Application of multiple linear regression and Bayesian belief network approaches to model life risk to beach users in the UK

Christopher Stokes, Gerhard Masselink, Matthew Revie, Timothy Scott, David Purves, Thomas Walters

Research output: Contribution to journalArticle

Abstract

A data-driven, risk-based approach is being pursued by the Royal National Lifeboat Institution (RNLI) to guide the selection of beaches for new lifeguard services around the UK coast. In this contribution, life risk to water users is quantified from the number and severity of life-threatening incidents at a beach during the peak summer tourist season, and this predictand is modelled using both multiple linear regression and Bayesian belief network approaches. First, the underlying levels of hazard and water-user exposure at each beach were quantified, and a dataset of 77 potential predictor variables was collated at 113 lifeguarded beaches. These data were used to develop exposure and hazard sub-models, and a final prediction of peak-season life risk wasmade at each beach from the product of the exposure and hazard predictions. Both the regression and Bayesian network algorithms identified that intermediate morphology is associated with increased hazard, while beaches with a slipway were predicted to be less hazardous than those without a slipway. Beaches with increased car parking area and beaches enclosed by headlands were associated with higher water-user numbers by both algorithms, and beach morphology type was seen to either increase water-user numbers (intermediate morphology in the regression model) or decrease water-user numbers (reflective morphology in the Bayesian network). Overall, intermediate beach morphology can be considered the most crucial contributor to water-user life risk, as it was linked to both higher hazard, and higher water-user exposure. The predictive skill of the regression and Bayesian network models are compared, and the benefits that each approach provides to beach risk managers are discussed.
LanguageEnglish
JournalOcean and Coastal Management
Publication statusAccepted/In press - 24 Jan 2017

Fingerprint

beaches
beach
hazard
beach morphology
water
Multiple linear regression
Water
Bayesian belief networks
risk managers
parking
prediction
Hazard
tourists
automobile
exposure
coast
Bayesian networks
summer
coasts

Keywords

  • Bayesian network
  • multiple linear regression
  • lifeguard
  • rip current
  • beach users

Cite this

@article{1b3e157a04eb463195e02649fd3df672,
title = "Application of multiple linear regression and Bayesian belief network approaches to model life risk to beach users in the UK",
abstract = "A data-driven, risk-based approach is being pursued by the Royal National Lifeboat Institution (RNLI) to guide the selection of beaches for new lifeguard services around the UK coast. In this contribution, life risk to water users is quantified from the number and severity of life-threatening incidents at a beach during the peak summer tourist season, and this predictand is modelled using both multiple linear regression and Bayesian belief network approaches. First, the underlying levels of hazard and water-user exposure at each beach were quantified, and a dataset of 77 potential predictor variables was collated at 113 lifeguarded beaches. These data were used to develop exposure and hazard sub-models, and a final prediction of peak-season life risk wasmade at each beach from the product of the exposure and hazard predictions. Both the regression and Bayesian network algorithms identified that intermediate morphology is associated with increased hazard, while beaches with a slipway were predicted to be less hazardous than those without a slipway. Beaches with increased car parking area and beaches enclosed by headlands were associated with higher water-user numbers by both algorithms, and beach morphology type was seen to either increase water-user numbers (intermediate morphology in the regression model) or decrease water-user numbers (reflective morphology in the Bayesian network). Overall, intermediate beach morphology can be considered the most crucial contributor to water-user life risk, as it was linked to both higher hazard, and higher water-user exposure. The predictive skill of the regression and Bayesian network models are compared, and the benefits that each approach provides to beach risk managers are discussed.",
keywords = "Bayesian network, multiple linear regression, lifeguard, rip current , beach users",
author = "Christopher Stokes and Gerhard Masselink and Matthew Revie and Timothy Scott and David Purves and Thomas Walters",
year = "2017",
month = "1",
day = "24",
language = "English",
journal = "Ocean and Coastal Management",
issn = "0964-5691",

}

Application of multiple linear regression and Bayesian belief network approaches to model life risk to beach users in the UK. / Stokes, Christopher; Masselink, Gerhard; Revie, Matthew; Scott, Timothy; Purves, David; Walters, Thomas.

In: Ocean and Coastal Management, 24.01.2017.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Application of multiple linear regression and Bayesian belief network approaches to model life risk to beach users in the UK

AU - Stokes, Christopher

AU - Masselink, Gerhard

AU - Revie, Matthew

AU - Scott, Timothy

AU - Purves, David

AU - Walters, Thomas

PY - 2017/1/24

Y1 - 2017/1/24

N2 - A data-driven, risk-based approach is being pursued by the Royal National Lifeboat Institution (RNLI) to guide the selection of beaches for new lifeguard services around the UK coast. In this contribution, life risk to water users is quantified from the number and severity of life-threatening incidents at a beach during the peak summer tourist season, and this predictand is modelled using both multiple linear regression and Bayesian belief network approaches. First, the underlying levels of hazard and water-user exposure at each beach were quantified, and a dataset of 77 potential predictor variables was collated at 113 lifeguarded beaches. These data were used to develop exposure and hazard sub-models, and a final prediction of peak-season life risk wasmade at each beach from the product of the exposure and hazard predictions. Both the regression and Bayesian network algorithms identified that intermediate morphology is associated with increased hazard, while beaches with a slipway were predicted to be less hazardous than those without a slipway. Beaches with increased car parking area and beaches enclosed by headlands were associated with higher water-user numbers by both algorithms, and beach morphology type was seen to either increase water-user numbers (intermediate morphology in the regression model) or decrease water-user numbers (reflective morphology in the Bayesian network). Overall, intermediate beach morphology can be considered the most crucial contributor to water-user life risk, as it was linked to both higher hazard, and higher water-user exposure. The predictive skill of the regression and Bayesian network models are compared, and the benefits that each approach provides to beach risk managers are discussed.

AB - A data-driven, risk-based approach is being pursued by the Royal National Lifeboat Institution (RNLI) to guide the selection of beaches for new lifeguard services around the UK coast. In this contribution, life risk to water users is quantified from the number and severity of life-threatening incidents at a beach during the peak summer tourist season, and this predictand is modelled using both multiple linear regression and Bayesian belief network approaches. First, the underlying levels of hazard and water-user exposure at each beach were quantified, and a dataset of 77 potential predictor variables was collated at 113 lifeguarded beaches. These data were used to develop exposure and hazard sub-models, and a final prediction of peak-season life risk wasmade at each beach from the product of the exposure and hazard predictions. Both the regression and Bayesian network algorithms identified that intermediate morphology is associated with increased hazard, while beaches with a slipway were predicted to be less hazardous than those without a slipway. Beaches with increased car parking area and beaches enclosed by headlands were associated with higher water-user numbers by both algorithms, and beach morphology type was seen to either increase water-user numbers (intermediate morphology in the regression model) or decrease water-user numbers (reflective morphology in the Bayesian network). Overall, intermediate beach morphology can be considered the most crucial contributor to water-user life risk, as it was linked to both higher hazard, and higher water-user exposure. The predictive skill of the regression and Bayesian network models are compared, and the benefits that each approach provides to beach risk managers are discussed.

KW - Bayesian network

KW - multiple linear regression

KW - lifeguard

KW - rip current

KW - beach users

UR - http://www.sciencedirect.com/science/journal/09645691

M3 - Article

JO - Ocean and Coastal Management

T2 - Ocean and Coastal Management

JF - Ocean and Coastal Management

SN - 0964-5691

ER -