Bite-score signals

The user-facing features/bite-score.md describes what the bite score is. This page documents the signals that compose it and the literature behind each.

We don’t publish the exact numerical weights — those get retuned as data accumulates — but the structure and the per-signal logic is open.

Signal: solunar timing

What it captures. Major and minor feeding periods derived from the moon’s position: moon overhead, moon underfoot, moonrise, moonset.

Why it’s in the model. The solunar theory has a mixed academic record but a strong practitioner record. Pelagic predator activity spikes during major periods are well-documented across reef fish, billfish, and inshore species. The signal is not load-bearing on its own — it’s one input among many — but ignoring it would be wrong.

How it’s used. Time-distance from the nearest major or minor period maps to a positive contribution that decays with hours from the period centre. Major periods contribute roughly twice the magnitude of minors.

Signal: barometric pressure trend

What it captures. The 24-hour change in surface barometric pressure, in hectopascals.

Why it’s in the model. Falling pressure ahead of a weather change is the most-cited “feeding spike” condition in fishing literature. Stable high pressure suppresses activity in many species; rapidly rising pressure after a front is associated with a 12–24 h lull.

How it’s used. Pressure trend (24 h delta) maps to a contribution that’s positive for falling pressure (ahead of a system), neutral for stable, and slightly negative for rapidly rising. The magnitude is species-modulated — gamefish respond more strongly than bottom species.

Signal: recent fronts

What it captures. Cold-front passage in the last 48 hours.

Why it’s in the model. Cold fronts depress activity for ~6–24 h post-passage, then often produce a strong “rebound” 24–36 h later as fish resume feeding under the new stable pattern. The signal is partly redundant with the barometric trend but captures the specific post-front rebound timing better.

How it’s used. A binary “front passed in last X hours” combined with a continuous “hours-since” curve produces a contribution that’s negative for the first 24 h post-front and positive in the 24–48 h window.

Signal: tide stage

What it captures. Rising / falling / slack state, plus the rate of water-level change.

Why it’s in the model. Most species feed actively on moving water — bait gets displaced, predators ambush, currents concentrate forage. Slack tends to suppress activity. The relationship varies by species and habitat: redfish work the falling tide on flats; tarpon feed the incoming through inlets; bottom species have weaker tide preferences.

How it’s used. Per-species tide preference (encoded as a curve over the tide cycle) is multiplied by the rate of water-level change. Spring tides amplify the signal; neap tides damp it.

Signal: water temperature versus species preference

What it captures. Current sea-surface temperature versus the target species’ published preferred range.

Why it’s in the model. Warm-water species become lethargic in cold water and vice versa. Activity drops sharply outside the species-specific comfort band, and very sharply outside the survival band.

How it’s used. Distance between observed SST and the centre of the species preference range maps to a smooth penalty. Within the preference range, no penalty. Beyond it, an exponential drop-off to zero contribution.

The species temperature ranges come from WoRMS (World Register of Marine Species) and FishBase, with hand-curation for species where those sources disagree.

Signal: species presence (seasonal range)

What it captures. Whether the target species is plausibly present in the area at the current season, based on iNaturalist / OBIS observation density.

Why it’s in the model. A spectacular bite-score for a species that isn’t in the area is a useless number. We treat species presence as a gating signal: if the seasonal range overlap is below a threshold, the score is suppressed and the app surfaces a “this species is unlikely-to-be-present here / now” note.

How it’s used. Presence is a multiplier on the rest of the score, not a separate additive contribution. If presence is near zero, the score collapses to near zero regardless of other signals.

Signal: time of day

What it captures. Civil dawn, civil dusk, solar noon, midnight — plus the sun-altitude curve through the day.

Why it’s in the model. Most predatory species are crepuscular — peak feeding within an hour or two of sunrise and sunset. Some species (swordfish, certain pelagics) are night-feeders; others (mahi, dolphin) are mid-day surface predators.

How it’s used. A per-species diel-activity curve maps the current time-of-day to a contribution. The bite-score for sailfish at solar noon is structurally different from the bite-score for bonefish at solar noon.

Signal: wind and surface conditions

What it captures. Wind speed, gust, wave height.

Why it’s in the model. Heavy chop suppresses surface and near-surface feeding. Some species (snook, redfish in the surf zone) actually feed better in moderate chop because bait is disoriented.

How it’s used. Per-species “preferred wind” curve, mostly penalising high wind, with some species-specific bumps for moderate chop.

What the model doesn’t include (yet)

Honest list of signals we know matter but haven’t integrated:

These gaps are honest, not hidden. As we get data, the model gets sharper.

The composition

Once each signal computes its (value, weight, reason) triple, the final score is:

score = (Σ value_i × weight_i) / (Σ weight_i)   × presence_multiplier
score = clamp(0, 10, score × 10)                 # scale to 0–10

Each contribution is visible in the breakdown. If you disagree with a weighting, the methodology question template exists for that.