Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 291, in _generate_tables
                  pa_table = paj.read_json(
                      io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
                  )
                File "pyarrow/_json.pyx", line 364, in pyarrow._json.read_json
                File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
                  return check_status(status)
                File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
                  raise convert_status(status)
              pyarrow.lib.ArrowInvalid: JSON parse error: Column(/overview_shapes_hw_block/[]/[]) changed from number to array in row 0
              
              During handling of the above exception, another exception occurred:
              
              Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 101, in _split_generators
                  pa_table = next(iter(self._generate_tables(**splits[0].gen_kwargs, allow_full_read=False)))[1]
                             ~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 306, in _generate_tables
                  batch = json_encode_fields_in_json_lines(original_batch, json_field_paths)
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 111, in json_encode_fields_in_json_lines
                  examples = [ujson_loads(line) for line in original_batch.splitlines()]
                              ~~~~~~~~~~~^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
                  return pd.io.json.ujson_loads(*args, **kwargs)
                         ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
              ValueError: Expected object or value
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

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.

Starman - Mars elevation, 200 m

Every elevation sample of the Mars MGS MOLA - MEX HRSC Blended DEM Global 200m, v2, at its native 200 m spacing, as a lossless Cloud-Optimized GeoTIFF. This is the sample-level companion to enzeeoai/starman, which holds per-cell terrain statistics on a permanent 5 km grid computed from this exact raster. (Both repos are currently private.)

Every Mars cell in starman points into this file: see Reading one grid cell below.

Renamed. This repo was published as enzeeoai/starman-mars-pixels and is now enzeeoai/starman-mars-elevation. The old URL redirects, and the file name, bytes and SHA-256 are unchanged. The name now says what the data is rather than how it is stored.

File What
Mars_HRSC_MOLA_BlendDEM_Global_200mp_v2_cog.tif The elevation model, Int16, DEFLATE + predictor 2, 512x512 tiles, overviews 2/4/8/16/32 (average)
Mars_HRSC_MOLA_BlendDEM_Global_200mp_v2.lbl The original PDS label, unmodified
source_manifest.json Our inspection of the source file, unmodified (see Caveats about the manifest)
cog_verification.json Machine-readable proof that the COG is pixel-identical to the original

Lossless conversion

The COG was made from the original uncompressed GeoTIFF with GDAL's COG driver (rasterio 1.5.2, GDAL 3.12.2; COMPRESS=DEFLATE PREDICTOR=2 BLOCKSIZE=512 OVERVIEW_RESAMPLING=AVERAGE OVERVIEW_COUNT=5). It was verified by reading both files in 512-row strips: identical shape, dtype, pixel values (all strips), affine transform (exact), CRS, nodata value and nodata count. Only the pixel storage differs; the overviews are additions at the end of the structure and do not alter the full-resolution band.

Original Mars_HRSC_MOLA_BlendDEM_Global_200mp_v2.tif This COG
Bytes 11,384,463,908 3,790,264,667 (3.00x smaller)
SHA-256 e8bfe4c3b9dee9d5fdb8c4d49798869654e552e7b154712c9e12d6d117464419 a78e40a79295d2cd0219362b257d103e788f895278ea4764ddd1aa1a89e9bc63

The original's SHA-256 is the value recorded in every row of the starman terrain table (source_checksum). It was computed by us; it was not compared with a publisher checksum (we found none published).

Raster properties

  • Size: 106,694 columns x 53,347 rows, one band, Int16. Pixel-is-area; row 0 is the north edge.
  • CRS: GCS_Mars_2000_Sphere (+proj=longlat +a=3396190 +b=3396190 +no_defs): planetocentric latitude, positive-east longitude, sphere radius 3,396,190 m.
  • Pixel size: 0.003374120830641 degrees (199.999991643332 m, 296.3735 pixels/degree).
  • Geotransform (a, b, c, d, e, f): (0.003374120830641, 0, -180.0, 0, -0.003374120830641, 90.0).
  • Extent: longitude -180 to 179.998448, latitude -89.999224 to 90. The raster does not close the globe: the east edge is ~92 m short of +180 and the south edge ~46 m short of -90.
  • Nodata: -32768. There are exactly 39,957 nodata pixels, all in the last column (x = 106693).
  • Valid pixels: 5,691,764,861 of 5,691,804,818. Valid range -8,528 to 21,226. Percentiles (1/5/25/50/75/95/99): -5,882 / -4,828 / -3,701 / -256 / 1,620 / 3,598 / 6,010.
  • Vertical unit and datum are inferred, not stated in the label. The label has no unit field for pixel values (only Base = 0.0, Multiplier = 1.0). We take the unit to be metres because the value range matches published MOLA relief, and the datum to be the areoid because values at the poles (about -1,957 m north, +3,785 m south) are not what a 3,396,190 m sphere reference would give. Olympus Mons reads 21,226 m against a published ~21,229 m. Treat both as assumptions until confirmed against the source documentation.

Source, citation and caveats

Source product: Mars MGS MOLA - MEX HRSC Blended DEM Global 200m, version v2 (name and version taken from the file name; the label records neither). Distributed by the USGS Astrogeology Science Center (Astropedia): https://astrogeology.usgs.gov/search/map/mars_mgs_mola_mex_hrsc_blended_dem_global_200m.

Recommended citation (as shown on that page): Fergason, R. L., Hare, T. M., & Laura, J. (2018). HRSC and MOLA Blended Digital Elevation Model at 200m v2. Astrogeology PDS Annex, U.S. Geological Survey.

The product blends ~50 m Mars Express HRSC DEMs onto the ~463 m MGS MOLA DEM and samples to 200 m. HRSC covers only part of Mars (about 44% when the page was written), so effective resolution outside it is MOLA-limited, and the page warns that the blend "did not always work well" in places. The page lists v2 as in work and to be superseded. The source has no mask saying which pixels are HRSC-derived.

Source URL. The download URL https://planetarymaps.usgs.gov/mosaic/Mars/HRSC_MOLA_Blend/Mars_HRSC_MOLA_BlendDEM_Global_200mp_v2.tif resolved on 2026-11-12 and served a file of exactly 11,384,463,908 bytes, the same length as our copy, last modified 2022-11-09. This confirms the location and size, not the content: we did not re-download the file or compare checksums.

License: unverified. We could not confirm the terms for this product. The Astropedia page could not be fetched by our tooling, the .lbl has no rights or credit field, and searches found no license text. USGS-authored work is generally public domain in the United States, but this product incorporates ESA/DLR/FU Berlin HRSC-derived elevation data whose terms we have not checked (ESA material is sometimes distributed under CC BY-SA 4.0; we have no evidence either way for this copy). license: other in this card means terms unverified, not "open". This repository is kept private until the terms are checked. Credit the sources above in any use.

Caveats about the manifest. source_manifest.json was written before the URL check and still labels the source URL, the unit and the datum as assumptions. It is published unmodified so it matches what the starman pipeline read; this card supersedes it only on the URL, which is now confirmed to the extent described above.

Reading it

Windowed read with rasterio (no need to fetch 3.8 GB). Download the file once with hf_hub_download, or point GDAL at the file's HTTPS URL if you have access. The repo id below is the new one; the old enzeeoai/starman-mars-pixels still resolves:

import numpy as np
import rasterio
from rasterio.windows import Window
from huggingface_hub import hf_hub_download

path = hf_hub_download("enzeeoai/starman-mars-elevation", "Mars_HRSC_MOLA_BlendDEM_Global_200mp_v2_cog.tif",
                       repo_type="dataset", revision="b182072b8e02a57f238108a67dc8f4d46e6ab67d")   # pinned, not "main"
with rasterio.open(path) as src:
    # 1,000 x 1,000 native pixels near the equator (rows run north to south)
    window = Window(col_off=50_000, row_off=26_000, width=1_000, height=1_000)
    z = src.read(1, window=window).astype("float64")
    z[z == src.nodata] = np.nan
    print(np.nanmin(z), np.nanmax(z), src.transform * (50_000, 26_000))
    coarse = src.read(1, out_shape=(src.height // 32, src.width // 32))   # uses the 32x overview

Reading one grid cell (with the starman grid)

Every one of the 5,797,982 Mars cells in enzeeoai/starman points at its own rectangle of samples in this file. Release terrain-mars-v0.3.1 publishes that mapping as grid/planet=mars/latband5km-v1_elevation_windows.parquet (config elevation_windows_mars), which records this repo and the exact commit it was built against as elevation_repo and elevation_revision. The windows are half-open, row 0 is the north edge and column 0 is -180 degrees east, and (row_stop - row_start) * (col_stop - col_start) equals the cell's n_native_samples in the terrain table. Summed over every cell they come to exactly 106,694 x 53,347: the windows tile this raster with no gap and no overlap.

import pandas as pd, rasterio, numpy as np
from rasterio.windows import Window

base = "hf://datasets/enzeeoai/starman@terrain-mars-v0.3.1"
windows = pd.read_parquet(f"{base}/grid/planet=mars/latband5km-v1_elevation_windows.parquet")
cell = windows[windows.region_id == "M_1067_01234"].iloc[0]
window = Window(cell.elev_raster_col_start, cell.elev_raster_row_start,
                cell.elev_raster_col_stop - cell.elev_raster_col_start,
                cell.elev_raster_row_stop - cell.elev_raster_row_start)
with rasterio.open(path) as src:          # `path` from the example above
    z = src.read(1, window=window).astype("float64")
z[z == -32768] = np.nan
print(cell.region_id, z.shape, np.nanmin(z), np.nanmax(z))

code/mars_terrain.py in that repo does the same thing for one cell without reading the table: resolve_region("M_1067_01234") returns this file's URL at the pinned revision plus row_off, col_off, n_rows and n_cols.

The Mars grid table also still carries the same four numbers under their original names (src_row_min, src_row_max, src_col_min, src_col_max), added at terrain-mars-v0.2.0 and unchanged since. They are the same windows; the separate file is the one to prefer, because it is the one Earth's equivalent matches.

These are the native, unaveraged samples. The starman terrain features were computed on east-west block-averaged samples, so for cells with ew_samples_averaged > 1 the raw min/max here are wider than the table's elev_min / elev_max; for ew_samples_averaged == 1 they are identical.

Downloads last month
41