Fix get_area_slices under-crop for cropped geostationary cross-CRS areas

pyresample/future/geometry/_subset.py

@@ -5,35 +5,61 @@
 from typing import TYPE_CHECKING, Any
 
 import numpy as np
+from pyproj import Proj
 
 # this caching module imports the geometries so this subset module
 # must be imported inside functions in the geometry modules if needed
 # to avoid circular dependencies
 from pyresample._caching import cache_to_json_if
 from pyresample.boundary import Boundary
-from pyresample.geometry import get_geostationary_bounding_box_in_lonlats, logger
+from pyresample.geometry import (
+    DEFAULT_AREA_SLICE_SAMPLE_STEPS,
+    get_geostationary_bounding_box_in_lonlats,
+    logger,
+)
 from pyresample.utils import check_slice_orientation
 
 if TYPE_CHECKING:
     from pyresample import AreaDefinition
 
 
+MAX_POINTS_PER_CHUNK = 600_000
+
+
 @cache_to_json_if("cache_geometry_slices")
 def get_area_slices(
         src_area: AreaDefinition,
         area_to_cover: AreaDefinition,
         shape_divisible_by: int | None,
+        sample_steps: int | None = DEFAULT_AREA_SLICE_SAMPLE_STEPS,
+        sample_grid: bool = False,
 ) -> tuple[slice, slice]:
-    """Compute the slice to read based on an `area_to_cover`."""
+    """Compute the slice to read based on an `area_to_cover`.
+
+    For geostationary source areas in cross-projection mode:
+    - ``sample_steps`` >= 2 and ``sample_grid=False`` samples edge points only.
+    - ``sample_steps`` >= 2 and ``sample_grid=True`` samples an interior grid.
+    - ``sample_steps`` <= 0 or ``None`` and ``sample_grid=False`` samples all edge points.
+    - ``sample_steps`` <= 0 or ``None`` and ``sample_grid=True`` samples all destination points.
+    - ``sample_steps`` == 1 raises ``ValueError``.
+    """
     if not _is_area_like(src_area):
         raise NotImplementedError(f"Only AreaDefinitions are supported, not {type(src_area)}")
     if not _is_area_like(area_to_cover):
         raise NotImplementedError(f"Only AreaDefinitions are supported, not {type(area_to_cover)}")
 
-    # Intersection only required for two different projections
-    proj_def_to_cover = area_to_cover.crs
-    proj_def = src_area.crs
-    if proj_def_to_cover == proj_def:
+    normalized_sample_steps = _normalize_sample_steps(sample_steps)
+
+    # Intersection is only required for two different projections.
+    src_crs_wkt = getattr(src_area, "crs_wkt", None)
+    dst_crs_wkt = getattr(area_to_cover, "crs_wkt", None)
+    if src_crs_wkt is not None and src_crs_wkt == dst_crs_wkt:
+        proj_def_to_cover = src_crs_wkt
+    else:
+        proj_def_to_cover = area_to_cover.crs
+        if proj_def_to_cover != src_area.crs:
+            proj_def_to_cover = None
+    if proj_def_to_cover is not None:
         logger.debug('Projections for data and slice areas are identical: %s',
                      proj_def_to_cover)
         # Get slice parameters
@@ -45,6 +71,16 @@ def get_area_slices(
         y_slice = _ensure_integer_slice(y_slice)
         return x_slice, y_slice
 
+    if src_area.is_geostationary:
+        coverage_slices = _get_covered_source_slices(
+            src_area,
+            area_to_cover,
+            sample_steps=normalized_sample_steps,
+            sample_grid=sample_grid,
+        )
+        if coverage_slices is not None:
+            return _finalize_slices(src_area, coverage_slices[0], coverage_slices[1], shape_divisible_by)
+
     data_boundary = _get_area_boundary(src_area)
     area_boundary = _get_area_boundary(area_to_cover)
     intersection = data_boundary.contour_poly.intersection(
@@ -57,6 +93,33 @@ def get_area_slices(
         np.rad2deg(intersection.lon), np.rad2deg(intersection.lat))
     x_slice = slice(np.ma.min(x), np.ma.max(x) + 1)
     y_slice = slice(np.ma.min(y), np.ma.max(y) + 1)
+
+    return _finalize_slices(src_area, x_slice, y_slice, shape_divisible_by)
+
+
+def _normalize_sample_steps(sample_steps: int | None):
+    """Normalize sampling config to sampled mode or ALL-mode fallback.
+
+    ``None`` and values ``<= 0`` map to ``ALL`` mode, whose exact behavior is
+    determined later by ``sample_grid``.
+    A value of ``1`` is rejected because it does not provide meaningful sampled
+    coverage for edge or grid modes.
+    """
+    if sample_steps is None:
+        return None
+    try:
+        normalized_sample_steps = int(sample_steps)
+    except (TypeError, ValueError) as err:
+        raise ValueError(f"sample_steps must be an integer or None, got {sample_steps!r}") from err
+    if normalized_sample_steps <= 0:
+        return None
+    if normalized_sample_steps == 1:
+        raise ValueError("sample_steps=1 is not supported; use <= 0/None for ALL mode or >= 2 for sampled modes.")
+    return normalized_sample_steps
+
+
+def _finalize_slices(src_area: AreaDefinition, x_slice: slice, y_slice: slice, shape_divisible_by: int | None):
+    """Normalize slice bounds and apply orientation/divisibility rules."""
     x_slice = _ensure_integer_slice(x_slice)
     y_slice = _ensure_integer_slice(y_slice)
     if shape_divisible_by is not None:
@@ -105,6 +168,137 @@ def _get_area_boundary(area_to_cover: AreaDefinition) -> Boundary:
         raise NotImplementedError("Can't determine boundary of area to cover") from err
 
 
+def _get_covered_source_slices(
+        src_area: AreaDefinition,
+        area_to_cover: AreaDefinition,
+        sample_steps: int | None,
+        sample_grid: bool,
+):
+    """Estimate covering source slices from sampled destination points.
+
+    Returns ``None`` when sampled points do not produce any valid source
+    coverage, allowing the caller to fall back to boundary intersection.
+    """
+    min_col = None
+    max_col = None
+    min_row = None
+    max_row = None
+    try:
+        src_proj = Proj(src_area.crs)
+        for destination_lons, destination_lats in _iter_destination_lonlat_samples(
+            area_to_cover=area_to_cover,
+            sample_steps=sample_steps,
+            sample_grid=sample_grid,
+        ):
+            source_xs, source_ys = src_proj(
+                destination_lons,
+                destination_lats,
+            )
+            source_cols, source_rows = src_area.get_array_indices_from_projection_coordinates(
+                source_xs,
+                source_ys,
+            )
+            valid = ~np.ma.getmaskarray(source_cols) & ~np.ma.getmaskarray(source_rows)
+            if not valid.any():
+                continue
+            chunk_cols = np.ma.getdata(source_cols)[valid]
+            chunk_rows = np.ma.getdata(source_rows)[valid]
+            chunk_min_col = int(chunk_cols.min())
+            chunk_max_col = int(chunk_cols.max())
+            chunk_min_row = int(chunk_rows.min())
+            chunk_max_row = int(chunk_rows.max())
+            min_col = chunk_min_col if min_col is None else min(min_col, chunk_min_col)
+            max_col = chunk_max_col if max_col is None else max(max_col, chunk_max_col)
+            min_row = chunk_min_row if min_row is None else min(min_row, chunk_min_row)
+            max_row = chunk_max_row if max_row is None else max(max_row, chunk_max_row)
+    except (RuntimeError, TypeError, ValueError):
+        logger.debug("Failed to estimate covered source slices from sampled destination points.", exc_info=True)
+        return None
+    if min_col is None or min_row is None or max_col is None or max_row is None:
+        return None
+    col_start = max(0, min_col)
+    col_stop = min(src_area.width, max_col + 1)
+    row_start = max(0, min_row)
+    row_stop = min(src_area.height, max_row + 1)
+    if col_start >= col_stop or row_start >= row_stop:
+        return None
+    return slice(col_start, col_stop), slice(row_start, row_stop)
+
+
+def _iter_destination_lonlat_samples(
+        area_to_cover: AreaDefinition,
+        sample_steps: int | None,
+        sample_grid: bool,
+):
+    """Yield destination lon/lat samples for ALL, grid, or edge mode."""
+    if sample_steps is None:
+        if sample_grid:
+            yield from _iter_dense_lonlat_samples(area_to_cover)
+            return
+        yield _get_edge_lonlat_samples(area_to_cover, sample_steps=None)
+        return
+    if sample_grid:
+        yield _get_grid_lonlat_samples(area_to_cover, sample_steps)
+        return
+    yield _get_edge_lonlat_samples(area_to_cover, sample_steps)
+
+
+def _iter_dense_lonlat_samples(area_to_cover: AreaDefinition):
+    """Yield full destination lon/lat coverage in row chunks."""
+    row_block_size = max(1, MAX_POINTS_PER_CHUNK // area_to_cover.width)
+    for row_start in range(0, area_to_cover.height, row_block_size):
+        row_stop = min(area_to_cover.height, row_start + row_block_size)
+        yield area_to_cover.get_lonlats(
+            data_slice=(slice(row_start, row_stop), slice(None)),
+            dtype=np.float32,
+        )
+
+
+def _get_grid_lonlat_samples(area_to_cover: AreaDefinition, sample_steps: int):
+    """Return one evenly spaced interior destination sample grid."""
+    sample_rows = _get_sample_indices(area_to_cover.height, sample_steps)
+    sample_cols = _get_sample_indices(area_to_cover.width, sample_steps)
+    return area_to_cover.get_lonlats(
+        data_slice=(sample_rows[:, None], sample_cols[None, :]),
+        dtype=np.float32,
+    )
+
+
+def _get_edge_lonlat_samples(area_to_cover: AreaDefinition, sample_steps: int | None):
+    """Return perimeter destination samples for edge-only sampling mode."""
+    if area_to_cover.is_geostationary:
+        # Use limb-aware geostationary boundary sampling instead of raw array
+        # corners/edges. Projection corners may be off-earth and can under-cover.
+        if sample_steps is None:
+            nb_points = max(4, area_to_cover.width * 2 + area_to_cover.height * 2 - 4)
+        else:
+            nb_points = max(4, sample_steps * 4)
+        return get_geostationary_bounding_box_in_lonlats(
+            area_to_cover,
+            nb_points=nb_points,
+        )
+    sample_rows = _get_sample_indices(area_to_cover.height, sample_steps)
+    sample_cols = _get_sample_indices(area_to_cover.width, sample_steps)
+    top_rows = np.zeros(sample_cols.size, dtype=np.int64)
+    top_cols = sample_cols
+    right_rows = sample_rows[1:]
+    right_cols = np.full(right_rows.size, area_to_cover.width - 1, dtype=np.int64)
+    bottom_cols = sample_cols[-2::-1]
+    bottom_rows = np.full(bottom_cols.size, area_to_cover.height - 1, dtype=np.int64)
+    left_rows = sample_rows[-2:0:-1]
+    left_cols = np.zeros(left_rows.size, dtype=np.int64)
+    edge_rows = np.concatenate((top_rows, right_rows, bottom_rows, left_rows))
+    edge_cols = np.concatenate((top_cols, right_cols, bottom_cols, left_cols))
+    return area_to_cover.get_lonlats(data_slice=(edge_rows, edge_cols), dtype=np.float32)
+
+
+def _get_sample_indices(axis_size: int, sample_steps: int | None):
+    """Return evenly spaced integer sample indices including both endpoints."""
+    if sample_steps is None or sample_steps >= axis_size:
+        return np.arange(axis_size, dtype=np.int64)
+    return (np.arange(sample_steps, dtype=np.int64) * (axis_size - 1)) // (sample_steps - 1)
+
+
 def _make_slice_divisible(sli, max_size, factor=2):
     """Make the given slice even in size."""
     rem = (sli.stop - sli.start) % factor
@@ -121,6 +315,7 @@ def _make_slice_divisible(sli, max_size, factor=2):
 
 
 def _ensure_integer_slice(sli):
+    """Round slice bounds outward to conservatively preserve target coverage."""
     start = sli.start
     stop = sli.stop
     step = sli.step

pyresample/geometry.py

@@ -68,6 +68,8 @@
 
 logger = getLogger(__name__)
 
+DEFAULT_AREA_SLICE_SAMPLE_STEPS = 21
+
 
 class DimensionError(ValueError):
     """Wrap ValueError."""
@@ -619,7 +621,13 @@ def overlap_rate(self, other):
         inter_area = get_polygon_area(self.intersection(other))
         return inter_area / other_area
 
-    def get_area_slices(self, area_to_cover):
+    def get_area_slices(
+            self,
+            area_to_cover,
+            shape_divisible_by=None,
+            sample_steps=DEFAULT_AREA_SLICE_SAMPLE_STEPS,
+            sample_grid=False,
+    ):
         """Compute the slice to read based on an `area_to_cover`."""
         raise NotImplementedError
 
@@ -2651,10 +2659,32 @@ def proj4_string(self):
                       "instead.", DeprecationWarning, stacklevel=2)
         return proj4_dict_to_str(self.proj_dict)
 
-    def get_area_slices(self, area_to_cover, shape_divisible_by=None):
-        """Compute the slice to read based on an `area_to_cover`."""
+    def get_area_slices(
+            self,
+            area_to_cover,
+            shape_divisible_by=None,
+            sample_steps=DEFAULT_AREA_SLICE_SAMPLE_STEPS,
+            sample_grid=False,
+    ):
+        """Compute the slice to read based on an `area_to_cover`.
+
+        Args:
+            area_to_cover: Destination area to crop around.
+            shape_divisible_by: Optional factor to make the resulting shape divisible by.
+            sample_steps: Number of edge/grid samples used for geostationary cross-projection coverage checks.
+                Use ``None`` or ``<= 0`` for ``ALL`` mode: all edge points when ``sample_grid=False``
+                and all destination pixels when ``sample_grid=True``.
+            sample_grid: If ``True``, sample an interior grid with ``sample_steps`` density.
+                If ``False``, sample edge points only.
+        """
         from .future.geometry._subset import get_area_slices
-        return get_area_slices(self, area_to_cover, shape_divisible_by)
+        return get_area_slices(
+            self,
+            area_to_cover,
+            shape_divisible_by,
+            sample_steps,
+            sample_grid,
+        )
 
     def crop_around(self, other_area):
         """Crop this area around `other_area`."""