Projection of forced changes in regional relative humidity and associated compound extremes constrained by homogenized observations

Compared to increasingly clear responses of temperature and precipitation to anthropogenic climate change, forced changes in relative humidity (RH) remain largely elusive. This is mainly because climate models failed to capture sharp decline in RH shown in both observations and reanalysis data. Despite growing attention to plausible drivers for the observation—model discrepancy, none of the theoretical reconstructions or dedicatedly-designed simulations could reproduce the drying as strong as observed.

We here propose another possibility, that is, observations/reanalysis (incl. HadISDH and ERA5) are wrong due to overlooked data inhomogeneity arising from region-wide changes in observing instruments, as showcased in China in the 2000s. Such simultaneous changes in instruments failed the detection of ‘break-points’ by automated methods via a neighbor-comparison scheme. By getting access to detailed meta data information provided by the National Meteorological Information Center of China, we now could adjust for the inhomogeneity through combining automated methods and manual checks. The newly homogenized data corrects previously estimated strong and significant drying trends into weak and insignificant ones across humid China, in consistent with most of CMIP and single model initial condition large ensemble (SMILE) simulations.

The homogenized RH dataset is then used for attribution and projection of forced changes in RH and associated compound events (extreme wet-bulb temperature and VPD) over Eastern China, within a Bayesian statistical framework. We find historical forcings of greenhouses gases and aerosols on RH nearly counteracted each other, leading to a weak and non-detectable regional trend. The constrained projection shows that raw CMIP6 projections underestimated the magnitude of forced drying of RH. The Bayesian constraint narrows the uncertainty range of projected RH changes by around ~30%, and effectively eliminates the possibly of wetting response at late 21^st century. Given the stronger-than-expected drying of RH, raw projections slightly overestimate future increases in extreme heat stress but substantially underestimate future rises in extreme VPD accordingly, with the inter-model spread in projections narrowed by 30~40% conditioned on the homogenized historical observational series.