Changes in hourly rainfall return levels due to temperature shifts: parametrization of the TENAX model across different climates

Extreme sub-daily precipitation is difficult to anticipate and has large impacts. It causes flash floods, urban floods and debris flows, resulting in casualties and damage to infrastructure, homes, and livelihoods. As temperatures increase, more moisture can be stored in the atmosphere, which means that there is potential for larger extreme precipitation events. Short-duration precipitation extremes are already increasing in magnitude, and return levels (i.e., magnitudes associated with low annual exceedance probabilities) are changing. Quantifying these return levels for the coming years is critical for decision making and for defining insurance premiums. However, the methods we typically use to derive rainfall return levels do not include the physics driving the processes, so they are not suitable for predicting future extremes. The TENAX model was recently proposed to address this issue. It is split into a temperature model which represents the temperature during precipitation events, and a magnitude model which incorporates the scaling rate of precipitation with temperature. It has been successfully applied to mid-latitude regions, but we do not currently know how it should be parameterized for other climates with different temperature conditions and different processes behind heavy precipitation. Here, we explore the parametrization of the model over a range of climate regions using a global hourly rainfall dataset (GSDR) and ERA5-land reanalysis temperature data. 

We find that the parameterisations of both models are highly region-dependent, varying with both longitude and latitude. In many cases, the temperature distribution is not well represented by the generalised normal distribution used so far. Rather, a skewed distribution or the combination of multiple distributions is required. Some regions present a temperature dependence of the tail heaviness of the intensity distribution, with a strong spatial dependence of its magnitude and sign. This means that quantiles associated with different exceedance probabilities may change with temperature at different rates.