Files in support of Gnanadesikan, Liu, Dutta, Feole, McCarthy and Qian, Machine learning methods suggest that large regional changes in phytoplankton biomass produced by Earth System Models do not reflect realistic responses to changing climate

Files in support of "Machine learning methods suggest that large regional changes in phytoplankton biomass produced by Earth System Models do not reflect realistic responses to changing climate", manuscript accepted to Global Biogeochemical Cycles. https://doi.org/10.1029/2025GB008761 We provide a total of 22 files here. These are 1. compute_RF_clim_change.m Processing file that loads all the datafiles and puts them in two structure (pictrl_data_uniform and ssp585_data_uniform) each of which is nominally 12 x 9 fields, each of which is either empty (obs. in the ssp585 case) or 360x180x12. Computes random forest emulators Mdl_rf_pictrl and Mdl_rf_ssp585 for each of these. Applies the emulators to compute random forest predictions. predict_phyc_pictrl: 1x9 structure with 9 predictions from random forest trained on pictrl inputs and fed those inputs. predict_phyc_ssp585_rfpictrl: 1x9 structure with 9 predictions from random forest trained on pictrl inputs and fed inputs from SSP585 scenario. predict_phyc_ssp585_rfssp585: 1x9 structure with 9 predictions from random forest trained on SSP585 inputs and fed inputs from SSP585 scenario. ... and so forth. The other important set of outputs are predict_phyc_pictrl_med predict_phyc_ssp585_med These are 11x9 structures which store predictions made using the random forest emulators, but with individual predictors replaced by their median value in the pictrl simulation. This allows us to map the impact of variations in different parameters. The file also computes accumulated local effects arrays for each dataset (this is the part that takes the most time, several hours). Makes all plots other than S6. file 20: RF_cphyto_bbp_zenodo.m runs the analysis on just one dataset (Phytoplankton biomass from backscatter) and makes plot S6 file 10: make_ale.m contains the algorithm we use to compute accumulated local effects We provide a set of 19 files with 777600 rows that can be mapped to 360x180x12 grids Files which incluce the string "_data_uniform" are uniformly gridded fields for each model in the pictrl and ssp585 scenarios 2. Cphyto_bbp_uniform.asc: Biomass calculated using backscatter from MODIS at 470nm using method of Graff et al., (2015) 3. GFDL-CM4_pictrl_data_uniform.asc 4. GFDL-CM4-ssp585_data_uniform.asc 5. GFDL-ESM4_pictrl_data_uniform.asc 6. GFDL-ESM4_ssp585_data_uniform.asc 7. IPSL-CM5A-LR_pictrl_data_uniform.asc 8. IPSL-CM5A-LR_ssp585_data_uniform.asc 9. kostadinov_data_uniform.asc 11. modis_data_uniform.asc 12. MPI-ESM1-2-LR_pictrl_data_uniform.asc 13.. MPI-ESM1-2-LR_ssp585_data_uniform.asc 14. NCAR-CESM2-WACCM_pictrl_data_uniform.asc 15. NCAR-CESM2-WACCM_ssp585_data_uniform.asc 16. NorESM2-LM_pictrl_data_uniform.asc 17. NorESM2-LM_ssp585_data_uniform.asc 21. UKESM1-0-LL_pictrl_data_uniform.asc 22. UKESM1-0-LL_ssp585_data_uniform.asc Are preindustrial climatologies of 12 variables. By column Col 1: Dissolved iron at ocean surface in mol/m3.(CMIP6 variable dfe) Col 2. Mixed layer depth in m (CMIP6 variable mltost) Col 3. Ammonium at ocean surface in mol/m^3 (CMIP6 variable nh4) Col 4. Nitrate at ocean surface in mol/m^3 (CMIP6 variable no3) Col 5. log10 Phytoplankton carbon biomass (CMIP6 variable phycos, log transformed for purposes of analysis) Observed data is filled at points where there is some data over the course of the year. Col 6. Phosphate at ocean surface in mol/m^3 (CMIP6 variable po4) Col 7. Net downward shortwave radiation at ocean surface in W/m^2 (CMIP6 variable rsntds) Col 8: Silicate at ocean surface in mol/m^3 (CMIP6 variable si) Col 9: Salinity at ocean surface in PSU (CMIP6 variable so) Col 10: Tempeature at ocean surface in C (CMIP6 variable tos) Col 11: Vertical velocity at grid point nearest 50m (CMIP6 variable wo) Col 12: Months since light was greater than 10 W/m^2 (can be changed in the processing file). Finally there are two files File 18. phyc_modis.asc, File 19. phyc_kost.asc which represent are raw climatologies of the log10 phytoplankton biomass from the the standard MODIS product of Stramski et al., 2008. and the size structured biomass product of Kostadinov et al. 2016 gridded to a one degree dataset. They have not been filled. Data is given as a single vector that can be mapped to a 360x180x12 grid