roms-tools 1.4.2__py3-none-any.whl → 1.6.0__py3-none-any.whl

roms_tools/setup/boundary_forcing.py

@@ -3,19 +3,24 @@ import numpy as np
 import pandas as pd
 import yaml
 import importlib.metadata
+import warnings
 from typing import Dict, Union, List
 from dataclasses import dataclass, field, asdict
 from roms_tools.setup.grid import Grid
-from roms_tools.setup.mixins import ROMSToolsMixins
+from roms_tools.setup.regrid import LateralRegrid, VerticalRegrid
 from datetime import datetime
 from roms_tools.setup.datasets import GLORYSDataset, CESMBGCDataset
 from roms_tools.setup.utils import (
-    nan_check,
-    substitute_nans_by_fillvalue,
     get_variable_metadata,
-    get_boundary_info,
     group_dataset,
     save_datasets,
+    get_target_coords,
+    rotate_velocities,
+    compute_barotropic_velocity,
+    transpose_dimensions,
+    one_dim_fill,
+    nan_check,
+    substitute_nans_by_fillvalue,
 )
 from roms_tools.setup.plot import _section_plot, _line_plot
 import matplotlib.pyplot as plt
@@ -23,7 +28,7 @@ from pathlib import Path
 
 
 @dataclass(frozen=True, kw_only=True)
-class BoundaryForcing(ROMSToolsMixins):
+class BoundaryForcing:
     """Represents boundary forcing input data for ROMS.
 
     Parameters
@@ -55,6 +60,10 @@ class BoundaryForcing(ROMSToolsMixins):
 
     model_reference_date : datetime, optional
         Reference date for the model. Default is January 1, 2000.
+    apply_2d_horizontal_fill: bool, optional
+        Indicates whether to perform a two-dimensional horizontal fill on the source data prior to regridding to boundaries.
+        If `False`, a one-dimensional horizontal fill is performed separately on each of the four regridded boundaries.
+        Defaults to `False`.
     use_dask: bool, optional
         Indicates whether to use dask for processing. If True, data is processed with dask; if False, data is processed eagerly. Defaults to False.
 
@@ -84,6 +93,7 @@ class BoundaryForcing(ROMSToolsMixins):
     source: Dict[str, Union[str, Path, List[Union[str, Path]]]]
     type: str = "physics"
     model_reference_date: datetime = datetime(2000, 1, 1)
+    apply_2d_horizontal_fill: bool = False
     use_dask: bool = False
 
     ds: xr.Dataset = field(init=False, repr=False)
@@ -91,52 +101,153 @@ class BoundaryForcing(ROMSToolsMixins):
     def __post_init__(self):
 
         self._input_checks()
-        lon, lat, angle, straddle = super()._get_target_lon_lat()
+        target_coords = get_target_coords(self.grid)
 
         data = self._get_data()
-        data.choose_subdomain(
-            latitude_range=[lat.min().values, lat.max().values],
-            longitude_range=[lon.min().values, lon.max().values],
-            margin=2,
-            straddle=straddle,
-        )
-
-        if self.type == "physics":
-            vars_2d = ["zeta"]
-            vars_3d = ["temp", "salt", "u", "v"]
-        elif self.type == "bgc":
-            vars_2d = []
-            vars_3d = data.var_names.keys()
-
-        data_vars = super()._regrid_data(data, vars_2d, vars_3d, lon, lat)
 
-        if self.type == "physics":
-            data_vars = super()._process_velocities(data_vars, angle, "u", "v")
-        object.__setattr__(data, "data_vars", data_vars)
+        if self.apply_2d_horizontal_fill:
+            data.choose_subdomain(
+                target_coords,
+                buffer_points=20,  # lateral fill needs good buffer from data margin
+            )
+            data.extrapolate_deepest_to_bottom()
+            data.apply_lateral_fill()
 
-        d_meta = get_variable_metadata()
+        variable_info = self._set_variable_info(data)
         bdry_coords = get_boundary_info()
+        ds = xr.Dataset()
 
-        ds = self._write_into_dataset(data, d_meta, bdry_coords)
-
-        # NaN values at wet points indicate that the raw data did not cover the domain, and the following will raise a ValueError
-        # this check works only for 2D fields because for 3D I extrapolate to bottom which eliminates NaNs
         for direction in ["south", "east", "north", "west"]:
             if self.boundaries[direction]:
-                if type == "physics":
-                    nan_check(
-                        ds[f"zeta_{direction}"].isel(bry_time=0),
-                        self.grid.ds.mask_rho.isel(**bdry_coords["rho"][direction]),
+
+                bdry_target_coords = {
+                    "lat": target_coords["lat"].isel(
+                        **bdry_coords["vector"][direction]
+                    ),
+                    "lon": target_coords["lon"].isel(
+                        **bdry_coords["vector"][direction]
+                    ),
+                    "straddle": target_coords["straddle"],
+                }
+
+                bdry_data = data.choose_subdomain(
+                    bdry_target_coords,
+                    buffer_points=3,
+                    return_copy=True,
+                )
+
+                if not self.apply_2d_horizontal_fill:
+                    bdry_data.extrapolate_deepest_to_bottom()
+
+                processed_fields = {}
+
+                # lateral regridding of vector fields
+                vector_var_names = [
+                    name for name, info in variable_info.items() if info["is_vector"]
+                ]
+                if len(vector_var_names) > 0:
+                    lon = target_coords["lon"].isel(**bdry_coords["vector"][direction])
+                    lat = target_coords["lat"].isel(**bdry_coords["vector"][direction])
+                    lateral_regrid = LateralRegrid(
+                        {"lat": lat, "lon": lon}, bdry_data.dim_names
+                    )
+                    for var_name in vector_var_names:
+                        if var_name in bdry_data.var_names.keys():
+                            processed_fields[var_name] = lateral_regrid.apply(
+                                bdry_data.ds[bdry_data.var_names[var_name]]
+                            )
+
+                # lateral regridding of tracer fields
+                tracer_var_names = [
+                    name
+                    for name, info in variable_info.items()
+                    if not info["is_vector"]
+                ]
+                if len(tracer_var_names) > 0:
+                    lon = target_coords["lon"].isel(**bdry_coords["rho"][direction])
+                    lat = target_coords["lat"].isel(**bdry_coords["rho"][direction])
+                    lateral_regrid = LateralRegrid(
+                        {"lat": lat, "lon": lon}, bdry_data.dim_names
+                    )
+                    for var_name in tracer_var_names:
+                        if var_name in bdry_data.var_names.keys():
+                            processed_fields[var_name] = lateral_regrid.apply(
+                                bdry_data.ds[bdry_data.var_names[var_name]]
+                            )
+
+                # rotation of velocities and interpolation to u/v points
+                if "u" in variable_info and "v" in variable_info:
+                    angle = target_coords["angle"].isel(
+                        **bdry_coords["vector"][direction]
                     )
-                elif type == "bgc":
-                    nan_check(
-                        ds[f"ALK_{direction}"].isel(bry_time=0, s_rho=-1),
-                        self.grid.ds.mask_rho.isel(**bdry_coords["rho"][direction]),
+                    (processed_fields["u"], processed_fields["v"],) = rotate_velocities(
+                        processed_fields["u"],
+                        processed_fields["v"],
+                        angle,
+                        interpolate=True,
                     )
 
+                # selection of outermost margin for u/v variables
+                for var_name in variable_info.keys():
+                    if var_name in processed_fields:
+                        location = variable_info[var_name]["location"]
+                        if location in ["u", "v"]:
+                            processed_fields[var_name] = processed_fields[
+                                var_name
+                            ].isel(**bdry_coords[location][direction])
+
+                if not self.apply_2d_horizontal_fill:
+                    processed_fields = apply_1d_horizontal_fill(processed_fields)
+
+                # vertical regridding
+                for location in ["rho", "u", "v"]:
+                    var_names = [
+                        name
+                        for name, info in variable_info.items()
+                        if info["location"] == location and info["is_3d"]
+                    ]
+                    if len(var_names) > 0:
+                        vertical_regrid = VerticalRegrid(
+                            self.grid.ds[f"layer_depth_{location}"].isel(
+                                **bdry_coords[location][direction]
+                            ),
+                            bdry_data.ds[bdry_data.dim_names["depth"]],
+                        )
+                        for var_name in var_names:
+                            if var_name in processed_fields:
+                                processed_fields[var_name] = vertical_regrid.apply(
+                                    processed_fields[var_name]
+                                )
+
+                # compute barotropic velocities
+                if "u" in variable_info and "v" in variable_info:
+                    for var_name in ["u", "v"]:
+                        processed_fields[
+                            f"{var_name}bar"
+                        ] = compute_barotropic_velocity(
+                            processed_fields[var_name],
+                            self.grid.ds[f"interface_depth_{var_name}"].isel(
+                                **bdry_coords[var_name][direction]
+                            ),
+                        )
+
+                # Reorder dimensions
+                for var_name in processed_fields.keys():
+                    processed_fields[var_name] = transpose_dimensions(
+                        processed_fields[var_name]
+                    )
+
+                # Write the boundary data into dataset
+                ds = self._write_into_dataset(direction, processed_fields, ds)
+
+        # Add global information
+        ds = self._add_global_metadata(data, ds)
+
+        self._validate(ds, variable_info, bdry_coords)
+
         # substitute NaNs over land by a fill value to avoid blow-up of ROMS
-        for var in ds.data_vars:
-            ds[var] = substitute_nans_by_fillvalue(ds[var])
+        for var_name in ds.data_vars:
+            ds[var_name] = substitute_nans_by_fillvalue(ds[var_name])
 
         object.__setattr__(self, "ds", ds)
 
@@ -187,37 +298,88 @@ class BoundaryForcing(ROMSToolsMixins):
 
         return data
 
-    def _write_into_dataset(self, data, d_meta, bdry_coords):
+    def _set_variable_info(self, data):
+        """Sets up a dictionary with metadata for variables based on the type of data
+        (physics or BGC).
 
-        # save in new dataset
-        ds = xr.Dataset()
+        The dictionary contains the following information:
+        - `location`: Where the variable resides in the grid (e.g., rho, u, or v points).
+        - `is_vector`: Whether the variable is part of a vector (True for velocity components like 'u' and 'v').
+        - `vector_pair`: For vector variables, this indicates the associated variable that forms the vector (e.g., 'u' and 'v').
+        - `is_3d`: Indicates whether the variable is 3D (True for variables like 'temp' and 'salt') or 2D (False for 'zeta').
 
-        for direction in ["south", "east", "north", "west"]:
-            if self.boundaries[direction]:
+        Returns
+        -------
+        dict
+            A dictionary where the keys are variable names and the values are dictionaries of metadata
+            about each variable, including 'location', 'is_vector', 'vector_pair', and 'is_3d'.
+        """
+        default_info = {
+            "location": "rho",
+            "is_vector": False,
+            "vector_pair": None,
+            "is_3d": True,
+        }
 
-                for var in data.data_vars.keys():
-                    if var in ["u", "ubar"]:
-                        ds[f"{var}_{direction}"] = (
-                            data.data_vars[var]
-                            .isel(**bdry_coords["u"][direction])
-                            .astype(np.float32)
-                        )
-                    elif var in ["v", "vbar"]:
-                        ds[f"{var}_{direction}"] = (
-                            data.data_vars[var]
-                            .isel(**bdry_coords["v"][direction])
-                            .astype(np.float32)
-                        )
-                    else:
-                        ds[f"{var}_{direction}"] = (
-                            data.data_vars[var]
-                            .isel(**bdry_coords["rho"][direction])
-                            .astype(np.float32)
-                        )
-                    ds[f"{var}_{direction}"].attrs[
-                        "long_name"
-                    ] = f"{direction}ern boundary {d_meta[var]['long_name']}"
-                    ds[f"{var}_{direction}"].attrs["units"] = d_meta[var]["units"]
+        # Define a dictionary for variable names and their associated information
+        if self.type == "physics":
+            variable_info = {
+                "zeta": {
+                    "location": "rho",
+                    "is_vector": False,
+                    "vector_pair": None,
+                    "is_3d": False,
+                },
+                "temp": default_info,
+                "salt": default_info,
+                "u": {
+                    "location": "u",
+                    "is_vector": True,
+                    "vector_pair": "v",
+                    "is_3d": True,
+                },
+                "v": {
+                    "location": "v",
+                    "is_vector": True,
+                    "vector_pair": "u",
+                    "is_3d": True,
+                },
+                "ubar": {
+                    "location": "u",
+                    "is_vector": True,
+                    "vector_pair": "vbar",
+                    "is_3d": False,
+                },
+                "vbar": {
+                    "location": "v",
+                    "is_vector": True,
+                    "vector_pair": "ubar",
+                    "is_3d": False,
+                },
+            }
+        elif self.type == "bgc":
+            variable_info = {}
+            for var_name in data.var_names.keys():
+                variable_info[var_name] = default_info
+
+        return variable_info
+
+    def _write_into_dataset(self, direction, processed_fields, ds=None):
+        if ds is None:
+            ds = xr.Dataset()
+
+        d_meta = get_variable_metadata()
+
+        for var_name in processed_fields.keys():
+            ds[f"{var_name}_{direction}"] = processed_fields[var_name].astype(
+                np.float32
+            )
+
+            ds[f"{var_name}_{direction}"].attrs[
+                "long_name"
+            ] = f"{direction}ern boundary {d_meta[var_name]['long_name']}"
+
+            ds[f"{var_name}_{direction}"].attrs["units"] = d_meta[var_name]["units"]
 
         # Gracefully handle dropping variables that might not be present
         variables_to_drop = [
@@ -235,59 +397,9 @@ class BoundaryForcing(ROMSToolsMixins):
             "lat_v",
             "lon_v",
         ]
-        existing_vars = [var for var in variables_to_drop if var in ds]
+        existing_vars = [var_name for var_name in variables_to_drop if var_name in ds]
         ds = ds.drop_vars(existing_vars)
 
-        ds = self._add_global_metadata(ds)
-
-        # Convert the time coordinate to the format expected by ROMS
-        if data.climatology:
-            ds.attrs["climatology"] = str(True)
-            # Preserve absolute time coordinate for readability
-            ds = ds.assign_coords(
-                {"abs_time": np.datetime64(self.model_reference_date) + ds["time"]}
-            )
-            # Convert to pandas TimedeltaIndex
-            timedelta_index = pd.to_timedelta(ds["time"].values)
-
-            # Determine the start of the year for the base_datetime
-            start_of_year = datetime(self.model_reference_date.year, 1, 1)
-
-            # Calculate the offset from midnight of the new year
-            offset = self.model_reference_date - start_of_year
-
-            # Convert the timedelta to nanoseconds first, then to days
-            bry_time = xr.DataArray(
-                (timedelta_index - offset).view("int64") / 3600 / 24 * 1e-9,
-                dims="time",
-            )
-
-        else:
-            # Preserve absolute time coordinate for readability
-            ds = ds.assign_coords({"abs_time": ds["time"]})
-            # TODO: Check if we need to convert from 12:00:00 to 00:00:00 as in matlab scripts
-            bry_time = (
-                (ds["time"] - np.datetime64(self.model_reference_date)).astype(
-                    "float64"
-                )
-                / 3600
-                / 24
-                * 1e-9
-            )
-
-        ds = ds.assign_coords({"bry_time": bry_time})
-        ds["bry_time"].attrs[
-            "long_name"
-        ] = f"days since {str(self.model_reference_date)}"
-        ds["bry_time"].encoding["units"] = "days"
-        ds["bry_time"].attrs["units"] = "days"
-        ds = ds.swap_dims({"time": "bry_time"})
-        ds = ds.drop_vars("time")
-        ds.encoding["unlimited_dims"] = "bry_time"
-
-        if data.climatology:
-            ds["bry_time"].attrs["cycle_length"] = 365.25
-
         return ds
 
     def _get_coordinates(self, direction, point):
@@ -325,7 +437,7 @@ class BoundaryForcing(ROMSToolsMixins):
 
         return layer_depth, interface_depth
 
-    def _add_global_metadata(self, ds=None):
+    def _add_global_metadata(self, data, ds=None):
 
         if ds is None:
             ds = xr.Dataset()
@@ -345,11 +457,125 @@ class BoundaryForcing(ROMSToolsMixins):
         ds.attrs["theta_b"] = self.grid.ds.attrs["theta_b"]
         ds.attrs["hc"] = self.grid.ds.attrs["hc"]
 
+        # Convert the time coordinate to the format expected by ROMS
+        if data.climatology:
+            ds.attrs["climatology"] = str(True)
+            # Preserve absolute time coordinate for readability
+            ds = ds.assign_coords(
+                {"abs_time": np.datetime64(self.model_reference_date) + ds["time"]}
+            )
+            # Convert to pandas TimedeltaIndex
+            timedelta_index = pd.to_timedelta(ds["time"].values)
+
+            # Determine the start of the year for the base_datetime
+            start_of_year = datetime(self.model_reference_date.year, 1, 1)
+
+            # Calculate the offset from midnight of the new year
+            offset = self.model_reference_date - start_of_year
+
+            # Convert the timedelta to nanoseconds first, then to days
+            bry_time = xr.DataArray(
+                (timedelta_index - offset).view("int64") / 3600 / 24 * 1e-9,
+                dims="time",
+            )
+
+        else:
+            # Preserve absolute time coordinate for readability
+            ds = ds.assign_coords({"abs_time": ds["time"]})
+            bry_time = (
+                (ds["time"] - np.datetime64(self.model_reference_date)).astype(
+                    "float64"
+                )
+                / 3600
+                / 24
+                * 1e-9
+            )
+
+        ds = ds.assign_coords({"bry_time": bry_time})
+        ds["bry_time"].attrs[
+            "long_name"
+        ] = f"days since {str(self.model_reference_date)}"
+        ds["bry_time"].encoding["units"] = "days"
+        ds["bry_time"].attrs["units"] = "days"
+        ds = ds.swap_dims({"time": "bry_time"})
+        ds = ds.drop_vars("time")
+        ds.encoding["unlimited_dims"] = "bry_time"
+
+        if data.climatology:
+            ds["bry_time"].attrs["cycle_length"] = 365.25
+
         return ds
 
+    def _validate(self, ds, variable_info, bdry_coords):
+        """Validate the dataset for NaN values at the first time step based on the fill
+        method used.
+
+        Parameters
+        ----------
+        ds : xarray.Dataset
+            The dataset to validate.
+
+        variable_info : dict
+            A dictionary containing metadata about the variables, including their locations (e.g., 'rho', 'u', 'v').
+
+        bdry_coords : dict
+            A dictionary containing the boundary coordinates for each variable location.
+
+        Raises
+        ------
+        ValueError
+            If NaN values are found in any of the specified variables at wet points,
+            indicating incomplete data coverage.
+
+        Notes
+        -----
+        Validation is performed on the initial boundary time step (`bry_time=0`) for each
+        variable in the dataset. If the `apply_2d_horizontal_fill` attribute is set to False,
+        a warning is issued instead of a strict NaN check, as the data may not be reliably validated.
+        Conversely, if `apply_2d_horizontal_fill` is True, a strict NaN check is performed, raising
+        a ValueError if any NaN values are detected.
+        """
+        if self.apply_2d_horizontal_fill:
+            # Strict NaN check with ValueError makes sense to be applied
+            for var_name in variable_info:
+                location = variable_info[var_name]["location"]
+
+                # Select the appropriate mask based on variable location
+                if location == "rho":
+                    mask = self.grid.ds.mask_rho
+                elif location == "u":
+                    mask = self.grid.ds.mask_u
+                elif location == "v":
+                    mask = self.grid.ds.mask_v
+                else:
+                    continue  # Skip if location is not recognized
+
+                for direction in ["south", "east", "north", "west"]:
+                    if self.boundaries[direction]:
+                        bdry_var_name = f"{var_name}_{direction}"
+
+                        # Check for NaN values at the first time step using the nan_check function
+                        nan_check(
+                            ds[bdry_var_name].isel(bry_time=0),
+                            mask.isel(**bdry_coords[location][direction]),
+                        )
+        else:
+            # Can't apply strict NaN check because land values haven't been filled before regridding step; instead warn user
+            for direction in ["south", "east", "north", "west"]:
+                if self.boundaries[direction]:
+                    for var_name in variable_info:
+                        bdry_var_name = f"{var_name}_{direction}"
+                        if ds[bdry_var_name].isel(bry_time=0).isnull().any().values:
+                            warnings.warn(
+                                f"NaN values detected in regridded variables along the {direction}ern boundary. This may indicate that the entire boundary is on land in the source data, or that the source data does not cover this boundary.",
+                                UserWarning,
+                            )
+                            # Break after the first warning for this direction to avoid duplicates
+                            break
+
     def plot(
         self,
-        varname,
+        var_name,
         time=0,
         layer_contours=False,
     ) -> None:
@@ -357,7 +583,7 @@ class BoundaryForcing(ROMSToolsMixins):
 
         Parameters
         ----------
-        varname : str
+        var_name : str
             The name of the boundary forcing field to plot. Options include:
 
             - "temp_{direction}": Potential temperature,
@@ -417,37 +643,37 @@ class BoundaryForcing(ROMSToolsMixins):
         Raises
         ------
         ValueError
-            If the specified varname is not one of the valid options.
+            If the specified var_name is not one of the valid options.
         """
 
-        if varname not in self.ds:
-            raise ValueError(f"Variable '{varname}' is not found in dataset.")
+        if var_name not in self.ds:
+            raise ValueError(f"Variable '{var_name}' is not found in dataset.")
 
-        field = self.ds[varname].isel(bry_time=time).load()
+        field = self.ds[var_name].isel(bry_time=time).load()
         title = field.long_name
 
         if "s_rho" in field.dims:
-            if varname.startswith(("u_", "ubar_")):
+            if var_name.startswith(("u_", "ubar_")):
                 point = "u"
-            elif varname.startswith(("v_", "vbar_")):
+            elif var_name.startswith(("v_", "vbar_")):
                 point = "v"
             else:
                 point = "rho"
-            direction = varname.split("_")[-1]
+            direction = var_name.split("_")[-1]
 
             layer_depth, interface_depth = self._get_coordinates(direction, point)
 
             field = field.assign_coords({"layer_depth": layer_depth})
 
         # chose colorbar
-        if varname.startswith(("u", "v", "ubar", "vbar", "zeta")):
+        if var_name.startswith(("u", "v", "ubar", "vbar", "zeta")):
             vmax = max(field.max().values, -field.min().values)
             vmin = -vmax
             cmap = plt.colormaps.get_cmap("RdBu_r")
         else:
             vmax = field.max().values
             vmin = field.min().values
-            if varname.startswith(("temp", "salt")):
+            if var_name.startswith(("temp", "salt")):
                 cmap = plt.colormaps.get_cmap("YlOrRd")
             else:
                 cmap = plt.colormaps.get_cmap("YlGn")
@@ -621,3 +847,95 @@ class BoundaryForcing(ROMSToolsMixins):
 
         # Create and return an instance of InitialConditions
         return cls(grid=grid, **boundary_forcing_data, use_dask=use_dask)
+
+
+def get_boundary_info():
+    """This function provides information about the boundary points for the rho, u, and
+    v variables on the grid, specifying the indices for the south, east, north, and west
+    boundaries.
+
+    Returns
+    -------
+    dict
+        A dictionary where keys are variable types ("rho", "u", "v"), and values
+        are nested dictionaries mapping directions ("south", "east", "north", "west")
+        to the corresponding boundary coordinates.
+    """
+
+    # Boundary coordinates
+    bdry_coords = {
+        "rho": {
+            "south": {"eta_rho": 0},
+            "east": {"xi_rho": -1},
+            "north": {"eta_rho": -1},
+            "west": {"xi_rho": 0},
+        },
+        "u": {
+            "south": {"eta_rho": 0},
+            "east": {"xi_u": -1},
+            "north": {"eta_rho": -1},
+            "west": {"xi_u": 0},
+        },
+        "v": {
+            "south": {"eta_v": 0},
+            "east": {"xi_rho": -1},
+            "north": {"eta_v": -1},
+            "west": {"xi_rho": 0},
+        },
+        "vector": {
+            "south": {"eta_rho": [0, 1]},
+            "east": {"xi_rho": [-2, -1]},
+            "north": {"eta_rho": [-2, -1]},
+            "west": {"xi_rho": [0, 1]},
+        },
+    }
+
+    return bdry_coords
+
+
+def apply_1d_horizontal_fill(processed_fields: dict) -> dict:
+    """Forward and backward fill NaN values in horizontal direction for open boundaries.
+
+    Parameters
+    ----------
+    processed_fields : dict
+        A dictionary of variables to be updated, where each value is an
+        `xarray.DataArray`.
+
+    Returns
+    -------
+    dict of str : xarray.DataArray
+        The updated dictionary of variables, with NaN values filled.
+
+    Raises
+    ------
+    ValueError
+        If more than one horizontal dimension is found or none at all.
+    """
+
+    horizontal_dims = ["eta_rho", "eta_v", "xi_rho", "xi_u"]
+
+    for var_name in processed_fields.keys():
+        selected_horizontal_dim = None
+        # Determine the horizontal dimension to fill
+        for dim in horizontal_dims:
+            if dim in processed_fields[var_name].dims:
+                if selected_horizontal_dim is not None:
+                    raise ValueError(
+                        f"More than one horizontal dimension found in variable '{var_name}'."
+                    )
+                selected_horizontal_dim = dim
+
+        if selected_horizontal_dim is None:
+            raise ValueError(
+                f"No valid horizontal dimension found for variable '{var_name}'."
+            )
+        # Forward and backward fill in the horizontal direction
+        filled = one_dim_fill(
+            processed_fields[var_name], selected_horizontal_dim, direction="forward"
+        )
+        processed_fields[var_name] = one_dim_fill(
+            filled, selected_horizontal_dim, direction="backward"
+        )
+
+    return processed_fields