The data was produced employing the Advanced Research Weather Research and Forecasting model (WRF) version 4.1.2 (Skamarock et al., 2019) for the dynamical downscaling of GCM data. WRF is a fully compressible non-hydrostatic atmospheric simulation system. Two sensitivity simulations were conducted using 15-year time slices for the present day and the mid-Pliocene simulated by ECHAM5 as initial and boundary conditions (Mutz et al., 2018; Botsyun et al., 2020). Except for the atmospheric forcing data, other parameters were the same in both simulations.

The model domain has a grid spacing of 30 km. In the vertical direction, 28 terrain-following eta-levels were used. The model time steps are 120 seconds with a 6 hourly data output and are aggregated to daily values in post processing. The boundary conditions were updated every 6 h. The daily re-initialization strategy from Maussion et al. (2011) and Maussion et al. (2014) were employed: each simulation starts at 12 UTC and contains 36 h, with the first 12 h as the spin-up time. This strategy kept the large-scale circulation patterns simulated by WRF closely constrained by the forcing data, while concurrently allowing WRF to develop the mesoscale atmospheric features. Physical parameterization schemes were consistent with the ones used for high-resolution dynamical downscaling in High Mountain Asia in Wang et al. (2021). The data format follows the guidelines of the [UC]² Data Standard ( http://www.uc2-program.org/uc2_data_standard.pdf ).