Backtest
Would this model have flagged past shark bites? This page checks one slice of the model against the historical bite record. It is a limited test, and it leans against the factors rather than for them. Please read what it covers before reading the results: “can’t tell apart from chance” here usually means the test is too blunt to see an effect, not that the effect isn’t there.
What this test covers
Only factors with weather or satellite records going back decades can be replayed on past bite days. That is mostly the rain, wind, cloud, pressure and moon factors. The factors the rating leans on most (live SharkSmart detections, the season, fishing and wildlife reports) are either missing or held fixed by the way the test works.
Tested
19%
of the model’s factor weight
Varies day to day within a month, with records back to the 1980s.
- Heavy rain in the past 8 days
- Estuary or lagoon plume after rain
- Any rain in the last 48 h (>2 mm)
- Rapid air pressure change
- Full moon
- Onshore wind (NE to SE)
- Heavy cloud cover (>90%)
Only partly tested
29%
of the model’s factor weight
Water-temperature factors. Comparing within the same month removes the seasonal swing, so only small week-to-week changes are tested.
- White shark water temperature (15-21°C)
- Cold upwelling below ~15°C
- Bull shark water temperature (>20°C)
- Late afternoon to night
- Tiger shark water temperature (>22°C)
Not tested
52%
of the model’s factor weight
No records going back far enough, or identical on both days being compared.
- Recent shark activity nearby
- Cooler-than-surroundings water (SST anomaly)
- Warmer than usual for the time of year
- Murky water (satellite turbidity)
- Chlorophyll bloom (satellite)
- Baitfish and fish runs (fishing reports)
- Whales, dolphins and diving birds
- Whale carcass nearby
- Moderate swell 1.8-2.8 m (bull sharks)
- Dawn and dusk
- Around high tide
- Small swell (<1.5 m)
- Warm water offshore, cooler band inshore
- Strong southward current (East Australian Current)
Factor weight is each factor’s maximum points, as listed on How it works.
Why bites are a hard yardstick
- A day with no bite is not a day with no shark. A bite needs a shark nearby, a person in the water near it, and the shark choosing to bite. The rating scores conditions for shark activity, but the test only sees the handful of days when all three lined up and someone recorded it. Sharks were very likely around on many of the “no bite” comparison days too. That dilutes any real effect and pulls every result towards 50%.
- Crowds follow the weather. Comparing the same weekday removes the weekly pattern of beach use, but not “it was raining, so nobody went in”. Rainy, cloudy and onshore-wind days have fewer people in the water, so a factor that brings sharks closer and also keeps swimmers away can look neutral here. The bias works against exactly the factors this test can check.
- Blurry inputs. Weather comes from a reanalysis grid about 25 km across, as daily averages; water temperature from 5 km satellite cells. The bites are from the whole NSW coast, not just the Northern Beaches. Measurement noise also pulls results towards 50%.
- Few bites. 133 usable bites over 1986–2022. That is enough to spot a large effect, but not a moderate one.
Small-looking numbers can still matter
Take a factor that turns up on one day in four. If it made a bite 1.5 times as likely, bite days would beat their matched days only about 54% of the time. If it doubled the chance of a bite, about 57%. So a result of 54% is not “barely better than a coin toss”: it is the size of effect a genuinely useful factor would produce.
With this many bites, the 95% ranges are about ±6 percentage points wide. In odds-ratio terms, a typical factor would need to make bites roughly 1.9 times as likely before this test could reliably confirm it. Effects smaller than that can be real and still show up here as “can’t tell”.
Results: bite days vs nearby days
The short version
- Overall, the testable part of the score picked the bite day over its matched days 55% of the time, against 50% for chance (95% range 49%–62%). The direction is right, but this many bites can’t confirm it. That fits the research, which finds day-to-day weather matters much less than place and season, and those are exactly what this test holds fixed.
- ▲ Bull shark water temperature (>20°C): bite days scored higher than their matched days 60% of the time for bull and whaler sharks (34 bites).
- Leaning one way but not confirmed: Any rain in the last 48 h (>2 mm) 1.43× (range 0.99–2.11); Onshore wind (NE to SE) 1.53× (range 0.93–2.58). Treat these as hints: with 7 factors tested, one or two would lean by chance alone.
- The live parts of the rating (SharkSmart detections, fishing reports, satellite water quality) couldn’t be tested, because those records don’t go back far enough.
How to read the results
- Matched comparison (statisticians call it a case-crossover design)
- Each bite day is compared with the same weekday in the same month and year, at the same location. Think of it as asking “was it rainier on the Saturday of the bite than on the other Saturdays that month, at that beach?”. Comparing a place with itself, in the same month and on the same day of the week, cancels out the season, the location and the weekly pattern of how busy the beach is. That is deliberate, because bites mostly track when people swim, but it also cancels out the season factors the rating relies on.
- Matched days
- The 2–4 comparison days for each bite: same place, same month and year, same weekday, no bite recorded. Not necessarily days without sharks.
- Odds ratio
- How much more often a factor was switched on for the bite day than for its matched days, as a multiplier. 1.00× means no difference; 1.50× means bites were about one and a half times as likely on days with that factor. For a single factor, this is the most direct answer to “does it matter?”.
- Concordance
- The share of bite-day vs matched-day comparisons where the bite day scored higher (a tie counts as half). If the score had no connection to bites, this would be about 50%, like a coin toss. It works for the overall score, but for a factor that is often switched off on both days it stays close to 50% even when the effect is real (see the box above).
- 95% range (also called a confidence interval)
- The range the true value plausibly lies in, given we only have 133 bites. If the whole range sits above 50% (or above 1.00× for an odds ratio), chance is an unlikely explanation. If the range includes it, the data can’t separate the effect from chance. That is not the same as showing there is no effect.
- Bootstrap
- How the ranges are worked out: the calculation is repeated 2,000 times, each time on a random re-draw of the bites (some picked twice, some left out), and the middle 95% of those results gives the range.
- Reading the strips
- The dot is the result; the shaded bar is the 95% range; the dashed line is 50% (chance). Green means the whole range is above 50%, red means it’s below, grey means it overlaps 50%.
Data: Australian Shark-Incident Database (Riley et al. 2022), NSW unprovoked bites by white, bull/whaler, tiger or unidentified sharks with a full date, 1986-2022. 133 of 136 incidents had usable weather and satellite water temperature (NOAA CoralTemp, 5 km).
Overall score (testable factors only)
All species · 133 bites
55%
95% range 49%–62%
Too few bites to tell apart from chance (range includes 50%)
white · 85 bites
54%
95% range 45%–62%
Too few bites to tell apart from chance (range includes 50%)
bull/whaler · 34 bites
53%
95% range 41%–66%
Too few bites to tell apart from chance (range includes 50%)
unidentified · 11 bites
65%
95% range 46%–83%
Too few bites to tell apart from chance (range includes 50%)
Factor by factor (all species)
| Factor | On for bite days | On for matched days | Odds ratio (95% range) | Concordance (95% range) | Interval strip |
|---|---|---|---|---|---|
| White shark water temperature (15-21°C) Medium confidence Within-month temperature changes only | 49% | 54% | 0.55× (0.24–1.10)needs ~3.0× to confirm | 47% (42%–51%) | |
| Cold upwelling below ~15°C Low confidence Within-month temperature changes only | 1% | 1% | too rare to estimate | 50% (50%–51%) | |
| Bull shark water temperature (>20°C) High confidence Within-month temperature changes only | 83% | 84% | 0.81× (0.24–2.90)needs ~5.9× to confirm | 55% (50%–60%) | |
| Heavy rain in the past 8 days Medium confidence | 29% | 26% | 1.19× (0.75–1.86)needs ~1.9× to confirm | 52% (47%–57%) | |
| Estuary or lagoon plume after rain Medium confidence | 29% | 27% | 1.16× (0.73–1.79)needs ~1.9× to confirm | 51% (46%–57%) | |
| Late afternoon to night Medium confidence Within-month temperature changes only | 73% | 74% | 0.79× (0.29–1.79)needs ~3.7× to confirm | 54% (48%–59%) | |
| Tiger shark water temperature (>22°C) Medium confidence Within-month temperature changes only | 61% | 59% | 1.41× (0.70–3.30)needs ~3.0× to confirm | 54% (49%–58%) | |
| Any rain in the last 48 h (>2 mm) Low confidence | 45% | 37% | 1.43× (0.99–2.11)needs ~1.7× to confirm | 55% (50%–60%) | |
| Rapid air pressure change Local lore | 75% | 76% | 0.98× (0.62–1.59)needs ~2.0× to confirm | 48% (42%–53%) | |
| Full moon Evidence disagrees | 31% | 32% | 0.96× (0.65–1.37)needs ~1.7× to confirm | 49% (44%–54%) | |
| Onshore wind (NE to SE) Local lore | 27% | 21% | 1.53× (0.93–2.58)needs ~2.1× to confirm | 53% (49%–57%) | |
| Heavy cloud cover (>90%) Local lore | 27% | 23% | 1.21× (0.76–1.92)needs ~1.9× to confirm | 51% (47%–55%) |
“On” means the factor added a meaningful amount to the score that day (for example, heavy rain was counted on 29% of bite days). “Needs ~N× to confirm” is the smallest effect this sample would pick up reliably (4 times in 5); anything smaller can be real and still show as uncertain. Dawn and dusk is left out because a bite day and its matched days cover the same hours. Read the ranges, not the headline numbers.
The water-temperature factors can lean the “wrong” way here for a reason unrelated to sharks: within a month, warmer days bring more swimmers, so bites cluster on the warmer days whatever the sharks are doing. The incidents also span the whole NSW coast, including the warmer far north coast, so the white shark temperature band (calibrated on Sydney-area presence data) is a looser fit statewide than it is for the Northern Beaches.
Water temperature on bite days
- white (85 bites): median 22.3°C (half of the bites happened between 20.4 and 24.3°C). 32% were in the 15–21°C white shark band and 84% above 20°C.
- bull/whaler (34 bites): median 22.3°C (half of the bites happened between 21.3 and 23.2°C). 15% were in the 15–21°C white shark band and 91% above 20°C.
These are descriptive. Most bites of both kinds happened in water above 20°C, when the most people are swimming, even though tagging and drumline data put white sharks mainly in 15–21°C water. That gap is exactly why the rating is built on presence evidence rather than bite records.
Time of day
Of 97 bites with a recorded time, 39% (95% range 30%–49%) happened within an hour of sunrise or sunset. If people were in the water evenly from an hour before sunrise to an hour after sunset, you’d expect 28%: those two-hour windows make up 4 hours of a 14-hour day on average. That baseline is our own arithmetic from each bite’s sunrise and sunset times, not a beach-use survey.
So bites near dawn or dusk are about 1.4 times what even use of the water would give, split about equally between the two (20% near sunrise, 20% near sunset). Whether that means anything depends on how many people are actually in the water at those hours. 78% of these bites were on surfers and other board riders, so the best yardstick is when surfers surf.
Measured against when surfers are actually out
A surf camera at Currumbin Alley on the Gold Coast counted surfers in the water every hour of daylight for 16 months (2,593 hours of video, Murray et al. 2023). Numbers were lowest around sunrise, climbed to a peak at 10–11am, dipped through the afternoon, rose slightly at 4–5pm after work and school, then fell fast into the evening. The first and last hours of light are among the quietest of the day, not the busiest.
Weighting each bite’s day by that hourly pattern, you’d expect only about 15% of bites within an hour of sunrise or sunset if sharks bit at the same rate all day. The observed 39% (range 30%–49%) is about 2.6 times that. Sliding the surfing pattern an hour earlier or later (Queensland has no daylight saving) gives an expected 10%–21%, still well below what was observed.
Caveats: the pattern is from one north-facing Queensland point break with a busy learner wave, read off the paper’s chart (median surfers per hour), and applied by local clock time. Swimmers follow a different pattern: the same paper cites a Sydney beach where general beach use peaked at 1–3pm, which would make dawn and dusk even quieter. Bite times in the database are often rounded to the hour or half hour.
Other evidence on when people are in the water
- Northern Beaches. Council’s Smart Beaches cameras count people every 15 minutes at Manly, Shelly, Freshwater and Dee Why, but the counts aren’t published. Lifeguard counts only cover patrol hours, roughly 9am to 5pm, so they miss dawn and dusk entirely.
- Gold Coast beaches. Drone surveys counted everyone in the water from 6am to 6pm (Herrera et al. 2024). Numbers were broadly similar in the morning, midday and afternoon blocks, but the flights didn’t cover the hour after a summer sunrise or the evening.
- Réunion Island. Planes counted surfers at 10:30am and 3:30pm: 51% of surfing was in the morning and 49% in the afternoon, yet 76% of bites happened after 4pm (Lagabrielle et al. 2018). That also points to more bites late in the day than surfing explains, there mostly bull and tiger sharks in murky water.
Taken together, this is the strongest support any factor gets on this page: the excess holds against both baselines and lines up with Réunion. It is still based on a borrowed exposure pattern from Queensland and about a hundred timed bites, so it supports keeping the dawn and dusk factor rather than proving its size.
| Activity | Timed bites | Near sunrise | Near sunset | Either (95% range) | If use were even | If use matched surfers |
|---|---|---|---|---|---|---|
| All | 97 | 20% | 20% | 39% (30%–49%) | 28% | 15% |
| Board riders | 76 | 18% | 22% | 41% (30%–52%) | 28% | 15% |
| Swimmers and others | 21 | 24% | 10% | 33% (17%–55%) | 27% | 14% |
Recorded times are often rounded to the hour or half hour. 2 timed bites fell at night, outside the comparison window, and are left out. With only 21 timed bites on swimmers and others, that row is too small to read much into.
How often each rating happens (Manly, conditions only)
Replaying the conditions-only part of the model over 12,783 days at Manly (1991–2025) gives the share of days in each band by month. Live detections and fishing reports add to these on top. This climatology is also what the “compared with a typical day this month” line on the Today page uses.
Blue Typical · amber Elevated · red High. Annual share: Typical 46%, Elevated 50%, High 4%.
Ocean temperature and current ideas
Several ideas about water temperature patterns and the East Australian Current (EAC) are tested here: the cool-pocket idea from the Macquarie thesis, ideas from the research review, and a Hunter shark fisherman’s October 2026 warning that unusually warm, fast-moving water was squeezing white sharks onto the beaches (Newcastle Herald). Each idea’s threshold was set before looking at the results.
Read these with care. There are about 66 comparisons below, so one or two will look “supported” by chance alone. Bites are a weak yardstick for the reasons above, and most of these ideas are about where sharks are, which bites only measure indirectly. A range that includes 1× means “can’t tell”, not “no effect”. Altimetry currents exist for 1993-2012 and 2017 onward, so current hypotheses use fewer bites. Altimetry cannot see the first ~20 km from shore: it measures the East Australian Current and its eddies, not surf-zone flow.
- Cooler than the last 30 days (0.5°C+ below, day to day, all species): bites were 0.56× as likely (95% range 0.30–0.97, 133 bites).
- Cooler than the last 30 days (0.5°C+ below, day to day, white sharks): bites were 0.40× as likely (95% range 0.14–0.88, 85 bites).
- Warm water offshore, cooler band inshore (front) (offshore 1°C+ warmer, day to day, all species): bites were 0.51× as likely (95% range 0.29–0.83, 134 bites).
- Warm water offshore, cooler band inshore (front) (offshore 1°C+ warmer, day to day, white sharks): bites were 0.40× as likely (95% range 0.19–0.77, 86 bites).
- Warm water offshore, cooler band inshore (front) (offshore 1°C+ warmer, day to day, bull and whaler sharks): bites were 0.38× as likely (95% range 0.10–0.93, 34 bites).
- Warmer than usual for the time of year (week averaged 0.5°C+ above normal, vs other years, white sharks): bites were 1.82× as likely (95% range 1.07–3.04, 85 bites).
Every clear bite result here points the same way: warmer water nearshore went with more bites, cooler water with fewer. People swim more when the water is warm, and these comparisons cannot fully separate that from shark behaviour, which is why the presence check below matters.
Day to day, within the month
Each bite day vs the same weekday in the same month and year at the same place (2-4 control days). Tests day-to-day changes; anything that is constant for a month cancels out. 136 bites. Each cell: how many times as likely a bite was on a day with the condition (95% range), and how often it was on for bite days vs matched days.
| Idea | All species | White | Bull / whaler |
|---|---|---|---|
| Inshore cooler than the coast either side 0.1°C+ cooler · Cool-pocket factor (now starts at 0.1°C) | 1.00× 0.54–1.81 134 bites · on 34% vs 34% | 1.04× 0.45–2.31 86 bites · on 34% vs 33% | 0.83× 0.21–2.31 34 bites · on 32% vs 35% |
| Inshore cooler than the coast either side 0.2°C+ cooler · Cool-pocket factor (earlier 0.2°C start) | 0.89× 0.45–1.61 134 bites · on 25% vs 27% | 0.93× 0.38–2.04 86 bites · on 27% vs 27% | 0.77× 0.17–2.44 34 bites · on 24% vs 27% |
| Inshore clearly cooler than the coast either side 0.5°C+ cooler · Lynch 2017 / Ryan 2019, at beach scale | 0.76× 0.27–1.82 134 bites · on 9% vs 10% | 0.88× 0.23–2.59 86 bites · on 11% vs 11% | 0.80× 0.00–5.30 34 bites · on 9% vs 11% |
| Cooler than the last 30 days 0.5°C+ below · Cool-pocket factor | 0.56× 0.30–0.97 133 bites · on 17% vs 24% | 0.40× 0.14–0.88 85 bites · on 13% vs 22% | 0.73× 0.22–1.97 34 bites · on 27% vs 32% |
| Cooler than normal for the date 0.5°C+ below normal · Lynch 2017 (bites at lower anomaly) | 0.69× 0.31–1.28 133 bites · on 14% vs 17% | 0.53× 0.14–1.33 85 bites · on 8% vs 12% | 0.81× 0.24–2.27 34 bites · on 27% vs 30% |
| Warmer than normal for the date 1°C+ above normal · Fisherman: water "way hotter" than usual | 0.81× 0.35–1.75 133 bites · on 13% vs 14% | 1.09× 0.38–2.62 85 bites · on 14% vs 14% | 1.06× 0.00–5.20 34 bites · on 12% vs 12% |
| Rapid warming over the past week +1.5°C in 7 days · Fisherman: current arrived "all at once" | too rare 133 bites · on 2% vs 2% | too rare 85 bites · on 1% vs 2% | too rare 34 bites · on 0% vs 3% |
| Sudden cooling (upwelling event) −1°C in 3 days · Ryan 2019: EAC pushes sharks into upwelled water | too rare 133 bites · on 1% vs 1% | too rare 85 bites · on 1% vs 1% | too rare 34 bites · on 0% vs 3% |
| Warm water offshore, cooler band inshore (front) offshore 1°C+ warmer · Fisherman: whites squeezed inshore; Lee 2021: whites search at fronts | 0.51× 0.29–0.83 134 bites · on 35% vs 45% | 0.40× 0.19–0.77 86 bites · on 36% vs 48% | 0.38× 0.10–0.93 34 bites · on 27% vs 43% |
| Strong southward current within ~110 km 0.5 m/s+ southward · Fisherman: currents "way stronger" | 1.49× 0.43–14 33 bites · on 49% vs 45% | 6.60× 18 bites · on 50% vs 40% | 0.31× 11 bites · on 36% vs 49% |
| Southward current close inshore (~25 km) 0.3 m/s+ southward · EAC encroaching on the coast | too rare 33 bites · on 0% vs 1% | too rare 18 bites · on 0% vs 2% | too rare 11 bites · on 0% vs 0% |
| Northward flow close inshore (eddy / counter-current) 0.2 m/s+ northward · Eddy-driven upwelling | 1.29× 0.00–6.78 33 bites · on 6% vs 5% | too rare 18 bites · on 0% vs 6% | 1.67× 11 bites · on 9% vs 5% |
| Warmer than normal and a strong southward current 0.5°C+ above normal and 0.5 m/s+ · Fisherman's combined warning | 1.18× 0.29–4.57 33 bites · on 18% vs 16% | 0.89× 0.00–9.60 18 bites · on 11% vs 13% | 2.00× 11 bites · on 18% vs 12% |
Unusual for the time of year
The week before each bite vs the same week 1-3 years either side at the same place. Holds season and place fixed; does not hold beach crowds fixed (a warm spring can bring more people into the water). 136 bites.
| Idea | All species | White | Bull / whaler |
|---|---|---|---|
| Warmer than usual for the time of year week averaged 0.5°C+ above normal · Fisherman: water "way hotter" than usual | 1.25× 0.82–1.89 133 bites · on 38% vs 34% | 1.82× 1.07–3.04 85 bites · on 44% vs 32% | 0.63× 0.23–1.36 34 bites · on 29% vs 41% |
| Cooler than usual for the time of year week averaged 0.5°C+ below normal · Lynch 2017: same place and month, adjacent years | 0.91× 0.50–1.49 133 bites · on 15% vs 16% | 0.63× 0.26–1.22 85 bites · on 12% vs 17% | 1.82× 0.56–4.90 34 bites · on 21% vs 14% |
| Stronger southward current than usual week averaged 0.5 m/s+ southward · Fisherman: currents "way stronger" than usual | 1.20× 0.47–4.16 23 bites · on 65% vs 61% | 4.17× 10 bites · on 80% vs 56% | 0.50× 0.06–2.54 9 bites · on 44% vs 64% |
| Warmer than usual and a strong current both of the above · Fisherman's combined warning | 1.40× 0.41–5.15 23 bites · on 30% vs 24% | 2.36× 0.00–14 10 bites · on 30% vs 18% | 0.69× 0.00–5.63 9 bites · on 22% vs 29% |
Shark presence: SharkSmart white shark detections, 2026-07-06 to 2026-10-14
Distinct tagged white sharks detected per day in each region, July to mid-October 2026. Days with each condition are compared with days without it in the same region and fortnight, so the seasonal run of white sharks cancels out. Detections depend on which receivers are working and which sharks carry tags.
Sydney (Cronulla to Palm Beach): 168 shark-days (1.66 a day) · Central Coast: 54 shark-days (0.53 a day) · Newcastle and Port Stephens: 537 shark-days (5.32 a day)
| Idea | Days on | White sharks per day, on vs off |
|---|---|---|
| Inshore cooler than the coast either side0.1°C+ cooler | 242 of 300 | 1.28× (0.87–2.17) |
| Inshore cooler than the coast either side0.2°C+ cooler | 173 of 300 | 1.11× (0.73–1.71) |
| Inshore clearly cooler than the coast either side0.5°C+ cooler | 40 of 300 | 0.68× (0.56–0.96) |
| Cooler than the last 30 days0.5°C+ below | 75 of 300 | 1.09× (0.87–1.28) |
| Cooler than normal for the date0.5°C+ below normal | 0 of 300 | too few days |
| Warmer than normal for the date1°C+ above normal | 29 of 300 | 0.98× (0.42–1.58) |
| Rapid warming over the past week+1.5°C in 7 days | 0 of 300 | too few days |
| Sudden cooling (upwelling event)−1°C in 3 days | 0 of 300 | too few days |
| Warm water offshore, cooler band inshore (front)offshore 1°C+ warmer | 160 of 300 | 1.27× (0.93–1.61) |
| Strong southward current within ~110 km0.5 m/s+ southward | 134 of 253 | 1.12× (0.78–1.57) |
| Southward current close inshore (~25 km)0.3 m/s+ southward | 24 of 253 | 0.82× (0.53–1.07) |
| Northward flow close inshore (eddy / counter-current)0.2 m/s+ northward | 111 of 253 | 0.96× (0.77–1.22) |
| Warmer than normal and a strong southward current0.5°C+ above normal and 0.5 m/s+ | 54 of 253 | 0.99× (0.64–1.43) |
| Southward current just off the beach (model)0.3 m/s+ southward | 16 of 303 | 1.58× (0.15–3.12) |
| Northward current just off the beach (model)0.1 m/s+ northward | 19 of 303 | 0.59× (0.33–0.83) |
Three and a half months of one season, with days that follow each other closely, so these ranges are wider than they look. This is the start of the presence record, not a verdict.
What comes next
The better test is against shark presence, not bites. Since October 2026 the daily run archives every SharkSmart detection, sighting and capture near the Northern Beaches. That sidesteps the two biggest problems above: it counts sharks whether or not anyone was swimming, and it covers the live and satellite factors this backtest can’t reach. Once a season or two has accumulated, we will check whether High days actually coincided with more sharks nearby, compare against a simple “what month is it” baseline, and adjust the weights. Until then, treat the weights as informed priors.
Backtest generated 2026-10-15.