Future increases in extreme precipitation exceed observed scaling rates

Models and physical reasoning predict that extreme precipitation will increase in a warmer climate due to increased atmospheric humidity. Observational tests using regression analysis have reported a puzzling variety of apparent scaling rates includingstrongrates inmidlatitude locationsbutweakor negative rates in the tropics. Here we analyse daily extreme precipitation events in several Australian cities to show that temporary local cooling associated with extreme events and associated synoptic conditions reduces these apparent scaling rates, especially in warmer climatic conditions. A regional climate projection ensemble for Australia, which implicitly includes these e ects, accurately and robustly reproduces the observed apparent scaling throughout the continent for daily precipitation extremes. Projections from the samemodel show future daily extremes increasing at rates faster than those inferred from observed scaling. The strongest extremes (99.9th percentile events) scale significantly faster than near-surface water vapour, between 5.7–15% C depending onmodel details. This scaling rate is highly correlated with the change in water vapour, implying a trade-o between a more arid future climate or one with strong increases in extreme precipitation. These conclusions are likely to generalize to other regions. Intensification of precipitation extremes is expected to occur in a warmer climate as the saturation water vapour pressure increases with temperature at a rate of roughly 6% C at the surface or 7–8% C in the column integral, as governed by the Clausius– Clapeyron (C–C) relation. Increase of extreme precipitation has been confirmed in numerous modelling studies, but the rate of this remains uncertain. Global climate models (GCMs) project daily precipitation extremes to increase at below the C–C rate in the extra-tropics, but there is large model range in the tropics, with some models predicting super C–C increases in some regions. GCMs poorly represent the distribution of rain rate and may underestimate the sensitivity of precipitation extremes in response to warming. Increases of daily and hourly precipitation extremes in convection-permittingmodels have generally shown rates around 7% C (refs 3,9,11), but a variety of scaling rates (including super C–C) have also been found depending onmicrophysics scheme and temperature. However, most convection-permittingmodel studies focusing on precipitation extremes have been conducted only in small regions that cannot capture synoptic-scale organizations. It would be desirable to constrain the dependency of extreme precipitation on temperature using observational records. Recently a simple ‘binning method’ has been widely used to relate daily or sub-daily extreme precipitation to local temperature. On the basis of this approach, different scaling behaviour has been found in different places. For hourly up to six-hourly extreme precipitation, increases at or above the C–C rate have been found in the Netherlands, Switzerland, Germany, the Mediterranean, the UK, southern Australia, North America and China, while in northern Australia strong negative rates have been observed. For daily extremes, monotonically positive scaling has been found at high latitudes and monotonically negative scaling in the tropics. This scaling also shows distinct seasonal characteristics, with positive rates in winter and negative rates in summer in Europe. The decrease in wet-event duration and decline in relative humidity at high temperatures have been proposed to explain the negative scaling relations. Differences in scaling relations have also been attributed to seasonal dependencies and changes in the frequency of large-scale versus convective rain. In contrast, long-term trends in observed annual maximum daily precipitation over land suggest higher scaling rates in the tropics than in the midlatitudes, although uncertainties arise from natural variability and limited data availability in the tropics. Higher scaling rates in the tropics have also been projected by some GCMs. Thus, the scaling relations remain debatable, as contradictory results have been found especially in the tropics. Since precipitation extremes do not necessarily scale with local surface air temperature and potential confounding factors discussed below that may affect temperature and precipitation are ignored in the binning method, the question remains: Do conclusions drawn from this method accurately represent the influence of warming on extreme precipitation? We start by estimating the spatial distribution of ‘apparent scaling rate’, or regression slope between 99th percentile wetday precipitation and daily mean temperature as reported in previous studies using the binning method. Daily precipitation and near-surface air temperature data are from the Australian Water Availability Project (AWAP). Similarly to previous studies, we find negative scaling rates of up to −50% C over northern Australia and positive rates of up to +16% C over southern Australia (Fig. 1a). In winter there are stronger positive scaling rates in southern Australia (Fig. 1b). In summer, negative scaling rates prevail in most regions, even in places where there are positive scaling rates when calculated over the entire year (Fig. 1c). To find out why there are opposite trends across the Australian continent and different seasonal trends, we focus on three specific locations, Darwin, Sydney andMelbourne. In tropical Darwin, there is a negative scaling rate (Fig. 1a), whereas in cooler Melbourne, the apparent scaling rate is positive at temperatures up to about 12 C, and flat or slightly declining at higher temperatures (Supplementary Fig. 2c,f). Although not obvious in the Sydney data (Supplementary Fig. 2b,e), a change from positive to negative scaling above some threshold temperature has been noted in both daily and hourly precipitation extremes in previous observational studies and in hourly precipitation extremes in a modelling study. Some studies have attributed the turnaround to relative humidity