Research Summary: Drowning Detection and Lifeguard Performance

RESEARCH

A meta-analysis of the key findings from thirty original, academic, and peer-reviewed papers on drowning detection performance by lifeguards in the surf, pool, and laboratory environments. We endeavour to key this list updated with any relevant publications we subsequently discover. 

 

Langendorfer, S., Pia, F. and Beale-Tawfeeq, B. 2022. Effective lifeguard scanning: a review. International Journal of Aquatic Research and Education, 13(4), pp.1-24. 

A meta-analysis of twenty-two articles on drowning detection and lifeguard performance included seventeen experimental studies, two meta-analyses and one observation study. The review concluded that a stronger focus on training experiences in visual scanning for lifeguard candidates was required. The use of hybrid delivery was to be encouraged to help new lifeguards identify drowning behaviour patterns amongst persons in the water. Three studies found that lifeguards were better than non-lifeguards at identifying hazards in the aquatic environment, whilst two studies found the inverse relationship. 

Five studies established experience and expertise as important to scan performance. The papers reviewed did not support reliance on teaching lifeguards artificial scanning patterns (e.g., geometric patterns or equivalent).

The literature reviewed reinforced the challenges with visual scanning and surveillance, including selective attention, intentional blindness, expectation violations, and vigilance degradations being responsible for false detections or failure to make a positive detection. 

Two interventions were identified as potentially helping lifeguards develop the required experience and expertise in drowning detection. Deering et al. proposed a 3-minute version of the Psychomotor Vigilance Test (PVT), and Sanchez-Lopez et al. proposed eye gaze-contingent attentional training (ECAT). 

 

Laxton, V., MacKenzie, A. and Crundall, D. 2022. An exploration into the contributing cognitive skills of lifeguard visual search. Applied Cognitive Psychology, 36(1), pp.216-227. 

Sixty lifeguards and non-lifeguards were assessed in two tasks: a Functional Field of View and a Multiple Object Avoidance task. The tests required participants to scan naturalistic videos of swimmers in a pool and identify any swimmer in distress. Ten clips were used, of which three did not contain a drowning swimmer. 

Lifeguards correctly responded to 67% of drowning targets, whilst non-lifeguards only 36% in the occlusion task. 83% of lifeguards, compared with 73% of non-lifeguards, made positive detections in the FFOV task. Lifeguards also performed better at the MOA task. 

The study hypothesised that whilst lifeguard experience may improve eye movement in scanning the visual scene, lifeguards' superior performance may be dependent on their ability to process a drowning target once fixated. 

 

Chavan, A., Dhake, S., Jadhav, S. and Mathew, J. 2022. Drowning detection system using LRCN approach. International Journal of Research in Applied Science & Engineering Technology, 10(4), pp.2980-2985. 

A scoping paper for a Long Term Recurrent Convolution (deep learning) network suited to video classification and action recognition. Videos used to validate the model were obtained from YouTube. The system learned how to detect drowning casualties with around 80% accuracy. 

 

Laxton, V., Guest, D., Howard, C. and Crundall, D. 2021. Search for a distressed swimmer in a dynamic, real-world environment. Journal of Experimental Psychology: Applied, 27(2), pp.352-368. 

Fifty lifeguards and non-lifeguards were assessed in a drowning detection task across 45 US wave pool video clips. Lifeguards made correct detections in 77% of cases compared with 64% for non-lifeguards. Lifeguards were faster to make a correct identification than non-lifeguards at 3.5 seconds vs 4.1 seconds. 

 

Meir, A., Hartmann, D. and Borowsky, A. 2021. Examining lifeguard's abilities to anticipate surf hazard instigators - an exploratory study. Safety Science, 143, 105421.

Fifteen lifeguards (pool and surf) and ten non-lifeguards were assessed in a drowning detection task across two practice clips and twenty-nine naturalistic hazard clips. Clips were taken from a camcorder located at a beach lifeguard tower, recording five days of footage. Surf lifeguards were significantly more likely to respond to breakwater, glare, rip currents and groins than pool lifeguards or non-lifeguards. Pool lifeguards fixated on waves and bathers for longer than controls. 

 

Laxton, V., Crundall, D., Guest, D. and Howard, C. 2020. Visual search for drowning swimmers: investigating the impact of lifeguarding experience. Applied Cognitive Psychology, 35, pp.215-231. 

Forty-two lifeguards and non-lifeguards were assessed in a drowning detection task across 45 staged video clips involving 3-9 swimmers and examples of active, passive and non-drowning responses. There was no significant difference between lifeguards and non-lifeguards in detection performance. Lifeguards made correct detections faster than non-lifeguards (4.2 vs 4.9 seconds). Passive drownings were detected faster than active drownings (4 seconds vs 5 seconds). As swimmer numbers increased, correct detections took longer to identify.  

 

Smith, J., Long, G., Dawes, P., Runswick, O. and Tipton, M. 2020. Changes in lifeguards' hazard detection and eye movements with experience: is one season enough? International Journal of Aquatic Research and Education, 13(1), pp.1-21. 

Forty-three UK beach lifeguards were assessed in a drowning detection task across two twenty-minute naturalistic video clips. Video clips were captured using a camera aligned to the lifeguard's field of view of a real-life beach scene with hazardous activities simulated by lifeguards. The study found no difference in the number of hazards detected by less experienced lifeguards at the beginning of the season compared with the end. Experienced lifeguards detected more hazards than less experienced lifeguards at the beginning and end of the season. 

There was no difference in the number of fixations or mean fixation duration of more compared with less experienced lifeguards either at the beginning or end of the season. There was no difference in the percentage of active responses by more compared with less experienced lifeguards. Less experienced lifeguards provided more active responses than experienced lifeguards. 

 

Vansteenkiste, P., Lenoir, M. and Bourgois, J. 2020. Gaze behaviour of experienced and inexperienced beach lifeguards - an exploratory in situ study. Applied Cognitive Psychology, 35, pp.251-257. 

Seven experienced lifeguards and nine novice lifeguards were assessed in a hazard detection task across a 45-minute real-world beach scene. Lifeguards wore eye-tracking hardware whilst lifeguarding a real-life beach scene. No correlation was found between the number of swimmers and the average fixation duration for the relevant or irrelevant areas. Less experienced lifeguards looked less at the swimming area as swimmer numbers increased. The average fixation frequency did not differ between the two lifeguard groups. Experienced lifeguards fixated on targets in the task-relevant region longer than non-experienced lifeguards. Both lifeguard groups spent about 33% of their time scanning the task-irrelevant region. 

 

Lanagan-Leitzel, L. 2020. Does incident severity influence surveillance by lifeguards in aquatic scenes? Applied Cognitive Psychology, 35, pp.181-191. 

Three lifeguards were assessed in a drowning detection task across twenty naturalistic video clips, each two minutes long, and a second set of 100 video clips, between 3-20 seconds long. The lifeguards were inconsistent in their reported severity scores for the incidents witnessed. The events least likely to be monitored were similar in salience to the entire scene. Those most likely to be monitored were lower in salience than the entire scene. How long an event is monitored was found to be determined by the number of people on the scene, with greater numbers requiring longer fixations. 

 

Laxton, V. 2019. Testing and training lifeguard visual search. Thesis: Nottingham Trent University, pp.1-329. 

One hundred and nineteen participants, including forty-two lifeguards, forty non-lifeguards, twenty-six lifesavers, and eleven lifeguard trainers, were assessed in a drowning detection task across forty-five naturalistic video clips of a US wave pool. 

Non-lifeguards made more false detections than lifeguards, lifeguard trainers, and lifesavers. Non-lifeguards were also the least sensitive to targets. Active drownings were responded to more slowly than passive drownings. Lifeguards made correct detections faster than non-lifeguards. More experience was associated with faster and more accurate detection performance. 

 

García, S., Villar, M., Manzano, J., García, J., Jiménez, M., Dios, R. and Gómez, C. 2019. The use of pupillometry in aquatic lifesaving. Vigilance as a key factor in drowning. Retos (Madrid), 36, p.461-467. 

A narrative review of the processes around lifeguard vigilance. 

 

Laxton, V. and Crundall, D. 2018. The effect of lifeguard experience upon the detection of drowning victims in a realistic dynamic visual search task. Applied Cognitive Psychology, 32(1), pp.14-23. 

Thirty lifeguards and thirty non-lifeguards were assessed in a drowning detection task across forty-five naturalistic video clips of a swimming pool shallow end with pool users simulating hazard scenarios. Lifeguards identified more drowning targets than controls. Active drowning targets were identified more often than passive targets. Lifeguards responded nearly one second faster than controls (3.5 vs 4.4 seconds). The response time gap between lifeguards and non-lifeguards narrowed as pool user numbers increased. 

 

Koon, W., Rowhani-Rahbar, A. and Quan, L. 2017. The ocean lifeguard drowning prevention paradigm. How and where do lifeguards intervene in the drowning process? Injury Prevention, 24, pp.296-299. 

Analysis of 423,071 unique lifeguard interventions recorded in post-intervention records between 1 January 2015 and 31 December 2016. 54% were preventative actions, 32% were interactions with members of the public, 6% were surfer warnings, and 1.9% were rescues. 73% of preventative actions and 70% of rescues occurred during the summer months of June, July, and August. 57% of preventative actions and 64% of rescues occurred between 13:00-16:00. 41% of preventative actions and 46% of rescues occurred on weekends. July had the highest rescue rate of 6.48 per 100,000 and September 6.3 per 100,000. There were 1158 co-occurring preventative actions and rescues for the same lifeguard. 

 

Sivakami, S., Janani, K., Janani, K. and Ranjana, R. 2017. Drowning prevention system - At sea level. 2nd International Conference on Computing and Communications Technologies, IEEE Xplore, pp. 370-373. 

A wearable drowning detection system for use on beaches. The system utilises water submersion detectors for those playing on the shore and is tracked using GPS. The water detector activates after ten seconds of contact with water. For those who swim, oxygen sensors are provided in armbands. The oxygen sensor activates after ten seconds of submersion. Swimmers are also equipped with a wristband with a floating aide. 

 

Ramos, W., Anderson, A. and Fletcher, A. 2015. Prevalence of inadequate hydration levels in aquatic safety personnel. A pilot study. International Journal of Aquatic Research and Education, 9(3), pp.329-341. 

Fifty-five lifeguards had their hydration levels assessed across seven test sites, including three waterparks and four traditional pools, over 2-3 hours. 24% of males and 29% of females were dehydrated. 64% of those working at indoor pools and 50% at outdoor pools were dehydrated. Dehydration is recognised as resulting in increased fatigue and decreased cognitive functions such as concentration and memory. 

 

Lanagan-Leitzel, L., Skow, E. and Moore, C. 2015. Great Expectations: Perceptual challenges of visual surveillance in lifeguarding. Applied Cognitive Psychology, 29, pp.425-435. 

A narrative analysis of the history and approaches to measuring lifeguard performance in drowning detection. 

 

Aminaee, A., Shahabi Kaseb, M. and Stiri, Z. 2014. Design and construction of lifeguards' vigilance in surveillance questionnaire. pp.105-126. 

Published only in Urdu. 

 

Hunsucker, J. and Davison, S. 2013. Scan time goals with analysis of scan times from aquatic facilities. International Journal of Aquatic Research and Education, 7(3), pp.227-237. 

Fifteen thousand seven hundred thirty-seven recorded lifeguard observations between 2005-2010 were analysed. 78% of lifeguards detected the simulated manikin 'casualty' in the pool within 30 seconds. The average scan time ranged between 19.18-24.4 seconds. 

 

Lanagan-Leitzel, L. 2012. Identification of critical events by lifeguards, instructors, and non-lifeguards. International Journal of Aquatic Research and Education, 6(3), pp.203-2014. 

Fifty-nine participants, including ten lifeguard instructors, seventeen lifeguards, twelve lifeguard candidates, and twenty non-lifeguards, were assessed in a drowning detection test involving twenty naturalistic videos, each two minutes in length. Lifeguard candidates reported more events at the end of the semester than at the beginning, suggesting that experience improved their hazard perception. 

 

United States Lifeguard Standards. 2011. An evidence-based review and report by the United States Lifeguard Standards Coalition. pp.1-67. 

A literature review that included scanning found that lifeguards developed their scanning strategies, and scanning became more efficient, albeit not more effective, with practice. Increased target homogeneity decreases the number of positive detections. A decrease in the number of targets increases detection rates. Sleep apnea, noise, the consumption of sugared beverages, long periods of supervision, a lack of interventions, tiredness, high environmental temperature, high humidity,  dehydration, and drug use all impaired detection performance. Small doses of caffeine were found to increase detection performance. 

 

Schwebel, D., Jones, H., Holder, E. and Marciani, F. 2011. The influence of simulated drowning audits on lifeguard surveillance and swimmer risk-taking at public swimming pools. International Journal of Aquatic Research and Education, 5(2), pp.210-218. 

The study found that simulated drowning audits may be an effective strategy to improve lifeguard surveillance and reduce swimmer risk-taking. However, the duration of the effect was only assessed up to one month after the visit. 

 

Hunsucker, J. and Davison, S. 2008. How lifeguards overlook a victim: Vision and signal detection. International Journal of Aquatic Research and Education, 2(1), pp.59-74. 

A narrative review of scanning patterns and drowning detection formed the basis for lifeguards to be taught scanning strategies (such as scanning in geometric patterns). Researchers have since tested and found no evidence that scanning strategies help improve drowning detection rates. 

 

Schwebel, D., Lindsay, S. and Simpson, J. 2007. Brief report. A brief intervention to improve lifeguard surveillance at a public swimming pool. Journal of Pediatric Psychology, 32(7), pp.862-868. 

An observational study of interventions in a swimming pool before and after a safety briefing to the lifeguard team based on the Health Belief Model. Lifeguards post-intervention were less distracted and spent more time scanning the pool than before the intervention. 

 

Lu, W. and Tan, Y. 2004. A vision-based approach to early detection of drowning incidents in swimming pools. IEEE Transactions on Circuits and Systems for Video Technology, 14(2), pp.159-178. 

A camera-based drowning detection system was proposed. The background scene of the entire pool area is modelled using a multivariate Gaussian mixture model of which the Gassian element consists of one scene class, colour distribution. Hue, saturation and value colour values of pixels belong to two different scene classes. A mean-shift clustering algorithm is used to identify the dominant scene classes of the monitored pool area. Background pixels are grouped into different scene classes by assigning each pixel to the class whose mode has the most similar colour to that of the pixel. A new swimmer is assumed if a sizeable connected region of non-background pixels has been detected. Turbulence can be separated from the background pixels and that of swimmers by contrasting the light intensity reflected by a swimmer compared with turbulence (the latter being much greater). 

The system stops tracking where swimmers collide until they become separate again. Test 1 is for moving speed. Test 2 identifies a circular shape instead of an ellipse as a swimmer in a verticle position and tracks motion to determine if the swimmer is in danger. Test 3 measures rapid changes in swimmer segmentation over a short duration, indicating to the system the struggles of an active drowning swimmer. Where Tests 1 and 2 are satisfied, the system will flag a possible drowning swimmer. Where Tests 1-3 are satisfied, alarms will be triggered. The system produced no false alarms or missed alarms out of 25 tests on the training set. On the test set, the system produced two false alarms and no missed alarms out of 69 tests on the test set. 

 

Harrell, A. and Boisvert, J. 2003. An information theory analysis of duration of lifeguards' scanning. Perceptual and Motor Skills, 97, pp.129-134. 

Six lifeguards across three indoor swimming pools were observed providing surveillance over pool users. Scan duration increased as the number of children in the pool increased, but only over the children, not their accompanying adults. 

 

Unknown. 2001. Lifeguard vigilance. Bibliographic study. Applied Anthropology, pp.1-34. 

A study of the factors that affect drowning detection performance, including the characteristics of the task, boredom, noise, sleep deprivation, air temperature, and humidity. 

 

Harrell, A. 2001. Does supervision by a lifeguard make a difference in rule violations? Effects of lifeguards' scanning. Psychological Reports, 89, pp.327-330. 

Five lifeguards at five separate indoor swimming pools were observed every five minutes for sixty minutes. Twenty-six rule violations occurred over the five hours observed. The most common rule violation was the misuse of equipment. More scans occurred later in the day and when child-to-adult ratios were low. More scans were associated with fewer, rather than greater, numbers of children. 

 

Fenner, P., Lealhy, S., Buhk, A. and Dawes, P. 1999. Prevention of drowning. Visual scanning and attention span in lifeguards. The Journal of Occupational Health and Safety - Australia and New Zealand, 15(1), pp.61-66. 

A narrative review of the factors that affect drowning detection performance, including boredom, gender, age, sleep disturbances, time of day, stress, workload, and diet. 

 

Harrell, A. 1999. Lifeguards' vigilance. Effects of child-adult ratio and lifeguard positioning on scanning by lifeguards. Psychological Reports, 84, pp.193-197. 

Four lifeguards were observed providing surveillance at three indoor swimming pools. Each lifeguard was observed twelve times over sixty minutes. An increase in the number of children resulted in a decrease in the number of scans. Observations later in the day were associated with fewer scans. Being positioned in the lifeguard tower resulted in more frequent scanning than patrolling on foot. 

 

Hansucker, J. and Jen-Gwo Chen, J. 1994. An examination of fatigue and vigilance during lifeguard training activities. Advances in Industrial Ergonomics and Safety VI, Taylors & Francis, pp.209-212. 

 

A narrative review of the studies on fatigue and its effect on lifeguard vigilance, much of which is now superseded. 

 

 

Citation. Jacklin, D. 2023. Research Summary: Drowning Detection and Lifeguard Performance. Water Incident Research Hub, 6 January; updated 29 April.