Diagnosing wet bulb globe temperature from numerical weather prediction model output using empirical, physics based, and machine learning methods

William Y. Y. Cheng, Jason C. Knievel, Rajesh Kumar, Gregory Roux

Discover Applied Sciences · 30 December 2025 · Météo-France (incl. CNRM), NSF NCAR / UCAR

In one sentence. Compares empirical, physics, regression and ML methods for diagnosing heat-stress WBGT from NWP output; a neural network scores best for the red flag category (CSI 0.191).

Abstract

Abstract The wet bulb globe temperature (WBGT) is a linear combination of the natural wet bulb temperature (NWBT), the black globe temperature (BGT), and the dry bulb temperature (DBT). WBGT is used as a criterion for heat stress advisories by the United States Department of Defense and in the sports community. Strenuous outdoor activities under the two most severe (red and black) WBGT flag categories can potentially lead to heat exhaustion and heat stroke. However, given the complexity of the variables, NWBT and BGT are not directly available from numerical weather prediction (NWP) output. In this study, we diagnose the WBGT from NWP output at the Phillips Army Airfield managed by Aberdeen Proving Ground in Maryland. Multiple diagnostic methods are used, including empirical formulas from the literature, a physics-based formulation, multiple linear regression (MLR), and machine learning methods such as extreme gradient boosting (XGB) and neural network (NN) models. For the red WBGT flag category, the top three methods with the highest critical success index (CSI) include a neural network model at 0.191 followed by MLR and extreme gradient boosting (XGB) with CSI at 0.150 and 0.147, respectively. Finally, for the black WBGT flag category, the top three methods with the highest CSI scores are MLR, NN model, and XGB with values of 0.349, 0.348, and 0.317, respectively.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.