Author(s): TAKAHASHI, K., HAYASHI, R.2, KIM, H.Y.3, INASHIMA, S.4, Institution: COMMUNITY AND SOCIAL WELFARE , Country: JAPAN, Abstract-ID: 372

INTRODUCTION:
In Japan, the Wet-Bulb Globe Temperature (WBGT) is widely adopted as the standard metric for heatstroke prevention. In the sporting context, the guidelines formulated by the Japan Sport Association are becoming increasingly utilised. These guidelines specify criteria across five levels; for instance, a WBGT of 31°C or higher constitutes “Exercise Prohibited”, whilst a WBGT of 28°C–31°C is designated as “Severe Warning”.
However, discrepancies arise between the WBGT values provided by the Ministry of the Environment at observation stations (“provided values”) and the actual field-measured WBGT (“observed values”).
This study aimed to propose a method utilising posterior distribution estimation via Bayesian statistical modelling (“Bayesian estimation”) to probabilistically predict the WBGT category at the fields based on the provided values.
By employing sequential Bayesian updating to integrate past data (2024) with the latest data (2025), this study constructed a prediction model of higher reliability.
METHODS:
To estimate the conditional probability of the field observation falling into a predetermined category, given that the value provided by the Ministry of the Environment is in a specific category, a Bayesian statistical model was constructed using a Dirichlet distribution as the prior and a multinomial distribution as the likelihood function.
Specifically, the joint frequencies of the provided and observed values across all five levels were assumed to follow a multinomial distribution. For the prior distribution of the occurrence probability parameters for each category, a uniform Dirichlet distribution (uninformative prior, α=0.01) was set, assuming no specific bias.
Markov Chain Monte Carlo methods were employed for parameter estimation. Based on the calculated posterior distribution, the expected value (posterior mean) of the conditional probability and the 95% Bayesian credible interval, indicating the uncertainty of the estimation, were derived.
RESULTS:
Although the Ministry of the Environment’s observation indicated a “Severe Warning”, the probability of the field environment reaching the “Exercise Prohibited” level was extremely high at 86.3% (95% credible interval [CI]: 77.3–93.5%). By contrast, the probability of the field being at the “Severe Warning” level was 12.1%. This suggests that relying solely on the values provided by the Ministry of the Environment entails a high risk of overlooking critical risks.
CONCLUSION:
The presentation of such conditional probabilities allows for the evaluation of thermal environmental risks using intuitive probabilities, such as understanding that, “If tomorrow’s forecast is “Severe Warning”, there is an 86.3% probability that the observation site will be at the “Exercise Prohibited” level”. As this method is applicable with only continuous observational data, it is anticipated to be applicable in sports facilities and educational settings.
This work was supported by JSPS KAKENHI Grant Number JP24K14570.