ResearchOnClimate

Dependencies

• Users are encouraged to use NCAR Casper Login*.
• Data is available from the sixth Coupled Model Intercomparison Product (CMIP6). It is publicly archived and available, but the data is readily available on Casper. Here is a link to the data on the Earth System Grid Federation Portal at Centre for Environmental Data Analysis and the cloud.
• Run pip install -r requirements.txt to download any missing Python dependencies.

(NCAR Casper Login Aside: You can create an account by following the directions on the Casper website.)

Exploratory Data Analysis

Run the explore_CMIP6_data.py file to generate a variety of graphs of the historical_r1i1p1f1 model output data. Can also use this file as an example of how to generate new spatial plots of any other variables.

Prepare Data for Model Input

Use the prepare_data.py file to generate training-ready data files for the emulators. The preprocessing includes variable selection, annual averaging, and feature derivations (ie. diurnal temperature range = tasmax - tasmin).

Data storage structure

/glade/collections/cmip/CMIP6/{MIP}/NCC/NorESM2-LM/{experiment}/{member}/{general variable category}/{variable}/*/*/{variable}/*.nc

• MIP (Model Intercomparison Projects) : CMIP, DAMIP, ScenarioMIP, AerChemMIP
• experiment
  • CMIP: 1pctCO2, abrupt-4xCO2, historical, piControl
  • DAMIP: hist-GHG', hist-aer
  • ScenarioMIP: ssp126, ssp245, ssp370, ssp370-lowNTCF, ssp585
• member
  • E.g., r1i1p1f1, r2i1p1d1, r2i1p1d1
  • r for realization, i for initialization, p for physics, and f for forcing
• general variable category

Acronym	Spelled Out Version
Amon	Atmospheric Month
Omon	Oceanic Month
day	Daily
Oday	Oceanic Daily
Eday	Earth Daily

• variable

List of Climate Variables

Acronym	Spelled Out Version
prc	Convective Precipitation
rtmt	Net Top-of-Model Radiation
hfls	Surface Latent Heat Flux
ch4global	Global Methane Concentration
hur	Relative Humidity
sfcWind	Surface Wind Speed
rlds	Surface Downwelling Longwave Radiation
prw	Water Vapor Path
hus	Specific Humidity
n2oglobal	Global Nitrous Oxide Concentration
clwvi	Column-Integrated Liquid Water
prsn	Snowfall Flux
va	Meridional Wind
cfc11global	Global CFC-11 Concentration
tauu	Zonal Wind Stress
rlutcs	Clear-Sky Upwelling Longwave Radiation
tas	Near-Surface Air Temperature
clt	Total Cloud Fraction
tauv	Meridional Wind Stress
cfc12global	Global CFC-12 Concentration
ci	Sea Ice Concentration
co2	Carbon Dioxide Concentration
clivi	Column-Integrated Ice Water
cl	Cloud Fraction
ua	Zonal Wind
hfss	Surface Sensible Heat Flux
ps	Surface Pressure
o3	Ozone Concentration
wap	Vertical Velocity
rsds	Surface Downwelling Shortwave Radiation
rlus	Surface Upwelling Longwave Radiation
rsutcs	Clear-Sky Upwelling Shortwave Radiation
rsdt	Top-of-Atmosphere Downwelling Shortwave Radiation
tasmin	Minimum Near-Surface Air Temperature
ta	Air Temperature
rlut	Top-of-Atmosphere Upwelling Longwave Radiation
hurs	Near-Surface Relative Humidity
co2mass	Total Mass of Carbon Dioxide
sbl	Sea Ice Thickness
rldscs	Clear-Sky Surface Downwelling Longwave Radiation
psl	Sea-Level Pressure
pr	Precipitation
tasmax	Maximum Near-Surface Air Temperature
clw	Cloud Liquid Water Content
huss	Near-Surface Specific Humidity
zg	Geopotential Height
evspsbl	Evaporation
rsuscs	Clear-Sky Surface Upwelling Shortwave Radiation
rsus	Surface Upwelling Shortwave Radiation
cli	Cloud Ice Water Content
rsut	Top-of-Atmosphere Upwelling Shortwave Radiation
ts	Surface Temperature
rsdscs	Clear-Sky Surface Downwelling Shortwave Radiation

Emulator Replication

Data Access

The processed training, validation and test data can be obtained from Zenodo.

• Download test.tar.gz and train_val.tar.gz.
• Decompressing the two files
• Upload all .nc files in train_val and test onto CASPER and place them in the same directory.

Models

• Download utils.py and upload onto Casper.
• Pattern Scaling
  • TODO
• Gaussian Process
  • TODO
• Random Forest Model
  • Download RF_model_ESEm.py and upload onto Casper.
  • Place utils.py and RF_model_ESEm.ipynb in the same directory as the .nc files.
  • Run the notebook to see the random forest model and outputs.