Situation Awareness and Maritime Accidents

emerging Updated: 2026-06-07
maritime-safety human-element situation-awareness accident-analysis leximancer endsley merchant-shipping human-error automation

Situation Awareness and Maritime Accidents

Status: emerging
Last updated: 2026-06-07
Sources: Human_Error_In_Maritime_Operations_Analyses_Of_Acc.Pdf
Tags: [maritime-safety, human-element, situation-awareness, accident-analysis, leximancer, endsley, merchant-shipping, human-error, automation]

Summary

Grech, Horberry and Smith (2002) argue that loss of Situation Awareness (SA) is a primary factor underlying maritime casualties, and that automated text analysis can measure it at scale. Analysing 177 merchant-shipping accident reports against a three-level SA taxonomy adapted from Endsley, they find that 71% of human errors were SA-related, most of them failures at the lowest level — failing to perceive available information. They also validate the Leximancer machine-learning tool against manual coding, finding the two comparable, which makes large-scale SA analysis of accident reports practical. The study brings the aviation concept of SA into the maritime domain and warns that increasing automation can itself erode SA.

Body

Context

Grech, Horberry and Smith (2002) examine the role of lost SA in maritime accidents and, in parallel, test a method for detecting it. Their corpus is 177 public-domain accident reports from eight countries (accidents 1987–2001), coded with a taxonomy collapsed from Endsley's (1995) model into three levels — failure to perceive, to comprehend, and to project — chosen for reliability and for compatibility with automated analysis (PDF p. 1). Within this knowledge base the article sits in the human-element strand of maritime safety alongside Simultaneous Tasks Maritime Accidents, with which it shares an author (Michelle Grech) and a thesis: that maritime casualties are rooted in the work and cognitive system rather than in isolated individual error. Where the simultaneous-tasks study traces socio-cultural drivers, this one quantifies a cognitive mechanism — SA breakdown — and supplies a measurement method.

Key Points

SA was under-studied at sea. The authors note that more than 75% of ship accidents were attributed to human and organisational error (IMO, 1994), yet SA research had been concentrated in aviation and medicine; very little maritime work existed, even though Endsley held SA to be equally important in any complex, dynamic environment (PDF p. 1). The study adapts Endsley's taxonomy into three levels and applies it to casualty reports.

The method pairs manual and machine coding. Two raters — one domain expert, one human factors specialist — independently hand-coded a 26-report pilot subset, which then trained the Leximancer tool; Leximancer uses machine learning to map text onto conceptual dimensions rather than keywords (PDF p. 2). The manual coding stage found that 71% of human errors were SA-related, distributed as 58.5% level 1 (perception), 32.7% level 2 (comprehension), and 8.8% level 3 (projection) (PDF p. 3). The dominance of level-1 failures means most SA breakdowns were failures to perceive information that was available — for example a mate preoccupied with private phone calls instead of monitoring course, speed and position (PDF p. 2).

The tool validated well on the controlled subset but degraded at scale. On the 26 abridged, trained-upon reports, Leximancer matched the manual coding with no significant difference (paired t-test, p > 0.05), with precision above 84% across all three SA levels and total recall of 89%. When the full, un-abridged 177-report set was analysed, precision fell to about 48%, which the authors attribute to repetitive and irrelevant text in the un-edited reports (PDF p. 3). They are candid that the favourable subset results were helped by a small sample, abridged reports, and classifying on the trained text.

The study also flags inconsistent reporting and an automation tension. Maritime accident reports lack a standard format and often capture proximate technical causes at the expense of human-factors causes, making it hard to link rare casualties to frequent incidents and near misses (PDF p. 4). Looking forward, the authors note their related work (Grech & Horberry, 2002) found that rising technology levels can themselves cause loss of SA, and that the decision whether to rely on automation is among the most important a crew makes — over-reliance leads to failures to monitor, while under-reliance defeats automation's purpose (PDF p. 4).

Conclusion

Grech, Horberry and Smith (2002) conclude that lack of SA is a serious and quantifiable problem in the maritime domain: a large share of human error in their corpus reduces to "loss of SA," with a distribution close to that found in aviation, which they take as evidence that SA breakdown is a primary cause of maritime casualties. Methodologically, they conclude that tools such as Leximancer can analyse accident reports quickly and accurately enough to be useful, while cautioning that report standardisation and larger, un-abridged validation are needed. The recommended directions — common reporting formats and study of automation's effect on shipboard SA — set an agenda this knowledge base can track.

  • Simultaneous Tasks Maritime Accidents — shares author Grech and the human-element thesis; multitasking as a driver of the task deviation that can precede SA loss
  • See also (cross-KB): crew-visual-behaviour-control-cabin — Yu et al. (2021) on crew visual behaviour and the "keyhole effect" in ship centralized control cabins (owned by human-centered-design-kb); a concrete display-level mechanism for the perception-level SA failures quantified here

References

Endsley, M.R. (1995) 'Measurement of situation awareness in dynamic systems', Human Factors, 37(1), pp. 65–84. To be validated.

Grech, M.R. and Horberry, T. (2002) 'Human Error in Maritime Operations: Situation Awareness and Accident Reports', 5th International Workshop on Human Error, Safety and Systems Development. Newcastle, Australia. To be validated.

Grech, M.R., Horberry, T. and Smith, A. (2002) 'Human Error in Maritime Operations: Analyses of Accident Reports Using the Leximancer Tool', Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 46(19), pp. 1718–1721. doi: 10.1177/154193120204601906. grech2002human

International Maritime Organization (1994) Better standards, training and certification: IMO's response to human error. IMO News. To be validated.

Open Questions

  • The favourable validation rested on a small (26-report), abridged subset classified on the trained text; precision halved on the full set. How well does machine SA coding generalise to large, un-edited casualty corpora? The authors call for, but do not provide, broader validation.
  • The corpus is accident reports only; near misses and normal operations (where most SA losses occur without consequence) were not captured. The true prevalence of SA breakdown at sea is therefore underestimated by an unknown amount.
  • The automation–SA tension is raised but not measured here. How increasing bridge automation changes the distribution of SA-level failures is an open thread, linking to the automation literature held in sibling knowledge bases.
  • Cross-domain link: the SA construct and its measurement are developed further in the human-factors corpus (Endsley's taxonomy). A cross-KB reference to the eye-tracking and human-centered-design treatments of situation awareness would connect this maritime application to its cognitive foundations.