Calculate the Averaged Temperature Anomaly#
Averaged Temperature Anomaly 2081-2100 vs 1971-2000 SSP585#
This example calculates the averaged temperature anomaly (using the TG indicator) for the period 2081-2100 compared to the reference 1971-2000 for SSP585 and several climate models.
We assume to have the tas variable in netCDF files in a ./data/latest folder. The data can be downloaded using the metalink provided with this notebook. The data described in a .metalink file can be downloaded with tools such as aria2 or a browser plugin such as DownThemAll! If you wish to use a different dataset, you can use the climate 4 impact
portal to search and select the data you wish to use and a metalink file to the ESGF data will be provided.
The data is read using xarray and a plot of the time series averaged over Europe is generated, as well as an average spatial map. Several output types examples are shown.
The datasets expected for this notebook are the tas variable needed to calculate TG for several climate models, for the historical period 1971-2000 and the SSP585 period 2081-2100, and for one member. Daily data is used. In C4I, you can find all of the data needed in the CMIP6 project, at the esgf-data3.ceda.ac.uk and esgf3.dkrz.de mirrors.
Preparation of the needed modules#
[ ]:
import datetime
from pathlib import Path
import pandas as pd
import xarray as xr
import icclim
%pip install icclim matplotlib nc_time_axis
[ ]:
# provides cftime axis in matplotlib
Datasets#
sudo apt install -y aria2.Specification of the parameters and period of interest#
The time period of interest as well as the reference period are defined here. A list of models is listed here as an example. Here we used Monthly data (Amon) but daily data could also be used. The corresponding datafiles must have been selected by the user, containing both the studied and referenced periods.
icclim is then executed for both periods for each climate model separately.
[ ]:
# data directory
data_dir = "data/latest"
# studied period
dt1 = datetime.datetime(2081, 1, 1, tzinfo=datetime.timezone.utc)
dt2 = datetime.datetime(2100, 12, 31, tzinfo=datetime.timezone.utc)
# reference period
dtr1 = datetime.datetime(1971, 1, 1, tzinfo=datetime.timezone.utc)
dtr2 = datetime.datetime(2000, 12, 31, tzinfo=datetime.timezone.utc)
models = ["CMCC-ESM2", "GFDL-ESM4"]
out_f = {}
out_hist_f = {}
for model in models:
out_f[model] = f"output/averaged_tas_anomaly/tg_icclim_{model}.nc"
out_hist_f[model] = f"output/averaged_tas_anomaly/tg_icclim_{model}_hist.nc"
filenames_hist = Path("data_dir").glob(f"tas_day_{model}_historical_*.nc")
filenames = Path("data_dir").glob(f"tas_day_{model}_ssp585_*.nc")
icclim.index(
index_name="TG",
in_files=[str(f) for f in filenames],
var_name="tas",
slice_mode="year",
time_range=[dt1, dt2],
out_file=out_f[model],
logs_verbosity="LOW",
)
icclim.index(
index_name="TG",
in_files=[str(f) for f in filenames_hist],
var_name="tas",
slice_mode="year",
time_range=[dtr1, dtr2],
out_file=out_hist_f[model],
logs_verbosity="LOW",
)
Data preparation#
Here all data is loaded in 2 separate variables, one containing all the historical periods for all the models, and the same for the future time period.
An example is shown on how to select a specific data location. But this is not used.
[ ]:
# Open datasets
tg = []
tg_hist = []
ds = []
ds_hist = []
xr.set_options(enable_cftimeindex=False)
for model in models:
dsl = xr.open_dataset(out_f[model], decode_times=False)
dsl["time"] = xr.decode_cf(dsl).time
dsl = dsl.assign_coords({"model_id": model})
tg.append(dsl["TG"])
dshl = xr.open_dataset(out_hist_f[model], decode_times=False)
dshl["time"] = xr.decode_cf(dshl).time
dshl = dshl.assign_coords({"model_id": model})
tg_hist.append(dshl["TG"])
# Select a single x,y combination from the data
longitude = tg[0]["lon"].sel(lon=3.5, method="nearest").values
latitude = tg[0]["lat"].sel(lat=44.2, method="nearest").values
Perform spacial average on all the geographical domain#
[ ]:
# Average different grids
for ii in range(len(tg)):
tg[ii] = tg[ii].sel(lat=[0.0, 90.0], method="nearest").mean(dim=["lon", "lat"])
for ii in range(len(tg_hist)):
tg_hist[ii] = (
tg_hist[ii].sel(lat=[0.0, 90.0], method="nearest").mean(dim=["lon", "lat"])
)
Define a function to align all different calendar types#
[ ]:
# Define function to align different calendars using annual data
def to_pandas_dt(da: "xr.DataArray") -> "xr.DataArray":
"""Takes an annual DataArray. Change the calendar to a pandas datetime."""
val = da.copy()
timev = [int(v) for v in da.time.dt.strftime("%Y")]
timevp = pd.to_datetime(timev, format="%Y")
time1 = xr.DataArray(data=timevp, dims=["time"])
time1.name = "time"
# We rename the time dimension and coordinate to time360 to make it clear it isn't
# the original time coordinate.
val = val.rename({"time": "timepd"})
time1 = time1.rename({"time": "timepd"})
return val.assign_coords({"timepd": time1})
Align calendars of all input data#
[ ]:
# Convert all calendars to annual precision (we have configured icclim to output yearly
# data)
ll = [to_pandas_dt(da) for da in tg]
ll_hist = [to_pandas_dt(da) for da in tg_hist]
# Concatenate all models into one
full_tg = xr.concat(ll, "model_id", join="outer")
full_tg_hist = xr.concat(ll_hist, "model_id", join="outer")
full_tg_anomaly = full_tg - full_tg_hist.mean(dim="timepd")
Plot a multi-model time series for the future time period#
Temperature for SSP585 of the period 2080-2100.
The multi-model average is shown in bold black line.
[ ]:
# Plot future time period and superimpose multi-model average in bold black line
full_tg.plot(hue="model_id")
full_tg.mean(dim="model_id").plot(color="black", linewidth=4)
Plot a time series of the anomaly of temperature#
Anomaly of temperature for SSP585 of the period 2080-2100 compared to 1971-2000.
The multi-model average is shown in bold black line.
[ ]:
# Plot temperature anomaly compared to historical period and superimpose multi-model
# average in bold black line
full_tg_anomaly.plot(hue="model_id")
full_tg_anomaly.mean(dim="model_id").plot(color="black", linewidth=4)