Skip to content

Jet finding

Jet detection, categorization, tracking and properties.

This probably too big module contains all the utilities relative to jet extraction from 2D fields, jet tracking, jet categorization and jet properties. All of the functions are wrapped by the convenience class JetFindingExperiment.

JetFindingExperiment

Bases: object

Convenience class that wraps basically all the functions in this module, applying it to the data held by its DataHandler and storing the results to avoid recomputing in the subfolder of its DataHandler.

I think it's a bit broken right now, fix coming at some point. In the mean time use "do_everything" or unwrap it yourself.

TODO: do_everything and this class's methods should be broken down into functions with a decorator for saving. Any function that has the argument "subpath", the decorator takes an argument "filename" a str, constructs the file path from DATADIR, subpath and filename, tries to load it and return it (parquet or nc) or does the computation defined by the function. The function does not know anything about saving.

Attributes:

Name Type Description
data_handler DataHandler

Stores data and provides a subfolder in which to store results that is uniquely defined by the data (see .data.DataHandler)
ds Dataset

shortcut to self.data_handler.da
path Path

shortcut to self.data_handler.path
metadata dict

shortcut to self.data_handler.metadata
time Dataset

shortcut to self.data_handler.get_sample_dims()["time"]
Source code in jetutils/jet_finding.py 
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
class JetFindingExperiment(object):
    """
    Convenience class that wraps basically all the functions in this module, applying it to the data held by its `DataHandler` and storing the results to avoid recomputing in the subfolder of its `DataHandler`.

    I think it's a bit broken right now, fix coming at some point. In the mean time use "do_everything" or unwrap it yourself.

    TODO: do_everything and this class's methods should be broken down into functions with a decorator for saving.
    Any function that has the argument "subpath", the decorator takes an argument "filename" a str, constructs the file path from DATADIR, subpath and filename, tries to load it and return it (parquet or nc) or does the computation defined by the function. The function does not know anything about saving.

    Attributes
    ----------
    data_handler : DataHandler
        Stores data and provides a subfolder in which to store results that is uniquely defined by the data (see `.data.DataHandler`)
    ds : xr.Dataset
        shortcut to `self.data_handler.da`
    path : Path
        shortcut to `self.data_handler.path`
    metadata : dict
        shortcut to `self.data_handler.metadata`
    time : xr.Dataset
        shortcut to `self.data_handler.get_sample_dims()["time"]`
    """

    def __init__(
        self,
        data_handler: DataHandler,
    ) -> None:
        self.ds = data_handler.da
        self.path = data_handler.path
        self.data_handler = data_handler
        self.metadata = self.data_handler.metadata
        self.time = data_handler.get_sample_dims()["time"]

    def find_low_wind(self):
        """
        Assuming the same file structure as me, finds the data corresponding to this Experiment's specs except for another `varname`: `"mid_wind"`.

        Returns
        -------
        ds: xr.Dataset
            A Dataset with the same specs as this object's `DataHandler`, but for another varname.
        """
        metadata = self.data_handler.metadata
        dataset = "CESM2" if "member" in metadata else "ERA5"
        dt = self.time[1] - self.time[0]
        resolution = "6H" if dt == np.timedelta64(6, "H") else "dailymean"
        ds_ = open_da(
            dataset,
            "plev",
            "mid_wind",
            resolution,
            metadata["period"],
            metadata["season"],
            *metadata["region"],
            metadata["levels"],
            None,
            None,
            None,
        )
        return ds_

    def find_jets(self, force: bool = False, **kwargs) -> DataFrame:
        """
        Wraps `find_all_jets(**kwargs)` and stores the output
        """
        ofile_ajdf = self.path.joinpath("all_jets_one_df.parquet")

        if ofile_ajdf.is_file() and not force:
            all_jets_one_df = pl.read_parquet(ofile_ajdf)
            return all_jets_one_df
        try:
            qs_path = self.path.joinpath("s_q.nc")
            qs = open_dataarray(qs_path).sel(quantile=0.7)
            kwargs["thresholds"] = qs.rename("s")
        except FileNotFoundError:
            pass
        else:
            print(f"Using thresholds at {qs_path}")

        all_jets_one_df = []
        if "member" in self.metadata or self.ds.nbytes > 8e10:
            several_years = 2
        else:
            several_years = 5
        iterator = iterate_over_year_maybe_member(
            da=self.ds, several_years=several_years
        )
        for indexer in iterator:
            ds_ = compute(self.ds.isel(**indexer), progress_flag=True)
            all_jets_one_df.append(find_all_jets(ds_, **kwargs))
        all_jets_one_df = pl.concat(all_jets_one_df)
        all_jets_one_df.write_parquet(ofile_ajdf)
        return all_jets_one_df

    def categorize_jets(
        self,
        low_wind: xr.Dataset | xr.DataArray | None = None,
        feature_names: tuple | None = None,
        force: int = 0,
        mode: (
            Literal["year"] | Literal["season"] | Literal["month"] | Literal["week"]
        ) = "week",
        n_components: int | Sequence = 2,
        n_init: int = 20,
        init_params: str = "random_from_data",
        v2: bool = True,
    ):
        """
        Makes sure the necessary columns are present in `jets`, then wraps `is_polar_gmix()` and and stores the output.

        Parameters
        ----------
        low_wind : xr.Dataset | xr.DataArray
            Wind at lower levels to compute the vertical wind shear
        force : int
            to recompute the categorization even if the jets already have a `is_polar` column. With `force > 1`, also re-interpolates `theta` and `ratio` on the jet points. By default 0
        mode : "all", "season", "month" or "week
            Trains one model if "all", otherwise train one independent model for each season, month or week of the year. By default "week".
        n_components : int
            Number of Gaussian components, by default 2
        n_init : int
            Number of repeated independent training with. Can massively increase run time. By default 20
        init_params : str, optional
            Type of init, by default "random_from_data"

        Returns
        -------
        DataFrame
            DataFrame of same length as the jets with the same index columns, the `feature_names` columns: "ratio" and "theta", and a new `is_polar` column, corresponding to the proability of each row to belong to the eddy-driven jet component.
        """
        jets = self.find_jets()
        ofile_ajdf = self.path.joinpath("all_jets_one_df.parquet")
        jets = categorize_jets(
            jets=jets,
            ds=self.ds,
            low_wind=low_wind,
            feature_names=feature_names,
            force=force,
            mode=mode,
            n_components=n_components,
            n_init=n_init,
            init_params=init_params,
            v2=v2,
        )
        jets.write_parquet(ofile_ajdf)
        return jets

    def compute_jet_props(
        self,
        force: bool = False,
    ) -> DataFrame:
        """
        Compute "raw" jet properties from the jets

        Parameters
        ----------
        force : bool
            To recompute the properties even if the file "props_as_df_raw.parquet" exists in the subfolder, by default False

        Returns
        -------
        DataFrame
            Jet properties for all jets
        """
        jet_props_incomplete_path = self.path.joinpath("props_as_df_raw.parquet")
        if jet_props_incomplete_path.is_file() and not force:
            return pl.read_parquet(jet_props_incomplete_path)
        all_jets_one_df = self.find_jets()
        props_as_df = compute_jet_props(all_jets_one_df)
        width = []
        if "s" not in self.ds:
            self.ds["s"] = np.sqrt(self.ds["u"] ** 2 + self.ds["v"] ** 2)
        da = self.ds["s"]
        indexer = iterate_over_year_maybe_member(all_jets_one_df, da)
        for idx1, idx2 in tqdm(list(indexer)):
            these_jets = all_jets_one_df.filter(*idx1)
            da_ = compute(da.sel(**idx2), progress_flag=False)
            width_ = compute_widths(these_jets, da_)
            width.append(width_)
        width = pl.concat(width)
        index_columns = get_index_columns(width)
        props_as_df = props_as_df.join(width, on=index_columns, how="inner").sort(
            index_columns
        )
        props_as_df.write_parquet(jet_props_incomplete_path)
        return props_as_df

    def props_as_df(self, categorize: bool = True, force: int = 0) -> DataFrame:
        """
        Compute full jet properties from the jets, with or without categorization

        Parameters
        ----------
        categorize: bool
            whether to return the uncatd or categorized jet properties. Will contain either the `jet ID` column and have a value for each jet at each timestep, or the `jet` category aggregating over the two or three categories of jets: EDJ, STJ and maybe hybrid.
        force : int
            To recompute the properties even if the file "props_as_df.parquet" exists in the subfolder, by default False. If set to *2* or more, also recomputes the `raw` props.s

        Returns
        -------
        DataFrame
            Jet properties for all jets
        """
        ofile_padu = self.path.joinpath("props_as_df_uncat.parquet")
        ofile_pad = self.path.joinpath("props_as_df.parquet")
        if ofile_padu.is_file() and not categorize and not force:
            return pl.read_parquet(ofile_padu)
        if ofile_pad.is_file() and categorize and not force:
            return pl.read_parquet(ofile_pad)
        if ofile_padu.is_file() and categorize and not force:
            props_as_df = average_jet_categories(pl.read_parquet(ofile_padu))
            props_as_df.write_parquet(ofile_pad)
            return props_as_df
        props_as_df = self.compute_jet_props(force=force > 1)
        if force == 1:
            jets = self.categorize_jets()
            index_columns = get_index_columns(jets)
            is_polar = jets.group_by(index_columns, maintain_order=True).agg(
                pl.col("is_polar").mean()
            )
            props_as_df = props_as_df.drop("is_polar").join(is_polar, on=index_columns)
        props_as_df.write_parquet(ofile_padu)
        props_as_df_cat = average_jet_categories(props_as_df)
        props_as_df_cat.write_parquet(ofile_pad)
        if categorize:
            return props_as_df_cat
        return props_as_df

    # def track_jets(
    #     self,
    #     dist_thresh: float = 2,
    #     overlap_min_thresh: float = 0.5,
    #     overlap_max_thresh: float = 0.6,
    #     dis_polar_thresh: float | None = 1.0,
    #     n_next: int = 1,
    #     force: int = 0,
    # ) -> DataFrame:
    #     """
    #     Wraps cross and summary

    #     Parameters
    #     ----------
    #     dist_thresh : float, optional
    #         _description_, by default 2
    #     overlap_min_thresh : float, optional
    #         _description_, by default 0.5
    #     overlap_max_thresh : float, optional
    #         _description_, by default 0.6
    #     dis_polar_thresh : float | None, optional
    #         _description_, by default 1.0
    #     force : int, optional
    #         _description_, by default 0

    #     Returns
    #     -------
    #     DataFrame
    #         _description_
    #     """
    #     cross_opath = self.path.joinpath("cross.parquet")
    #     summary_opath = self.path.joinpath("summary.parquet")
    #     all_jets_one_df = self.find_jets().cast({"time": pl.Datetime("ms")})
    #     if not cross_opath.is_file() or force > 1:
    #         cross = track_jets(all_jets_one_df)
    #         cross.write_parquet(cross_opath)
    #     else:
    #         cross = pl.read_parquet(cross_opath)
    #     if not summary_opath.is_file() or force:
    #         cross, summary = connected_from_cross(
    #             all_jets_one_df,
    #             cross,
    #             dist_thresh,
    #             overlap_min_thresh,
    #             overlap_max_thresh,
    #             dis_polar_thresh,
    #         )
    #         summary.write_parquet(summary_opath)
    #     else:
    #         summary = pl.read_parquet(summary_opath)
    #     return cross, summary

    def jet_position_as_da(self, force: bool = False):
        """
        Creates and saves a DataArray with the same dimensions as `self.ds`. Its values are NaN where no jet is present, and if a jet is present the value is this jet's local `is_polar` value.
        """
        ofile = self.path.joinpath("jet_pos.nc")
        if ofile.is_file() and not force:
            return open_dataarray(ofile)

        all_jets_one_df = self.find_jets()
        da_jet_pos = jet_position_as_da(all_jets_one_df)
        to_netcdf(da_jet_pos, ofile)
        return da_jet_pos

    def compute_extreme_clim(self, varname: str, subsample: int = 5):
        """
        Computes this object's data's extreme climatology.

        Parameters
        ----------
        varname : str
            If `self.ds` is a dataset, the variable to look at.
        subsample : int, optional
            Don't take every year in the data but only one every `subsample`, by default 5
        """
        da = self.ds[varname]
        time = Series("time", self.time)
        years = time.dt.year().to_numpy()
        mask = np.isin(years, np.unique(years)[::subsample])
        opath = self.path.joinpath(f"{varname}_q_clim.nc")
        compute_extreme_climatology(da.isel(time=mask), opath)
        quantiles_clim = open_dataarray(opath)
        quantiles = xr.DataArray(
            np.zeros((len(self.ds.time), quantiles_clim.shape[0])),
            coords={
                "time": self.ds.time.values,
                "quantile": quantiles_clim.coords["quantile"].values,
            },
        )
        for qcl in quantiles_clim.transpose():
            dayofyear = qcl.dayofyear
            quantiles[quantiles.time.dt.dayofyear == dayofyear, :] = qcl.values
        to_netcdf(quantiles, self.path.joinpath(f"{varname}_q.nc"))

categorize_jets(low_wind=None, feature_names=None, force=0, mode='week', n_components=2, n_init=20, init_params='random_from_data', v2=True)

Makes sure the necessary columns are present in jets, then wraps is_polar_gmix() and and stores the output.

Parameters:

Name Type Description Default
low_wind Dataset | DataArray

Wind at lower levels to compute the vertical wind shear

 None
force int

to recompute the categorization even if the jets already have a is_polar column. With force > 1, also re-interpolates theta and ratio on the jet points. By default 0

 0
mode "all", "season", "month" or "week

Trains one model if "all", otherwise train one independent model for each season, month or week of the year. By default "week".

 'week'
n_components int

Number of Gaussian components, by default 2

 2
n_init int

Number of repeated independent training with. Can massively increase run time. By default 20

 20
init_params str

Type of init, by default "random_from_data"

 'random_from_data'

Returns:

Type Description
DataFrame

DataFrame of same length as the jets with the same index columns, the feature_names columns: "ratio" and "theta", and a new is_polar column, corresponding to the proability of each row to belong to the eddy-driven jet component.
Source code in jetutils/jet_finding.py 
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
def categorize_jets(
    self,
    low_wind: xr.Dataset | xr.DataArray | None = None,
    feature_names: tuple | None = None,
    force: int = 0,
    mode: (
        Literal["year"] | Literal["season"] | Literal["month"] | Literal["week"]
    ) = "week",
    n_components: int | Sequence = 2,
    n_init: int = 20,
    init_params: str = "random_from_data",
    v2: bool = True,
):
    """
    Makes sure the necessary columns are present in `jets`, then wraps `is_polar_gmix()` and and stores the output.

    Parameters
    ----------
    low_wind : xr.Dataset | xr.DataArray
        Wind at lower levels to compute the vertical wind shear
    force : int
        to recompute the categorization even if the jets already have a `is_polar` column. With `force > 1`, also re-interpolates `theta` and `ratio` on the jet points. By default 0
    mode : "all", "season", "month" or "week
        Trains one model if "all", otherwise train one independent model for each season, month or week of the year. By default "week".
    n_components : int
        Number of Gaussian components, by default 2
    n_init : int
        Number of repeated independent training with. Can massively increase run time. By default 20
    init_params : str, optional
        Type of init, by default "random_from_data"

    Returns
    -------
    DataFrame
        DataFrame of same length as the jets with the same index columns, the `feature_names` columns: "ratio" and "theta", and a new `is_polar` column, corresponding to the proability of each row to belong to the eddy-driven jet component.
    """
    jets = self.find_jets()
    ofile_ajdf = self.path.joinpath("all_jets_one_df.parquet")
    jets = categorize_jets(
        jets=jets,
        ds=self.ds,
        low_wind=low_wind,
        feature_names=feature_names,
        force=force,
        mode=mode,
        n_components=n_components,
        n_init=n_init,
        init_params=init_params,
        v2=v2,
    )
    jets.write_parquet(ofile_ajdf)
    return jets

compute_extreme_clim(varname, subsample=5)

Computes this object's data's extreme climatology.

Parameters:

Name Type Description Default
varname str

If self.ds is a dataset, the variable to look at.

 required
subsample int

Don't take every year in the data but only one every subsample, by default 5

 5
Source code in jetutils/jet_finding.py 
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
def compute_extreme_clim(self, varname: str, subsample: int = 5):
    """
    Computes this object's data's extreme climatology.

    Parameters
    ----------
    varname : str
        If `self.ds` is a dataset, the variable to look at.
    subsample : int, optional
        Don't take every year in the data but only one every `subsample`, by default 5
    """
    da = self.ds[varname]
    time = Series("time", self.time)
    years = time.dt.year().to_numpy()
    mask = np.isin(years, np.unique(years)[::subsample])
    opath = self.path.joinpath(f"{varname}_q_clim.nc")
    compute_extreme_climatology(da.isel(time=mask), opath)
    quantiles_clim = open_dataarray(opath)
    quantiles = xr.DataArray(
        np.zeros((len(self.ds.time), quantiles_clim.shape[0])),
        coords={
            "time": self.ds.time.values,
            "quantile": quantiles_clim.coords["quantile"].values,
        },
    )
    for qcl in quantiles_clim.transpose():
        dayofyear = qcl.dayofyear
        quantiles[quantiles.time.dt.dayofyear == dayofyear, :] = qcl.values
    to_netcdf(quantiles, self.path.joinpath(f"{varname}_q.nc"))

compute_jet_props(force=False)

Compute "raw" jet properties from the jets

Parameters:

Name Type Description Default
force bool

To recompute the properties even if the file "props_as_df_raw.parquet" exists in the subfolder, by default False

 False

Returns:

Type Description
DataFrame

Jet properties for all jets
Source code in jetutils/jet_finding.py 
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
def compute_jet_props(
    self,
    force: bool = False,
) -> DataFrame:
    """
    Compute "raw" jet properties from the jets

    Parameters
    ----------
    force : bool
        To recompute the properties even if the file "props_as_df_raw.parquet" exists in the subfolder, by default False

    Returns
    -------
    DataFrame
        Jet properties for all jets
    """
    jet_props_incomplete_path = self.path.joinpath("props_as_df_raw.parquet")
    if jet_props_incomplete_path.is_file() and not force:
        return pl.read_parquet(jet_props_incomplete_path)
    all_jets_one_df = self.find_jets()
    props_as_df = compute_jet_props(all_jets_one_df)
    width = []
    if "s" not in self.ds:
        self.ds["s"] = np.sqrt(self.ds["u"] ** 2 + self.ds["v"] ** 2)
    da = self.ds["s"]
    indexer = iterate_over_year_maybe_member(all_jets_one_df, da)
    for idx1, idx2 in tqdm(list(indexer)):
        these_jets = all_jets_one_df.filter(*idx1)
        da_ = compute(da.sel(**idx2), progress_flag=False)
        width_ = compute_widths(these_jets, da_)
        width.append(width_)
    width = pl.concat(width)
    index_columns = get_index_columns(width)
    props_as_df = props_as_df.join(width, on=index_columns, how="inner").sort(
        index_columns
    )
    props_as_df.write_parquet(jet_props_incomplete_path)
    return props_as_df

find_jets(force=False, **kwargs)

Wraps find_all_jets(**kwargs) and stores the output

Source code in jetutils/jet_finding.py 
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
def find_jets(self, force: bool = False, **kwargs) -> DataFrame:
    """
    Wraps `find_all_jets(**kwargs)` and stores the output
    """
    ofile_ajdf = self.path.joinpath("all_jets_one_df.parquet")

    if ofile_ajdf.is_file() and not force:
        all_jets_one_df = pl.read_parquet(ofile_ajdf)
        return all_jets_one_df
    try:
        qs_path = self.path.joinpath("s_q.nc")
        qs = open_dataarray(qs_path).sel(quantile=0.7)
        kwargs["thresholds"] = qs.rename("s")
    except FileNotFoundError:
        pass
    else:
        print(f"Using thresholds at {qs_path}")

    all_jets_one_df = []
    if "member" in self.metadata or self.ds.nbytes > 8e10:
        several_years = 2
    else:
        several_years = 5
    iterator = iterate_over_year_maybe_member(
        da=self.ds, several_years=several_years
    )
    for indexer in iterator:
        ds_ = compute(self.ds.isel(**indexer), progress_flag=True)
        all_jets_one_df.append(find_all_jets(ds_, **kwargs))
    all_jets_one_df = pl.concat(all_jets_one_df)
    all_jets_one_df.write_parquet(ofile_ajdf)
    return all_jets_one_df

find_low_wind()

Assuming the same file structure as me, finds the data corresponding to this Experiment's specs except for another varname: "mid_wind".

Returns:

Name Type Description
ds Dataset

A Dataset with the same specs as this object's DataHandler, but for another varname.
Source code in jetutils/jet_finding.py 
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
def find_low_wind(self):
    """
    Assuming the same file structure as me, finds the data corresponding to this Experiment's specs except for another `varname`: `"mid_wind"`.

    Returns
    -------
    ds: xr.Dataset
        A Dataset with the same specs as this object's `DataHandler`, but for another varname.
    """
    metadata = self.data_handler.metadata
    dataset = "CESM2" if "member" in metadata else "ERA5"
    dt = self.time[1] - self.time[0]
    resolution = "6H" if dt == np.timedelta64(6, "H") else "dailymean"
    ds_ = open_da(
        dataset,
        "plev",
        "mid_wind",
        resolution,
        metadata["period"],
        metadata["season"],
        *metadata["region"],
        metadata["levels"],
        None,
        None,
        None,
    )
    return ds_

jet_position_as_da(force=False)

Creates and saves a DataArray with the same dimensions as self.ds. Its values are NaN where no jet is present, and if a jet is present the value is this jet's local is_polar value.

Source code in jetutils/jet_finding.py 
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
def jet_position_as_da(self, force: bool = False):
    """
    Creates and saves a DataArray with the same dimensions as `self.ds`. Its values are NaN where no jet is present, and if a jet is present the value is this jet's local `is_polar` value.
    """
    ofile = self.path.joinpath("jet_pos.nc")
    if ofile.is_file() and not force:
        return open_dataarray(ofile)

    all_jets_one_df = self.find_jets()
    da_jet_pos = jet_position_as_da(all_jets_one_df)
    to_netcdf(da_jet_pos, ofile)
    return da_jet_pos

props_as_df(categorize=True, force=0)

Compute full jet properties from the jets, with or without categorization

Parameters:

Name Type Description Default
categorize bool

whether to return the uncatd or categorized jet properties. Will contain either the jet ID column and have a value for each jet at each timestep, or the jet category aggregating over the two or three categories of jets: EDJ, STJ and maybe hybrid.

 True
force int

To recompute the properties even if the file "props_as_df.parquet" exists in the subfolder, by default False. If set to 2 or more, also recomputes the raw props.s

 0

Returns:

Type Description
DataFrame

Jet properties for all jets
Source code in jetutils/jet_finding.py 
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
def props_as_df(self, categorize: bool = True, force: int = 0) -> DataFrame:
    """
    Compute full jet properties from the jets, with or without categorization

    Parameters
    ----------
    categorize: bool
        whether to return the uncatd or categorized jet properties. Will contain either the `jet ID` column and have a value for each jet at each timestep, or the `jet` category aggregating over the two or three categories of jets: EDJ, STJ and maybe hybrid.
    force : int
        To recompute the properties even if the file "props_as_df.parquet" exists in the subfolder, by default False. If set to *2* or more, also recomputes the `raw` props.s

    Returns
    -------
    DataFrame
        Jet properties for all jets
    """
    ofile_padu = self.path.joinpath("props_as_df_uncat.parquet")
    ofile_pad = self.path.joinpath("props_as_df.parquet")
    if ofile_padu.is_file() and not categorize and not force:
        return pl.read_parquet(ofile_padu)
    if ofile_pad.is_file() and categorize and not force:
        return pl.read_parquet(ofile_pad)
    if ofile_padu.is_file() and categorize and not force:
        props_as_df = average_jet_categories(pl.read_parquet(ofile_padu))
        props_as_df.write_parquet(ofile_pad)
        return props_as_df
    props_as_df = self.compute_jet_props(force=force > 1)
    if force == 1:
        jets = self.categorize_jets()
        index_columns = get_index_columns(jets)
        is_polar = jets.group_by(index_columns, maintain_order=True).agg(
            pl.col("is_polar").mean()
        )
        props_as_df = props_as_df.drop("is_polar").join(is_polar, on=index_columns)
    props_as_df.write_parquet(ofile_padu)
    props_as_df_cat = average_jet_categories(props_as_df)
    props_as_df_cat.write_parquet(ofile_pad)
    if categorize:
        return props_as_df_cat
    return props_as_df

average_jet_categories(props_as_df, polar_cutoff=None, allow_hybrid=False)

For every timestep, member and / or cluster (whichever applicable), aggregates each jet property (with a different rule for each property but usually a mean) into a single number for each category: subtropical, eddy driven jet and potentially hybrid, summarizing this property fo all the jets in this snapshot that fit this category, based on their mean is_polar value and a threshold given by polar_cutoff.

E.g. on the 1st of January 1999, there are two jets with is_polar < polar_cutoff and one with is_polar > polar_cutoff. We pass allow_hybrid=False to the function. In the output, for the row corresponding to this date and jet=STJ, the value for the "mean_lat" column will be the mean of the "mean_lat" values of two jets that had is_polar < polar_cutoff.

Parameters:

Name Type Description Default
props_as_df DataFrame

Uncategorized jet properties, that contain at least the jet ID column.

 required
polar_cutoff float | None

Cutoff, by default None

 None
allow_hybrid bool

Whether to output two or three jet categories (hybrid jet between EDJ and STJ), by default False

 False

Returns:

Type Description
props_as_df

Categorizes jet properties. The columns jet ID does not exist anymore, and a new column jet with two or three possible values has been added. Two possible values if allow_hybrid=False: "STJ" or "EDJ". If allow_hybrid=True, the third hybrid category can also be found in the output props_as_df.
Source code in jetutils/jet_finding.py 
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
def average_jet_categories(
    props_as_df: DataFrame,
    polar_cutoff: float | None = None,
    allow_hybrid: bool = False,
):
    """
    For every timestep, member and / or cluster (whichever applicable), aggregates each jet property (with a different rule for each property but usually a mean) into a single number for each category: subtropical, eddy driven jet and potentially hybrid, summarizing this property fo all the jets in this snapshot that fit this category, based on their mean `is_polar` value and a threshold given by `polar_cutoff`.

    E.g. on the 1st of January 1999, there are two jets with `is_polar < polar_cutoff` and one with `is_polar > polar_cutoff`. We pass `allow_hybrid=False` to the function. In the output, for the row corresponding to this date and `jet=STJ`, the value for the `"mean_lat"` column will be the mean of the `"mean_lat"` values of two jets that had `is_polar < polar_cutoff`.

    Parameters
    ----------
    props_as_df : DataFrame
        Uncategorized jet properties, that contain at least the `jet ID` column.
    polar_cutoff : float | None, optional
        Cutoff, by default None
    allow_hybrid : bool, optional
        Whether to output two or three jet categories (hybrid jet between EDJ and STJ), by default False

    Returns
    -------
    props_as_df
        Categorizes jet properties. The columns `jet ID` does not exist anymore, and a new column `jet` with two or three possible values has been added. Two possible values if `allow_hybrid=False`: "STJ" or "EDJ". If `allow_hybrid=True`, the third `hybrid` category can also be found in the output `props_as_df`.
    """
    if allow_hybrid and polar_cutoff is None:
        polar_cutoff = 0.15
    elif polar_cutoff is None:
        polar_cutoff = 0.5
    if allow_hybrid:
        props_as_df = props_as_df.with_columns(
            pl.when(pl.col("is_polar") > 1 - polar_cutoff)
            .then(pl.lit("EDJ"))
            .when(pl.col("is_polar") < polar_cutoff)
            .then(pl.lit("STJ"))
            .otherwise(pl.lit("Hybrid"))
            .alias("jet")
        )
    else:
        props_as_df = props_as_df.with_columns(
            pl.when(pl.col("is_polar") >= polar_cutoff)
            .then(pl.lit("EDJ"))
            .otherwise(pl.lit("STJ"))
            .alias("jet")
        )
    index_columns = get_index_columns(
        props_as_df, ("member", "time", "cluster", "jet ID", "spell", "relative_index")
    )
    other_columns = [
        col for col in props_as_df.columns if col not in [*index_columns, "jet"]
    ]
    agg = {
        col: (pl.col(col) * pl.col("int")).sum() / pl.col("int").sum()
        for col in other_columns
    }
    agg["int"] = pl.col("int").mean()
    agg["is_polar"] = pl.col("is_polar").mean()
    agg["s_star"] = pl.col("s_star").max()
    agg["lon_ext"] = pl.col("lon_ext").max()
    agg["lat_ext"] = pl.col("lat_ext").max()
    agg["njets"] = pl.col("int").len().cast(pl.UInt8())

    gb_columns = get_index_columns(
        props_as_df, ("member", "time", "cluster", "jet", "spell", "relative_index")
    )
    props_as_df_cat = (
        props_as_df.group_by(gb_columns, maintain_order=True)
        .agg(**agg)
        .sort(gb_columns)
    )

    if "member" in index_columns:
        dummy_indexer = (
            props_as_df_cat["member"]
            .unique(maintain_order=True)
            .to_frame()
            .join(
                props_as_df_cat["time"].unique(maintain_order=True).to_frame(),
                how="cross",
            )
            .join(
                props_as_df_cat["jet"].unique(maintain_order=True).to_frame(),
                how="cross",
            )
        )
    elif "cluster" in index_columns:
        dummy_indexer = (
            props_as_df_cat["cluster"]
            .unique(maintain_order=True)
            .to_frame()
            .join(
                props_as_df_cat["jet"].unique(maintain_order=True).to_frame(),
                how="cross",
            )
        )
    else:
        dummy_indexer = (
            props_as_df_cat["time"]
            .unique(maintain_order=True)
            .to_frame()
            .join(
                props_as_df_cat["jet"].unique(maintain_order=True).to_frame(),
                how="cross",
            )
        )
    new_index_columns = get_index_columns(
        props_as_df_cat, ("member", "time", "cluster", "jet", "spell", "relative_index")
    )

    sort_descending = [False] * len(new_index_columns)
    sort_descending[-1] = True
    props_as_df_cat = dummy_indexer.join(
        props_as_df_cat, on=[pl.col(col) for col in new_index_columns], how="left"
    ).sort(new_index_columns, descending=sort_descending)
    props_as_df_cat = props_as_df_cat.with_columns(pl.col("njets").fill_null(0))
    return props_as_df_cat

average_jet_categories_v2(props_as_df)

For every timestep, member and / or cluster (whichever applicable), aggregates each jet property (with a different rule for each property but usually a mean) into a single number for each category: subtropical or eddy driven, summarizing this property for all the jets in this snapshot that fit this category. This version does it as a weighted mean over all jets, with weights either ?is_polar or 1 - is_polar. Honestly nonsensical, don't use except to prove that v1 is better.

Parameters:

Name Type Description Default
props_as_df DataFrame

Uncategorized jet properties, that contain at least the jet ID column.

 required

Returns:

Type Description
props_as_df

Categorized jet properties. The columns jet ID does not exist anymore, and a new column jet with two or three possible values has been added. Two possible values if allow_hybrid=False: "STJ" or "EDJ". If allow_hybrid=True, the third hybrid category can also be found in the output props_as_df.
Source code in jetutils/jet_finding.py 
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
def average_jet_categories_v2(props_as_df: DataFrame):
    """
    For every timestep, member and / or cluster (whichever applicable), aggregates each jet property (with a different rule for each property but usually a mean) into a single number for each category: subtropical or eddy driven, summarizing this property for all the jets in this snapshot that fit this category. This version does it as a weighted mean over all jets, with weights either ?`is_polar` or `1 - is_polar`. Honestly nonsensical, don't use except to prove that v1 is better.

    Parameters
    ----------
    props_as_df : DataFrame
        Uncategorized jet properties, that contain at least the `jet ID` column.

    Returns
    -------
    props_as_df
        Categorized jet properties. The columns `jet ID` does not exist anymore, and a new column `jet` with two or three possible values has been added. Two possible values if `allow_hybrid=False`: "STJ" or "EDJ". If `allow_hybrid=True`, the third `hybrid` category can also be found in the output `props_as_df`.
    """

    def polar_weights(is_polar: bool = False):
        return pl.col("is_polar") if is_polar else 1 - pl.col("is_polar")

    index_columns = get_index_columns(
        props_as_df, ("member", "time", "cluster", "spell", "relative_index")
    )
    other_columns = [
        col for col in props_as_df.columns if col not in [*index_columns, "jet"]
    ]
    agg = [
        {
            col: weighted_mean_pl(
                pl.col(col).fill_nan(0.0), pl.col("int") * polar_weights(bool(i))
            )
            for col in other_columns
        }
        for i in range(2)
    ]
    for i in range(2):
        agg[i]["int"] = weighted_mean_pl("int", polar_weights(bool(i)))
        agg[i]["is_polar"] = weighted_mean_pl("is_polar", polar_weights(bool(i)))
        agg[i]["njets"] = (pl.col("is_polar") < 0.5).sum().cast(pl.UInt8())

    jet_names = ["STJ", "EDJ"]
    props_as_df_cat = [
        props_as_df.group_by(index_columns)
        .agg(**agg[i])
        .with_columns(jet=pl.lit(jet_names[i]))
        for i in range(2)
    ]
    props_as_df_cat = pl.concat(props_as_df_cat)

    if "member" in index_columns:
        dummy_indexer = (
            props_as_df_cat["member"]
            .unique()
            .sort()
            .to_frame()
            .join(
                props_as_df_cat["time"].unique().sort().to_frame(),
                how="cross",
            )
            .join(
                props_as_df_cat["jet"].unique().sort(descending=True).to_frame(),
                how="cross",
            )
        )
    elif "cluster" in index_columns:
        dummy_indexer = (
            props_as_df_cat["cluster"]
            .unique()
            .sort()
            .to_frame()
            .join(
                props_as_df_cat["jet"].unique().sort(descending=True).to_frame(),
                how="cross",
            )
        )
    else:
        dummy_indexer = (
            props_as_df_cat["time"]
            .unique()
            .sort()
            .to_frame()
            .join(
                props_as_df_cat["jet"].unique().sort(descending=True).to_frame(),
                how="cross",
            )
        )
    new_index_columns = get_index_columns(
        props_as_df_cat, ("member", "time", "cluster", "jet", "spell", "relative_index")
    )

    sort_descending = [False] * len(new_index_columns)
    sort_descending[-1] = True
    props_as_df_cat = dummy_indexer.join(
        props_as_df_cat, on=[pl.col(col) for col in new_index_columns], how="left"
    ).sort(new_index_columns, descending=sort_descending)
    props_as_df_cat = props_as_df_cat.with_columns(pl.col("njets").fill_null(0))
    return props_as_df_cat

coarsen_pl(df, coarsen_map)

Coarsening for polars DataFrame

Source code in jetutils/jet_finding.py 
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
def coarsen_pl(df: DataFrame, coarsen_map: dict[str, float]) -> DataFrame:
    """
    Coarsening for polars DataFrame
    """
    index_columns: list[str] = get_index_columns(df)
    other_columns = [
        col for col in df.columns if col not in [*index_columns, *list(coarsen_map)]
    ]
    by = [
        *index_columns,
        *[pl.col(col).floordiv(val) for col, val in coarsen_map.items()],
    ]
    agg = [pl.col(col).max() for col in other_columns]
    # agg = [*[pl.col(col).alias(f"{col}_").mean() for col, val in coarsen_map.items()], *agg]
    df = df.group_by(by, maintain_order=True).agg(*agg)
    # df = df.drop(list(coarsen_map)).rename({f"{col}_": col for col in coarsen_map})
    return df

compute_jet_props(jets)

Computes all basic jet properties from a DataFrame containing many jets.

Source code in jetutils/jet_finding.py 
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
def compute_jet_props(jets: DataFrame) -> DataFrame:
    """
    Computes all basic jet properties from a DataFrame containing many jets.
    """
    position_columns = [col for col in ["lat", "lev", "theta"] if col in jets.columns]

    def dl(col):
        return pl.col(col).max() - pl.col(col).min()

    mean_lon = circular_mean("lon", "s").alias("mean_lon")

    diff_lon = pl.col("lon").diff()
    diff_lon = (
        pl.when(diff_lon > 180)
        .then(diff_lon - 360)
        .when(diff_lon <= -180)
        .then(diff_lon + 360)
        .otherwise(diff_lon)
    )

    unraveled_lon = pl.col("lon").first() + diff_lon.cum_sum().fill_null(0.0)

    aggregations = [
        mean_lon,
        *[weighted_mean_pl(col, "s").alias(f"mean_{col}") for col in position_columns],
        pl.col("s").mean().alias("mean_s"),
        *[
            pl.col(col).get(pl.col("s").arg_max()).alias(f"{col}_star")
            for col in ["lon", "lat", "s"]
        ],
        *[dl(col).alias(f"{col}_ext") for col in ["lon", "lat"]],
        (
            pds.lin_reg_report(unraveled_lon, target=pl.col("lat"), add_bias=True)
            .struct.field("beta")
            .first()
            .alias("tilt")
        ),
        (
            1
            - pds.lin_reg_report(unraveled_lon, target=pl.col("lat"), add_bias=True)
            .struct
            .field("r2")
            .first()
        ).alias("waviness1"),
        (pl.col("lat") - pl.col("lat").mean()).pow(2).sum().alias("waviness2"),
        (
            pl.col("lat").gather(pl.col("lon").arg_sort()).diff().abs().sum()
            / dl("lon")
        ).alias("wavinessR16"),
        (
            jet_integral_haversine(pl.col("lon"), pl.col("lat"), x_is_one=True)
            / pl.lit(RADIUS)
            / pl.col("lat").mean().radians().cos()
            / dl("lon")
        ).alias("wavinessDC16"),
        (
            ((pl.col("v") - pl.col("v").mean()) * pl.col("v").abs() / pl.col("s")).sum()
            / pl.col("s").sum()
        ).alias("wavinessFV15"),
        jet_integral_haversine(pl.col("lon"), pl.col("lat"), pl.col("s")).alias("int"),
        jet_integral_haversine(
            pl.col("lon").filter(pl.col("lon") > -10),
            pl.col("lat").filter(pl.col("lon") > -10),
            pl.col("s").filter(pl.col("lon") > -10),
        ).alias("int_over_europe"),
        pl.col("is_polar").mean(),
    ]

    aggregations = [agg.replace([float("-inf"), float("inf"), float("nan")], None).first() for agg in aggregations]

    jets_lazy = jets.lazy()
    index_columns = get_index_columns(jets)
    if "member" not in index_columns:
        gb = jets_lazy.group_by(index_columns, maintain_order=True)
        props_as_df = gb.agg(*aggregations)
        props_as_df = props_as_df.collect()
    else:
        # streaming mode doesn't work well
        collected = []
        for member in tqdm(jets["member"].unique(maintain_order=True).to_numpy()):
            gb = jets_lazy.filter(pl.col("member") == member).group_by(
                index_columns, maintain_order=True
            )
            props_as_df = gb.agg(*aggregations)
            collected.append(props_as_df.collect())
        props_as_df = pl.concat(collected).sort("member")
    which_jet = "jet" if "jet" in index_columns else "jet ID"
    index_columns.remove(which_jet)
    unique_lon_over_europe = jets["lon"].unique()
    unique_lon_over_europe = unique_lon_over_europe.filter(unique_lon_over_europe > -10)
    dji = jets.group_by([*index_columns, "lon"]).agg(
        pl.col(which_jet).n_unique()
    )
    dji = unique_lon_over_europe.to_frame().join(dji, on="lon", how="left").group_by(index_columns).agg(double_jet_index=(pl.col(which_jet).fill_null(0) >= 2).mean())
    props_as_df = props_as_df.join(dji, on=index_columns)
    props_as_df = standardize_polars_dtypes(props_as_df)
    return props_as_df

compute_widths(jets, da)

Computes the width of each jet using normally interpolated wind speed on either side of the jet.

Source code in jetutils/jet_finding.py 
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
def compute_widths(jets: DataFrame, da: xr.DataArray):
    """
    Computes the width of each jet using normally interpolated wind speed on either side of the jet.
    """
    jets = gather_normal_da_jets(jets, da, half_length=1.3e6, dn=5e4, delete_middle=True, in_meters=True)

    index_columns = get_index_columns(
        jets, ("member", "time", "cluster", "spell", "relative_index", "jet ID", "jet")
    )

    # Expr half_width
    below = pl.col("s_interp") <= pl.max_horizontal(pl.col("s") / 4 * 3, pl.lit(10))
    stop_up = below.arg_max()
    nlo_up = pl.col("normallon").gather(stop_up)
    nla_up = pl.col("normallat").gather(stop_up)
    half_width_up = haversine(
        nlo_up, nla_up, pl.col("lon").first(), pl.col("lat").first()
    ).cast(pl.Float32)
    half_width_up = pl.when(below.any()).then(half_width_up).otherwise(float("nan"))

    stop_down = below.len() - below.reverse().arg_max() - 1
    nlo_down = pl.col("normallon").gather(stop_down)
    nla_down = pl.col("normallat").gather(stop_down)

    agg_out = {ic: pl.col(ic).first() for ic in ["lon", "lat", "s"]}

    half_width_down = haversine(
        nlo_down, nla_down, pl.col("lon").first(), pl.col("lat").first()
    ).cast(pl.Float32)
    half_width_down = pl.when(below.any()).then(half_width_down).otherwise(float("nan"))

    half_width = (
        pl.when(pl.col("side") == -1)
        .then(pl.col("half_width_down"))
        .otherwise(pl.col("half_width_up"))
        .list.first()
    )

    first_agg_out = agg_out | {
        "half_width_up": half_width_up,
        "half_width_down": half_width_down,
    }
    second_agg_out = agg_out | {"half_width": pl.col("half_width").sum()}
    third_agg_out = agg_out | {
        "width": (pl.col("half_width").fill_nan(None) * pl.col("s")).sum()
        / pl.col("s").sum()
    }

    jets = (
        jets.lazy()
        .group_by([*index_columns, "index", "side"], maintain_order=True)
        .agg(**first_agg_out)
        .with_columns(half_width=half_width)
        .drop(["half_width_up", "half_width_down", "side"])
        .group_by([*index_columns, "index"])
        .agg(**second_agg_out)
        .group_by(index_columns)
        .agg(**third_agg_out)
        .drop("lon", "lat", "s")
        .cast({"width": pl.Float32})
        .sort(index_columns)
    )
    return jets.collect()

do_everything(ds, save_path, bias_correct_realspace=False, do_smooth_spline=False, track_large=True, feature_names=None, n_init=10, **find_jets_kwargs)

More easily maintainable than object-oriented approach for quick testing of the whole pipeline

Parameters:

Name Type Description Default
ds Dataset

description

 required
save_path Path

description

 required
bias_correct_realspace bool

description, by default False

 False
do_smooth_spline bool

description, by default False

 False

Returns:

Type Description
_type_

description
Source code in jetutils/jet_finding.py 
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
def do_everything(ds: xr.Dataset, save_path: Path, bias_correct_realspace: bool = False, do_smooth_spline: bool = False, track_large: bool = True, feature_names: tuple | None = None, n_init: int = 10, **find_jets_kwargs):
    """
    More easily maintainable than object-oriented approach for quick testing of the whole pipeline

    Parameters
    ----------
    ds : xr.Dataset
        _description_
    save_path : Path
        _description_
    bias_correct_realspace : bool, optional
        _description_, by default False
    do_smooth_spline : bool, optional
        _description_, by default False

    Returns
    -------
    _type_
        _description_
    """
    save_path.mkdir(exist_ok=True)
    ds = standardize(ds)
    jets_path = save_path.joinpath("jets.parquet")
    props_path = save_path.joinpath("props.parquet")
    props_final_path = save_path.joinpath("props_full.parquet")
    cross_path = save_path.joinpath("cross.parquet")
    bc_path = save_path.joinpath("bias_correct.parquet")
    if not jets_path.is_file():
        iterator = iterate_over_year_maybe_member(da=ds, several_years=1)
        jets = []
        for indexer in tqdm(list(iterator)):
            ds_ = compute(ds.isel(**indexer), progress_flag=False)
            these_jets = find_all_jets(ds_, **find_jets_kwargs)
            if bias_correct_realspace:
                these_jets = online_bias_correct(these_jets, ds_)
            if do_smooth_spline:
                these_jets = spline_smooth(these_jets, s=2, factor=1)
            if bias_correct_realspace or do_smooth_spline:
                these_jets = join_on_ds(these_jets, ds_, fix_id=True)
            jets.append(these_jets)
        jets = pl.concat(jets)
        if "int" not in jets.columns:
            index_columns = get_index_columns(jets)
            int_expr = jet_integral_haversine(pl.col("lon"), pl.col("lat"), pl.col("s"))
            int_expr = int_expr.over(index_columns)
            jets = jets.with_columns(int=int_expr)
        jets = standardize_polars_dtypes(jets)
        jets.write_parquet(jets_path)
    else:
        jets = standardize_polars_dtypes(pl.read_parquet(jets_path))
    if "is_polar" not in jets.columns:
        if feature_names is None:
            feature_names = ("s", "theta")
        jets = is_polar_gmix(jets, feature_names=feature_names, n_init=n_init)
        jets.write_parquet(jets_path)

    phat_jets = to_one_large(jets, int_EDJ_threshold=1.3e8)
    phat_filter = (pl.col("is_polar") < 0.5) | ((pl.col("is_polar") > 0.5) & (pl.col("int") > 1.3e8))
    if not cross_path.is_file():
        if track_large:
            cross = track_jets(phat_jets)
        else:
            cross = track_jets(jets)
        cross.write_parquet(cross_path)
    else:
        cross = standardize_polars_dtypes(pl.read_parquet(cross_path))

    pers = pers_from_cross(cross)    

    if not props_path.is_file():
        props = compute_jet_props(jets)
        width = []
        da = ds["s"]
        indexer = iterate_over_year_maybe_member(jets, da)
        for idx1, idx2 in tqdm(list(indexer)):
            these_jets = jets.filter(*idx1)
            da_ = compute(da.sel(**idx2), progress_flag=False)
            width_ = compute_widths(these_jets, da_)
            width.append(width_)
        width = pl.concat(width)
        index_columns = get_index_columns(width)
        props = props.join(width, on=index_columns, how="left").sort(index_columns)
        props.write_parquet(props_path)
    else:
        props = standardize_polars_dtypes(pl.read_parquet(props_path))
        index_columns = get_index_columns(props)

    if not props_final_path.is_file():
        if not track_large:
            props = props.join(pers, on=index_columns)
        phat_props = props.filter(phat_filter)

        phat_props = average_jet_categories(phat_props, polar_cutoff=0.5)
        index_columns = get_index_columns(phat_props)
        if track_large:
            phat_props = phat_props.join(pers, on=index_columns)
        phat_props.write_parquet(props_final_path)
    else:
        phat_props = standardize_polars_dtypes(pl.read_parquet(props_final_path))

    if not bias_correct_realspace and not bc_path.is_file():
        bc = create_bias_correction(phat_jets, ds)
        bc.write_parquet(bc_path)

    return jets, phat_jets, props, phat_props

extract_features(df, feature_names=None, season=None)

Tiny wrapper to extract columns and subset time from a polars DataFrame.

Source code in jetutils/jet_finding.py 
670
671
672
673
674
675
676
677
678
679
680
681
682
def extract_features(
    df: DataFrame,
    feature_names: Sequence | None = None,
    season: list | str | tuple | int | None = None,
) -> DataFrame:
    """
    Tiny wrapper to extract columns and subset time from a polars DataFrame.
    """
    df = extract_season_from_df(df, season)
    if feature_names is None:
        feature_names = ["mean_lon", "mean_lat", "s_star"]

    return df[feature_names]

find_all_jets(ds, n_coarsen=3, smooth_func=window_smooth_func, thresholds=None, base_s_thresh=0.5, alignment_thresh=0.6, int_thresh_factor=0.6, hole_size=10, **smooth_kwargs)

Main function to find all jets in a polars DataFrame containing at least the "lon", "lat", "u", "v" and "s" columns. Will group by any potential index columns to compute jets independently for every, timestep, member and / or cluster. Any other non-index column present in df (like "theta" or "lev") will be interpolated to the jet cores in the output.

Thresholds passed as a DataArray are wind speed thresholds. This Dataarray needs to have one value per timestep present in df. If not passed, base_s_thresh is used for all times.

The jet integral threshold is computed from the wind speed threshold.

Source code in jetutils/jet_finding.py 
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
def find_all_jets(
    ds: xr.Dataset,
    n_coarsen: int = 3,
    smooth_func: Callable = window_smooth_func,
    thresholds: xr.DataArray | None = None,
    base_s_thresh: float = 0.5,
    alignment_thresh: float = 0.6,
    int_thresh_factor: float = 0.6,
    hole_size: int = 10,
    **smooth_kwargs,
):
    """
    Main function to find all jets in a polars DataFrame containing at least the "lon", "lat", "u", "v" and "s" columns. Will group by any potential index columns to compute jets independently for every, timestep, member and / or cluster. Any other non-index column present in `df` (like "theta" or "lev") will be interpolated to the jet cores in the output.

    Thresholds passed as a DataArray are wind speed thresholds. This Dataarray needs to have one value per timestep present in `df`. If not passed, `base_s_thresh` is used for all times.

    The jet integral threshold is computed from the wind speed threshold.
    """
    # process input
    ds, ds_orig = preprocess_ds(ds, n_coarsen=n_coarsen, smooth_func=smooth_func, **smooth_kwargs)
    df = xarray_to_polars(ds_orig)
    x_periodic = has_periodic_x(ds)
    index_columns = get_index_columns(
        df,
        (
            "member",
            "number",
            "forecast_init",
            "time",
            "cluster",
            "spell",
            "relative_index",
            "sample_index",
            "inside_index",
        ),
    )

    # thresholds
    dl = np.radians(df["lon"].max() - df["lon"].min())
    base_int_thresh = (
        RADIUS * dl * base_s_thresh * np.cos(np.pi / 4) * int_thresh_factor
    )
    ic_nomemb = [ic for ic in index_columns if ic not in ["member", "number"]]
    if base_s_thresh <= 1.0:
        thresholds = df.group_by(ic_nomemb).agg(
            pl.col("s").quantile(base_s_thresh).alias("s_thresh")
        )
        base_s_thresh = thresholds["s_thresh"].mean()  # disgusting
        base_int_thresh = (
            RADIUS * dl * base_s_thresh * np.cos(np.pi / 4) * int_thresh_factor
        )
    elif thresholds is not None:
        thresholds = (
            xarray_to_polars(thresholds)
            .drop("quantile")
            .cast({"s": pl.Float32})
            .rename({"s": "s_thresh"})
        )
    if thresholds is not None:
        df = df.join(
            thresholds, on=ic_nomemb
        )  # ugly
        df = df.with_columns(
            int_thresh=base_int_thresh * pl.col("s_thresh") / base_s_thresh
        )
        condition_expr = (pl.col("s") > pl.col("s_thresh")) & (
            pl.col("alignment") > alignment_thresh
        )
        condition_expr2 = pl.col("int") > pl.col("int_thresh")
    else:
        condition_expr = (pl.col("s") > base_s_thresh) & (
            pl.col("alignment") > alignment_thresh
        )
        condition_expr2 = pl.col("int") > base_int_thresh

    # contours
    repeat_lons = 120 if x_periodic else 0.
    all_contours = detect_contours_lonlat(
        (ds["sigma"] > 0).astype(np.uint8), [0.0], processes=N_WORKERS, repeat_lons=repeat_lons,
        do_round=False,
    ).drop("level")

    int_expr = jet_integral_haversine(pl.col("lon"), pl.col("lat"), pl.col("s"))
    int_expr = int_expr.over([*index_columns, "jet ID"])
    distance_ends_expr = haversine(
        pl.col("lon").first(),
        pl.col("lat").first(),
        pl.col("lon").last(),
        pl.col("lat").last(),
    )
    distance_ends_expr = distance_ends_expr.over([*index_columns, "jet ID"])

    for index_column in index_columns:
        try:
            all_contours = all_contours.cast({index_column: df[index_column].dtype})
        except ColumnNotFoundError:
            pass
    all_contours = all_contours.join(df, on=[*index_columns, "lon", "lat"], how="left")
    all_contours = compute_alignment(all_contours, x_periodic)

    # jets from contours
    ## consecutive runs of contour points respecting both point wise conditions, allowing holes of size up to three
    # return all_contours, condition_expr, index_columns, hole_size
    valids = (
        explode_rle(
            do_rle(
                all_contours.with_columns(condition=condition_expr),
                [*index_columns, "contour"],
            )
            .with_columns(value=pl.col("value").fill_null(False))
            .filter(
                (pl.col("value").not_() & (pl.col("len") < hole_size)) | pl.col("value")
            )
            .filter(
                pl.col("value").cum_sum().over(*index_columns, "contour") > 0,
                ~(
                    ~pl.col("value")
                    & (
                        pl.col("start")
                        == pl.col("start").max().over(*index_columns, "contour")
                    )
                ),
            )
        )
        .drop("value", "len", "start")
        .group_by([*index_columns, "contour"], maintain_order=True)
        .agg(pl.col("index"), len=pl.col("index").len())
        .filter(pl.col("len") > 5)
        .explode("index")
    )
    jets = valids.join(all_contours, on=[*index_columns, "contour", "index"])

    jets = (
        sort_by_index_then_difflon(
            jets.with_columns(len=pl.len().over([*index_columns, "contour"])).filter(
                pl.col("len") > 6
            ),
            index_columns,
            "contour",
        )
        .with_columns(diff=diff_exp().over([*index_columns, "contour"]))
        .with_columns(
            contour=(pl.col("contour") + 0.01 * (pl.col("diff") > 3).cum_sum())
            .rle_id()
            .over(index_columns)
        )
        .rename({"contour": "jet ID"})
        .drop("cyclic", "len", "len_right")
    )
    jets = (
        sort_by_index_then_newindex(
            jets.with_columns(len=pl.len().over([*index_columns, "jet ID"])).filter(
                pl.col("len") > 6
            ),
            index_columns,
            "jet ID",
        )
        .with_columns(
            int=int_expr,
            distance_ends=distance_ends_expr,
        )
        .filter(condition_expr2, pl.col("distance_ends") > 1e6)
        .drop("distance_ends")
        .with_columns(pl.col("jet ID").rle_id().over([*index_columns]))
    )
    jets = sort_by_index_then_newindex(jets, index_columns, "jet ID")
    jets = standardize_polars_dtypes(jets)
    return jets

gaussian_smooth_func(da, sigma_lon=1.0, sigma_lat=1.0)

Paper-thin wrapper around scipy.ndimage.gaussian_smooth

Parameters:

Name Type Description Default
da DataArray

description

 required
sigma_lon float

description, by default 1.

 1.0
sigma_lat float

description, by default 1.

 1.0

Returns:

Type Description
_type_

description
Source code in jetutils/jet_finding.py 
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
def gaussian_smooth_func(da: xr.DataArray, sigma_lon: float = 1., sigma_lat: float = 1.):
    """
    Paper-thin wrapper around `scipy.ndimage.gaussian_smooth`

    Parameters
    ----------
    da : xr.DataArray
        _description_
    sigma_lon : float, optional
        _description_, by default 1.
    sigma_lat : float, optional
        _description_, by default 1.

    Returns
    -------
    _type_
        _description_
    """
    sigmas = tuple(
        [0] * len([dim for dim in da.dims if dim not in ["lon", "lat"]])
    )
    sigmas = sigmas + (sigma_lat, sigma_lon)
    return gaussian_filter(da.values, sigmas)

get_double_jet_index(jet_pos_da, diff_cat=False)

Adds a new columns to props_as_df; "double_jet_index", by checking, for all longitudes, if there are at least two jet core points along the latitude, then averaging this over longitudes above 20° West.

Source code in jetutils/jet_finding.py 
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
def get_double_jet_index(jet_pos_da: xr.DataArray, diff_cat: bool = False):
    """
    Adds a new columns to props_as_df; `"double_jet_index"`, by checking, for all longitudes, if there are at least two jet core points along the latitude, then averaging this over longitudes above 20° West.
    """
    if diff_cat:
        overlap = (jet_pos_da >= 0.5).any("lat") & (jet_pos_da < 0.5).any("lat") 
    else:
        overlap = (~np.isnan(jet_pos_da)).sum("lat") >= 2
    overlap = overlap.rename("double_jet_index")
    return xarray_to_polars(overlap.sel(lon=slice(-20, None, None)).mean("lon"))

has_periodic_x(df)

Checks if the lon column contains both sides of the +-180 line. Only makes sense if data went through .data.standardize().

Parameters:

Name Type Description Default
df DataFrame

A DataFrame containing the lon column.

 required

Returns:

Type Description
bool
Source code in jetutils/jet_finding.py 
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
def has_periodic_x(df: DataFrame | xr.Dataset | xr.DataArray) -> bool:
    """
    Checks if the `lon` column contains both sides of the +-180 line. Only makes sense if data went through `.data.standardize()`.

    Parameters
    ----------
    df : DataFrame
        A DataFrame containing the `lon` column.

    Returns
    -------
    bool
    """
    if isinstance(df, DataFrame):
        lon = df["lon"].unique().sort().to_numpy()
    else:
        lon = np.sort(df["lon"].values)
    dx = lon[1] - lon[0]
    return (-180 in lon) and ((180 - dx) in lon)

is_polar_gmix(jets, feature_names, mode='week', n_components=2, n_init=20, init_params='random_from_data', v2=True)

Trains one or several Gaussian Mixture model independently, depending on the mode argument.

Parameters:

Name Type Description Default
df DataFrame

DataFrame to cluster

 required
feature_names list

Which columns to cluster on

 required
mode "all", "season", "month" or "week

Trains one model if "all", otherwise train one independent model for each season, month or week of the year. By default "week".

 'week'
n_components int

Number of Gaussian components, by default 2

 2
n_init int

Number of repeated independent training with. Can massively increase run time. By default 20

 20
init_params str

Type of init, by default "random_from_data"

 'random_from_data'

Returns:

Type Description
DataFrame

DataFrame of same length as df with the same index columns, the feature_names columns and a new is_polar column, corresponding to the proability of each row to belong to the eddy-driven jet component.
Source code in jetutils/jet_finding.py 
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
def is_polar_gmix(
    jets: DataFrame,
    feature_names: list,
    mode: (
        Literal["all"] | Literal["season"] | Literal["month"] | Literal["week"]
    ) = "week",
    n_components: int | Sequence = 2,
    n_init: int = 20,
    init_params: str = "random_from_data",
    v2: bool = True,
) -> DataFrame:
    """
    Trains one or several Gaussian Mixture model independently, depending on the `mode` argument.

    Parameters
    ----------
    df : DataFrame
        DataFrame to cluster
    feature_names : list
        Which columns to cluster on
    mode : "all", "season", "month" or "week
        Trains one model if "all", otherwise train one independent model for each season, month or week of the year. By default "week".
    n_components : int
        Number of Gaussian components, by default 2
    n_init : int
        Number of repeated independent training with. Can massively increase run time. By default 20
    init_params : str, optional
        Type of init, by default "random_from_data"

    Returns
    -------
    DataFrame
        DataFrame of same length as `df` with the same index columns, the `feature_names` columns and a new `is_polar` column, corresponding to the proability of each row to belong to the eddy-driven jet component.
    """
    # TODO: assumes at least one year of data, check for season / month actually existing in the data, figure out output
    kwargs = dict(n_init=n_init, init_params=init_params)
    gmix_fn = one_gmix_v2 if v2 else one_gmix
    if "time" not in jets.columns:
        mode = "all"
    if mode == "all":
        X = extract_features(jets, feature_names)
        kwargs["n_components"] = n_components
        probas = gmix_fn(X, **kwargs)
        return jets.with_columns(is_polar=probas)
    index_columns = get_index_columns(jets, ["member", "number", "forecast_init", "time", "jet ID", "index"])
    to_concat = []
    if mode == "season":
        if isinstance(n_components, int):
            n_components = [n_components] * 4
        else:
            assert len(n_components) == 4
        for season, n_components_ in zip(
            tqdm(["DJF", "MAM", "JJA", "SON"]), n_components
        ):
            X = extract_season_from_df(jets, season)
            X = X[[*index_columns, *feature_names]]
            kwargs["n_components"] = n_components_
            probas = gmix_fn(X[feature_names], **kwargs)
            to_concat.append(X.with_columns(is_polar=probas).drop(feature_names))
    elif mode == "month":
        months = jets["time"].dt.month().unique().sort().to_numpy()
        if isinstance(n_components, int):
            n_components = [n_components] * len(months)
        else:
            assert len(n_components) == len(months)
        for month, n_components_ in zip(tqdm(months), n_components):
            X = extract_season_from_df(jets, month)
            X = X[[*index_columns, *feature_names]]
            kwargs["n_components"] = n_components_
            probas = gmix_fn(X[feature_names], **kwargs)
            to_concat.append(X.with_columns(is_polar=probas).drop(feature_names))
    elif mode == "week":
        weeks = jets["time"].dt.week().unique().sort().to_numpy()
        if isinstance(n_components, int):
            n_components = [n_components] * len(weeks)
        else:
            assert len(n_components) == len(weeks)
        for week, n_components_ in zip(tqdm(weeks, total=len(weeks)), n_components):
            X = jets.filter(pl.col("time").dt.week() == week)
            X = X[[*index_columns, *feature_names]]
            kwargs["n_components"] = n_components_
            probas = gmix_fn(X[feature_names], **kwargs)
            to_concat.append(X.with_columns(is_polar=probas).drop(feature_names))

    return jets.join(pl.concat(to_concat), on=index_columns).sort(index_columns)

jet_position_as_da(jets)

Constructs a DataArray of dimensions (*index_columns, lat, lon) from jets. The DataArray starts with NaNs everywhere. Then, for every jet point, the DataArray is filled with the jet point's "is_polar" value.

Source code in jetutils/jet_finding.py 
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
def jet_position_as_da(
    jets: DataFrame,
) -> xr.DataArray:
    """
    Constructs a `DataArray` of dimensions (*index_columns, lat, lon) from jets. The DataArray starts with NaNs everywhere. Then, for every jet point, the DataArray is filled with the jet point's `"is_polar"` value.
    """
    index_columns = get_index_columns(
        jets, ("member", "time", "cluster", "spell", "relative_index")
    )
    all_jets_pandas = (
        jets.group_by([*index_columns, "lon", "lat"], maintain_order=True)
        .agg(pl.col("is_polar").mean())
        .to_pandas()
    )
    da_jet_pos = xr.Dataset.from_dataframe(
        all_jets_pandas.set_index([*index_columns, "lat", "lon"])
    )["is_polar"]
    return da_jet_pos

join_on_ds(jets, ds, fix_id=False)

Wrapper around .join, neareast_mapping and potentially xarray_to_polars if you provided a xarray object. For now we're doing nearest interpolation but we want to support bilinear interpolation in the future too, since we already have the code for it in geospatial.interp_from_other.

Parameters:

Name Type Description Default
jets _type_

description

 required
ds Dataset | DataArray | DataFrame

description

 required
fix_id bool

description, by default False

 False

Returns:

Type Description
_type_

description
Source code in jetutils/jet_finding.py 
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
def join_on_ds(jets, ds: xr.Dataset | xr.DataArray | pl.DataFrame, fix_id: bool = False):
    """
    Wrapper around `.join`, `neareast_mapping` and potentially `xarray_to_polars` if you provided a `xarray` object. For now we're doing nearest interpolation but we want to support bilinear interpolation in the future too, since we already have the code for it in `geospatial.interp_from_other`. 

    Parameters
    ----------
    jets : _type_
        _description_
    ds : xr.Dataset | xr.DataArray | pl.DataFrame
        _description_
    fix_id : bool, optional
        _description_, by default False

    Returns
    -------
    _type_
        _description_
    """
    index_columns = get_index_columns(jets, ["member", "number", "cluster", "time"])
    other_indexer = get_index_columns(jets, ["jet", "jet ID", "contour", "level"])
    if isinstance(ds, xr.DataArray | xr.Dataset):
        ds = xarray_to_polars(ds)
    lo = nearest_mapping(jets, ds, "lon")
    la = nearest_mapping(jets, ds, "lat")

    jets = (
        jets
        .join(lo, on="lon")
        .drop("lon")
        .rename({"lon_": "lon"})
        .join(la, on="lat")
        .drop("lat")
        .rename({"lat_": "lat"})
        .unique([*index_columns, *other_indexer, "lon", "lat"], maintain_order=True)
        .join(ds, on=[*index_columns, "lon", "lat"], how="left")
    )
    if fix_id:
        jets = jets.with_columns(pl.col("index").rle_id().over([*index_columns, *other_indexer]))

    return jets

one_gmix(X, n_components=2, n_init=20, init_params='random_from_data')

Trains one Gaussian Mixture model, and outputs the predicted probability of all data points on the component identified as the eddy-driven jet.

Parameters:

Name Type Description Default
X DataFrame

Input data

 required
n_components int

Number of Gaussian components, by default 2

 2
n_init int

Number of repeated independent training with. Can massively increase run time. By default 20

 20
init_params str

Type of init, by default "random_from_data"

 'random_from_data'

Returns:

Type Description
ndarray

Probabilities of every point on the Gaussian component identified as the eddy-driven jet.
References

https://scikit-learn.org/stable/modules/generated/sklearn.mixture.GaussianMixture.html

Source code in jetutils/jet_finding.py 
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
def one_gmix(
    X,
    n_components=2,
    n_init=20,
    init_params="random_from_data",
):
    """
    Trains one Gaussian Mixture model, and outputs the predicted probability of all data points on the component identified as the eddy-driven jet.

    Parameters
    ----------
    X : DataFrame
        Input data
    n_components : int
        Number of Gaussian components, by default 2
    n_init : int
        Number of repeated independent training with. Can massively increase run time. By default 20
    init_params : str, optional
        Type of init, by default "random_from_data"

    Returns
    -------
    ndarray
        Probabilities of every point on the Gaussian component identified as the eddy-driven jet.

    References
    ----------
    https://scikit-learn.org/stable/modules/generated/sklearn.mixture.GaussianMixture.html
    """
    model = GaussianMixture(
        n_components=n_components, init_params=init_params, n_init=n_init
    )
    if "ratio" in X.columns:
        X = X.with_columns(ratio=pl.col("ratio").fill_null(1.0).fill_nan(1.0))
    model = model.fit(X)
    if X.columns[1] == "theta":
        return 1 - model.predict_proba(X)[:, np.argmax(model.means_[:, 1])]
    elif X.columns[1] == "lat":
        return 1 - model.predict_proba(X)[:, np.argmin(model.means_[:, 1])]
    return 1 - model.predict_proba(X)[:, np.argmax(model.means_[:, 0])]

one_gmix_v2(X, n_components=2, n_init=20, init_params='random_from_data')

Trains one Gaussian Mixture model, and outputs the predicted probability of all data points on the component identified as the eddy-driven jet.

Parameters:

Name Type Description Default
X DataFrame

Input data

 required
n_components int

Number of Gaussian components, by default 2

 2
n_init int

Number of repeated independent training with. Can massively increase run time. By default 20

 20
init_params str

Type of init, by default "random_from_data"

 'random_from_data'

Returns:

Type Description
ndarray

Probabilities of every point on the Gaussian component identified as the eddy-driven jet.
References

https://scikit-learn.org/stable/modules/generated/sklearn.mixture.GaussianMixture.html

Source code in jetutils/jet_finding.py 
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
def one_gmix_v2(
    X,
    n_components=2,
    n_init=20,
    init_params="random_from_data",
):
    """
    Trains one Gaussian Mixture model, and outputs the predicted probability of all data points on the component identified as the eddy-driven jet.

    Parameters
    ----------
    X : DataFrame
        Input data
    n_components : int
        Number of Gaussian components, by default 2
    n_init : int
        Number of repeated independent training with. Can massively increase run time. By default 20
    init_params : str, optional
        Type of init, by default "random_from_data"

    Returns
    -------
    ndarray
        Probabilities of every point on the Gaussian component identified as the eddy-driven jet.

    References
    ----------
    https://scikit-learn.org/stable/modules/generated/sklearn.mixture.GaussianMixture.html
    """
    model = GaussianMixture(
        n_components=n_components, init_params=init_params, n_init=n_init
    )
    if "ratio" in X.columns:
        X = X.with_columns(ratio=pl.col("ratio").fill_null(1.0))
    model = model.fit(X)
    scores = []
    for mean, covar in zip(model.means_, model.covariances_):
        covar = sqrtm(np.linalg.inv(covar))
        x = X.to_numpy() - mean[None, :]
        score = np.linalg.norm(np.einsum("jk,ik->ij", covar, x), axis=1)
        scores.append(score)
    if X.columns[1][:5] == "theta":
        order = np.argsort(model.means_[:, 1])
    else:
        order = np.argsort(model.means_[:, 1])[::-1]
    otherscores = np.sum([scores[k] for k in order[:-1]], axis=0)
    return 1 / (1 + otherscores / scores[order[-1]])

online_bias_correct(jets, ds)

This is a fun idea but it does not work great. Use the bias correction function to find how far the actual sigma contour is around the detected (biased) jet is, then send a ray there using the formulae for normallon and normallat, and define the new jet at these new points. You shoud spline-interpolate after this for good measure.

For now, use biased jets and only bias correct the jet-centred composites, it's fine. Issue is of course sigma contours are super noisy, and while the smoothing works much better in jet-centred composites it still biases the position so you can't do too much of it. You're left with the same problem as in real space almost.

Parameters:

Name Type Description Default
jets DataFrame

description

 required
ds Dataset

description

 required

Returns:

Type Description
DataFrame

description
Source code in jetutils/jet_finding.py 
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
def online_bias_correct(jets: pl.DataFrame, ds: xr.Dataset) -> pl.DataFrame:
    """
    This is a fun idea but it does not work great. 
    Use the bias correction function to find how far the actual sigma contour is around the detected (biased) jet is, then send a ray there using the formulae for normallon and normallat, and define the new jet at these new points. You shoud spline-interpolate after this for good measure.

    For now, use biased jets and only bias correct the jet-centred composites, it's fine. Issue is of course sigma contours are super noisy, and while the smoothing works much better in jet-centred composites it still biases the position so you can't do too much of it. You're left with the same problem as in real space almost.

    Parameters
    ----------
    jets : pl.DataFrame
        _description_
    ds : xr.Dataset
        _description_

    Returns
    -------
    pl.DataFrame
        _description_
    """
    index_columns = get_index_columns(jets)
    jets_orig = jets.clone()
    jets = bias_correct(jets, ds, same_len=False)

    arc_distances = pl.col("n").first() / RADIUS
    lon = pl.col("lon").first().radians()
    lat = pl.col("lat").first().radians()
    angle = pl.col("angle").first()

    newlat = (
        lat.sin() * arc_distances.cos() + lat.cos() * arc_distances.sin() * angle.sin()
    ).arcsin()
    newlat = newlat.degrees()

    newlon = lon + pl.arctan2(
        angle.cos() * arc_distances.sin() * lat.cos(),
        arc_distances.cos() - lat.sin() * newlat.sin(),
    )
    newlon = newlon.degrees()

    jets = (
        jets
        .group_by(*index_columns, "index", maintain_order=True)
        .agg(
            lon=newlon,
            lat=newlat,
        )
    )
    idx_new = jets.select(index_columns).unique(index_columns)
    idx_old = jets_orig.select(index_columns).unique(index_columns)
    left_behind = idx_old.join(idx_new, on=index_columns, how="anti")
    jets = pl.concat([jets, left_behind.join(jets_orig, on=index_columns).select(*jets.columns)]).sort([*index_columns, "index"])
    return jets

preprocess_ds(ds, n_coarsen=1, smooth_func=window_smooth_func, **smooth_kwargs)

Lego-type preprocessing, coarsens then applies smooth_func and also compute sigma the horizontal normal wind shear. It returns the unprocessed data too because it's useful. Extra kwargs are passed to smooth_func but you should really be building it with functools.partial because it's more explicit.

Parameters:

Name Type Description Default
ds Dataset

description

 required
n_coarsen int

description, by default 1

 1
smooth_func Callable

description, by default window_smooth_func

 window_smooth_func

Returns:

Type Description
_type_

description
Source code in jetutils/jet_finding.py 
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
def preprocess_ds(ds: xr.Dataset, n_coarsen: int = 1, smooth_func: Callable = window_smooth_func, **smooth_kwargs):
    """
    Lego-type preprocessing, coarsens then applies `smooth_func` and also compute sigma the horizontal normal wind shear. It returns the unprocessed data too because it's useful. Extra kwargs are passed to smooth_func but you should really be building it with `functools.partial` because it's more explicit.

    Parameters
    ----------
    ds : xr.Dataset
        _description_
    n_coarsen : int, optional
        _description_, by default 1
    smooth_func : Callable, optional
        _description_, by default window_smooth_func

    Returns
    -------
    _type_
        _description_
    """
    ds = standardize(ds)
    if "s" not in ds:
        ds["s"] = np.sqrt(ds["u"] ** 2 + ds["v"] ** 2)
    ds_orig = ds.copy()
    ds = coarsen_da(ds, n_coarsen=n_coarsen, reduce_func=np.mean)
    to_smooth = ["u", "v", "s", "theta"]
    for var in to_smooth:
        ds[var] = ds[var].copy(data=smooth_func(ds[var], **smooth_kwargs))
    ds["sigma"] = compute_norm_derivative(ds, "s")
    ds["sigma"] = ds["sigma"].copy(data=smooth_func(ds["sigma"], **smooth_kwargs))
    if "sigma" not in ds_orig:
        ds_orig["sigma"] = compute_norm_derivative(ds_orig, "s")
    return ds, ds_orig

round_half(x)

Round to the nearest integer or half integer

Parameters:

Name Type Description Default
x Number or array

Array-like to round

 required

Returns:

Type Description
Same as input

Rounded input
Source code in jetutils/jet_finding.py 
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
def round_half(x):
    """
    Round to the nearest integer or half integer

    Parameters
    ----------
    x : Number or array
        Array-like to round

    Returns
    -------
    Same as input
        Rounded input
    """
    return np.round(x * 2) / 2

round_polars(col, factor=2)

Generates an Expression that rounds the given column to a given base, one over the factor.

Source code in jetutils/jet_finding.py 
111
112
113
114
115
def round_polars(col: str, factor: int = 2) -> Expr:
    """
    Generates an Expression that rounds the given column to a given base, one over the factor.
    """
    return (to_expr(col) * factor).round() / factor

spline_smooth(jets, s=0.0, factor=3)

This works but was mostly useful in conjunction with bias_correct.

One day, polars-splines will be updated and this function will use it to be even faster instead of looping. maybe.

Parameters:

Name Type Description Default
jets DataFrame

description

 required
s float

description, by default 0.

 0.0
factor int

description, by default 3

 3

Returns:

Type Description
DataFrame

description
Source code in jetutils/jet_finding.py 
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
def spline_smooth(jets: pl.DataFrame, s: float = 0., factor: int = 3) -> pl.DataFrame:
    """
    This works but was mostly useful in conjunction with bias_correct. 

    One day, polars-splines will be updated and this function will use it to be even faster instead of looping. maybe.

    Parameters
    ----------
    jets : pl.DataFrame
        _description_
    s : float, optional
        _description_, by default 0.
    factor : int, optional
        _description_, by default 3

    Returns
    -------
    pl.DataFrame
        _description_
    """
    index_columns = get_index_columns(jets)
    newjets = []
    oldjets = []
    for indexer, newjet in tqdm(jets.group_by(index_columns), disable=True):
        lo, la = newjet["lon"].to_numpy(), newjet["lat"].to_numpy()
        try:
            tck, u = splprep([lo, la], s=s)
        except ValueError:
            oldjets.append(newjet[*index_columns, "index", "lon", "lat"])
            continue
        unew = np.linspace(u.min(), u.max(), len(u) * factor)
        los, las = splev(unew, tck)
        indexer = dict(zip(index_columns, indexer))
        df = indexer | {"index": np.arange(len(unew)), "lon": los, "lat": las}
        newjets.append(pl.DataFrame(df))
    newjets = standardize_polars_dtypes(pl.concat(newjets)).cast({"jet ID": pl.UInt32()})
    if len(oldjets) > 0:
        newjets = pl.concat([newjets, pl.concat(oldjets)])
    newjets =  newjets.unique([*index_columns, "index"]).sort(*index_columns, "index")
    return newjets

track_jets(all_jets_one_df, yearly=False, monthly=False)

Iterates over maybe years and maybe members and performs explicit jet tracking. Only use yearly and monthly if severely memory bound, it makes everything much slower

Parameters:

Name Type Description Default
all_jets_one_df DataFrame

Data source

 required

Returns:

Type Description
DataFrame

description
Source code in jetutils/jet_finding.py 
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
def track_jets(all_jets_one_df: DataFrame, yearly: bool = False, monthly: bool = False) -> DataFrame:
    """
    Iterates over maybe years and maybe members and performs explicit jet tracking. Only use yearly and monthly if severely memory bound, it makes everything much slower

    Parameters
    ----------
    all_jets_one_df : DataFrame
        Data source

    Returns
    -------
    DataFrame
        _description_
    """
    cross = []
    do_member = "member" in all_jets_one_df.columns
    if do_member:
        members = all_jets_one_df["member"].unique()
    else:
        members = [None]

    if yearly:
        years = all_jets_one_df["time"].dt.year().unique()
    else:
        years = [None]

    if monthly:
        months = all_jets_one_df["time"].dt.year().unique()
    else:
        months = [None]
    iterator = list(product(members, years, months))
    total = len(iterator)

    for member, year, month in tqdm(iterator, total=total):
        cross.append(
            track_jets_(
                all_jets_one_df,
                member,
                year,
                month,
            )
        )
    cross = pl.concat(cross)
    return cross

track_jets_(jets, member=None, year=None, month=None)

Performs jet tracking for real

Parameters:

Name Type Description Default
jets DataFrame

description

 required
year int

description

 None
member str | None

description, by default None

 None

Returns:

Name Type Description
cross DataFrame

cross-jet distance metrics from a jet and its best successor
Source code in jetutils/jet_finding.py 
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
def track_jets_(
    jets: DataFrame, member: str | None = None, year: int | None = None, month: int | None = None,
) -> DataFrame:
    """
    Performs jet tracking for real

    Parameters
    ----------
    jets : DataFrame
        _description_
    year : int
        _description_
    member : str | None, optional
        _description_, by default None

    Returns
    -------
    cross: DataFrame
        cross-jet distance metrics from a jet and its best successor
    """
    jets = jets.cast({"time": pl.Datetime("ms")})
    if "len_right" in jets.columns:
        jets = jets.drop("len_right")
    deltas = ["s", "theta"]
    if "is_polar" in jets.columns:
        deltas.append("is_polar")
    deltas2 = [f"d{col}" for col in deltas]
    which_jet = "jet" if "jet" in jets.columns else "jet ID"
    typical_group_by = ["time", "time_right", which_jet, f"{which_jet}_right"]
    filter_ = pl.lit(True) 
    if year is not None:
        filter_ = filter_ & (pl.col("time").dt.year() == year)
    if month is not None:
        filter_ = filter_ & (pl.col("time").dt.month() == month)
    if year is not None and (year + 1) in jets["time"].dt.year().unique() and (month is None or month == 12):
        filter_ = filter_ | pl.col("time").is_in(
            pl.col("time")
            .filter(pl.col("time").dt.year() == year + 1)
            .unique()
            .bottom_k(1)
            .implode()
        )
    if "member" in jets.columns and member is not None:
        filter_ = filter_ & (pl.col("member") == member)
        typical_group_by.insert(0, "member")

    cols_i_want = get_index_columns(jets, ["member", "time", "jet ID", "jet", "lon", "lat"])
    jets_current = (
        jets.filter(filter_)
        .select(*cols_i_want, *deltas)
        .with_columns(
            len=pl.col("lon").len().over(["time", which_jet]),
            index=pl.int_range(0, pl.col("lon").len()).over(["time", which_jet]),
        )
    )
    dt = jets_current["time"].unique().bottom_k(2).sort()
    dt = dt[1] - dt[0]
    jets_next = jets_current.with_columns(time_shifted=pl.col("time") - dt)
    jets_current = jets_current.filter(filter_)

    aggs_first = {"points": pl.concat_arr("lon", "lat"), "s": weighted_mean_pl("s")} | {
        d: weighted_mean_pl(d, "s") for d in deltas[1:]
    }

    memb = ["member"] if "member" in jets_current.columns else []

    jets_current = jets_current.group_by(*memb, "time", which_jet).agg(**aggs_first)
    jets_next = jets_next.group_by(*memb, "time", "time_shifted", which_jet).agg(**aggs_first)
    cross = jets_current.join(
        jets_next, left_on=[*memb, "time"], right_on=[*memb, "time_shifted"], how="left"
    )
    if which_jet == "jet":
        cross = cross.filter(pl.col(which_jet) == pl.col(f"{which_jet}_right"))
    cross = cross.filter(pl.col("points").is_not_null(), pl.col("points_right").is_not_null())
    overlap = _lon_overlap()
    more_aggs = {
        "dist": _frechet_st(),
        "lon_overlap": overlap,
    } | {
        d2: (pl.col(d1) - pl.col(f"{d1}_right")).abs()
        for d1, d2 in zip(deltas, deltas2)
    }
    cross = (
        cross.with_columns(**more_aggs)
        .drop(*[f"{name}{suffix}" for name in ["points"] for suffix in ["", "_right"]])
        .drop_nulls("lon_overlap")
        .sort(*typical_group_by)
    )
    return standardize_polars_dtypes(cross)