3.1 Experimental design
The main simulations presented here were the five members from the Atmospheric Model Inter-comparison Project (AMIP) protocol simulation (for the period 1979–2014), three groups of simulations named CTL, LC, and LGC (corresponding to the use of the FTF, FTF and HWR, FTF, HWR and GLF modules), and an additional single member simulation of the land-atmosphere coupled simulation named NP (Table 1). The AMIP simulations were performed using the time-varying external forcing recommended by CMIP6 (https://esgf-node.llnl.gov/search/input4mips/). The atmosphere and land resolution was 2 in both latitude and longitude in these simulations, and the monthly observations of the sea surface temperatures (SSTs) and sea ice concentration from the HadISST data set (Rayner et al. 2003) were used for prescribing the ocean surface conditions. The forcing for the AMIP runs included the monthly mean total solar irradiance (TSI) (Matthes 2017), greenhouse gas (GHG) concentrations, including latitudinal changes and seasonality (Meinshausen et al. 2017), ozone concentrations, anthropogenic aerosol optical properties and the associated Twomey effect (Stevens 2017), land use changes (Land Use Harmonization v2, http://luh.umd.edu/), and historical stratospheric aerosols. In addition, the anthropogenic groundwater exploitation forcing data from 1965–2014 were added (Zeng et al. 2017).
For the AMIP runs, there were five members that shared the same setup, but which had different initial conditions. The simulation time for the AMIP runs was 1979–2014. For the CTL, LC and LGC simulations, they generally followed a similar setup to the AMIP runs, except that the simulation time for these runs was for a 41-year duration from 1976 to 2010 (using the period 1970–75 for spin-up). To reduce the internal noise, and enhance the forced signal caused by the HWR and GLF, ensemble averages were used here, where each group of experiments contained three ensemble simulations using different initial conditions, the results of which were then averaged over the ensemble for evaluation (Koster et al. 2002, 2006). In the current study, three individual simulations, typically differing only in their initial atmospheric and land surface conditions, comprised each ensemble (CTL, GC, or LGC). Simulations with corresponding ensemble members in the three sets of simulations typically shared the same initial conditions (for instance, CTL1, LC1, and LGC1 shared the same initial conditions). For the NP transport simulation, it generally followed a similar setup to the AMIP runs, except that the bio-geochemical module and nitrogen transport module were opened to obtain the nitrogen transport-related variables. The simulation time for this simulation was 1960–2014.