What drives the catch

Two views per river: a horizontal bar of feature importance (what the model explains the variance with) and a line chart of the next 30 days (expected catch rate, with the top 5 dates highlighted in emerald). R² values are reported honestly — most are slightly negative, which means catch rate is mostly noise at the fishery-day grain. Treat the bars as a directional signal, the dates as a "where to look first" hint, not a tip.

Top 5 rivers by total catches: Ätran (4137), Mörrumsån (2110), Byskeälven (1270), Kalixälven (328), Lögdeälven (184)

• WhatWhen does each river peak across the season? Surfaces the seasonal timing pattern per river, year by year.
• DataTop 5 rivers by total catches (Ätran, Mörrumsån, Byskeälven, Kalixälven, Lögdeälven). One row per year, columns are ISO weeks 14–39 — the salmon season window.
• MethodPer (river × year × week), sum reported catches. Heatmap intensity scales against the per-river maximum so each river is comparable to itself across years.

Rows with temp + catch_rate present: 3,194 of 3,363

• WhatDoes air temperature affect catch rate? Tests the classic "13–16 °C is optimal" finding from Severn studies on Swedish rivers.
• Data3,175 fishery-days with both temperature and catch_rate. Buckets: cold / cool / optimal (13–16 °C) / warm / hot.
• MethodBucket each day by daily mean air temperature, then compare catch_rate distributions across buckets with Kruskal-Wallis (non-parametric — catch_rate is a heavily skewed 0–1 ratio).

Rows with flow_regime + catch_rate present: 1,110

• WhatDoes the amount of water moving through a river predict catches? Tests the "sweet spot" hypothesis around normal flow.
• Data1,092 fishery-days with paired flow regime + catch_rate. Each day classified by per-river flow percentiles (drought / low / normal / high / flood).
• MethodPer-river flow percentiles avoid one big river dominating the buckets. Kruskal-Wallis tests whether catch_rate distributions differ across regimes.

Rows with Δp_24h + catch_rate: 3,130

• WhatThe folklore: pressure dropping = fish are biting. Test against real data per river.
• Data3,111 fishery-days with 24-hour pressure change and catch_rate. Top 3 rivers shown individually plus global summary.
• MethodPearson r between Δpressure_24h and catch_rate. 95% CI via Fisher z-transform. Each river tested independently — folklore could be local.

• WhatDo fly and spin anglers catch in different conditions? Tests whether method choice and weather are independent.
• Data4,998 reported catches with both method and temperature recorded (fly: 4,005 / spin: 993). Fly is the dominant method.
• MethodChi-square test of independence on the (method × temperature bucket) contingency table. Cramér's V scales the effect size.

• WhatWhere in the calendar does each river peak? Multi-year overlay shows whether peak timing is stable or drifts.
• DataAll catches 2021–2026 for top 3 rivers (Ätran, Mörrumsån, Byskeälven). Each year drawn as a separate line.
• MethodSum catches per day-of-year, smooth with a 7-day rolling mean to suppress noise. Display weeks 60–300 of the year (early March to late October) so off-season zeros don't flatten the curve.

Aggregated (river, day) rows: 1,191

• WhatWhen a fish is counted at the migration camera, is it caught upstream the same day?
• Data1,187 (river, day) pairs across Ätran and Byskeälven — the two rivers where camera and fishery data overlap.
• MethodSpearman ρ (non-parametric, robust to outlier spike days) between same-day migration_count and caught_count totals per river.

Bars: green = positive ρ, red = negative. Annotation: *** p<0.001, ** p<0.01, * p<0.05, · n.s. Significance after BH-FDR correction is in findings.json.

• WhatWhich environmental signals correlate with migration count? Ranks candidate drivers per river.
• DataPer-river daily metrics × 12 candidate drivers (temperatures, flows, pressure, humidity, wind, clouds, precipitation, season position, moon phase). ~370–800 days per river.
• MethodSpearman ρ between each driver and migration_count, computed independently per river. Bars sorted by |ρ| so the strongest signal sits at the top.

• WhatAfter a fish passes the camera, how many days before it reaches the fishery upstream? Cross-correlation by lag.
• DataPer-river paired daily totals (migration upstream + catches downstream), lag tested 0–14 days.
• MethodFor each lag N, compute Spearman ρ between today's catches and migration N days ago. Plot the curve.

Camera counts cover the whole calendar year while catch reports are season-skewed — totals are comparable per river over time but the absolute migration/catch ratio is not a literal 'catch share'.

• WhatDo big-migration years coincide with big-catch years? Yearly totals comparison.
• DataCamera totals vs catch totals per (river, year). 5–6 years per river — small-n by design.
• MethodSpearman ρ on yearly totals. With ≤6 data points per river this is exploratory only — no robust statistical claim possible.

Normalised by number of fishery-days per direction → corrects for wind prevalence bias. Proper head/tail/cross decomposition requires river flow orientation (geojson linestring) — deferred.

• WhatDo specific wind directions correlate with productive days per river?
• DataTop 3 rivers × 8 cardinal directions (N, NE, E, SE, S, SW, W, NW). Day-counts vary per direction per river.
• MethodNormalise catches by number of fishery-days in each direction (corrects for the prevalence bias — a common wind direction would otherwise dominate). Kruskal-Wallis tests whether the per-direction distribution differs.

ephem.Moon.phase returns % illumination only — waxing and waning are not distinguished here. For tide-relevant analyses (Mörrum / Ätran estuaries) a full lunar-day model would be needed.

• WhatDo lunar phases line up with catch peaks? Especially relevant for sea trout and estuarial salmon.
• DataTop 4 rivers × moon phase buckets (new / waxing-waning / gibbous / full). ~150–1,400 fishery-days per river.
• MethodSpearman ρ between moon illumination (0–1) and catch_rate per river. Grouped bar chart for mean catches per (river × phase).

Buckets: dry/light/medium/heavy. Lag-1 tests the 'fresh-water push' hypothesis: yesterday's rain triggers an upstream run today.

• WhatDoes rain trigger upstream runs? Test "fresh-water push" hypothesis — yesterday's rain → today's catch.
• Data~150–1,250 fishery-days per river with 24-hour precipitation totals. Buckets: dry / light / medium / heavy.
• MethodSpearman ρ on (precipitation vs catch_rate) at lag 0 (same day) and lag 1 (yesterday). Plus Kruskal-Wallis across rain buckets to check if heavy-rain days are systematically different.