Remove rioxarray dependency

Author: dcherian

docs/conf.py

@@ -155,7 +155,8 @@
 napoleon_type_aliases = {
     # general terms
     "sequence": ":term:`sequence`",
-    "iterable": ":term:`iterable`",
+    "Hashable": ":term:`sequence`",
+    "iterable": "~collections.abc.Hashable",
     "callable": ":py:func:`callable`",
     "dict_like": ":term:`dict-like <mapping>`",
     "dict-like": ":term:`dict-like <mapping>`",

docs/rasterize/exactextract.ipynb

@@ -28,9 +28,10 @@
    "outputs": [],
    "source": [
     "import xarray as xr\n",
+    "import xproj  # noqa\n",
     "\n",
     "ds = xr.tutorial.open_dataset(\"eraint_uvz\")\n",
-    "ds = ds.rio.write_crs(\"epsg:4326\")\n",
+    "ds = ds.proj.assign_crs(spatial_ref=\"epsg:4326\")\n",
     "ds"
    ]
   },

docs/rasterize/geometry_mask.ipynb

@@ -28,9 +28,10 @@
    "outputs": [],
    "source": [
     "import xarray as xr\n",
+    "import xproj  # noqa\n",
     "\n",
     "ds = xr.tutorial.open_dataset(\"eraint_uvz\")[[\"u\"]]\n",
-    "ds = ds.rio.write_crs(\"epsg:4326\")\n",
+    "ds = ds.proj.assign_crs(spatial_ref=\"epsg:4326\")\n",
     "ds"
    ]
   },

docs/rasterize/rasterio.ipynb

@@ -28,9 +28,10 @@
    "outputs": [],
    "source": [
     "import xarray as xr\n",
+    "import xproj  # noqa\n",
     "\n",
     "ds = xr.tutorial.open_dataset(\"eraint_uvz\")\n",
-    "ds = ds.rio.write_crs(\"epsg:4326\")\n",
+    "ds = ds.proj.assign_crs(spatial_ref=\"epsg:4326\")\n",
     "ds"
    ]
   },

pyproject.toml

@@ -54,7 +54,6 @@ docs = [
   "pooch",
   "dask-geopandas",
   "rasterio",
-  "rioxarray",
   "exactextract",
   "sparse",
   "netCDF4",

src/rasterix/raster_index.py

@@ -21,6 +21,7 @@
 
 from rasterix.odc_compat import BoundingBox, bbox_intersection, bbox_union, maybe_int, snap_grid
 from rasterix.rioxarray_compat import guess_dims
+from rasterix.utils import get_affine
 
 T_Xarray = TypeVar("T_Xarray", "DataArray", "Dataset")
 
@@ -37,10 +38,14 @@ def assign_index(
 ) -> T_Xarray:
     """Assign a RasterIndex to an Xarray DataArray or Dataset.
 
+    By default, the affine transform is guessed by first looking for a ``GeoTransform`` attribute
+    on a CF "grid mapping" variable (commonly ``"spatial_ref"``). If not present, then the affine is determined from 1D coordinate
+    variables named ``x_dim`` and ``y_dim`` provided to this function.
+
     Parameters
     ----------
     obj : xarray.DataArray or xarray.Dataset
-        The object to assign the index to. Must have a rio accessor with a transform.
+        The object to assign the index to.
     x_dim : str, optional
         Name of the x dimension. If None, will be automatically detected.
     y_dim : str, optional
@@ -53,22 +58,37 @@ def assign_index(
     xarray.DataArray or xarray.Dataset
         The input object with RasterIndex coordinates assigned.
 
+    Notes
+    -----
+    The "grid mapping" variable is determined following the CF conventions:
+
+    - If a DataArray is provided, we look for an attribute named ``"grid_mapping"``.
+    - For a Dataset, we pull the first detected ``"grid_mapping"`` attribute when iterating over data variables.
+
+    The value of this attribute is a variable name containing projection information (commonly ``"spatial_ref"``).
+    We then look for a ``"GeoTransform"`` attribute on this variable (following GDAL convention).
+
+    References
+    ----------
+    - `CF conventions document <http://cfconventions.org/Data/cf-conventions/cf-conventions-1.11/cf-conventions.html#grid-mappings-and-projections>`_.
+    - `GDAL docs on GeoTransform <https://gdal.org/en/stable/tutorials/geotransforms_tut.html>`_.
+
     Examples
     --------
     >>> import xarray as xr
-    >>> import rioxarray  # Required for rio accessor
+    >>> import rioxarray  # Required for reading TIFF
     >>> da = xr.open_dataset("path/to/raster.tif", engine="rasterio")
     >>> indexed_da = assign_index(da)
     """
-    import rioxarray  # noqa
-
     if x_dim is None or y_dim is None:
         guessed_x, guessed_y = guess_dims(obj)
     x_dim = x_dim or guessed_x
     y_dim = y_dim or guessed_y
 
+    affine = get_affine(obj, x_dim=x_dim, y_dim=y_dim, clear_transform=True)
+
     index = RasterIndex.from_transform(
-        obj.rio.transform(),
+        affine,
         width=obj.sizes[x_dim],
         height=obj.sizes[y_dim],
         x_dim=x_dim,

src/rasterix/rasterize/rasterio.py

@@ -15,7 +15,8 @@
 from rasterio.features import geometry_mask as geometry_mask_rio
 from rasterio.features import rasterize as rasterize_rio
 
-from .utils import XAXIS, YAXIS, clip_to_bbox, get_affine, is_in_memory, prepare_for_dask
+from ..utils import get_affine
+from .utils import XAXIS, YAXIS, clip_to_bbox, is_in_memory, prepare_for_dask
 
 F = TypeVar("F", bound=Callable[..., Any])
 
@@ -161,7 +162,7 @@ def rasterize(
         obj = clip_to_bbox(obj, geometries, xdim=xdim, ydim=ydim)
 
     rasterize_kwargs = dict(
-        all_touched=all_touched, merge_alg=merge_alg, affine=get_affine(obj, xdim=xdim, ydim=ydim), env=env
+        all_touched=all_touched, merge_alg=merge_alg, affine=get_affine(obj, x_dim=xdim, y_dim=ydim), env=env
     )
     # FIXME: box.crs == geometries.crs
 
@@ -325,7 +326,7 @@ def geometry_mask(
         obj = clip_to_bbox(obj, geometries, xdim=xdim, ydim=ydim)
 
     geometry_mask_kwargs = dict(
-        all_touched=all_touched, affine=get_affine(obj, xdim=xdim, ydim=ydim), env=env
+        all_touched=all_touched, affine=get_affine(obj, x_dim=xdim, y_dim=ydim), env=env
     )
 
     if is_in_memory(obj=obj, geometries=geometries):

src/rasterix/utils.py

@@ -0,0 +1,61 @@
+import xarray as xr
+from affine import Affine
+
+
+def get_grid_mapping_var(obj: xr.Dataset | xr.DataArray) -> xr.DataArray | None:
+    grid_mapping_var = None
+    if isinstance(obj, xr.DataArray):
+        if maybe := obj.attrs.get("grid_mapping", None):
+            if maybe in obj.coords:
+                grid_mapping_var = maybe
+    else:
+        # for datasets, grab the first one for simplicity
+        for var in obj.data_vars.values():
+            if maybe := var.attrs.get("grid_mapping"):
+                if maybe in obj.coords:
+                    # make sure it exists and is not an out-of-date attribute
+                    grid_mapping_var = maybe
+                    break
+    if grid_mapping_var is None and "spatial_ref" in obj.coords:
+        # hardcode this
+        grid_mapping_var = "spatial_ref"
+    if grid_mapping_var is not None:
+        return obj[grid_mapping_var]
+    return None
+
+
+def get_affine(obj: xr.Dataset | xr.DataArray, *, x_dim="x", y_dim="y") -> Affine:
+    """
+    Grabs an affine transform from an Xarray object.
+
+    This method will first look for the ``"GeoTransform"`` attribute on a variable named
+    ``"spatial_ref"``. If not, it will auto-guess the transform from the provided ``x_dim``,
+    and ``y_dim``.
+
+    Parameters
+    ----------
+    obj: xr.DataArray or xr.Dataset
+    x_dim: str, optional
+        Name of the X dimension coordinate variable.
+    y_dim: str, optional
+        Name of the Y dimension coordinate variable.
+
+    Returns
+    -------
+    affine: Affine
+    """
+    grid_mapping_var = get_grid_mapping_var(obj)
+    if grid_mapping_var is not None and (transform := grid_mapping_var.attrs.get("GeoTransform")):
+        return Affine.from_gdal(*map(float, transform.split(" ")))
+    else:
+        x = obj.coords[x_dim]
+        y = obj.coords[y_dim]
+        if x.ndim != 1:
+            raise ValueError(f"Coordinate variable {x_dim=!r} must be 1D.")
+        if y.ndim != 1:
+            raise ValueError(f"Coordinate variable {y_dim=!r} must be 1D.")
+        dx = (x[1] - x[0]).item()
+        dy = (y[1] - y[0]).item()
+        return Affine.translation(
+            x[0].item() - dx / 2, (y[0] if dy < 0 else y[-1]).item() - dy / 2
+        ) * Affine.scale(dx, dy)