Datasets:

nafizrahaman
/

geopulse-bench

Error code:   StreamingRowsError
Exception:    TypeError
Message:      Couldn't cast array of type
struct<georgia_pines_2019: string, oregon_coast_2020: string, michael_blowdown_2018: string, sweden_vasterbotten_2019: string, para_novo_progresso_2020: string>
to
{'midwest_hamburg_2019': Value('string'), 'harvey_brazos_2017': Value('string'), 'idai_buzi_2019': Value('string'), 'florence_lumberton_2018': Value('string'), 'ian_arcadia_2022': Value('string'), 'emilia_conselice_2023': Value('string'), 'pakistan_dadu_2022': Value('string')}
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2312, in cast_table_to_schema
                  cast_array_to_feature(
                  ~~~~~~~~~~~~~~~~~~~~~^
                      table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                      feature,
                      ^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
                  return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
                                           ~~~~^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2158, in cast_array_to_feature
                  raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
              TypeError: Couldn't cast array of type
              struct<georgia_pines_2019: string, oregon_coast_2020: string, michael_blowdown_2018: string, sweden_vasterbotten_2019: string, para_novo_progresso_2020: string>
              to
              {'midwest_hamburg_2019': Value('string'), 'harvey_brazos_2017': Value('string'), 'idai_buzi_2019': Value('string'), 'florence_lumberton_2018': Value('string'), 'ian_arcadia_2022': Value('string'), 'emilia_conselice_2023': Value('string'), 'pakistan_dadu_2022': Value('string')}

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

GeoPulse-Bench v0.1

Event-split benchmarks for multimodal Earth-change mapping, built by GeoPulse from Sentinel-1 RTC, Sentinel-2 L2A and Copernicus DEM GLO-30 (Microsoft Planetary Computer). Each folder is one benchmark:

Folder	Task	Events (train / val / test)	Tiles	Labels
`geopulse-bench-flood`	flood	5 / 1 / 1 (test: Pakistan 2022)	236	weak: S1+S2 consensus
`geopulse-bench-wildfire`	burned area	5 / 1 / 1 (test: Martin Fire 2018, Nevada sagebrush)	212	MTBS burn severity
`geopulse-bench-vegetation`	stand-replacing forest loss	3 / 1 / 1 (test: Pará, Brazil 2020)	176	Hansen GFC v1.12 loss year

Format

Each tile is <event>/<row>_<col>.npz, 256 × 256 px at 10 m on a UTM grid, float16, normalised as GeoPulse's to_tensors does (see geopulse/model.py::NORM):

s1_pre, s1_post (2 × H × W: VV, VH dB), s2_pre, s2_post (6 × H × W: B02 B03 B04 B08 B11 B12), each with a *_valid mask (1 × H × W); dem (2 × H × W: height above the AOI median / 20, slope / 10).
labels (2 × H × W, uint8): [task target, generic change], 0 / 1, 255 = ignore.

index.json lists every tile with its event and split, the manifest SHA-256, label source per event, scene IDs used per event and period, and event bounds.

How it was built

geopulse dataset build configs/data/<task>_bench.yaml (manifests are in the GeoPulse repository): STAC search per event window, per-task compositing (flood post = earliest valid, otherwise median; vegetation samples both windows evenly across the same season), reprojection straight onto a snapped 10 m grid, then tiling. Tiles with < 2 % labelled pixels are dropped.

Labels and known issues

Flood labels are weak: pixels where SAR and optical agree confidently. Scores against them measure agreement with that consensus, not ground truth; the ambiguous pixels (cloud, flooded vegetation, urban) are ignored.
Wildfire: MTBS low/moderate/high = burned; outside mapped fires and "increased greenness" = unburned; "unburned to low" and non-mapping areas = ignored. Fires below MTBS's size threshold are not mapped.
Vegetation: Hansen loss in the label year = 1; forest (≥ 30 % cover in 2000) with no loss 2001–2024 = 0; loss in other years = ignored. Partial disturbance (thinning, partial windthrow) counts as "no loss" in Hansen.
Positive fractions vary widely by event (vegetation: 1–49 %). Change labels are conservative consensus.

Licenses and required notices

Mixed open data; details in DATA_LICENSES.md:

Contains modified Copernicus Sentinel data (2017–2023).
Copernicus DEM GLO-30 © DLR e.V. 2010–2014 and © Airbus Defence and Space GmbH 2014–2018, provided under COPERNICUS by the European Union and ESA.
MTBS: U.S. Government work (USGS / USDA Forest Service), public domain.
Hansen et al. (2013) Global Forest Change v1.12, CC BY 4.0.