diff --git a/CHANGELOG.md b/CHANGELOG.md
index fffa9f41..c1ced2f1 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -35,6 +35,8 @@ and start a new "In Progress" section above it.
     - `run_udf` on the result of `read_vector`/`get_geometries` no longer passes the geometries as feature collection to the UDF, but follows the generic vector cube handling (like `load_uploaded_files`).
 - Use `save_result` node id in synchronous download filename hint (content-disposition header), and align filename extension handling with batch jobs ([#548](https://github.com/Open-EO/openeo-python-driver/issues/548))
 - Support extraction of default job options from predefined (and "custom") processes, and do basic merging ([#483](https://github.com/Open-EO/openeo-python-driver/issues/483))
+- `load_stac` with STAC metadata where "bands" dimension is erroneously missing: "bands" dimension will still be forced in dry run if "bands" argument is given to `load_stac` ([Open-EO/openeo-geopyspark-driver#1907](https://github.com/Open-EO/openeo-geopyspark-driver/issues/1907))
+
 
 ## 0.141.0
 
diff --git a/openeo_driver/dry_run.py b/openeo_driver/dry_run.py
index 44598825..78944998 100644
--- a/openeo_driver/dry_run.py
+++ b/openeo_driver/dry_run.py
@@ -446,7 +446,7 @@ def load_result(self, job_id: str, arguments: dict) -> "DryRunDataCube":
 
     @staticmethod
     @lru_cache(maxsize=20)
-    def _stac_metadata(stac_ref):
+    def _stac_metadata(stac_ref) -> CubeMetadata:
         try:
             return openeo_driver.stac.datacube.stac_to_cube_metadata(stac_ref=stac_ref)
         except Exception as e:
@@ -474,6 +474,14 @@ def load_stac(
 
         metadata = DryRunDataTracer._stac_metadata(stac_ref=url)
 
+        if arguments.get("bands") and not metadata.has_band_dimension():
+            _log.warning(
+                "No band dimension detected in dry-run `load_stac` metadata extraction, but bands were requested. Forcing 'bands' dimension in metadata."
+            )
+            band_dimension = BandDimension(name="bands", bands=[Band(name=b) for b in arguments["bands"]])
+            # TODO: possible to avoid usage of `_dimensions` internal?
+            metadata = CubeMetadata(dimensions=metadata._dimensions + [band_dimension])
+
         cube = DryRunDataCube(traces=[trace], data_tracer=self, metadata=metadata)
         if "temporal_extent" in arguments:
             cube = cube.filter_temporal(*arguments["temporal_extent"])
diff --git a/openeo_driver/stac/datacube.py b/openeo_driver/stac/datacube.py
index 08835709..f3a2d0d0 100644
--- a/openeo_driver/stac/datacube.py
+++ b/openeo_driver/stac/datacube.py
@@ -41,6 +41,7 @@ def stac_to_cube_metadata(stac_ref: StacRef) -> openeo.metadata.CubeMetadata:
     """
     Parse STAC metadata and convert it to :py:class:`openeo.metadata.CubeMetadata`.
     """
+    # TODO: replace/unify with dimension parsing from python client introduced by https://github.com/Open-EO/openeo-python-client/pull/867
     # Dimensions as pystac objects
     pystac_dimensions = _get_dimensions(stac_ref=stac_ref)
 
diff --git a/tests/test_dry_run.py b/tests/test_dry_run.py
index fb2a5fe5..e696582a 100644
--- a/tests/test_dry_run.py
+++ b/tests/test_dry_run.py
@@ -1659,6 +1659,77 @@ def test_load_stac_spatial_extent_requires_a_polygon(
         evaluate(pg, dry_run_env)
 
 
+@pytest.mark.parametrize(
+    ["load_stac_kwargs", "expected"],
+    [
+        ({"bands": ["B01", "B02"]}, "warning"),
+        ({}, "error"),
+        ({"bands": None}, "error"),
+    ],
+)
+def test_load_stac_no_bands_in_metadata_but_user_forced(
+    dry_run_tracer, backend_implementation, tmp_path, caplog, load_stac_kwargs, expected
+):
+    """
+    No bands in the STAC metadata, but user forced a band selection.
+    https://github.com/Open-EO/openeo-geopyspark-driver/issues/1907
+    """
+    stac_ref = tmp_path / "stac.json"
+    stac_metadata = StacDummyBuilder.collection(
+        cube_dimensions={
+            "x": {"type": "spatial", "axis": "x", "extent": [0, 10], "reference_system": 4326},
+            "y": {"type": "spatial", "axis": "y", "extent": [0, 20], "reference_system": 4326},
+            "t": {"type": "temporal", "extent": ["2020-01-01T00:00:00Z", "2020-02-01T00:00:00Z"]},
+        }
+    )
+    stac_ref.write_text(json.dumps(stac_metadata))
+
+    pg = {
+        "loadstac1": {
+            "process_id": "load_stac",
+            "arguments": {
+                "url": str(stac_ref),
+                **load_stac_kwargs,
+            },
+        },
+        "reducebands": {
+            "process_id": "reduce_dimension",
+            "arguments": {
+                "data": {"from_node": "loadstac1"},
+                "dimension": "bands",
+                "reducer": {
+                    "process_graph": {
+                        "mean": {
+                            "process_id": "mean",
+                            "arguments": {"data": {"from_parameter": "data"}},
+                            "result": True,
+                        }
+                    }
+                },
+            },
+            "result": True,
+        },
+    }
+
+    dry_run_env = EvalEnv(
+        {ENV_DRY_RUN_TRACER: dry_run_tracer, "backend_implementation": backend_implementation, "version": "2.0.0"}
+    )
+
+    if expected == "warning":
+        # `reduce_dimension` of "bands" works, but with warning
+        evaluate(pg, dry_run_env)
+        assert "Forcing 'bands' dimension in metadata" in caplog.text
+    elif expected == "error":
+        # Fails with error about missing bands dimension in metadata
+        with pytest.raises(
+            ProcessParameterInvalidException,
+            match=r"parameter.*dimension.*is invalid.*Must be one of.*'x', 'y', 't'.* but got 'bands'",
+        ):
+            evaluate(pg, dry_run_env)
+    else:
+        raise ValueError(f"Unexpected expected value: {expected}")
+
+
 @pytest.mark.parametrize(
     ["arguments", "expected"],
     [