Regional dynamical and statistical downscaling temperature, humidity and wind speed for the Beijing region under stratospheric aerosol injection geoengineering

Abstract. We use four Earth system models (ESMs) to simulate climate under the modest greenhouse emissions RCP4.5 (Representative Concentration Pathway), the “business-as-usual” RCP8.5 and the stratospheric aerosol injection G4 geoengineering scenarios. These drive a 10 km resolution dynamically downscaled model (Weather Research and Forecasting, WRF) and a statistically bias-corrected (Inter-Sectoral Impact Model Intercomparison Project, ISIMIP) and downscaled simulation in a 450×330 km domain containing the Beijing Province, ranging from 2000 m elevation to sea level. The 1980s simulations of surface temperatures, humidities and wind speeds using statistical bias correction make for a better estimate of mean climate determined by ERA5 reanalysis data than does the WRF simulation. However correcting the WRF output with quantile delta mapping bias correction removes the offsets in mean state and results in WRF better reproducing observations over 2007–2017 than ISIMIP bias correction. The WRF simulations consistently show 0.5 ∘C higher mean annual temperatures than from ISIMIP due both to the better resolved city centres and also to warmer winter temperatures. In the 2060s WRF produces consistently larger spatial ranges of surface temperatures, humidities and wind speeds than ISIMIP downscaling across the Beijing Province for all three future scenarios. The WRF and ISIMIP methods produce very similar spatial patterns of temperature with G4 and are always cooler than RCP4.5 and RCP8.5, by a slightly larger amount with ISIMIP than WRF. Humidity scenario differences vary greatly between ESMs, and hence ISIMIP downscaling, while for WRF the results are far more consistent across ESMs and show only small changes between scenarios. Mean wind speeds show similarly small changes over the domain, although G4 is significantly windier under WRF than either RCP scenario.