Supplement of Development of a sequential tool, LMDZ-NEMO-med-V1, to conduct global-to-regional past climate simulation for the Mediterranean basin: an Early Holocene case study

Abstract. Recently, major progress has been made in the simulation of the ocean
dynamics of the Mediterranean using atmospheric and oceanic models with
high spatial resolution. High resolution is essential to accurately capture the
synoptic variability required to initiate intermediate- and deep-water
formation, the engine of the Mediterranean thermohaline circulation (MTC).
In paleoclimate studies, one major problem with the simulation of regional
climate changes is that boundary conditions are not available from
observations or data reconstruction to drive high-resolution regional
models. One consistent way to advance paleoclimate modelling is to use a
comprehensive global-to-regional approach. However, this approach needs
long-term integration to reach equilibrium (hundreds of years), implying
enormous computational resources. To tackle this issue, a sequential
architecture of a global–regional modelling platform has been developed for
the first time and is described in detail in this paper. First of all, the
platform is validated for the historical period. It is then used to
investigate the climate and in particular, the oceanic circulation, during
the Early Holocene. This period was characterised by a large reorganisation
of the MTC that strongly affected oxygen supply to the intermediate and deep
waters, which ultimately led to an anoxic crisis (called sapropel). Beyond
the case study shown here, this platform may be applied to a large number of
paleoclimate contexts from the Quaternary to the Pliocene, as long as
regional tectonics remain mostly unchanged. For example, the climate
responses of the Mediterranean basin during the last interglacial period (LIG), the
Last Glacial Maximum (LGM) and the Late Pliocene all present interesting
scientific challenges which may be addressed using this numerical platform.


This section is intended as a user manual to explain how to compile and run LMDZ-NEMO-med on a 36 Linux system. It is not, however, a detailed description of the source code. Files relevant to the 37 running of the pre-industrial control simulation presented in the article have been archived and made 38 publicly available for downloading: https://zenodo.org/record/3258410 (Vadsaria et al., 2019). The compiling environment is MODIPSL, a convention for code compilation when the code is 47 distributed into different directories. The following directory should be consulted: 48 "cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/util" 49 Edit the "AA_make.gdef": the user should create a new entry to fit its computational architecture. 50 Compiler options have been set up in this file and will be propagated to "Makefile" at different places. 51

52
It is recommended that all previous configurations be cleared by typing "./clr_make". A new 53 configuration to match the right computer platform can then be created: 54 "./ins_make -t NAME_OF_YOUR_ARCHITECTURE_SYSTEM" 55 Before code compilation, the library netcdf and a Fortran compiler need to be installed. FCM (Flexible 56 Configuration Management: https://metomi.github.io/fcm/doc/), a tool developed by the UK Met 57 Office to manage the dependence among different subroutines of a complex code is also required. 58 Compiling options for FCM are stored under "machine/arch.path" and "machine/arch.fcm". They need 59 to be coherent with what stored under "AA_make.gdef" and "Makefile". 60 To compile the code, the following directory needs to be consulted: 61 "cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/config/LMDZ" 62 Then, with the help of "Makefile", the following can be compiled: 63 "gmake lmdz96x71global" 64 where "lmdz96x71global" is a keyword defined in the "AA_make" script allowing a configuration to 65 be chosen. 66 If the compilation is successful, then the executable codes "create_etat0_limit.e", 67 "make_relax_times.e" and "gcm.e" are stocked at the following directory: 68 "cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/modipsl/bin" 69 70 1.2 Running the model 71 72 The first step is the creation of boundary conditions for the global atmospheric model. The supporting 73 files needed for this step can be found in the following directory: 74 "cd vadsaria_et_al_model/files_and_boundary_conditions_for_LMDZ_global/start_limit" 75 76 A boundary condition file is already provided in this directory: "limit_picontrol_debiais.nc". It is 77 based on a bias-corrected file for SST and SIC data (following the procedure described in the main 78 article) derived from the IPSLCM5 model for the pre-industrial simulation. The procedure to generate 79 this boundary condition file is the following: 80 -Prepare a netcdf file with SST and SIC bias-corrected data, interpolated on a 1°x1° grid: "CM5-81 piControl-pseudo_amip_1x1_tos_sic.3600-3699_climato.after_correction.nc" (in the sub-82 directory "/interpol", a code to generate a 1°x1° "AMIP" grid is provided : 83 "interpol_ipslcm5_amip_tos_sic. This execution is based on a few ".nc" files containing information on topography, surface albedo, etc. 94 It also takes relevant information from definition files of the model (gcm.def, physic.def and 95 orchidee.def). It should create a "limit.nc" file. 96 After creating the initial states and boundary conditions, we are now ready to run the model with an 97 example from the following directory 98 "cd vadsaria_et_al_model/files_and_boundary_conditions_for_LMDZ_global" 99 100 The bash script "launch_picontrol_run_global_type" is an example of how to run the atmospheric 101 global model. The script firstly organizes files for boundary conditions and initial state (all presented 102 in the current directory), and then executes the model "gcm.e" to generate outputs. This script was 103 initially created for use in the supercomputing centre TGCC. It contains some TGCC-specific 104 instructions for the management of environmental variables, including the necessary pathways for the 105 model's preferences and allocation of computing resources. The script is executed with a time step of 106 one month. 107

108
To start the execution of the model: 109 ./launch_picontrol_run_global_make 1 110 111 "1" being the first month. It will create the launch_picontrol_run_global_launcher bash file. The 112 user should then execute this file according to the actual operating system. If the script works, it will 113 automatically generate the next iteration (the next month) until the maximum iteration is reached, 114 denoted as the "stop" variable in the "launch_picontrol_run_global_type" file, set here at 360 115 months (30 years). 116 117 2 Atmospheric regional model 118

Compiling the model 119
The code of this model is identical to that of the global version, but in "Makefile", the key word 120 should be changed from "lmdz96x71global" to "lmdz200120_oneway" 121 Go to the following directory: 122

"cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/config/LMDZ" 123
Then compile the code through Makefile: 124 gmake lmdz200120_oneway 125 where "lmdz200120_oneway" is a keyword defined in the "AA_make" script allowing a 126 configuration to be chosen. 127 If the compilation is successful, executable files are stored in the following directory: 128 "cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/modipsl/bin" 129 130 2.2 Running the model 131 The first step is to create the boundary conditions for the regional atmospheric model. A boundary 132 condition file, "limit_picontrol_debiais.nc", is already provided in the following directory: 133 It is of course different from that of the global model, but it is also obtained from the same bias-135 corrected SST and SIC data, derived from the IPSLCM5 global coupled model for the pre-industrial 136 simulation. The procedure to generate this boundary condition file is the same as described for the 137 global version. 138 To run the model, an example is given in the following directory 139 "cd vadsaria_et_al_model/files_and_boundary_conditions_for_LMDZ_regional" 140 The example bash script "launch_picontrol_run_regional_type" shows how to run the atmospheric 141 regional model. Unlike the global model, additional files are needed to nudge the regional model with 142 the global output. "biline_poids_s.nc'', "biline_poids_u.nc" and "biline_poids_v.nc" (presented in 143 the current directory) are interpolation files allowing efficient transformation of global variables for 144 the regional model grid. Nudged forcing, with a 2-hour time step, from the global model is stored in 145 "sortie_histfrq.nc. 146 Since the global and regional models share a common structure, their launch is also very similar, 147 although with different configuration files. 148 149 3 Mediterranean oceanic model 150 NEMOMED8 is the Mediterranean regional version of the NEMO ocean modelling platform.

151
Documentation of the model can be found at: http://forge.ipsl.jussieu.fr/nemo/wiki/Users 152 Compiling the model 153 The compilation of NEMOMED8 is managed entirely through MODIPSL, so the generation of 154 Makefile is the same as described earlier for LMDZ. The keyword to be used in the argument of 155 "gmake" is "nemomed8". The compilation procedure is simply the following: 156 "cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/config/NEMOMED8" 157 "gmake nemomed8" 158

"cd vadsaria_et_al_model/LMDZ_and_NEMOMED8_models/modipsl/modipsl/bin" 159
If the compilation is successful, then it creates the executable file, "opa". In our study, NEMOMED8 160 is compiled to run with 121 cores in parallel mode. 161

Running the model 162
Before running the model, the 3D boundary conditions for salinity and potential temperature over the 163 buffer zone in the Atlantic close to the Gibraltar need to be generated. This operation is conducted in 164 the following directory: 165 "cd vadsaria_et_al_model/files_and_boundary_conditions_for_NEMOMED8" 166 These boundary conditions are found in the files 167 "data_1m_potential_temperature_nomask_picontrol_debiais_climato.nc" and 168 "data_1m_salinity_nomask_picontrol_debiais_climato.nc", bias-corrected from the IPSLCM5 pre-169 industrial simulation. The grid of the NEMOMED8 model ("meshmask_med8.nc") is provided 170 allowing the user to interpolate their own boundary conditions from this grid. 171 The second step is to generate the surface fluxes from the atmospheric regional model. For this 172 purpose, an interpolation is used to convert the LMDZ4 air-sea fluxes into the NEMOMED8 grid 173 (bilinear for wind stress and conservative remapping for other fluxes). For NEMOMED8, the water, 174 radiative, latent, sensible fluxes and wind stress are required. In the sub-directory "/lmdz_to_nemo", a 175 code is provided to generate the bilinear interpolation scheme: 176 "interpol_between_lmdz_et_nemo.F90". During the execution of the executable file, a weight file is 177 required ("opalmdmo", also provided in the sub-directory). 178 "sst_picontrol_debiais.nc.000101", 179 "flx_picontrol_debiais.nc.000101", 180 "taux_picontrol_debiais.nc.000101" and 181 "tauy_picontrol_debiais.nc.000101". 182 Finally, the bash script "launch_picontrol_run_mediterranean_ocean_type" is an example of the 183 instructions necessary to run the oceanic regional model. The procedure is similar to the global and 184 regional atmospheric model. general biases that need to be corrected before running our regional system for paleo periods (Early 192 Holocene). In the following, the bias-correction method for the oceanic 3-D structures, SST and SIC, as 193 well as the freshwater discharges from rivers, is described.

3D temperatures and salinities in the buffer-zone 213
The 3-D fields of oceanic temperature and salinity (over the whole water column) in the Atlantic buffer 214 zone has been adjusted in the same way as for SST. We used the World Ocean Atlas (WOA) (Locarnini 215 et al., 2013) as a reference to correct the outputs from the IPSL-CM5A historical simulation. 216

River runoff to the Mediterranean Sea 217
Freshwater discharge from rivers around the Mediterranean Sea is an important factor controlling the 218 overturning circulation of the Mediterranean. Due to the high sensitivity of oceanic circulation to this 219 variable, we decided to apply a bias-correction to calibrate the river discharges produced by LMDZ-220 regional. Actually, the atmospheric model, coupled to the land surface model ORCHIDEE, tends to 221 overestimate the amount of freshwater runoffs compared to present-day observations ( Figure S1). The 222 bias-corrected that we applied is based on the observed climatological runoff (Ludwig et al. 2009;223 Vorosmarty et al., 1998) and the differences between the Early Holocene simulation and present-day 224 simulation. When the difference is relatively not significant, the corrected runoff is set to the 225 climatology, mainly to avoid negative values 1 . However, in order to stay consistent with the 226 methodology for SST and SIC bias correction, we chose the absolute difference correction method for 227 the river runoff.

Continental precipitation 232
The reconstructed data used for comparison with the EHOL simulation is taken from Dormoy et al.  Table S1). In summer, the model shows a more 246 contrasted response, with negative anomalies in summer temperatures (Table S1) due to the homogenous 247 drought (Fig 10d in the main article). However, this comparison cannot reflect the precipitation changes 248 for the entire continent. Indeed, in north of Lake Accesa we see positive summer anomalies (Fig 10d in  249 the main article). March, Figure S2, f) are a bit lower than the reconstruction especially for the Eastern basin (-1 to -2 °C). 257 The simulated summer SSTs (July to September, Figure S3 perturbation induced by enhanced river fluxes. As a reference for comparison, we use a synthesis of SSS 270 sampled from the S1 deposition, and provided by Kallel et al. (1997). Our EHOL simulation takes the 271 Nile river enhancement into account, that is an annual river discharge of 13000 m 3 .s -1 , against 2930 m 3 .s -272 1 for the pre-industrial value). The North-East rivers (Buyukmenderes, Vardar, Acheloos, Vjosa, 273 Semanit, Shkumbin, Durres, Mat and Drini) have their annual fresh water discharges increasing from 274 1082 m 3 .s -1 at pre-industrial level to 1622 m 3 .s -1 . The fresh water discharge from February to May 275 increases even more, from 1619 m 3 .s -1 at pre-industrial level to 3228 m 3 .s -1 for EHOL. Our EHOL 276 simulation, even with a significant increase of freshwater inputs, still cannot reproduce a sufficient 277 decrease of SSS to match the reconstructed values, as shown in Figure S3. . Rohling (1999Rohling ( , 2000 278 pointed out that this mismatch can be partly attributed to uncertainties in salinity reconstruction. It is not 279 always straightforward to interpret the isotopic composition of oxygen in terms of salinity. Finally, it is 280 likely that an additional non-negligible fresh water source is missing. To explain the substantial SSS