VH-MDX

On the night of 9 August 1981 a Cessna 210 carrying five men vanished over the Barrington Tops in New South Wales, and forty-five years of searching, radar geometry, sixteen thousand simulated flights and airborne lidar still cannot say exactly where it lies.

Barrington Tops from the south · terrain model, height exaggeratedAbout 100 × 80 km · north at the back
Not found.

If the Wittenoom paper is right that Williamtown's antenna was tilted up 1°, the radar's floor over the estimated fade point sat near 6,200 ft. On the tape log, Hutchins came down through that height at 19:36–19:37. That makes the last radar position the point where VH-MDX dropped below the beam, not where it crashed, and it flew on for at least three minutes. Sixteen thousand simulated flights that fit the radio log spread the likely impact widely: the most likely half covers about 1,000 km², and only about a quarter of it lies inside the box searched in 1981. The public 2018 lidar under the fade point averages 0.02 ground points per square metre. That is too sparse to see a light aircraft under 40 m eucalypts, and air photos from 1972 and 1983 show no fresh scar at any lidar candidate. The same methods also failed to pick out three wrecks whose fate is known, so a blank result says little. Finding the wreck needs a narrower corridor and a dedicated survey.

Last call
19:39
5,000 ft, below safe altitude
Radar floor at fade
≈6,200 ft
with a 1° beam floor · model
In the 1981 box
26%
of modelled impact mass · 16,000 flights
Ground returns
0.02/m²
2018 public lidar, two tiles
Found
0
searches since 1981 · 9 major
Looking up through a dense Antarctic beech canopy, branches and moss-covered trunks closing out the sky.
Antarctic beech canopy, Barrington Tops — From above, this is what searchers see.Doug Beckers · CC BY-SA 2.0 · Wikimedia Commons

A tired pilot with a failing instrument flew into cloud, ice and a gale over the ranges at night. The radio log records nineteen minutes of it, and it is the only hard evidence of where the aircraft went.

On 9 August 1981 Mike Hutchins was flying four Sydney workmates home. The men had spent about a week on a boat in the Whitsunday Passage. Hutchins was 52 and had more than 3,000 hours. He left Proserpine a little after ten in the morning, refuelled at Coolangatta, and took off again at 17:02 for Bankstown. The aircraft was VH-MDX, a Cessna 210M, a retractable-gear single. It was approved for instrument flight but not for flight into known icing, and it had no de-icing equipment.

Aboard were Ken Price, a NSW Police inspector and former head of the Newcastle Water Police, plus Rhett Bosler (33), Noel Wildash (40) and Philip Pembroke (43). Price's age is given as 54 or 55 depending on the source.

The fuel attendant at Coolangatta, Gordon Grieg, later said the pilot looked pale and tired. Hutchins mentioned a gyro or electrical problem and was not interested in staying the night. The engine took several attempts to start. Each detail was minor. Together they describe a tired pilot in a hurry, with a known instrument fault, heading into the night.

Strong west-south-west flow over northern NSW. Freezing level 4,000–7,000 ft, with moderate icing in cloud above it. There was a SIGMET for occasional severe turbulence below 12,000 ft east of the ranges. Winds at 9,000 ft were later reported at 70–76 kt.

At 18:50 he called Sydney Flight Information Service. He was cruising at 8,000 ft, estimated Singleton at 19:30, and wanted a clearance through RAAF Williamtown's airspace on the coast. What happened next is still argued about. One account says Williamtown declined night visual clearance because the weather was unsuitable. Another, drawing on the transcripts, says a clearance was available and was withheld. At 18:56 Hutchins chose to track inland via the Craven reporting point instead of waiting.

The tape log, paraphrased, in local time:

Fear of Landing records Williamtown's refusal on weather grounds. Owen Zupp in Australian Aviation (2016) calls the "denied and pushed inland" version a misconception, citing transcripts.

  1. 18:508,000 ftCalls Sydney FIS. Requests clearance through Williamtown airspace.
  2. 18:568,000 ftElects to route via Craven rather than wait for the coastal option.
  3. 19:198,000 ftReports over Craven, 36 nm from Singleton. Considerable turbulence and downdrafts.
  4. 19:23→10,000In cloud. Asks to climb. Reports loss of the artificial horizon and directional indicator. Asks for a radar steer to Bankstown.
  5. 19:25—ADF wandering. Confirms he is not in visual conditions. Uncertainty phase declared at 19:26.
  6. 19:28<8,500 ftRadar identifies him 36 nm north of Singleton. Struggling to climb. Asks for a vector to Maitland and turns south.
  7. 19:34—Ice. A 1,000 ft/min downdraft. Declines the Maitland diversion. Mentions a brief suspected cockpit fire.
  8. 19:357,500 ftVacuum pump failed. Climbing and descending erratically. Standby compass swinging.
  9. 19:36—Last radar position, over the Barrington Tops.
  10. 19:376,500 ftLosing height fast. FIS gives the lowest safe altitude: 6,000 ft.
  11. 19:395,000 ftLast transmission. Gives his height. Repeated calls go unanswered.

Gyros without vacuum

Instrument panel with a tumbled attitude indicator and drifting heading gyro, fed by a failed engine-driven vacuum pump, beside a working electric turn coordinator and wet compass.
An engine-driven pump pulls air through the attitude indicator and the directional gyro, and that airflow keeps their rotors spinning. When the pump fails, the rotors spin down. The attitude indicator tumbles and the directional gyro drifts, so the pilot loses both his horizon and his heading. The turn coordinator runs on electricity and the wet compass needs no power, so both kept working, though the compass was reported swinging. Hutchins reported losing the horizon and heading indicator at 19:23 and named the vacuum pump as the cause at 19:35. One analysis suggests the pump may have been failing well before Craven.

A flying instructor on a night training flight nearby was listening. He said the voice on the frequency started out casual and ended close to screaming. The 19:39 call put the Cessna at 5,000 ft, a thousand feet below the safe altitude for that country. The plateau tops out near 1,580 m, a little over 5,000 ft. Several popular retellings merge the ice report into the final call. On the tape log, the ice came five minutes earlier.

The rule under each entry is scaled to reported altitude against the 8,000 ft cruise. Entries with no altitude are drawn full.

Sinking air in the lee

Cross-section of wind streamlines rising over the Barrington plateau and dipping down its lee slope, with an icing layer above the freezing level and the Cessna sinking at 1,000 ft per minute.
A strong west-south-west wind, about 70 knots at 9,000 ft, crossed the Barrington plateau and poured down its eastern side. Air descending a lee slope carries an aircraft down with it. At 19:34 Hutchins reported ice and a downdraft of 1,000 ft a minute. The forecast freezing level was 4,000–7,000 ft, with moderate icing in cloud above it. Here the dots thicken above 7,000 ft, where ice was certain, and thin out across the band where it depended on the actual freezing level. Schematic section, heights exaggerated. The terrain is a simplified profile, not a survey.

The last call fixes a time and a height, but not a place. Where the Cessna was when it fell silent depends on a route that was never flown as planned.

Route, radar and search areas

Barrington TopsYBCGYSBKYWLMYSTWTaree0 — 100 km1981 search areaHugh Gooding Dingo Tops areaGloucester Tops Most Probable AreaRadar fade 19:36 (est.)Craven 19:19Mt BarringtonPolblueGloucester TopsMt RoyalCobarkDingo TopsCaparraMt AllynRawdon Vale151.25°E 32.30°S152.30°E 31.60°S
  • Filed route
  • Schematic deviation (inferred)
  • Search area (approximate)
  • Radar bearing arc, 45 nm from Williamtown
  • Landmark / fix
Coordinates come from published accounts. The radar-fade point is an estimate: 45 nm from Williamtown on the published bearing lines. Search boxes are illustrative, because no source gives their boundaries. Hover, focus or tap any feature for its note, confidence and source.

The 1981 search was one of the largest of its time, and it found nothing. Its failure shaped every search since.

The search started the next morning in gale-force wind and freezing temperatures. It ran nine days, to 18 August, and it was large for its time. Fixed-wing aircraft and helicopters flew over 400 hours between them. Police, forestry crews, the Water Board, bushwalkers and the SES worked the ground on foot, by four-wheel drive and on trail bikes. RAAF F-111s made infrared runs. Landsat and US Air Force imagery were checked. The aircraft's emergency locator beacon was never heard.

The weather that took the Cessna also worked on the search. One helicopter had a rotor-blade failure during the operation.

Days
9
air and ground search, 10–18 Aug 1981
Fixed-wing sorties
80
191 flying hours
Helicopter sorties
109
175 flying hours
ELT signals
0
detected

They found older wrecks and not this one. In September NSW Police ran a separate Operation Barrington Tops, which also found nothing. After that the work passed largely to volunteers, chiefly the Bushwalkers Wilderness Rescue Squad (now BSAR NSW). It has run search exercises on the Tops more or less every year since.

Each of those searches had to decide where to look, and each answer rests on a different reading of the same thin record.

Twisted snow gums on open plateau near Mount Barrington.
Snow gums near Mount Barrington — On the open plateau.Peter Woodard · CC BY-SA 3.0
A dirt road through dark cool temperate rainforest, trees closing overhead.
Gloucester Tops Road — Through cool temperate rainforest.Doug Beckers · CC BY-SA 2.0

Nine organised searches over 45 years have covered different ground, because each trusted a different piece of evidence. None has found the aircraft.

Every serious search since 1981 rests on the same thin evidence: the tape times, a handful of radar returns, and what a few people on the ground heard. Each theory of where the aircraft lies comes down to a theory of which of those to trust.

The Wittenoom paper notes the Williamtown antenna was tilted 1° upward. At 45 nm that leaves a blind band low over the ranges. So the radar fade marks where the Cessna went below coverage, not where it hit.

WhenWhoAreaResult
Aug 1981DCA, police, RAAF, SESRanges around Mt BarringtonNothing; older wrecks found
Sep 1981NSW PoliceFollow-up ground searchNothing
1981–BWRS / BSARAnnual exercises on the TopsOngoing
Oct 2013Op Wittenoom, Police RescueArchive-narrowed box (unverified)Nothing
2013–14Op PhoenixRe-timed tapes, new siteNot searched
2014BRG / BSARGloucester Tops, 10 nm eastOngoing
Sep 2016BSAR and agencies9 km² core, Gloucester TopsNothing
Sep 2018Multi-agency exerciseMore remote sensingNothing reported
NeverGooding, Dingo TopsDingo Tops / Caparra, 60 km NENever searched

The Operation Wittenoom paper, hosted by the Barrington Research Group, raises a problem with the Craven fix itself. If the vacuum pump was already failing near Nabiac, the directional gyro could have drifted before Hutchins reported Craven. He may have been north of where he thought he was. That moves the whole picture. It is an inference, and the paper presents it as one, but it explains why serious searchers have drifted east over the years and away from the 1981 ground.

The Newcastle Herald's 45th-anniversary piece makes the plainer point. Many aircraft lost in this region were found by accident, often by bushwalkers, often years later, and often away from any planned search line.

The Herald counts nine air crashes in the area between 1945 and 1987. VH-MDX is the only one never found.

What Williamtown could see

Radar sees in straight lines, so mountains and the curve of the earth hide low aircraft. Over the Barrington Tops the hidden band was deep enough that the radar lost the Cessna while it was still flying.

Everything placed after 19:36 depends on one question: was the Cessna gone, or just out of sight? A ground radar sees nothing below the line from its antenna across the highest terrain in between. Over the Barrington Tops that line sits high. The ranges block the view, and the earth curves away beneath them. Add the 1° upward antenna tilt the Wittenoom paper describes, and the floor rises again.

The figure calculates that floor for every square kilometre of the ranges, then replays the descent from 19:19 to 19:39 at four different beam floors, so the moment the Cessna slips out of sight can be watched.

The floor is calculated from a terrain model, a standard allowance for the slight bending of radio waves, and an assumed 25 m antenna. It is a reconstruction, not a reading from the 1981 radar.

Radar shadow from RAAF Williamtown

3D · drag to orbit · vertical ×4 · translucent sheet = radar floor · outlined plane = aircraft altitude
tape log replay · beam floor 0.0° → 1.5°
19:19
CravenFade 19:36 (est.)Mt BarringtonGloucester TopsMt AllynPolblueMt RoyalWilliamtown radar ↘ 60–130 km
—Hover the map for the lowest visible altitude at a point
  • Radar waves from Williamtown, where it sees this altitude
  • Edge of radar shadow: the waves stop here
  • Terrain, 100 m contours
  • Radar floor, in section
  • Hidden at this altitude, in section
  • Fix / landmark
  • 1981 search area (approx.)
Radar shadow from RAAF Williamtown, 151.834°E 32.795°S, at the chosen altitude and beam floor. At 0° only the terrain blocks the view. The Wittenoom paper says the antenna was tilted up 1°; the beam's true lower edge depends on beamwidth, which is not on the public record.Terrain: SRTM / Geoscience Australia, CC BY 4.0 · Radar floor: d/rksci reconstruction

The tilt matters more than the terrain. With terrain alone, the lowest altitude Williamtown could see over the estimated fade point is about 3,150 ft. Only 15% of the 1981 search box would hide an aircraft at 5,000 ft. With a 1° beam floor, the radar's floor over that same point rises to about 6,200 ft. At 5,000 ft, 94% of the search box is hidden, and 41% even at 6,500 ft.

That fits the tape log uncomfortably well. At 19:35 Hutchins reported about 7,500 ft, and the last radar position came at 19:36. At 19:37 he reported 6,500 ft and losing height. A 1° floor puts the radar's loss of the Cessna right where that descent would cross it. So the 19:36 position marks where VH-MDX went below the beam. It is not where the aircraft came down. It was still flying, and talking, for at least three more minutes.

The radar's beamwidth, tilt setting and clutter filtering on the night are unknown. The ATSB file may hold them.

Below the beam

Cross-section from Williamtown radar to the Barrington Tops, showing a 1° beam floor rising over a curved earth and the Cessna's descent crossing below it at the last radar return, then continuing unseen to 5,000 ft.
A radar sees along straight lines. The earth curves away beneath them, so the lowest height it can see rises with distance. Williamtown's antenna was reportedly tilted up 1°, which lifts that floor further. Over the estimated fade point, 45 nautical miles out, the floor sits near 6,200 ft. Hutchins reported 7,500 ft at 19:35. The last radar position came at 19:36, and at 19:37 he reported 6,500 ft, just under a floor that by then stood near 6,600 ft. The radar lost the Cessna where it went below the beam, not where it came down. Earth curvature and heights are exaggerated, and the range axis is broken between 4 and 30 nautical miles. The beam floor uses a standard refraction model with an assumed 25 m antenna height. The beamwidth on the night is unknown. Heights are the pilot's own reports, read in icing. Ranges after 19:35 are schematic. Terrain beyond 33 nautical miles is sampled along the radial from Williamtown; nearer terrain is drawn low and schematic. Radar identification at 19:28 is not drawn: on this geometry it falls 52 nautical miles out, beyond the 48-mile display range, a conflict noted in the record.

Three minutes in a falling aircraft is a lot of ground. The next question is how much.

A computer flew the last twenty minutes tens of thousands of times, each with different unknowns, and kept only the flights that could have produced the radio log. Where those flights came down is a map of where the aircraft could be. It does not mark where it is.

The radio log is a constraint as much as a story. Any account of the last twenty minutes has to fit what was recorded: 8,000 ft at Craven at 19:19, the instrument failures at 19:23, radar identification at 19:28, ice and a downdraft at 19:34, 7,500 ft at 19:35, 6,500 ft at 19:37 and 5,000 ft at 19:39. Then silence. The simulation is a simple flight model of a C210 with failed gyros, picking up ice in a strong west-south-west wind over this terrain. Each run draws random values for what nobody knows: where the gyro drift had really put him at Craven, his heading after 19:28, the wind, the ice rate, the radar's beam floor, and how control was lost. Only the flights that could have produced that log are kept.

Out of 197,522 attempts, 16,000 fit, about 8.1%. Each surviving flight is weighted by how well it matches the two radar fixes, at 19:28 and the fade at 19:36. The weights are deliberately soft, so a flight 30 nm from the estimated fade still counts for 40%. The figure shows where those weighted flights hit the ground, and cycles on its own through the outcomes and weightings; brushing a histogram tests any single assumption.

Of the accepted runs, 57.2% end in a stall or spin, 22.7% in a spiral dive, 20.0% in an in-flight break-up and 0.1% in controlled flight into terrain. A break-up spreads its debris downwind along a line up to 2.4 km long.

Keeping flights that fit

A funnel thinning 197,522 simulated flights through radio-log checks to 16,000 accepted, then weighting them by two radar fixes into a halftone map of impact density across a widened model area, with the earlier smaller area dotted.
The simulation flew 197,522 possible versions of the last twenty minutes. Each had to pass every check from the radio log: 8,000 ft over Craven at 19:19, not climbing at 19:28, ice and a downdraft at 19:34, then 7,500, 6,500 and 5,000 ft at 19:35, 19:37 and 19:39. It also had to stay airborne until the last call and come down within ten minutes of it. Only 16,000 passed, 8.1% of attempts. Each survivor is then weighted by how close it ran to the two radar fixes, at 19:28 and 19:36 (crosses). Below the funnel is the weighted density of where they came down, drawn as a halftone across a model area widened to the west and south (the dotted outline marks the earlier, smaller area). The density no longer piles against an edge. About half of the weighted density still lies in the northern part, inside the earlier frame, and the rest falls south of it, in lower country between the Barrington plateau and the Hunter. Illustrative model output, not a search determination.

Where 16,000 simulated flights came down

state 1 of 9

All outcomes · weighted by both radar fixes · break-up debris spread downwind

Mt BarringtonGloucester TopsCravenLoading 16,000 runs…
  • Impact density
  • 50 / 80 / 95% region
  • 1981 search area
  • Gloucester Tops focus (2014–)
  • Lidar tiles analysed
  • Terrain, 100 m contours
  • Radar fade 19:36 (est.)
  • Outer 2 km of the model area
Each run is a full flight from the Craven call to impact, kept only if it fits every altitude and time on the tape log. The density is smoothed over about 500 m. The model area was widened west and south for this run, and only 0.5% of attempts now leave it, so the map no longer piles up against its edges; the hatched rim marks the outer 2 km.Model: d/rksci scenario simulator, 16,000 accepted of 197,522 attempts · Terrain: SRTM / Geoscience Australia

With both radar weights on, about 26% of the modelled impact mass lies inside the box the 1981 search covered. About 1% falls in the Gloucester Tops area that later searches moved to.

That isn't a reason to send anyone back to 1981's ground. The 1981 search was hampered by weather and canopy, not by the wrong box, so "inside the box" and "already cleared" are different things. The distribution is also wide. The single most likely 500 m cell holds about 0.05%, and the most likely half of the mass is spread over about 1,000 km². Roughly half of it lies north of 32.3° S, near the plateau; the rest falls south, in lower country towards the Hunter. The model is one interpretation of the log with stated assumptions, and it should be read that way. Its job is to rank ground by likelihood, so that a dense lidar survey or a ground team can start where it matters most.

Only 0.2% of the modelled mass falls inside the two tiles analysed in part two. Those tiles were placed on the radar-fade estimate, before this model existed.

Eight ways it could have gone

The sixteen thousand flights are not one blur. They fall into eight families, each a recognisable story of heading, drift and descent, and each ends in a different patch of country.

Grouping the accepted flights by their heading after 19:28, how fast the heading gyro drifted, where and when they dropped below the radar, how they ended and where they came down gives eight families. The two largest, together about 39% of the weighted runs, hold west-south-west or south-west and come down in or near the 1981 box. The smaller families turn further south and end in the lower valleys towards the Hunter and Williamtown, where nobody has searched. The map opens with the flights themselves piling up, then settles into the density they make and each family's most typical path.

Each family's line is its most typical single run; the band around it holds the middle 80% of the family at each moment. Shares are weighted by fit to the two radar fixes. Select a row, a small map or a line to load that family's run in the cockpit replay below.

Scenario families

Loading the scenario families…

The map opens as a weighted sample of 3,000 of the accepted flights is drawn, coloured by family, under the tape-log clock; they then fade into the smoothed impact density and its 50, 80 and 95% contours, and each family's typical path appears. Any tap or key skips to the end. Small maps show where each family comes down; the table gives its share, outcome mix, heading, time below the radar and a satellite link to its impact region.Model: d/rksci scenario simulator · Terrain: SRTM / Geoscience Australia

A family is a statistic. To see what one of these flights might have been like from the inside, the model can be replayed second by second.

Inside one of them

The replay below follows one simulated flight from Craven to the ground, with the instruments, the view and a transcript. It shows what the workload might have been, not what happened.

Cockpit replay

Loading the replay…

The most typical run of the family selected above, starting from the Craven call at 19:19:19. It plays while on screen; with reduced motion, Play steps from one transcript line to the next. The window draws model terrain, not anything that could have been seen at night in cloud. Ink lines are the record; dotted lavender lines are synthetic.Model: d/rksci scenario simulator · Record: tape log as published

The model treats the loss as weather, instruments and terrain. Not everyone has.

A missing police inspector in a corrupt decade invited darker explanations. None of them has any evidence behind it, and the people they hurt most are the families.

A plane carrying a police inspector disappears in a decade when the NSW Police force was, as the 1997 Wood Royal Commission later found, broadly corrupt. People were always going to fill that silence. The ABC's November 2026 feature, built around the Expanse podcast, collects the versions. Former news cameraman Glenn Cook says the most common one has a gang bombing the aircraft to kill Price. Then there are the smuggling versions, with gold, drugs or money and a cover-up. Another has a bookmaker aboard with $600,000. And of course aliens.

The ABC found no evidence for any of them. Kate FitzGerald, a QUT researcher who studies conspiracy belief, doesn't endorse the theories, but she explains their pull. The corruption was real, so a story that leans on it feels plausible. She also notes that the families have never had closure.

"Status" records what the sources support. A theory nobody has ruled out is not thereby supported.

Best supported
Weather, instruments, loss of control

Vacuum failure took out the attitude and heading gyros in cloud, ice and severe turbulence. This is the official view and fits the tape log.

Plausible, untested
In-flight break-up

The Sydney FIS controller thought overstress could have scattered compacted wreckage across separate sites, which would make it harder to spot.

No evidence
Bomb aimed at Inspector Price

The most common rumour, according to cameraman Glenn Cook. Nothing in the record points to it.

No evidence
Police smuggling and a cover-up

It draws on real 1970s–80s police corruption findings. The ABC found no evidence linking Price or the flight to it.

Unsourced
Bookmaker's $600,000

It appears in later retellings. No primary source has been found for it.

No
Abduction by aliens

Cook mentions it. Nothing more needs saying.

Her father, she says, took his work seriously, and the corruption talk is wrong.— Merrilea Price, paraphrased from ABC News, 17 Nov 2026

Merrilea Price was 23 when her father vanished. The rumours still upset her. A reader who signs himself as Phillip Pembroke's son wrote on the Fear of Landing site that his mother was pregnant at the time. The family has a memorial plaque and no grave. Those are the people a find would actually matter to. Whether a laptop and public survey data can do better than nine searches is the question that remains.

Polblue Swamp, an open sedge wetland ringed by eucalypt forest under a pale sky.
Polblue Mountain and Swamp, about 1,450 m — The open ground on the plateau is the exception.Peter Woodard · CC BY-SA 3.0 · Wikimedia Commons

Airborne laser scanning can map the forest floor under the canopy. The public survey over the Tops is not dense enough to show a small, old wreck.

Airborne lidar fires laser pulses from an aircraft and records every return. The first returns come off the top of the canopy, more come off branches on the way down, and some get through gaps to the ground. Sort the points, keep the ones that hit the ground, and the result is a bare-earth model of terrain no one can see from the air. Archaeologists have used this to find roads, terraces and whole towns under rainforest. The question for VH-MDX is narrower and harder. Would a 45-year-old light aircraft wreck show up at all?

NSW point clouds follow the LAS classes: 2 ground, 3–5 low/medium/high vegetation, 7 noise. Older NSW captures aimed for about 1 ground point per m². The current spec asks for at least 4 pts/m² overall.

Lidar under 40 m trees

Cutaway of tall eucalypts with laser pulses from a survey aircraft stopping in the crowns, a few reaching the ground, a dashed interpolated bare-earth line, and aircraft debris under the canopy with no return.
A survey aircraft fires laser pulses down through the forest. Most end in the crowns of 40 m eucalypts as first returns. A few slip through gaps to the ground. The bare-earth surface is drawn by joining those few ground hits, so between them it is a guess. A crumpled light aircraft under the canopy, a few metres across, can easily get no return at all. The 2018 survey over the radar fade averages 0.02 ground hits per square metre, about one per 50 m². Schematic. Pulse count and spacing are illustrative, not to scale.

The data exists. NSW Spatial Services publishes classified point clouds through the national ELVIS portal, free to download in 2 × 2 km tiles. Over the Barrington Tops the ELVIS inspection tool returns one survey, UPPER MANNING 201805, captured in 2018. It also returns Geoscience Australia's 5 m Manning River DEM mosaic, which is far too coarse for this.

Point density and what it can show

1 pt/m² · 22% to groundOlder NSW captures under wet forest. Only a few ground points per 10 × 10 m patch.

4 pts/m² · 22% to groundThe current NSW minimum. A wing would catch a handful of hits.

20 pts/m² · 22% to groundA dedicated low-altitude or drone capture. Debris starts to have a shape.

Illustration, not data. Each panel is a 10 × 10 m patch. Dark points reached the ground and faint points stopped in the canopy. The outline is a Cessna 210's 11.2 m wingspan for scale. The ground fraction is an assumed figure for dense wet forest.

An intact Cessna 210 is about 8.6 m long with an 11.2 m span. After 45 years under wet sclerophyll forest it is more likely a crumpled, partly overgrown debris field a few metres across, sitting on a slope and possibly in a creek line. On a 2018 survey at around 1–4 ground points per square metre, that is a handful of points, which won't resolve an aircraft. What might show is a secondary signature: a linear gap or broken-crown scar along a descent line, an anomalous mound in the bare-earth model, or a hole in ground-return density where something solid sits under the leaves.

Two tiles from the 2018 Camberwell survey, downloaded and counted: 0.66–0.76 points per m² in total, but only 0.016–0.02 ground points per m². That's one ground hit per 50 m² or so.

An open toolkit

A short chain of standard steps turns a raw laser survey into maps of bare earth, canopy height and small bumps and hollows, then ranks places worth a look.

The processing uses standard open tools to read a laser survey and grid it into maps, and every step can be rerun on any tile from the public portal. Nothing in it is secret or expensive; the limit is the data.

The two Barrington tiles analysed here hold 5.7 million laser returns between them, and each takes about 40 seconds to process.

Pipeline, step by step

01 · FetchTileDownload a LAZ or COPC tile from ELVIS into a local cache
02 · ClassifySplit returnsCount points by class and grid the ground-point density, so you know where the ground is invisible
03 · SurfacesDTM · DSM · CHMBare earth, top of canopy, and canopy height as the difference
04 · ReliefLRM · SVFA local relief model and sky-view factor bring out small bumps and hollows that hillshade hides
05 · RankAnomaliesScore canopy gaps and ground mounds, then write candidates to GeoJSON for checking on the ground
Select a step to see its output for the two tiles. Steps 05 to 07 are the second pass, described below.Data: NSW Spatial Services, CC BY 4.0 · Processing: d/rksci

The methods aren't new. They come from forest archaeology, where maps of tiny bumps and hollows reveal old roads and walls under the trees. What is specific to this search is how the candidates get ranked. A small aircraft coming down steeply into tall timber tends to leave a short, straight gap in the canopy pointing along its heading. So a gap that lines up with plausible approach headings from the radar fade should outrank a round gap from a fallen tree. Barrington has plenty of fallen trees.

The best version of this needs two looks at the forest, one before 1981 and one after, to show where a gap appeared. No pre-1981 lidar exists, but historical air photos do, and they are put to that test further on.

RAAF Macchi MB-326H A7-079 was lost on the Barrington plateau on 3 Aug 1987. Bushwalkers found it in April 1995, in cool temperate rainforest near a creek head, a few hundred metres from the forest road. Even a jet lying near a road took eight years to find.

First pass under the radar fade

Two survey tiles under the estimated radar fade were downloaded and processed. They show how little of the laser ever reached the ground.

The two 2 × 2 km tiles covering the estimated radar-fade point, near Mount Allyn and Burraga Swamp, come from NSW Spatial Services' Camberwell survey, flown with a Leica ALS80 between 7 and 23 May 2018. This is Category 2 data with level 3 classification. Stated accuracy is 0.3 m vertical and 0.8 m horizontal at 95% confidence. At 31 MB for 8 km², the download was small, the first sign of how sparse the survey is.

3546446 and 3546444, MGA zone 56, easting 354–356 km, northing 6444–6448 km. Ground elevation runs from 391 m in the valley to 968 m on the ridges.

All classes
0.71
points per m², both tiles
Ground
0.02
ground points per m²
Cells with ground
6.1%
of 2 m cells with any ground return
Canopy
23.6 m
mean height (95th percentile 40.8 m)

Four views of the same strip

Four monochrome panels of the same 2 by 4 km strip: a hillshaded bare-earth model with a river valley on the left, a speckled canopy-height map, a local relief model showing every gully, and a near-black map of ground returns with sparse white dots.
Tiles 3546446 (top) and 3546444 (bottom), stacked, north up. The bare-earth model looks crisp because it is interpolated. The right-hand panel shows how little real ground the laser actually touched. Zoom any panel with the buttons, a double-click or Ctrl-scroll, and drag to pan; the others follow, and the readout gives position and values under the crosshair.Data: NSW Spatial Services, CC BY 4.0 · Processing: d/rksci

The bare-earth model looks convincing, but it shouldn't be trusted at this scale. More than nine in ten 2 m cells hold no ground return at all, so the surface between hits is triangulated guesswork. A crumpled fuselage on a slope would fall between points far more often than it would land on one.

One of the survey's flight lines left no ground-classified points in either tile; all 808,070 of its returns were filed as overlap. Counting those within half a metre of the surface adds about 92,000 ground points, lifting the density to 0.02–0.03 per m², still nowhere near enough.

Point slicer

Loading 286,347 points ...
A 20 m-wide side view through the real point sample: every ground return (solid) and a random subset of the rest (faint). Move the slice and see how rarely the laser reaches the floor.286,347 of 5.66 million points

The anomaly ranking did what a first pass should. It found the obvious things. The top-scoring candidates in both tiles follow the river corridor and the larger gullies, which are long, open and low, exactly what a canopy-gap detector is built to notice. None of them is a credible wreck signature. A second pass needed to remove streams and their banks before ranking, and to look for features the size of a light aircraft, not a landslip.

So, on this data: public 2018 lidar can map the terrain the Cessna would have crossed, and it can flag disturbances of tens of metres, such as a fresh scar or a run of snapped crowns. It can't see a 45-year-old Cessna 210 under 40 m eucalypts. That calls for a dedicated capture, low and slow, at twenty points per m² or more, flown over a corridor that a scenario model has already narrowed down.

The top 6 candidates in tile 3546446 all sit within 400 m of easting 354.2 km, on the main valley floor. Their bearings run north–south, along the river. The second pass removes them.

Second pass with a planted aircraft

Before looking for the Cessna in the survey, a model of one was planted in it to see whether anything would come back. Almost nothing did.

An AI-assisted review of the measured data ranked fifteen analysis methods by how much each could add, and every data claim in it was checked by hand. Its first choice was the plainest. Before trying to find a Cessna in this survey, plant one in it and see whether anything comes back. That test was run, with the two cheapest fixes: a stream mask built from the terrain itself, and a filter on size and shape matched to a light aircraft rather than a landslip.

About half of every tile's returns are filed as overlap between flight lines. One flight line has no ground-classified points at all. The "building" cluster in tile 3546446 is canopy 38 m above the ground, misclassified.

Fifteen methods, ranked by expected value

Loading
A language model's own expected-value scores for the 2018 data, not for ideal data, with every data claim checked against the files. Select a row for the reasoning.Ranking: AI-assisted methods review, checked by hand

The stream mask covers 14.5% of the strip. Of the first pass's top 25 candidates, 16 now fall inside it and two more are rejected as gaps that are still open. Those were the rivers. The return count is the important number. An intact C210 outline, wings 11.2 m and fuselage 8.6 m, was dropped at 2,000 random spots under canopy taller than 20 m. Then the real 2018 laser returns that reached below its top were counted.

Plant an aircraft, count the hits

Plan view of real lidar ground returns scattered sparsely around a planted Cessna 210 outline with one return inside, beside a histogram showing 68.5% of 2,000 placements get zero returns.
A real 40 × 25 m patch of the 2018 survey under canopy averaging 38 m, seen from above. Faint dots stand for the roughly 0.6 laser pulses per square metre. Green dots are every real return classified as ground, 25 in the patch. A Cessna 210 outline, 11.2 m span and 8.6 m long, is planted among them, and one ground return falls inside it. Repeated at 2,000 random spots under tall canopy, 68.5% of placements catch no return at all, and only 9.1% catch three or more. The histogram shows returns per placement. The outline is the simple wing-and-fuselage shape used in the test. The histogram counts every return reaching below the airframe's 2 m top, not only those classified as ground.
Intact airframe
68.5%
placements with zero returns
Intact, ≥3 returns
9.1%
mean 0.81 returns per airframe
Broken field, ≥3
45.6%
40 plates over 30 × 60 m
Scar in top 25
0.0%
planted regrowth scars recovered

Planted aircraft and planted scars

Loading
Left: real 2018 returns that would have struck a planted airframe. Right: how often the second-pass detector recovers a planted canopy scar, at any rank, in the top 100 and in the top 25.Data: NSW Spatial Services, CC BY 4.0 · Processing: d/rksci

At about 0.6 laser pulses per square metre, the 2018 survey cannot see the aircraft. Even a fresh open gap would only make the top 25 two times in a hundred.

The ranked list still has ten new candidates, all canopy patches 100–190 m² and 5–15 m tall inside 20–40 m forest. Their scores sit within 12% of each other. Each is the kind of feature a treefall leaves, and the plant test says a 1981 scar among them would rank no higher. They are places a ground team could check in passing, not leads. The survey's real use is as terrain and canopy input to the scenario model, and as a guide to where a new survey dense enough to see the aircraft would pay off.

The ranked methods review was written with an AI model. Every data claim in it was checked against the files, and one was corrected: a cluster of points had been read as water, but it is noise along the tile edge.

Candidates, with and without the stream mask

Loading candidates ...
The ranking cycles between the first pass and the second pass, with and without the stream mask, and the list leaves the river. Candidates are places to check, not finds.Data: NSW Spatial Services, CC BY 4.0 · Processing: d/rksci

Every candidate, then and now

Each place the lidar flagged was checked in air photos taken before and after 1981. None shows a fresh gap in the forest after the loss, but the photos cannot see a gap as narrow as a light aircraft would cut.

Every place named so far is collected here, with the reasoning behind each assessment. Each location opens Google Maps in satellite view at a zoom that suits it. The ten lidar candidates are canopy patches the 2018 survey flagged after the stream mask and size filter. In the 2015 20 cm orthophoto all ten sit under closed canopy, and most line up with crown shadow or treefall, not open ground. None of them is a lead. The model cells are the simulation's three most likely 500 m squares. The rest are fixes, theory sites and search areas, listed for context.

Historical air photos, aligned to a modern orthophoto, give each lidar candidate a before and after: 1964 and 1972 before the loss, June 1983 (22 months after it) and October 1992 after. No candidate shows a fresh opening in 1983 that was absent in 1972. At L3 and L8 an open patch shows in 1972 and has regrown by 1983, a disturbance from before 1981, which shows the photos can resolve one tens of metres across. L6 sits beside a long, straight pale band that appears between 1972 and 1983, a track, snig line or slip, too long and straight for a light aircraft.

Synthetic patches painted into a 1964 frame put the floor at about 40 m: none of the 20 m patches was found, a third of the 40 m ones and most at 80 m. A light aircraft's swathe is about its 11.2 m wingspan, so the absence of a 1983 scar is weak evidence.

Candidate dossier

Loading the candidate dossier …

Each candidate's 160 m window, opening on the 2015 photo, with every dated photo and lidar layer beside it. Before / after lays one date over another under a divider; 1972 against 1983 spans the loss. The known wrecks follow the candidates in the same row, and their windows are about ten times larger. Photos are aligned to within the error shown; frames flagged unreliable are marked. Detail finer than 2–3 m in the old photos is interpolation.Air photos: NSW Spatial Services historical imagery · Orthophoto: NSW Imagery · Lidar: NSW Spatial Services, CC BY 4.0

Every site named in this piece

IDSitePositionBasisVerdictWhat is there
Lidar candidate
L1Lidar #1 (satellite view, opens in a new tab)32.1019°S 151.4679°ECanopy patch 148 m²LowClosed forest in every photo from 1964 to 1992. What the lidar flagged is the shadow of a tall crown, in canopy averaging 29 m.
L2Lidar #2 (satellite view, opens in a new tab)32.1096°S 151.4667°ECanopy patch 152 m²LowClosed canopy in every year. Nothing at the patch changes between the photos taken before 1981 and after.
L3Lidar #3 (satellite view, opens in a new tab)32.1255°S 151.4539°ECanopy patch 140 m²LowAn open patch just to the west shows in 1972 and has grown back by 1983. It came before the loss, and it shows the old photos can see a clearing tens of metres across.
L4Lidar #4 (satellite view, opens in a new tab)32.1248°S 151.4530°ECanopy patch 140 m²LowA bright strip shows to the south-east in 1972. The patch itself never changes.
L5Lidar #5 (satellite view, opens in a new tab)32.1013°S 151.4687°ECanopy patch 148 m²LowClosed canopy throughout. The 1964 photo could not be aligned reliably here, so the earliest frame to trust is 1972.
L6Lidar #6 (satellite view, opens in a new tab)32.0994°S 151.4552°ECanopy patch 104 m²LowBeside a long, straight pale band that opens between 1972 and 1983, widens by 1992 and is still open in the 2018 lidar. Too long and straight for a light aircraft: a track, snig line or slip.
L7Lidar #7 (satellite view, opens in a new tab)32.1018°S 151.4675°ECanopy patch 112 m²LowClosed canopy every year, under the tallest trees of the ten candidates, 31 m on average.
L8Lidar #8 (satellite view, opens in a new tab)32.1255°S 151.4542°ECanopy patch 108 m²LowThirty metres from L3, with the same clearing: open in 1972, grown back by 1983, so older than the loss.
L9Lidar #9 (satellite view, opens in a new tab)32.0962°S 151.4710°ECanopy patch 184 m²LowIn hillside shadow in every photo. Nothing changes across the dates.
L10Lidar #10 (satellite view, opens in a new tab)32.1221°S 151.4662°ECanopy patch 188 m²LowNo change at the patch. The straight lines across the 1983 frame are scratches on the film, not marks on the ground.
Model cell
M1Model cell #1 (satellite view, opens in a new tab)32.2275°S 151.4475°E0.055% of impact massModel onlyThe most likely 500 m square in the simulation, on the south edge of the 1981 search box. A model result, not a sighting.
M2Model cell #2 (satellite view, opens in a new tab)32.2325°S 151.4525°E0.052% of impact massModel onlyThe next most likely square, about 700 m south-east of M1 on the same edge of the box.
M3Model cell #3 (satellite view, opens in a new tab)32.2275°S 151.4525°E0.050% of impact massModel onlyThe square just east of M1. The top three together hold about 0.16% of the modelled mass, which shows how widely the model spreads the impact.
Fix
R1Radar fade 19:36 (satellite view, opens in a new tab)32.1152°S 151.4605°EEstimate, 45 nm from WilliamtownBelow the beamWhere Williamtown last saw the aircraft. With the antenna tilted up 1°, this is where the Cessna dropped below the beam, not where it came down; it flew on for at least three minutes.
R2Craven fix 19:19 (satellite view, opens in a new tab)32.1267°S 151.7667°EReporting pointMay be offHutchins reported here at 8,000 ft. If the vacuum pump was already failing, the directional gyro may have drifted and put him north of where he thought he was.
Theory site
S1Mt Allyn / Burraga (satellite view, opens in a new tab)32.1253°S 151.4293°EChichester–Mt Allyn theoryIn the 1981 boxThe ground under the radar fade, inside the box searched in 1981. The two 2018 lidar tiles in this piece were flown over it.
S2Gloucester Tops (satellite view, opens in a new tab)32.0613°S 151.6101°EBRG focus since 2014≈1% of massAbout 10 nm east, where searches moved from 2014. A 2016 search of a 9 km² core found nothing, and about 1% of the modelled impact mass falls in the area.
S3Dingo Tops (satellite view, opens in a new tab)31.6658°S 152.1433°EGooding, witness reportsNever searchedAbout 60 km north-east, near Caparra, from witness reports. Never searched, and the model puts no impact mass nearby.
Known wreck
K1Mosquito A52-70 (satellite view, opens in a new tab)32.0352°S 151.4488°ELost 1945, found Jan 1946Not detectedA known wreck used as a test. The position is a memorial plaque, ±500 m. No clearing shows in any photo, and no highly ranked lidar candidate fell within 500 m.
K2Hudson VH-SML (satellite view, opens in a new tab)32.1734°S 151.6706°ELost 1954, found Dec 1956Not detectedThe second test. The position is a derived point, ±5 km, so the window may miss the wreck; lidar candidates near it were no more common than chance.
Lidar candidates, model cells, radar-era fixes, theory sites and the two known wrecks. Select a row to open the place in Google Maps satellite view.

If these methods cannot find a wreck whose fate is known, a blank result for VH-MDX means little. Three known wrecks in the same ranges were put through the same tests, and none was picked out.

Nine aircraft are known to have come down in or near the Barringtons since 1945. Three of them have positions good enough, and forest similar enough, to serve as controls: the RAAF Mosquito A52-70, lost in 1945; the Hudson VH-SML, lost in 1954; and the RAAF Macchi A7-079, lost in 1987. Each was run through the same lidar ranking and air-photo comparison used for VH-MDX.

The lidar did not point at either wreck it covered, even with the area known to about half a kilometre. That was not for want of data. Both control blocks came from the same 2018 programme as the VH-MDX strip, and far more of their laser pulses reached the ground: 16% at the Mosquito site and 9% at the Hudson, against 3% under the radar fade. A planted airframe would catch three or more returns two times in three on the Mosquito block, against fewer than one in ten for VH-MDX. Still, no top-ranked candidate fell within 500 m of either site, and the hit rate around each was no better than chance.

Every control position comes from a published description, not a survey fix. The Mosquito point is a memorial plaque, good to about 500 m, and little wreckage is recorded there now. The Hudson point is derived, good to about 5 km. The Macchi site is sensitive and is held only as a coarse zone; nothing that could locate it is published here.

Known wrecks and the VH-MDX model

Loading the wreck map…

Circles show each position's stated uncertainty, and each point links to a satellite view. The Macchi site is drawn only as its coarse published search zone, with no point and no link. Dashed boxes are the 2018 lidar blocks analysed. The shading is the VH-MDX model's impact density, an illustrative model, not a search determination.Positions: published accounts · Model: d/rksci scenario simulator

The air photos fared no better. The national index holds no frame from before 1952 near any control, so the designed test, a frame before the loss against one after, could only be run for the Macchi. Across the 41% of its search zone the frames cover, the 1987 against 1992 comparison found no more new open patches than areas where nothing crashed. For the two older wrecks the earliest photos came years after the loss, and showed no clearing at the Mosquito site in any frame.

All three were found by people who were not looking for them: a stockman mustering cattle nine months after the Mosquito came down, an airline crew who saw a glint two years after the Hudson, and bushwalkers eight years after the Macchi.

Planted patches in the old photos were found none of the time at 20 m across, a third of the time at 40 m and four times in five at 80 m. A swathe cut by an 11.2 m wingspan sits well below that floor.

Knowing the area to within half a kilometre still did not make a known wreck stand out.— Paraphrase of the control findings

Lost to found

Loading the timeline…

One row per wreck, from loss to discovery. Six of the nine were found within days. The three used as controls took one, two and eight years. The rose row is VH-MDX, still not found.Sources: ABC News, ADF-Serials, ATSB, Pacific Wrecks, as listed in the end matter
SiteLostFoundHow foundGround hitsAirframe ≥3 hitsLidar top 25 within 500 mPhotos show a scarNearest track
A52-70 Mosquito19451946Stockman mustering cattle16.2%66%0 (chance 0.3)No; first frame 19529 m
VH-SML Hudson19541956Airline crew saw a glint9.2%12%0 (chance 0.3)No; first frame 19641.1 km
A7-079 Macchi19871995Bushwalkers, by chance––no lidarNo excess over backgroundWithheld
VH-MDX1981–Not found3.0%9%–None at candidates691 m

Ground hits: share of last returns classified as ground. Airframe ≥3 hits: share of 2,000 planted Cessna-sized outlines catching three or more returns. Nearest track: today's network, from the recorded position (VH-MDX: from the top lidar candidate). The VH-MDX link opens that candidate.

What is left to see

A three-bladed propeller mounted upright on a red rock with a bronze plaque, in a park under a blue sky.

Memorial to the crash of DC-3 VH-ANK in September 1948. Dolphin51, 27 Aug 2024, CC0 1.0, via Wikimedia Commons ↗

Bronze memorial plaque reading ANA DC3 Lutana VH-ANK crashed into Square Peak Nundle, 2nd September 1948, with a list of names and a Lions Club emblem.

Plaque on memorial to the crash of DC-3 VH-ANK. Dolphin51, 27 Aug 2024, CC0 1.0, via Wikimedia Commons ↗

The Lutana memorial at Nundle, a propeller from DC-3 VH-ANK mounted on a rock by the Nundle District Lions Club in 1983, and its plaque recording the 2 September 1948 crash into Square Peak with the names of those aboard. Few of the region's wrecks have openly licensed photographs; the rest are linked below, not copied.Both photographs: Dolphin51, 27 Aug 2024, CC0 1.0, via Wikimedia Commons (plaque)

A known wreck under good data stayed hidden. That sets the bar for what would find this one.

Public data has done what it can: it narrows the ground and shows how weak a blank result is. Finding the aircraft needs a denser survey over a narrower corridor, and the records that would place the known wrecks precisely.

Elevation data that matters here

DatasetHolderFormDensityUse here
Camberwell201805NSW Spatial ServicesLAZ C3, 2 km tiles0.7 pts/m², ground 0.02Two tiles analysed
Camberwell201803NSW Spatial ServicesLAZ C3, 2 km tilesSame programmeModel's likely area
UPPER MANNING 201805NSW Spatial ServicesLAZ, 2 km tilesLikely similarNorthern plateau
NSW 1–2 m DEMNSW Spatial ServicesGeoTIFF1–2 mTerrain context
2025–26 releasesNSW Spatial ServicesLAZ≥4 pts/m² specCoverage to check
State forests 2022–23NSW NRCLAZ + 1 m DEMUnknownForest fringes, unconfirmed
Australia 5 m DEMGeoscience AustraliaGeoTIFF5 mRegional modelling
LPI sweep, Oct 2013NSW LPIUnknownUnknownNever published

Lidar alone is unlikely to find this aeroplane. What it can do cheaply is shrink the map. It can rank a few dozen square kilometres of forest down to a list of coordinates, each with a reason attached, that a ground team with a GPS can check in a weekend. The 1981 search had none of that, only hundreds of flying hours and a cloud base on the ridges.

Open items: the ATSB investigation file (198101477) could not be retrieved. Several Newcastle Herald and Manning River Times pieces are known only from search snippets. The Wittenoom box location is unconfirmed. The scenario model is illustrative: it ranks ground by likelihood under stated assumptions and is not a search determination. Precise positions for the known wrecks, held by NPWS, would sharpen the control tests. Corrections are welcome.

Sources

  1. Kokalj & Hesse (2017), Airborne laser scanning raster data visualization. Zakšek et al. (2011), sky-view factor. Doneus (2013), openness.
  2. Known wrecks: ABC News (13 Nov 2026) list of crashes in the Barringtons; ADF-Serials (A3, A7 series); Pacific Wrecks and ozatwar (A52-70, VH-SML); ATSB reports 197501331, 197601498 and 198201430. Positions are from published descriptions, not survey fixes.
  3. NSW Spatial Services historical aerial photography (1952–1993 frames) and NSW Imagery orthophotos (2012–2015), used for the candidate and control comparisons.
  4. Photographs: Wikimedia Commons, authors and licences as credited. The Lutana memorial photographs are by Dolphin51, CC0. No freely licensed photograph of VH-MDX itself is known.

Press photographs

Photographs of the five men, the aircraft and the 1981 search are held by news outlets and the families. They are copyright, and Australian fair dealing for news or review is narrower than US fair use, so they are cited and linked here rather than reproduced.