Comparative accuracy of instrumented and app-based WBGT for heat-stress screening in agricultural environments in Taiwan.

This pilot research assessed the reliability of wet-bulb globe temperature (WBGT) screening tools and examined how different microenvironments might affect worker-level WBGT in Taiwanese agriculture. In July 2024, WBGT was recorded at six locations using a QUESTemp monitor with sensors positioned at the head, abdomen, and ankle (weighted 1:2:1), along with portable Kestrel and AZ (AZ) devices and the AIHA® app. The AZ devices were evaluated for within-site variations across various work environments, including soil, shade, ground-cover cloth, grassland, asphalt, and ridges. Agreement analyses and Bland-Altman plots were used to compare methods. Median WBGT values ranged in the low to mid-30s°C, with four sites having 50% or more of measurements at or above 32 °C. The Kestrel device demonstrated the closest agreement (bias +0.32 °C; RMSE 0.80 °C), the AZ device tended to underestimate (-0.94 °C; RMSE 1.28 °C), and the AIHA app showed the largest errors (bias -0.83 °C; RMSE 1.82 °C). Microenvironmental factors caused WBGT shifts between -1.25 and +2.87 °C, with decreases caused by working in shaded areas and increases caused by working in reflective ground covers. When screened against the operational WBGT benchmarks (30 °C and 32 °C), the QUESTemp monitor frequently recorded values at or above these thresholds, indicating high-heat conditions. However, the systematic underestimation by the wearable AZ device resulted in a significantly lower detection rate of the high-heat conditions, highlighting the risk of false-negative safety assessments when using uncorrected personal monitors. Overall, the portable/wearable Kestrel and AZ devices showed closer agreement with the QUEST than the weather-station-based app estimates. Given the limitations of regional weather data observed in this pilot study, effective heat-risk management in agriculture requires site-specific monitoring to account for microenvironmental variations, rather than relying solely on app-based estimates.