Hourly dataset production¶

With the model(s) from train_fluxnet_models.ipynb we can now produce the hourly datasets for GPP/NEE/respiration.

To do so, you select the model folder corresponding to the trained model and specify to which output directory the results should be written to.

In [ ]:

from dask.distributed import Client
from excited_workflow.produce_fluxnet_dataset import produce_dataset
from pathlib import Path

client = Client(n_workers=2, threads_per_worker=2)

output_dir = Path("/data/volume_2/produced_models/test")

produce_dataset(  # takes ~40 minutes for 5 years of data
    model_dir="/data/volume_2/trained_models/fluxnet_gpp-lightgbm-2024-02-23_09_55",
    output_dir=output_dir,
)

from dask.distributed import Client from excited_workflow.produce_fluxnet_dataset import produce_dataset from pathlib import Path client = Client(n_workers=2, threads_per_worker=2) output_dir = Path("/data/volume_2/produced_models/test") produce_dataset( # takes ~40 minutes for 5 years of data model_dir="/data/volume_2/trained_models/fluxnet_gpp-lightgbm-2024-02-23_09_55", output_dir=output_dir, )

To demonstrate the results, we can open the resulting (multi-file) dataset and plot the GPP for one timestep:

In [4]:

import xarray as xr
ds = xr.open_mfdataset(output_dir.glob("*.nc"))
ds["GPP_NT_VUT_REF"].isel(time=3000).plot()

import xarray as xr ds = xr.open_mfdataset(output_dir.glob("*.nc")) ds["GPP_NT_VUT_REF"].isel(time=3000).plot()

Out[4]:

<matplotlib.collections.QuadMesh at 0x7f61fafe5180>

The time series (daily average) for a single location looks like the following:

In [4]:

one_loc = ds["GPP_NT_VUT_REF"].sel(latitude=40, longitude=-100, method="nearest")
one_loc.resample(time="1D").mean().plot()

one_loc = ds["GPP_NT_VUT_REF"].sel(latitude=40, longitude=-100, method="nearest") one_loc.resample(time="1D").mean().plot()

Out[4]:

[<matplotlib.lines.Line2D at 0x7ff5257c97e0>]

We can also study the average diurnal cycle at the same location:

In [5]:

one_loc.groupby(one_loc["time"].dt.hour).mean().plot()

one_loc.groupby(one_loc["time"].dt.hour).mean().plot()

Out[5]:

[<matplotlib.lines.Line2D at 0x7ff50adc77c0>]

In [ ]: