Extreme precipitation temperature scaling: disentangling causality and covariation

Warmer temperatures are expected to cause more intense rainfall, primarily due to the rise in atmospheric moisture at the rate of 7%/K, as indicated by the Clausius-Clapeyron (CC) equation. To evaluate this effect, studies use a statistical approach known as precipitation-temperature scaling that involves fitting an exponential regression between observations of extreme rainfall events and local temperatures, resembling how saturation-vapor pressure scales with temperature. However, the estimated sensitivities (also called scaling rates), exhibit notable deviations from the CC scaling (7%/K). These rates remain mostly negative in the tropics as the rainfall extremes exhibit a general monotonic decrease with temperature and hook-shape structures in most parts of tropics and mid-latitudes. Here we show that most of the variability in the observed scaling rates arises from the confounding radiative effect of clouds associated with rainfall events. Clouds substantially reduce the net radiative heating of the surface during the storms by up to 100 W/m2 in the tropics, leading to the cooling of surface temperatures by up to 8K. This cloud-induced cooling results in a covariation between precipitation and local temperature, inducing a two-way causality in the observed scaling rates. To isolate this cooling effect, we used a thermodynamically constrained surface energy balance model and force it with radiative fluxes under both "clear" and "cloudy" sky conditions. We then quantified the changes in surface temperatures due to clouds and remove it from temperature observations during rainy days. After removing this effect, we found positive scaling across the global land areas, closely aligning with CC rates of 7%/K. We demonstrate that cloud radiative effects alone can explain the observed negative and hook-shaped relationships found in precipitation-temperature scaling. Our findings imply that projected intensification of rainfall extremes with temperature by climate models is consistent with observations after the cloud-cooling effect is corrected for. Our results emphasize on making a clear distinction between causality and covariation by explicitly separating the temperatures before the rainfall event that are shaped by less clouds from temperature during the rainfall event which include clouds. This adds a crucial effect to the debate of interpreting observed precipitation - temperature scaling rates. Furthermore, our methodology of removing cloud effects on temperatures can be extended to estimate climate sensitivities from observations beyond precipitation extremes.