Evaluation of National Digital Forecast Database Wet Bulb Globe Temperatures in North Carolina

Wet bulb globe temperature (WBGT) is a heat stress metric that has been gaining increased attention due to its ability to more comprehensively represent the impacts of heat stress on the human body. As a result, the National Digital Forecast Database (NDFD) developed an experimental WBGT product in 2020 that has since transitioned to operational use as of June 1^st, 2022. However, one challenge when validating WBGT is its dependence on black globe temperature (BGT) that must either be estimated using other meteorological parameters or measured using a BGT sensor. In this work, we leverage the unique deployment of BGT sensors across the 44 stations of North Carolina’s Environment and Climate Observing Network (ECONET) to evaluate NDFD WBGT forecasts during the heat seasons (May-September) of 2020-2023. Our specific objectives are to 1) determine how well the NDFD product captures the spatial/temporal patterns of WBGT and extreme events in North Carolina; 2) quantify the error from the individual meteorological parameters that comprise WBGT to aid future improvements to the NDFD product; and 3) determine the reduction in error from using BGT observations in place of estimated BGT in the calculation of WBGT to elucidate the possible value BGT sensors might have to the National Weather Enterprise.

To accomplish these goals, we pair observations from the ECONET stations to their corresponding valid forecast times and nearest grid cells from the NDFD gridded dataset. Evaluations are conducted separately for the first 24 hrs. (i.e., Day 1) of each forecast and hrs. 25-72 (i.e., Days 2-3). Once the data has been paired, we compute various standard evaluation metrics such as Pearson’s correlation coefficient, root mean square error, and mean bias to determine how well the NDFD WBGT forecasts perform. To determine the error from each WBGT component, we utilize a “Leave-One-Out” approach where we recompute the WBGT using the NDFD algorithm by replacing all but one forecasted parameter with ECONET observations. These recalculations are done for each forecasted component (i.e., temperature, dew point, wind speed, solar radiation, and BGT) and compared against observation derived WBGT from the ECONET stations. To determine the value of BGT sensors, we compare WBGT calculated using an estimated BGT to one in which BGT is observed. Preliminary results from the Day 1 evaluation show that NDFD WBGT biases tend to be small (~1°F or less) but this relatively good performance is the result of compensation errors from overpredicted air and natural wet bulb temperatures and underpredictions in BGTs. In terms of individual components, the solar radiation values from the NDFD product tend to have the largest errors due to the use of a Gaussian distribution for estimating hourly solar flux values. This error is persistent across all stations and is one of the driving factors in an underprediction of NDFD WBGT between sunrise and noon. More results including the error from individual WBGT components will be presented.