NetCDF: Method Overview

NetCDF: Method Overview#

This page describes various approaches to reading NetCDF climate data using xarray. It explains different use cases, strengths, and limitations of each method.

📂 Example Datasets#

File Name

Description

slp.ncep.194801-202504.nc

Monthly sea level pressure (multi-decade, single file)

hgt_ncep_daily.2018.nc

Daily geopotential height (2018)

hgt_ncep_daily.2019.nc

Daily geopotential height (2019)

hgt_ncep_daily.2020.nc

Daily geopotential height (2020)

sst.oisst_high.198109-202504.nc

Monthly high-resolution SST

📘 Approach 1: Reading a Single NetCDF File#

Use xr.open_dataset() when the dataset is already consolidated into a single file with all variables and time steps. This is ideal for simple, complete datasets.

    import xarray as xr

    ds_slp = xr.open_dataset("../data/slp.ncep.194801-202504.nc")
    ds_slp.slp.isel(time=0).plot()

📘 Approach 2: Reading Multiple Files with a Wildcard#

Use xr.open_mfdataset() to automatically combine many files (e.g., one per year) that follow a naming pattern. This is common for reanalysis or observational archives.

    ds_hgt = xr.open_mfdataset("../data/hgt_ncep_daily.20*.nc", combine="by_coords")
    ds_hgt.hgt.sel(level=500).isel(time=0).plot()

Make sure the files are consistent in structure and metadata.

📘 Approach 3: Reading NCEP Files by Year Range#

If you want explicit control over which files to include, use a year range to construct filenames like hgt_ncep_daily.YYYY.nc. This is useful when skipping specific years or handling missing files.

    years = list(range(2018, 2021))
    file_list = [file_template.replace("YYYY", str(y)) for y in years]
    ds = xr.open_mfdataset(file_list, combine="by_coords")

📘 Approach 4: Downsampling High-Resolution SST#

High-resolution datasets (e.g., 0.25° daily SST) can be downsampled for faster analysis. There are two main methods for reducing spatial resolution: subsampling using isel() and block averaging using coarsen().

🔹 Option A: Subsampling using isel()#

This method selects every N-th grid point along latitude and longitude. It’s fast but does not preserve averages.

    ds_lowres = ds_full.isel(
        lat=slice(None, None, coarsen_factor),
        lon=slice(None, None, coarsen_factor)
    )

  • ✅ Fast and simple

  • ⚠️ May miss finer-scale structure

  • 🔄 Use when quick preview or approximate results are sufficient

🔹 Option B: Block averaging using coarsen()#

This method divides the grid into blocks (e.g., 4×4) and computes the mean in each block. It better preserves physical meaning.

    dataL = data.coarsen(
        latitude=coarsen_factor,
        longitude=coarsen_factor,
        boundary='trim'
    ).mean()

  • ✅ Preserves spatial means

  • ⚠️ Slightly slower, but more accurate

  • 🔄 Use for climate analysis, long-term statistics, or model comparisons

Both methods can be combined with .resample(time="1MS").mean() if the input is daily and monthly aggregation is desired.

🔗 See Also#

👉 View the companion notebook: io_netcdf_code.ipynb

Contents