Assessing the Uncertainty from Methodological Choices in Multivariate Heat Indicator Calculations

As climate change intensifies, extreme heat event projections are becoming increasingly crucial tools for near-term and longer-term public health planning and intervention. Various diagnostics are used to quantify the risk of extreme heat exposure, including single-variable indicators that focus solely on daily maximum or minimum temperature, and multivariate indicators including the Heat Index (HI) and Wet Bulb Globe Temperature (WBGT) that combine temperature with other factors such as humidity and wind speed. However, because these multivariate heat indicators are not typically direct outputs of climate models, certain methodological choices and approximations are needed to derive or estimate the indicators of interest. Additionally, bias correction / statistical downscaling of climate model outputs is often performed to generate information at the local scale, deemed more suitable for applied research studies that support decision-making. These additional value-added processing steps can introduce underappreciated uncertainties on top of the inherent uncertainties in large-scale climate models and scenario selection.

Here, to illustrate choices and challenges associated with generating local-scale projections of multi-variate indicators, we quantify a set of uncertainties in the calculation of multivariate daily maximum HI. HI is typically calculated using co-varying temperature and humidity. However, in climate models that lack hourly data for these variables, daily maximum temperature and minimum humidity have been used to estimate daily maximum HI (HImaxest). For a given day, this approximation can either exactly match the true daily maximum HI or underestimate it. We analyze hourly observations of temperature and humidity from selected weather across the US to calculate two versions of HI: HImax24, computed as the maximum of 24 hourly HI, and HImaxest, derived from daily maximum temperature and minimum humidity. We compare the frequency of extreme heat days at each station using these different HI calculations. We also quantify the impact of different bias correction techniques on HImaxest. Using downscaled data from the same station locations, we compare univariate versus multivariate and direct versus indirect bias correction approaches to understand how they affect HI values and extreme heat index day counts. The relative magnitude of uncertainties associated with these approximation and data processing choices is compared with the more commonly considered scenario and choice of climate model uncertainties.