Source code for pygmt.src.grd2xyz

"""
grd2xyz - Convert grid to data table
"""
import warnings

import pandas as pd
import xarray as xr
from pygmt.clib import Session
from pygmt.exceptions import GMTInvalidInput
from pygmt.helpers import (
    GMTTempFile,
    build_arg_string,
    fmt_docstring,
    kwargs_to_strings,
    use_alias,
)


[docs]@fmt_docstring @use_alias( R="region", V="verbose", o="outcols", ) @kwargs_to_strings(R="sequence", o="sequence_comma") def grd2xyz(grid, output_type="pandas", outfile=None, **kwargs): r""" Convert grid to data table. Read a grid and output xyz-triplets as a :class:`numpy.ndarray`, :class:`pandas.DataFrame`, or ASCII file. Full option list at :gmt-docs:`grd2xyz.html` {aliases} Parameters ---------- grid : str or xarray.DataArray The file name of the input grid or the grid loaded as a :class:`xarray.DataArray`. This is the only required parameter. output_type : str Determine the format the xyz data will be returned in [Default is ``pandas``]: - ``numpy`` - :class:`numpy.ndarray` - ``pandas``- :class:`pandas.DataFrame` - ``file`` - ASCII file (requires ``outfile``) outfile : str The file name for the output ASCII file. {R} Adding ``region`` will select a subsection of the grid. If this subsection exceeds the boundaries of the grid, only the common region will be output. {V} {o} Returns ------- ret : pandas.DataFrame or numpy.ndarray or None Return type depends on ``outfile`` and ``output_type``: - None if ``outfile`` is set (output will be stored in file set by ``outfile``) - :class:`pandas.DataFrame` or :class:`numpy.ndarray` if ``outfile`` is not set (depends on ``output_type``) """ if output_type not in ["numpy", "pandas", "file"]: raise GMTInvalidInput( "Must specify 'output_type' either as 'numpy', 'pandas' or 'file'." ) if outfile is not None and output_type != "file": msg = ( f"Changing 'output_type' of grd2xyz from '{output_type}' to 'file' " "since 'outfile' parameter is set. Please use output_type='file' " "to silence this warning." ) warnings.warn(message=msg, category=RuntimeWarning, stacklevel=2) output_type = "file" elif outfile is None and output_type == "file": raise GMTInvalidInput("Must specify 'outfile' for ASCII output.") if "o" in kwargs and output_type == "pandas": raise GMTInvalidInput( "If 'outcols' is specified, 'output_type' must be either 'numpy'" "or 'file'." ) # Set the default column names for the pandas dataframe header dataframe_header = ["x", "y", "z"] # Let output pandas column names match input DataArray dimension names if isinstance(grid, xr.DataArray) and output_type == "pandas": # Reverse the dims because it is rows, columns ordered. dataframe_header = [grid.dims[1], grid.dims[0], grid.name] with GMTTempFile() as tmpfile: with Session() as lib: file_context = lib.virtualfile_from_data(check_kind="raster", data=grid) with file_context as infile: if outfile is None: outfile = tmpfile.name arg_str = " ".join([infile, build_arg_string(kwargs), "->" + outfile]) lib.call_module("grd2xyz", arg_str) # Read temporary csv output to a pandas table if outfile == tmpfile.name: # if user did not set outfile, return pd.DataFrame result = pd.read_csv( tmpfile.name, sep="\t", names=dataframe_header, comment=">" ) elif outfile != tmpfile.name: # return None if outfile set, output in outfile result = None if output_type == "numpy": result = result.to_numpy() return result