movement series ends at WK17, September 26. Reporting vessels and observed hulls answer different questions and should not be treated as the same measure.
The latest catch per hull-day is 2.36 MT. The published accounting for the weekly move is a 116% rise against the revised WK17 catch. Catch per hull-day accounts for 102% of the log change, hulls present 2%, and days per hull −3% (rounded). WK18 hull-days are estimated at 519.2, within 514–522, from six of seven days; October 3 is missing. Separating hulls present, days per hull and catch per hull-day helps show whether a change in volume was accompanied by a change in participation, time observed or catch rate.
A hull-day is a unit of observed vessel presence. It does not prove that the vessel caught fish throughout that day. The accounting is therefore an explanation of the observed totals, rather than an estimate of fish abundance or a causal test of why catch changed.
E41 Catch rose 116% in week 18. Catch per hull-day did the moving.
Read data completeness beside the rate. A rate based on partial position coverage needs the corresponding limitation even when the weekly catch report itself is complete.
WK18 · September 27–October 3, 2026 · assessment dated October 6
What the remaining path requires
The base case requires a remaining catch rate of 2.18 times the trailing four-week catch per hull-day, under the measured effort path to week 22, compared with the registered recent-rate window. The historical comparison is two of six seasons (2019–2024) achieved the required catch-rate multiple after WK18. The 2024 rate was cut short by its quota closure. Two of six is a historical comparison, not a probability of success. These are consistency checks on the season estimate, not a forecast-error score.
Continuing at the trailing four-week catch per hull-day gives 20,192 MT. That yardstick asks what the season would total if the measured recent pace continued under the stated fleet and end-date assumptions. It should be read beside the model estimate, with both sets of assumptions visible.
The remaining-week path uses the season estimate’s published allocation. It does not independently qualify a prediction for each week. A change in fleet return timing can change the total and the time available to catch it; the base case and timing sensitivity keep that distinction explicit.
E40 The point asks catch per hull-day to run at x2.18 this month's; 2 of 6 past seasons did after week 18.
Read the comparison window, effort basis and remaining fishing time. A larger required multiple states how demanding the path is; it does not give the probability of achieving it.
WK18 · September 27–October 3, 2026 · assessment dated October 6
How the season estimate is constructed
The framework combines two questions: how far through the season the fleet appears to be, and what its weekly catch pace implies for the remaining season. The prior arm combines the completion ratio (CR) and FRA survey relationship; the likelihood arm integrates the weekly catch profile. Both arms contribute to the published season estimate. SPAN and time-to-fill describe timing and do not add catch tonnage.
Other methods can describe timing, environmental conditions or fishing-ground rankings without contributing catch tonnage to the estimate. Their presence in the evidence library does not mean they have passed the criterion for a published catch forecast.
Weather is presented separately as observed conditions and provider forecasts as issued. No weather adjustment is fed into the catch figures in this edition. Registered alternatives, including failed improvements, are reported at the level needed to understand the framework’s limits.
How to read the indicative range
- Native upper-end disclosure
- THE TOP OF THIS BAND IS A PHYSICAL LIMIT, NOT A FORECAST. Under the censored treatment the upper bound IS the binding ceiling, the fleet's capacity: the catch landed plus every vessel fishing at the panel's maximum sustained daily rate on every day left to the last fishing day, so the bound carries no probability content and no model uncertainty. Only the lower bound is a model quantile. P(fill) is the quantity that says how likely that limit is to be reached. This week that limit is 46,354.4 MT, the fleet's capacity to 31 October (week 22, the base case's last fishing week); the allocation, 59,148 MT, sits 12,793.6 MT above it and cannot be reached in the base case.
- Measured misses and applicable scope
- ONE-SIDED. The upper bound is the fleet's capacity to 31 October (week 22, the base case's last fishing week), which the base case cannot exceed, so this interval cannot miss high. The coverage record measured the band with the allocation at its top: on the 29 backfilled fold-anchors it missed 4 times LOW and 0 times high. No anchor has been measured with capacity at the top.
- Native width and treatment criterion
- this week's band is 100.0% of point, with capacity at its top. The measured mean width, 91.5% of point across the 29 anchors against 60.9% for the truncated alternative it beat, is the allocation-topped band's. Criterion 4 measures width and was NOT consulted: the pre-registration ranked coverage above width before any number existed, and criterion 2 decided.
- Change in width through the season
- THE WIDTH IS THE BINDING CEILING MINUS THE LOWER BOUND. In the base case the ceiling is the fleet's capacity to 31 October (week 22, the base case's last fishing week), which moves each week with the catch landed and the days left; this week it is 46,354.4 MT against a lower bound of 20,072.6 MT, 26,281.8 MT apart.
- Band collapse condition
- WHEN THE POSTERIOR MEDIAN SITS ABOVE THE BINDING CEILING, THIS BAND COLLAPSES. The lower bound is the 20th percentile and the upper is the fleet's capacity to 31 October (week 22, the base case's last fishing week), so if the median passes capacity the two meet. This week the median is 25,355.8 MT, 20,998.6 MT below capacity. No floor is applied here; the width is published so the condition is visible in the object.
E06 The prior and the weekly model are 3.6% apart. That is a gap between two numbers, not an error bar.
Distinguish a model quantile from a physical or regulatory limit. Validation of an allocation-topped band does not automatically validate a live capacity-topped endpoint.
WK18 · September 27–October 3, 2026 · assessment dated October 6
Size composition and biological context
The season-to-date premium share is 9.27%, changing by −0.41 percentage points in the latest report. Premium means commercial grade no. 2 and above, at least 100 g. The current grade-call status is the forward-call register contains four calls, one resolved and three open, with none void. The premium-share call resolves at season end. A single resolved call cannot establish accuracy; the register covers several kinds of call, not only grades.
Size gives useful information about the reported catch mix, but a commercial grade is not a direct age measurement. Fish of the same age can differ in weight, and survey size observations concern the survey’s place and date. Those boundaries matter when relating the reported mix to cohort development.
FRA’s pre-season survey and JAFIC’s in-season outlook provide biological and geographical context. Japanese survey or coastal-ground expectations should not be presented as a direct estimate of the Taiwanese high-seas fleet’s catch. The issue keeps those published expectations separate from its own recorded grade calls.
E03 The premium tier is 9.27% of the season so far. Weekly shares swing; the season figure counts.
Read grade shares as commercial composition. Any age or growth explanation must retain the source’s size definition, population, geography and date.
WK18 · September 27–October 3, 2026 · assessment dated October 6
Grounds, conditions and regulation
The observed fleet movement and spread are a net northeast movement across 17 complete weeks, with a median half-effort radius of 39 nautical miles. The six complete weeks from WK12 to WK17 show 208 nautical miles of net eastward movement; that is not a new WK18 displacement, over September 27–October 2, captured October 6; six of seven WK18 days. The complete weekly movement series ends at WK17, September 26. Water conditions are 96% of observed hull-days in the 14.8–19.7°C band; water at the fleet cells was 3.1°C above normal. The other-fleet comparison ends September 26. Copernicus Marine Service’s own forecast is valid October 13, as captured October 6. The ground-ranking call and its recorded performance are the October 6 reference, trained through WK17 using the September 27–October 2 observation window. At issue time, HGB-L captures 29.8% against 27.9% for the prior footprint and wins one of three held-out seasons. It remains unqualified; the full-week retrospective comparison is not information available at issue time. These distinguish observed activity, surrounding conditions and a forward ranking.
The weather outlook is the October 6 00 UTC cycle from both issuers, through October 13 (+168 hours). The highest significant wave height at each issuer’s native grid point nearest the fleet is NOAA 6.97 m and ECMWF 5.64 m, cited as issued. NOAA and ECMWF retain responsibility for the forecasts as issued; RSMC Tokyo remains the issuer of any cited cyclone advisory. A forecast of sea temperature from the E.U. Copernicus Marine Service is that provider’s forecast, with its own issue and valid dates.
The allocation reading is 25.4% when Taiwan-origin reported season catch is divided by the 59,148 MT allocation. This arithmetic is not established as Convention Area uptake; the capacity limit, 46,354.4 MT, is lower than the allocation. Taiwan’s allocation and the NPFC Convention Area tracker have different accounting purposes. A catch total whose geographic basis is not established should not be relabelled as Convention Area catch or subtracted from the area tracker.
Fleet behavior, ocean conditions and management restrictions can constrain what the season can produce. The evidence describes those constraints without converting an association into a causal claim or presenting unadopted adjustments as part of the accepted catch estimate.
Cyclone advisories as issued
RSMC Tokyo - Typhoon Center, Japan Meteorological Agency · RSMC Tokyo tropical cyclone advisory WTPQ50 RJTD 060600
Captured: 2026-10-06T08:25:17.439012+00:00
RSMC Tokyo - Typhoon Center, Japan Meteorological AgencyWTPQ50 RJTD 060600
RSMC TROPICAL CYCLONE ADVISORY
NAME TY 2627 CHOI-WAN (2627)
ANALYSIS
PSTN 060600UTC 31.4N 146.3E GOOD
MOVE NNE 16KT
PRES 960HPA
MXWD 075KT
GUST 105KT
50KT 150NM EAST 90NM WEST
30KT 450NM EAST 300NM WEST
FORECAST
12HF 061800UTC 36.3N 149.3E 25NM 70%
MOVE NNE 27KT
PRES 965HPA
MXWD 070KT
GUST 100KT
24HF 070600UTC 42.4N 156.5E 42NM 70% EXTRATROPICAL LOW =
Please note that information issued by the RSMC Tokyo - Typhoon Center represents neither official analysis/forecasts nor warnings for the areas concerned. Such official information is issued by the National Meteorological Services of individual countries.
RSMC Tokyo - Typhoon Center, Japan Meteorological Agency · RSMC Tokyo tropical cyclone advisory WTPQ52 RJTD 060600
Captured: 2026-10-06T08:25:17.439012+00:00
RSMC Tokyo - Typhoon Center, Japan Meteorological AgencyWTPQ52 RJTD 060600
RSMC TROPICAL CYCLONE ADVISORY
NAME TY 2628 NOLO (2628)
ANALYSIS
PSTN 060600UTC 25.6N 171.2E GOOD
MOVE W 17KT
PRES 965HPA
MXWD 085KT
GUST 120KT
50KT 60NM
30KT 210NM NORTH 150NM SOUTH
FORECAST
12HF 061800UTC 25.5N 166.6E 30NM 70%
MOVE W 21KT
PRES 975HPA
MXWD 075KT
GUST 105KT
24HF 070600UTC 26.1N 161.9E 42NM 70%
MOVE W 21KT
PRES 980HPA
MXWD 070KT
GUST 100KT
48HF 080600UTC 27.8N 154.6E 85NM 70%
MOVE WNW 17KT
PRES 985HPA
MXWD 060KT
GUST 085KT
72HF 090600UTC 30.5N 150.8E 140NM 70%
MOVE NW 11KT
PRES 985HPA
MXWD 060KT
GUST 085KT
96HF 100600UTC 32.9N 151.6E 200NM 70%
MOVE NNE 06KT
PRES 990HPA
MXWD 055KT
GUST 080KT
120HF 110600UTC 34.4N 153.8E 250NM 70% EXTRATROPICAL LOW =
Please note that information issued by the RSMC Tokyo - Typhoon Center represents neither official analysis/forecasts nor warnings for the areas concerned. Such official information is issued by the National Meteorological Services of individual countries.
Central Weather Administration · CWA 颱風消息 民國115年10月06日14時
Captured: 2026-10-06T08:25:17.439012+00:00
Central Weather Administration中度颱風彩雲編號第 26 號國際命名 CHOI-WAN 6日14時的中心位置在北緯 31.7 度，東經 146.4 度，以每小時45公里速度，向北北東進行。中心氣壓965百帕，近中心最大風速每秒35公尺，瞬間最大陣風每秒 45 公尺，七級風平均暴風半徑 250 公里(西北側 260 公里、東北側 280 公里、西南側 220 公里、東南側 250 公里)，十級風平均暴風半徑 80 公里(西北側 80 公里、東北側 80 公里、西南側 80 公里、東南側 80 公里)。Central Weather Administration · CWA 颱風消息 民國115年10月06日14時
Captured: 2026-10-06T08:25:17.439012+00:00
Central Weather Administration6日14時的中心位置在北緯 25.6 度，東經 171.1 度，以每小時35公里速度，向西進行。中心氣壓945百帕，近中心最大風速每秒45公尺，瞬間最大陣風每秒 55 公尺，七級風平均暴風半徑 220 公里(西北側 220 公里、東北側 250 公里、西南側 180 公里、東南側 220 公里)，十級風平均暴風半徑 100 公里(西北側 100 公里、東北側 100 公里、西南側 90 公里、東南側 100 公里)。E04 The fleet has moved north-east across 17 weeks of 2026; median half-effort radius 39 nm.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
WK18 · September 27–October 3, 2026 · assessment dated October 6
Reliability and what would change the assessment
The current forecasting record is season-total held-out log RMSE 0.2013 on anchors 15–19 in 2022–2024, with each season’s actual final fishing week supplied. That end-date protocol does not validate the 2026 end-of-October field assumption. The dated one-week record is 13 calls, log RMSE 0.9545 and bias -0.2536 (underprediction), versus baseline 0.9716; its typical error is a factor of 2.6. The capability statement is v0.2-ungated: no qualified next-week catch point or forward ground forecast. The version register shows 47 attempts, six earned and zero of two version gates met; the separate test register shows 36 tests, two passed. Inside the weekly component, 32.1% of its raw projected remainder depends on the unqualified persistence forecaster; this is not the share of the published season remainder. A retrospective test and a dated forward call answer different questions: one tests held-out seasons under a fixed protocol; the other records what was said before the outcome became known.
Short samples, measured misses and methods that do not beat their baseline remain visible. A more detailed map or a narrower-looking band does not itself establish a more accurate forecast. The relevant baseline and uncertainty interpretation travel with every result.
The next assessment will examine whether catch per hull-day recovers, whether complete-day fleet participation and movement change, and whether the cumulative premium share moves in the direction anticipated by the registered calls. The end-of-October return assumption remains explicit throughout those checks.
The companion WK18 brief focuses on the latest observations and the week ahead. This assessment supplies the broader season explanation; both retain their own dated disclosures, sources and notices.
E19p 13 calls: season RMSE 0.955 log. Condition 1 is to beat the baseline, 0.972; it is below it.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
WK18 · September 27–October 3, 2026 · assessment dated October 6
Supporting evidence
Exact weekly catch is shown only for WK18 and WK17. Earlier catch history is presented through approved charts and permitted aggregates.
E012026 is running at 46.7% of the panel mean. The fan says it does not catch up.
E01 2026 is running at 46.7% of the panel mean. The fan says it does not catch up.
Begin with the catch already reported, then examine the remaining path. The return assumption is part of the estimate, rather than an observation that the fleet has already returned.
Check the series basis, week and issue vintage. Historical weekly lines remain visible as approved plates, without exporting protected point values to the browser.
E41Catch rose 116% in week 18. Catch per hull-day did the moving.
E41 Catch rose 116% in week 18. Catch per hull-day did the moving.
Read data completeness beside the rate. A rate based on partial position coverage needs the corresponding limitation even when the weekly catch report itself is complete.
E40The point asks catch per hull-day to run at x2.18 this month's; 2 of 6 past seasons did after week 18.
E40 The point asks catch per hull-day to run at x2.18 this month's; 2 of 6 past seasons did after week 18.
Read the comparison window, effort basis and remaining fishing time. A larger required multiple states how demanding the path is; it does not give the probability of achieving it.
E42Week 18: 80 roster hulls were in the box in 2026. In 2019–2023 the same week held 49 to 78.
E42 Week 18: 80 roster hulls were in the box in 2026. In 2019–2023 the same week held 49 to 78.
Read data completeness beside the rate. A rate based on partial position coverage needs the corresponding limitation even when the weekly catch report itself is complete.
E15Catch since 2014: down by roughly three quarters. Fleet down by roughly a sixth: 91 vessels to 75.
E15 Catch since 2014: down by roughly three quarters. Fleet down by roughly a sixth: 91 vessels to 75.
Check the series basis, week and issue vintage. Historical weekly lines remain visible as approved plates, without exporting protected point values to the browser.
E03The premium tier is 9.27% of the season so far. Weekly shares swing; the season figure counts.
E03 The premium tier is 9.27% of the season so far. Weekly shares swing; the season figure counts.
Read grade shares as commercial composition. Any age or growth explanation must retain the source’s size definition, population, geography and date.
E27The commercial convention puts age-1 at No.2 · 2號. FRA's 2026 L50: No.4 · 4號, two grades down.
E27 The commercial convention puts age-1 at No.2 · 2號. FRA's 2026 L50: No.4 · 4號, two grades down.
Read grade shares as commercial composition. Any age or growth explanation must retain the source’s size definition, population, geography and date.
E04The fleet has moved north-east across 17 weeks of 2026; median half-effort radius 39 nm.
E04 The fleet has moved north-east across 17 weeks of 2026; median half-effort radius 39 nm.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E04b6 weeks of the fleet: 208 nm east from week 12 to week 17; 93% of hull-days in the band.
E04b 6 weeks of the fleet: 208 nm east from week 12 to week 17; 93% of hull-days in the band.
E16Week 19: within 163 nm of week 17 half the time, 437 nm 95% of the time. No direction forecast.
E16 Week 19: within 163 nm of week 17 half the time, 437 nm 95% of the time. No direction forecast.
E19Persistence's 80% ring for week 17: 290 nm. The fleet went 147 nm, inside its 50% ring.
E19 Persistence's 80% ring for week 17: 290 nm. The fleet went 147 nm, inside its 50% ring.
E38Most weeks, the fleet moves further than it spreads. Median move 94 nm; median half-effort radius 25 nm.
E38 Most weeks, the fleet moves further than it spreads. Median move 94 nm; median half-effort radius 25 nm.
E39The fleet is where it was: off by 184 nm a week ahead, the baseline. No better one is established.
E39 The fleet is where it was: off by 184 nm a week ahead, the baseline. No better one is established.
E5296% of our hull-days sat in the 14.8–19.7 °C band; its cold edge held at 171°E.
E52 96% of our hull-days sat in the 14.8–19.7 °C band; its cold edge held at 171°E.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E53The water at our cells was 3.1 °C warmer than normal this week.
E53 The water at our cells was 3.1 °C warmer than normal this week.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E05Our fleet sat 2.3° south of the band's cold edge; the 100 m 5 °C isotherm lay 1.6° north of its mean.
E05 Our fleet sat 2.3° south of the band's cold edge; the 100 m 5 °C isotherm lay 1.6° north of its mean.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E54Other fleets entered cells near ours 20 times in three nights to 09-26, against 7 a week earlier.
E54 Other fleets entered cells near ours 20 times in three nights to 09-26, against 7 a week earlier.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E55Copernicus Marine Service's forecast of sea temperature for 13 October 2026.
E55 Copernicus Marine Service's forecast of sea temperature for 13 October 2026.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E51The reference has not beaten last week's footprint on the information an issue actually has.
E51 The reference has not beaten last week's footprint on the information an issue actually has.
E50Full-week upper bound: HGB 38.8% vs 34.4%, 3 of 3. At issue time: HGB-L 29.8% vs 27.9%, 1 of 3.
E50 Full-week upper bound: HGB 38.8% vs 34.4%, 3 of 3. At issue time: HGB-L 29.8% vs 27.9%, 1 of 3.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
E06The prior and the weekly model are 3.6% apart. That is a gap between two numbers, not an error bar.
E06 The prior and the weekly model are 3.6% apart. That is a gap between two numbers, not an error bar.
Distinguish a model quantile from a physical or regulatory limit. Validation of an allocation-topped band does not automatically validate a live capacity-topped endpoint.
E242% of the pre-ceiling posterior sits above the fleet's capacity to October's end.
E24 2% of the pre-ceiling posterior sits above the fleet's capacity to October's end.
Distinguish a model quantile from a physical or regulatory limit. Validation of an allocation-topped band does not automatically validate a live capacity-topped endpoint.
E336.0 of 6 days observed were fishable at the fleet. Forecast maxima, 7 days: NOAA 6.97 m, ECMWF 5.64 m.
E33 6.0 of 6 days observed were fishable at the fleet. Forecast maxima, 7 days: NOAA 6.97 m, ECMWF 5.64 m.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E45Week 19: 2,944 MT at the band's point, 1,322 to 8,194 MT across it. Not a forecast.
E45 Week 19: 2,944 MT at the band's point, 1,322 to 8,194 MT across it. Not a forecast.
Read the comparison window, effort basis and remaining fishing time. A larger required multiple states how demanding the path is; it does not give the probability of achieving it.
E19p13 calls: season RMSE 0.955 log. Condition 1 is to beat the baseline, 0.972; it is below it.
E19p 13 calls: season RMSE 0.955 log. Condition 1 is to beat the baseline, 0.972; it is below it.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
E4934% of weeks carry 89% of the error. Without them, last week's rate is off by 0.399 log.
E49 34% of weeks carry 89% of the error. Without them, last week's rate is off by 0.399 log.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
E08The allocation does not clip the published interval: the bull path ends at 46,354 MT of 59,148.
E08 The allocation does not clip the published interval: the bull path ends at 46,354 MT of 59,148.
Each exhibit carries its own dates and coverage. Keep the fleet’s observation time, the forecast issue cycle and the forecast valid time separate.
E354 calls made. 1 resolved. Here is every one.
E35 4 calls made. 1 resolved. Here is every one.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
E3647 attempted, 6 earned; 0 of 2 version gates met. Each row sits under the gate it serves.
E36 47 attempted, 6 earned; 0 of 2 version gates met. Each row sits under the gate it serves.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
E4336 registered tests. 2 passed. 9 of 31 refusals have a registered next route.
E43 36 registered tests. 2 passed. 9 of 31 refusals have a registered next route.
Check the series basis, week and issue vintage. Historical weekly lines remain visible as approved plates, without exporting protected point values to the browser.
Read what each test actually measures and the number of observations available. An unqualified next-week reference or ground ranking remains labelled with its role and measured record.
E26The band narrows because less season remains — and ends above the likelihood floor, 0.0711.
E26 The band narrows because less season remains — and ends above the likelihood floor, 0.0711.
Distinguish a model quantile from a physical or regulatory limit. Validation of an allocation-topped band does not automatically validate a live capacity-topped endpoint.
Sources and coverage
Catch and commercial grades: TSSFA weekly vessel reporting. The issue distinguishes a corrected source figure from a new week of catch; all final values follow the approved issue vintage.
Fleet observations: GFW. Complete observation days, the capture date and the serving delay are stated beside the relevant exhibits. Vessel presence describes reported activity and does not measure fish abundance.
Ocean conditions: NOAA and the E.U. Copernicus Marine Service. Forecast sea temperatures remain the provider’s forecast as issued, with their own cycle and valid date.
Sea state: NOAA/NCEP GFS-Wave and ECMWF open-data forecasts as issued; cyclone advisories from RSMC Tokyo and Taiwan’s Central Weather Administration. The issue cites the October 6 00 UTC cycle on both wave sources. No substitute cycle or derived forecast of fishable days is presented.
Survey and published outlook context: FRA and JAFIC. Their findings concern the surveys, fleets and grounds they describe; they are not a direct forecast of Taiwan-origin catch.
Management context: NPFC publications and Taiwan’s competent authority. Season catch and Convention Area uptake are kept distinct where their geographic bases differ.
ECMWF attribution as supplied with the data
www.ecmwf.int
This output is based on data and products of the European Centre for Medium-Range Weather Forecasts (ECMWF).
© 2026 European Centre for Medium-Range Weather Forecasts (ECMWF)
ECMWF does not accept any liability whatsoever for any error or omission in the data, their availability, or for any loss or damage arising from their use.
Not modified: each value is the one ECMWF issued at its native grid point nearest the fleet, at ECMWF's own time steps.
This data is published under a Creative Commons Attribution 4.0 International (CC BY 4.0). https://creativecommons.org/licenses/by/4.0/
Notices and terms of use
The English notices govern. The Chinese text is the approved summary.
Important Notices, Licensing & Terms of Use
This publication is prepared by Global Trade Business Group (BX-TRADE) Commodity Research & Strategy, part of Baixian (百鮮). It is issued for general distribution and is not offered for sale or by subscription. Its distribution grants no licence to, and discloses none of, the proprietary models, methods, code, parameters, data and records that produce it.
Important Disclosures
Nature of this document. This research document is a commodity research publication. It is provided for informational purposes only and does not constitute investment advice, a solicitation, an offer, or a recommendation to buy, sell, or hold any commodity, financial instrument, or derivative position. This document is not a securities research report within the meaning of applicable securities regulations in any jurisdiction.
No reliance. The information, opinions, forecasts, and reference figures contained in this document are based on historical data spanning the 2019–2025 observation period (seven complete Pacific saury TW-origin seasons) and proprietary statistical models. While BX-TRADE has exercised reasonable care in preparing this document, it makes no representation or warranty, express or implied, as to the accuracy, completeness, reliability, or timeliness of any information herein. Readers should not rely on this document as the sole basis for any commercial or investment decision and should conduct their own independent analysis and due diligence. BX-TRADE accepts no obligation to update, revise, or correct this document following its publication.
Forward-looking statements. This document contains forward-looking statements, including catch projections, reference figures, probability estimates, and season-level assessments. These statements are inherently subject to significant uncertainties and contingencies, many of which are beyond the control of BX-TRADE. Actual results may differ materially from any projections or estimates contained herein due to, among other factors, changes in oceanographic conditions, fleet behavior, regulatory interventions, weather events, market conditions, and biological variability. Forward-looking statements speak only as of the date of this document. Past performance of the fishery, individual seasons, or model forecasts is not a reliable indicator of future outcomes.
Limitation of liability. To the fullest extent permitted by applicable law, BX-TRADE, Baixian, and their respective directors, officers, employees, agents, and affiliates shall not be liable for any loss, damage, cost, or expense (whether direct, indirect, incidental, consequential, special, or punitive) arising from or in connection with any person's use of, reliance on, or inability to use this document or any information contained herein, regardless of whether such liability arises in contract, tort (including negligence), strict liability, or any other legal theory, and regardless of whether BX-TRADE has been advised of the possibility of such loss or damage.
Regulatory Notices
Taiwan. BX-TRADE is not a licensed securities investment trust enterprise, securities investment consulting enterprise, or futures advisory enterprise under the Securities and Exchange Act, the Securities Investment Trust and Consulting Act, or the Futures Trading Act. This document does not constitute investment analysis, investment advice, or a recommendation within the meaning of regulations promulgated by the Financial Supervisory Commission. Readers in Taiwan should be aware that this document is provided strictly as commodity market research and should not be construed as advice to engage in any securities or futures transaction.
Mainland China. This document is not intended for distribution to, or use by, any person or entity in Mainland China where such distribution or use would be contrary to applicable law or regulation. BX-TRADE is not registered with or licensed by the China Securities Regulatory Commission or any other regulatory authority in Mainland China as a securities or futures advisory service provider. This document does not constitute investment advice, securities research, or futures advisory services within the meaning of the Securities Law, the Futures and Derivatives Law, or measures promulgated thereunder. Readers in Mainland China should not treat this document as a basis for any investment or trading decision.
General. Distribution of this document in other jurisdictions may be restricted by law. Persons who come into possession of this document should inform themselves of and observe all applicable legal and regulatory restrictions. BX-TRADE accepts no liability for any breach of such restrictions by any person.
Permitted Use
Forwarding. This publication may be forwarded to others only in its complete and unaltered form, including these notices.
No reproduction. No part of this publication, including its text, tables, figures and exhibits, may be reproduced, republished, excerpted, adapted, translated, distributed in part, or incorporated into any other document, presentation, database, product or service, by any means, without the prior written permission of Baixian.
No derivative modelling or data mining. This publication may not be used to reconstruct, reverse-engineer, replicate or calibrate the models, methods or parameters that produce it, or for text and data mining, or to train, develop or validate any model or automated system, without the prior written permission of Baixian.
Citation. Any reference to this publication must identify it as follows: Baixian (2026). 2026 Pacific Saury Catch Trajectory and Season Outlook, as of week 18, October 6, 2026.
What Is Protected
Copyright. © 2026 Baixian. All rights reserved. The text, tables, exhibits, design and compilation of this publication are protected by the Copyright Act of Taiwan and the Berne Convention for the Protection of Literary and Artistic Works. The Baixian name and logo identify the source of this publication.
Proprietary methodology. This publication describes the forecasting framework at the level needed to read and check its results: the model families, the validation protocols and the measured record, including what the models did not achieve. It does not disclose the source code, parameter values, feature sets, calibration windows, threshold derivations, training selections or internal records that implement the framework. Those elements are kept confidential under access controls and are protected as trade secrets under the Trade Secrets Act of Taiwan and equivalent laws elsewhere. This publication grants no licence to them.
Enforcement. Baixian reserves all rights and remedies under the Copyright Act, the Trade Secrets Act, the Fair Trade Act and other applicable laws in Taiwan, and under the applicable laws of Mainland China and other jurisdictions.
Third-Party Data and Licences
Each third-party dataset used in this publication remains subject to its own terms. Those terms govern the third-party data only; they do not extend to, and grant no rights in, BX-TRADE's text, analysis, models, exhibits or compilation, which remain subject to the reservation of rights above. BX-TRADE does not guarantee the accuracy or completeness of data obtained from external sources, and the use of third-party data does not imply endorsement by the data providers of BX-TRADE's analysis, methodology or conclusions.
Global Fishing Watch (GFW). Vessel presence and fishing-effort data: Copyright 2026, Global Fishing Watch, Inc., https://globalfishingwatch.org/our-apis/. Licensed under the Creative Commons Attribution-NonCommercial 4.0 International licence (CC BY-NC 4.0), https://creativecommons.org/licenses/by-nc/4.0/. BX-TRADE has aggregated and transformed these data into hull-day counts, weekly fleet positions, effort radii, position-forecast errors and capture scores; the exhibits built on them are marked "Powered by Global Fishing Watch". GFW provides its data as is, without warranty, and does not endorse this analysis. The licence applies to GFW's data, which are available from GFW under its terms; it grants no rights in BX-TRADE's analysis, exhibits or compilation.
Use of GFW data. GFW data are used solely in preparing this research publication. They are not sold, licensed or supplied to any person as a product or service, and the underlying GFW data are not redistributed.
Other sources.
Catch, vessel and grade data: TSSFA (台灣區遠洋魷魚暨秋刀魚公會), weekly vessel reporting, 2019–2026. Exact weekly catch is printed for the latest two weeks only; earlier weeks appear as multi-week or season totals.
Historical catch series: TSSFA public annual statistics, ROC years 66–114 (1977–2025).
Pre-season survey: Japan Fisheries Research and Education Agency (FRA), サンマ長期漁海況予報 (annual, late July), 2018–2026 editions. Facts are stated in our own words; no FRA text, table or chart is reproduced.
In-season outlook: Japan Fisheries Information Service Center (JAFIC), サンマ中短期漁況予報, 2026 editions, stated in our own words.
Quota and management measures: North Pacific Fisheries Commission (NPFC), conservation and management measures, public meeting documents, cumulative-catch tracker and vessel registry; Ministry of Agriculture, Republic of China (Taiwan).
Ocean fields: Copernicus Marine Service (E.U. Copernicus Marine Service Information); NOAA (OISST and wave fields, public domain).
Forecast sea temperatures in Exhibit E55 are the E.U. Copernicus Marine Service's own model forecast, shown as issued on its native grid without processing, with the issuer and issue time stated. They are not a BX-TRADE forecast, and BX-TRADE makes no representation as to their accuracy.
Coastline: Natural Earth (public domain).
Length–weight and age-at-length: Suyama et al. (2015); Huang (2007); Ginting & Huang (2023). Further literature is cited where it is used.
Contact
For questions regarding this document or permissions, please contact: Global Trade Business Group (BX-TRADE), Commodity Research & Strategy Division (國際貿易事業群 大宗商品研究與策略分析組 BX-TRADE-CRS) · Email: crs@baixiangroup.com · Tel: +886 7 536 8337
Baixian (2026). 2026 Pacific Saury Catch Trajectory and Season Outlook, as of week 18, October 6, 2026.