The state of
outdoor work
technical edition

Scope, and what this document is not#

This is a secondary synthesis of published Australian national datasets and peer-reviewed literature. It contains no primary data collection, no original survey and no new statistical modelling. Where a quantity is derived, the derivation is shown.

It is prepared by a commercial party. PocketMode builds workforce software for outdoor crews and therefore has an interest in outdoor work being understood as a distinct and hazardous category. That interest is disclosed here rather than buried, and the mitigation is that every figure is traceable to a named public source with a stated vintage, and that the limitations section names the places where the evidence does not support the argument.

Data sources and their vintages#

Mixing vintages is the most common failure mode in reports of this kind. Every source used, with its reference period and its known revision behaviour.

Estimating the outdoor workforce#

Method#

Outdoor work is not a category in the Australian and New Zealand Standard Industrial Classification, nor in the Australian and New Zealand Standard Classification of Occupations. No Australian statistical instrument asks whether work is performed outdoors.

The estimate here is an industry-division proxy: the sum of employed persons in the ANZSIC divisions where the majority of frontline work is performed outdoors. Two definitions are reported. The four-division definition covers Construction; Transport, postal and warehousing; Agriculture, forestry and fishing; and Mining. The five-division definition adds Electricity, gas, water and waste services.

Error analysis#

The proxy carries two opposing errors of unknown magnitude.

• Inclusion error (over-count). Non-field roles inside outdoor divisions: freight schedulers and distribution centre staff in transport, control room and corporate functions in mining, estimators and site administration in construction.
• Exclusion error (under-count). Field roles inside indoor divisions: emergency services within Public administration and safety, groundskeeping within Education and training, couriers within Retail trade and Accommodation and food services, and outdoor labour engaged through Administrative and support services labour hire.

Neither error has been quantified in the Australian literature. The report therefore states a range rather than a point estimate.

Independent cross-check#

The Australian Work Exposures Study provides an exposure-based rather than industry-based measurement. It was a cross-sectional computer-assisted telephone survey of 5,023 Australian workers aged 18 to 65, fielded from October 2011 to late 2012, funded by the National Health and Medical Research Council and Cancer Council Western Australia. Of those respondents, 1,100 (22 per cent) were assessed as exposed to solar radiation at work.

A note on persons versus jobs#

The ABS Labour Account publishes both employed persons (15,109,700) and total filled jobs (16,529,000) for the March quarter 2026, a gap of about 9 per cent arising from multiple job holding. This report uses employed persons throughout. Any comparison with a jobs-based figure will be inflated by roughly a tenth.

Ultraviolet radiation: effect sizes and attributable fractions#

Meta-analytic effect sizes#

The main report states that outdoor workers carry elevated keratinocyte cancer risk. The underlying pooled estimates are:

Report these as 1.77 and 1.43. The common paraphrases, "nearly double" and "about 1.5 times", round in the same direction and overstate the basal cell figure in particular.

The attributable fraction figures are modelled, not counted#

The estimate of approximately 200 melanomas and 34,000 non-melanoma skin cancers per year attributable to occupational exposure in Australia derives from attributable-fraction modelling, not from registry counts. Non-melanoma skin cancer is not a notifiable disease in Australia, so a registry count is not possible. The estimate should be cited as modelled.

Two protection datasets, not one#

Two distinct instruments are commonly conflated in Australian sun-safety reporting.

• Australian Work Exposures Study, fieldwork 2011-12. Source for: about 95 per cent used at least one form of protection; 8.7 per cent fully protected (hat, sunscreen, clothing and shade for more than half of outdoor working time); protective clothing 80.4 per cent; hats 72.2 per cent.
• National Hazard Exposure Worker Surveillance survey, 2008, Australian Safety and Compensation Council. Source for: 21 per cent of workers in direct sunlight scheduled work outside peak ultraviolet hours; 17 per cent reported that neither they nor their employer did anything to prevent sun-related harm.

These are separate surveys, four years apart, with different sampling frames. They should not be presented as one dataset.

Heat: a measurement problem before it is a safety problem#

Why the claims data cannot be read as incidence#

WorkSafe ACT reports 1,774 heat-related workers compensation claims accepted nationally between 2009-10 and 2018-19, of which 1,679 (95 per cent) arose from working in the sun rather than in hot indoor conditions.

Why heat is systematically undercounted#

Heat impairs cognitive and motor performance before it produces a diagnosable heat illness. Where heat impairment contributes to a fall, a struck-by event or a vehicle rollover, the coded mechanism of injury is the fall, the struck-by or the vehicle incident. There is no coding pathway in the Australian compensation data for heat as a contributing rather than proximate cause. The magnitude of this displacement has not been estimated in the Australian literature and should not be guessed at.

The projection basis#

The National Climate Risk Assessment, delivered by the Australian Climate Service and released 15 September 2025, projects 700,000 to 2.7 million additional working days lost annually by 2061, and a $135 to $423 billion reduction in economic output by 2063 corresponding to a labour productivity decline of 0.2 to 0.8 per cent. The ranges reflect emissions scenario and adaptation assumptions. The two years differ because they are drawn from different modelled series; this is not a transcription error.

Regulatory position as at August 2026#

There is no national heat standard, no model WHS regulation and no model code of practice for heat. Safe Work Australia publishes a guide only, and explicitly declines to set a stop-work temperature on the grounds that a single threshold cannot account for humidity, air flow, work intensity, duration, fitness and acclimatisation.

The Australian Capital Territory is the sole jurisdiction with an approved code: notifiable instrument NI2025-607, approved 6 November 2025 under section 274 of the WHS Act 2011 (ACT), commenced 14 November 2025. It sets no numeric threshold and creates no new legal obligations. New South Wales codes commencing February 2026 cover healthcare and social assistance, fatigue and respirable crystalline silica, not heat.

Regulatory instruments: a dated register#

Commencement dates, not passage dates. The two are frequently confused, and the gap is often more than six months.

Corporate penalties are frequently expressed in penalty units, which are indexed annually. Victoria's workplace manslaughter maximum is 100,000 penalty units, approximately $20.9 million at the 2026-27 unit value of $209.10. Confirm current values with the relevant regulator before relying on them.

Documented evidence gaps#

The most useful section of a report like this is the list of things it could not establish. Each gap below was searched deliberately, and the search method is stated so it can be repeated or refuted.

Gap 1: distraction and attention in non-road outdoor work#

Search method. Crossref, OpenAlex, Europe PMC and PubMed, plus the complete published-works list of AgHealth Australia, plus general web search, across combinations of distraction, inattention, mobile phone and smartphone with farm, agriculture, construction and outdoor work, restricted to Australian settings and authors.

What exists instead. Australian distraction research is confined to road and driving contexts, principally through the Monash University Accident Research Centre and the Centre for Accident Research and Road Safety Queensland. The nearest agricultural loss-of-control study, Milosavljevic et al. (2011), International Journal of Industrial Ergonomics 41(3):317-321, is a New Zealand study of quad bike operation that examines rider height, body mass, terrain, speed, distance and whole-body vibration, and does not address distraction or device use.

Why the gap matters. Vehicle incidents account for 42 per cent of Australian worker deaths and 66 per cent involve a vehicle, and 72 per cent of vehicle-involved deaths are single-vehicle events. The international driving literature establishes a strong causal relationship between handheld device interaction and crash risk. Whether that relationship transfers to low-speed agricultural and civil plant operation, or to workers on foot around moving plant, is unstudied in Australia.

Gap 2: fatigue in Australian agriculture#

Summers, Peachey and Lower (2023) conducted a narrative review of fatigue in agriculture, screening 6,031 papers and including 33. The included literature "unanimously agreed that fatigue contributes to occupational injury in agriculture and related industries", but the authors identified "a scarcity of specific literature to Australia or agriculture" as limiting any conclusion about the true relationship. Fatigue is therefore a plausible major contributor to Australian farm injury that cannot currently be quantified.

Gap 3: indoor versus outdoor splits in compensation data#

No Australian workers compensation dataset is published split by whether work is performed indoors or outdoors. Every indoor-outdoor comparison in this report, including the headline 79.8 per cent fatality share, is derived from industry division as a proxy. The mental health analysis in section 6 of the main report is inference from the documented claim drivers, not a measured finding, and is labelled as such.

Gap 4: the exposure prevalence figure is ageing#

The 22 per cent solar ultraviolet exposure prevalence rests on fieldwork completed in 2012. Safe Work Australia commissioned a 2025 Australian Worker Exposure Survey through the Social Research Centre; results are not yet published. Until they are, no current national exposure prevalence exists for any occupational hazard measured by that instrument.

Gap 5: heat as a contributing cause#

See section 5. There is no coding pathway for heat as a contributing rather than proximate cause of injury in Australian compensation data, and no published Australian estimate of the resulting displacement.

A proposed research agenda#

Section 7 names five gaps. This section sets out what closing the first of them would involve, in enough detail that a research group can judge whether it is worth their time. It is written as an invitation, not a plan we intend to execute ourselves.

8.1 What this research is not for#

There is also a structural point about what needs proving, and it is the one that makes this affordable. We do not need to establish that phone use causes worksite injuries. That is a decade-long epidemiological programme requiring tens of thousands of worker-shifts, because serious injuries are rare events and a pilot-scale study has no hope of detecting a difference in them.

This is exactly the logic of a lead indicator, which is why the Harrison and colleagues Delphi study is the right frame for the whole enterprise. A lead indicator does not have to cause the injury. It has to predict it early enough to act.

8.2 The questions, in ascending order of difficulty#

Q1 is worth answering on its own and carries no risk to anyone. Nobody has a distribution. Not a mean, not a range, not a variance. An epidemiologist cannot build an exposure model without one, and whatever the number turns out to be, it is the first number.

Q2 is where study design decides whether the project succeeds. Injuries are the wrong outcome variable at this scale. The workable outcomes are intermediate and already validated in adjacent literature: near-miss and hazard reports, vehicle telematics events such as harsh braking and speeding, response latency, and hazard-perception performance. The precedent is Cambridge Mobile Telematics, who measured behaviour change across 100,000 drivers and then modelled a 5.5 per cent bodily injury claim reduction from established behaviour-to-claim relationships rather than attempting to observe the claims directly. That is the template.

8.3 Why it is now tractable#

The instrument exists and is validated in adjacent settings. Inferring phone handling from a device's own inertial measurement unit is established method. Work published in Sensors in 2025 demonstrated real-time detection of distracted walking from smartphone accelerometer and gyroscope data alone, with personalised modelling and no camera, microphone or content capture. Cambridge Mobile Telematics holds a granted United States patent, US 11,932,257 B2 with a 2019 priority date, covering determination and scoring of driver phone distraction from three-axis accelerometer, gyroscope, screen state, proximity and tap detection, and has operated that sensing across a large insured population. The measurement problem is solved. It has simply never been pointed at a paddock.

The Australian safety system is asking for exactly this class of measure. Harrison, Peachey, Mesa-Castrillon, Lyle, Franklin and Lower are running a four-round modified Delphi study to develop lead indicators for Australian agriculture, testing 36 candidate indicators across seven ISO 45001 components. Minutes of handheld phone interaction per tracked shift-hour is a candidate of precisely that shape, and unlike most candidates it can be collected passively and continuously rather than by audit.

8.4 The framing that makes this a systems study, not a compliance study#

The most useful correction to our own initial thinking came from reading Professor Sharon Newnam's work, and it changes what the study should measure.

Newnam's argument across a decade is that work-related road safety is an organisational and systemic problem, not an individual behaviour problem, and that the field's habitual focus on the driver is a category error. Her 2021 Safety Science paper makes the case for systems-thinking surveillance of workplace road safety, using surveillance in the epidemiological sense of continuous population monitoring of a hazard. Her 2022 systematic review in the Journal of Safety Research applies the same lens to what interventions for work-related drivers have actually achieved.

Read that way, the interesting variance in Q1 is not between workers. It is between crews, tasks, rosters and sites. High exposure during machine cycle waits is a job design finding. High exposure on long lone drives is a rostering finding. High exposure at the end of a compressed weather-shortened week is a scheduling finding. Each of those points at a control an organisation can change, and none of them points at an individual to discipline.

That reframing is also what makes the study ethically and industrially viable, because it removes the individual blame frame that the electronic monitoring literature identifies as the source of most of the harm. We think it makes this a Newnam-shaped project rather than a vendor-shaped one, and she is the first person we intend to approach.

8.5 What the surrounding evidence already establishes#

A study of this kind does not start from zero. Each of the three dominant harm mechanisms in Australian outdoor work has an established phone-related impairment pathway.

The Pelicioni result is worth restating precisely because it is the closest analogue to a worksite fall. Fifty young adults walked a 10 metre perturbation walkway with progressive slip hazards. Texting reduced walking speed from 1.06 to 0.80 metres per second and increased step time variability from 80.8 to 128.9 milliseconds. Critically, texting showed no significant trunk angle effect during normal walking (P = 0.85) or under anticipated threat (P = 0.41), but did during an actual slip (P = 0.03). The impairment appears at the moment of the perturbation, not before it.

8.6 A proposed design#

A sketch, offered so a supervisor can see the shape of a feasible project rather than as a finished protocol.

Phase 1, descriptive, 6 to 12 months. Passive measurement of handheld phone interaction across outdoor work shifts in two or three industries, most usefully agriculture, civil construction and traffic management. Primary measure: minutes of handheld interaction per tracked shift-hour, gated to the shift window. Covariates at the level Newnam's framing implies: task type, crew size, lone versus crewed work, plant operation, roster pattern, schedule compression, time of day, ambient temperature. Output: the first Australian exposure distribution for this variable, and a paper publishable on its own.

Phase 2, associative. Link the exposure distribution to intermediate outcomes at crew level, per 8.2. This does not establish causation and should not claim to, but it establishes whether the association exists at a magnitude worth pursuing, and produces the effect size a later trial needs to be powered on.

Phase 3, interventional. A randomised or stepped-wedge trial of feedback design, drawing on what already works in driving. The precedent is unambiguous: private feedback with a distant consequence produced no significant change in a 2024 randomised trial, whereas feedback combined with social comparison produced a 21 per cent reduction that persisted 65 days after the intervention ended, and feedback plus a small frequent stake produced 28 per cent. The open question is whether that transfers from a private vehicle to a crew, where the social comparison is with people you will see again tomorrow.

8.7 Scope exclusion: productivity#

8.8 The ethical and industrial dimension is not an afterthought#

Any study of this kind is a study of workplace monitoring and should be designed as one. That means Human Research Ethics Committee approval, genuine and revocable consent, worker-first data access, and an explicit account of what the employer can and cannot see. It also means measuring the harms, not only the benefits: a study that reported the safety effect and not the psychosocial cost would be answering half the question, and under the New South Wales digital work systems duty it would be answering the less legally relevant half.

8.9 What PocketMode offers, and what it does not want#

• What we can contribute: a sensing platform that measures handheld interaction from inertial sensors with shift-window gating, offline buffering and hardware attestation; access to farm and civil worksites in northern New South Wales and south-east Queensland; and the founder's own farm as a first site.
• What we are not asking for: control of the research question, review rights over findings, or any restriction on publication.
• Plausible funding routes: an industry-linked PhD or an APR.Intern placement for a scoping phase, progressing to an Australian Research Council Linkage Project or a Rural Safety and Health Alliance round if phase 1 warrants it.

Why the instrument is a rostering app

A reader assessing this document as evidence is entitled to ask why a workforce scheduling company is the one publishing it, and whether that compromises the material. The honest answer is that the choice of instrument is not incidental to the research agenda in section 8. It is the reason the agenda is feasible at all.

Every design in section 8 requires exposure data at the level of the individual shift: hours actually worked, consecutive days on, crew size, location, ambient conditions, and time spent on foot. No Australian data collection currently captures any of those alongside outcomes. A purpose-built research instrument could capture them, and several international studies have tried, but they run into the same wall each time. Participants have no reason to carry the instrument once the novelty passes, so retention collapses, and the sample that remains is the sample that was most motivated to begin with.

This cuts both ways and the limitation should be stated plainly. A sample recruited through employers who have bought a workforce application is not a random sample of Australian outdoor work. It will over-represent businesses organised enough to buy software and under-represent the smallest operators, who are also, on the evidence in section 3, the most dangerous. Any published analysis has to characterise that selection rather than assume it away, and section 8.4 sets out how we propose to do so.

The second consequence is about what gets measured. Because the platform is a rostering system rather than a safety system, the intermediate outcomes proposed in section 8.2 are recorded whether or not anyone is running a study: roster lead time, consecutive-day counts, geofence-derived shift duration, weather at the site, near-miss reports, and handheld phone minutes per tracked hour. That is unusual. In most occupational research the measurement imposes a burden that itself changes behaviour. Here the measurement is the timesheet record, and the behaviour it changes is the behaviour the employer was already trying to change.

Peer-reviewed references#

Government statistical publications are listed in the sources section of the main report. The peer-reviewed literature behind both documents is listed here in full, with a line on why each one matters. Where a citation could not be fully verified it says so, because a reference list that quietly includes an unchecked entry is worse than a shorter one.

1. Anon (2025). Real-time detection of distracted walking using smartphone IMU sensors with personalised and emotion-aware modelling. Sensors 25(16):5047. doi.org/10.3390/s25165047Demonstrates detection of distracted walking from inertial sensors alone. Author list unverified, confirm before citing by name.
2. Bauer A, Diepgen TL, Schmitt J (2011). Is occupational solar ultraviolet irradiation a relevant risk factor for basal cell carcinoma? A systematic review and meta-analysis of the epidemiological literature. British Journal of Dermatology 165(3):612-25. doi.org/10.1111/j.1365-2133.2011.10425.xPooled OR 1.43 (95% CI 1.23 to 1.66) for outdoor workers.
3. Carey RN, Glass DC, Peters S, Reid A, Benke G, Driscoll TR, Fritschi L (2014). Occupational exposure to solar radiation in Australia: who is exposed and what protection do they use?. Australian and New Zealand Journal of Public Health 38(1):54-9. doi.org/10.1111/1753-6405.12174Source of the 22 per cent exposure prevalence and the 8.7 per cent fully protected figure. Fieldwork 2011-12.
4. Ebner M, Fetzer T, Bullmann M, Deinzer F, Grzegorzek M (2020). Recognition of typical locomotion activities based on the sensor data of a smartphone in pocket or hand. Sensors 20(22):6559. doi.org/10.3390/s20226559Source for the amplitude relationship in figure 5. Pocket carriage produces far larger excursions than in-hand use.
5. Fritschi L, Driscoll T (2006). Cancer due to occupation in Australia. Australian and New Zealand Journal of Public Health 30(3):213-9.Underlying attributable-fraction model for the occupational skin cancer estimates.
6. Harrison LJ, Peachey K-L, Mesa-Castrillon C, Lyle D, Franklin R, Lower T (2026). Development of lead indicators to reduce injury in Australian agriculture: protocol for a modified Delphi study. Safety 12(2):42. doi.org/10.3390/safety12020042Protocol only, no findings yet. Tests 36 candidate lead indicators across seven ISO 45001 components.
7. Hoy RF, Dimitriadis C, Abramson M, Glass DC, Gwini S, Hore-Lacy F, Jimenez-Martin J, Walker-Bone K, Sim MR (2023). Prevalence and risk factors for silicosis among a large cohort of stone benchtop industry workers. Occupational and Environmental Medicine 80(8):439-446. doi.org/10.1136/oemed-2023-108892117 of 544 screened workers with silicosis.
8. Kaviani F, Robards B, Young KL, Koppel S (2022). Using nomophobia severity to predict illegal smartphone use while driving. Computers in Human Behavior Reports 5:100158. doi.org/10.1016/j.chbr.2022.100158Attachment to the phone predicts illegal use better than the perceived risk of enforcement does.
9. Koppel S, Bugeja L, Hua P, Osborne R, Stephens AN, Young KL, Chambers R, Hassed C (2021). Do mindfulness interventions improve road safety? A systematic review. Accident Analysis and Prevention 123:88-98.Included for method. Verify before citing in detail.
10. Lower T, Peachey K-L, Rolfe M (2023). Farm-related injury deaths in Australia (2001-20). Australian Journal of Rural Health 31(1):52-60. doi.org/10.1111/ajr.129061,584 farm fatalities across twenty years. Motor vehicles 39 per cent, machinery 26 per cent.
11. Mesa-Castrillon C, Peachey K-L, Lower T (2025). Farm injury deaths and workers' compensation claims in Australia and their economic costs. Australian Journal of Rural Health 33(5):e70087. doi.org/10.1111/ajr.70087About $355 million a year in combined death and injury compensation costs.
12. Milosavljevic S, McBride DI, Bagheri N, Vasiljev RM, Carman AB, Rehn B, Moore D (2011). Factors associated with quad bike loss of control events in agriculture. International Journal of Industrial Ergonomics 41(3):317-321. doi.org/10.1016/j.ergon.2011.02.010New Zealand. Terrain, speed and vibration rather than distraction. The nearest agricultural analogue that exists.
13. Newnam S, Blower D, Molnar L, Eby D, Koppel S (2022). Applying systems thinking to improve the safety of work-related drivers: a systematic review of the literature. Journal of Safety Research 82:98-110. doi.org/10.1016/j.jsr.2022.05.006State of the evidence on interventions for work-related drivers, read through a systems lens.
14. Newnam S, Goode N, Salmon P, Stevenson M (2021). Reforming the future of workplace road safety using systems-thinking workplace road safety surveillance. Safety Science 138:105232. doi.org/10.1016/j.ssci.2021.105232Argues for continuous, systems-level surveillance of workplace road safety in the epidemiological sense. The closest published statement of what this report proposes to measure.
15. Newnam S, Watson B (2011). Work-related driving safety in light vehicle fleets: a review of past research and the development of an intervention framework. Safety Science 49(3):369-381.Foundational statement of the organisational rather than individual framing.
16. Osborne R, Horberry T, Young KL (2020). Pedestrian distraction from smartphones. Monash University Accident Research Centre, Report 349.Observed 20 per cent of pedestrians using smartphones while crossing roads, with significantly higher critical event rates.
17. Oviedo-Trespalacios O, Nandavar S, Newton JDA, Demant D, Phillips JG (2019). Problematic use of mobile phones in Australia, is it getting worse?. Frontiers in Psychiatry 10:105.Australian prevalence context for problematic phone use.
18. Oviedo-Trespalacios O, Rodriguez-Hernandez JM, Salas-Zapata W (2023). Evaluating the effectiveness of apps designed to reduce mobile phone use and prevent maladaptive mobile phone use: multimethod study. Journal of Medical Internet Research 25:e42541. doi.org/10.2196/42541Evaluates the intervention category PocketMode sits inside. Read before claiming novelty for the intervention.
19. Oviedo-Trespalacios O, Rubie E, Haque MM (2023). Is distraction on the road associated with maladaptive mobile phone use? A systematic review. Accident Analysis and Prevention 178:106872. doi.org/10.1016/j.aap.2022.106872The citation of record for the link between habitual phone use and road distraction.
20. Pelicioni PHS, Chan LLY, Shi S, Wong K, Kark L, Okubo Y, Brodie MA (2023). Impact of mobile phone use on accidental falls risk in young adult pedestrians. Heliyon 9(8):e18366. doi.org/10.1016/j.heliyon.2023.e18366Australian, n = 50. Texting impaired balance recovery during an actual slip. The closest published analogue to a worksite fall.
21. Schmitt J, Seidler A, Diepgen TL, Bauer A (2011). Occupational ultraviolet light exposure increases the risk for the development of cutaneous squamous cell carcinoma: a systematic review and meta-analysis. British Journal of Dermatology 164(2):291-307. doi.org/10.1111/j.1365-2133.2010.10118.xPooled OR 1.77 (95% CI 1.40 to 2.22).
22. Summers J-L, Peachey K-L, Lower T (2023). A narrative review of fatigue in agriculture and its impact on injury and fatality in Australia. Journal of Agromedicine 28(4):621-639. doi.org/10.1080/1059924X.2023.2215221Documents the Australian agricultural fatigue evidence gap. The direct precedent for the distraction gap in section 7.

This document synthesises publicly available Australian government and research data. It is general information and is not legal or medical advice. Prepared August 2026 by PocketMode Pty Ltd, Australia.