Experimental prototype — not an official warning system

Methodology

How vShark estimates relative shark interaction risk — and what it does not claim.

What vShark measures

vShark combines ecological observations, species-specific presence/hazard estimates, human exposure and consequence assumptions into a relative interaction risk assessment for a coastal location and time. It is experimental decision-support infrastructure, not a prediction of a specific shark attack.

Core formula

R(x,t) = E(x,t) × Σs [ Ws × Ps(x,t) × Cs ]
  • E — human exposure
  • Ps — species presence/hazard (logistic + thermal + detection decay)
  • Ws, Cs — configurable species weight and consequence scalar

The public 0–100 index is a development mapping of raw R (currently × 120, clamped 0–100). It is not calibrated to incidents, expert rankings, or official NSW scales. Do not treat the number as a scientific or operational rating.

Risk bands (provisional)

0–24LOW
25–49MODERATE
50–74ELEVATED
75–100HIGH

Thresholds are versioned development defaults. We never use “SAFE”.

Confidence

Public UI shows one combined confidence (HIGH / MODERATE / LOW). It reflects data freshness, completeness, mock/stale feeds and coverage — not false precision. Missing inputs reduce confidence; they do not silently imply low risk.

Development parameters

Thermal optima, species weights and consequence scalars are development seeds stored in the database and model version record. They are not published biological constants. See RISK_MODEL.md.

What we never say

  • SAFE TO SWIM
  • NO SHARKS PRESENT / SHARK-FREE
  • Guaranteed protection
  • Prediction of a specific attack

Known limits (First Build)

  • Model version 0.1.0-dev — development seeds, not validated biology.
  • Open-Meteo is live as regional centroids (Northern Beaches; six WA corridor centroids; east-coast bull-shark corridor), not beach-specific SST.
  • NSW Beachwatch water quality is live as labelled context only. It does not enter the shark-risk score and is not a vShark swim clearance.
  • iNaturalist, GDELT media volume, and whale/salmon climatologies are labelled context. They do not enter the shark-risk score.
  • NSW SharkSmart, WA SharkSmart / DPIRD, BOM and IMOS are not connected. MOCK detections do not enter the score. No WA detections are invented.
  • Without a live detection feed, public confidence cannot be HIGH.
  • Species weights and the ×120 public scale are uncalibrated.

Geography

Locations are configuration: country → jurisdiction → coastal region → location. NSW Northern Beaches remain on /nsw/northern-beaches/…. Western Australia (Esperance → Gnaraloo) uses /wa and /location/[slug]. Adding a beach should not require a new application.

White shark telemetry

vShark is the intelligence layer over fragmented trackers, not a copy of OCEARCH. OCEARCH tracker data are viewing-only and are not ingested. NSW/WA SharkSmart, Sharktivity, OTN, ATAP and ETN are registered with licence and API status. IMOS ATF public QC detections are periodic/historical, not live pings. A ping is an observation; a line between pings is not a path; no ping is not absence. See white shark intelligence and coverage.

East-coast bull sharks

A separate development layer (bull-shark-0.1-dev) estimates ecological conditions for Carcharhinus leucas along Moreton Bay → Sydney, with a Victorian temperate range-edge that is not a pulse extension. It is not the mixed-species 0–100 index and does not include human exposure. East Australian Current is a candidate variable and is omitted until a current product is connected. Acoustic receiver health is UNKNOWN. See the corridor and the Moreton Bay investigation.

WA species weights

There is no generic SHARK_SCORE. WA regions overlay hypothesized species weights on the same 0.1.0-dev engine (white / tiger / bull). Those overlays are labelled HYPOTHESIS — they encode a temperate-to-tropical transition to be tested, not a fact. NSW global weights are unchanged.

Evidence vs confidence

Direct recent detection outranks verified sighting, which outranks ecological proxies, which outrank habitat suitability and seasonal priors. Confidence is displayed separately from the risk band. Official closures always outrank the model.

Always follow

Beach closures, lifeguard instructions, NSW or WA Government shark-safety information, organisational safety procedures and professional judgement take precedence over any vShark output. vShark is not NSW DPI / WA DPIRD / SharkSmart, Surf Life Saving Australia, Surf Life Saving WA, Northern Beaches Council, City of Joondalup, or the Bureau of Meteorology.