feat: wider PDS ODE dataset roster (Phase 2)

docs/index.md

@@ -44,6 +44,30 @@ for batch in loader:
     batch["image"]  # torch.Tensor (16, C, H, W)
 ```
 
+## Beyond the STAC catalog
+
+Some instruments (LROC, LOLA, Mini-RF, Diviner, ShadowCam, Clementine, ...)
+aren't in the USGS ARD STAC catalog at all; those datasets search the NASA
+PDS Orbital Data Explorer instead, or read a single fixed mosaic URL, behind
+the exact same interface. The full roster is in
+[Instrument datasets](reference/datasets.md); two examples:
+
+```python
+import astrofetch as af
+
+# LROC NAC stereo DTM sites are a few hundred named sites, not global
+# coverage, so sampled windows are drawn from inside a real site by default.
+nac = af.LROCNACDTM(products=["dtm", "ortho"], bbox=(3.0, 25.0, 4.5, 26.5))
+sample = nac[0]
+
+# A global 100 m WAC mosaic and a global LOLA DEM, channel-stacked with `&`.
+terrain = af.LROCWACMosaic(resolution=100) & af.LOLA(resolution=100)
+
+# ShadowCam mosaics of permanently shadowed polar craters, another
+# site-based instrument -- footprint sampling applies the same way.
+shadowcam = af.ShadowCam(products=["mosaic"])
+```
+
 ## Discovering what data exists
 
 The `MOON` catalog enumerates probes, instruments, and products, and points at

docs/reference/catalog.md

@@ -1,8 +1,11 @@
 # Catalog & layer registry
 
 `MOON` is the discovery catalog — enumerate probes, instruments, and products,
-and reach the dataset classes. `LAYERS` maps each layer id to its provenance and
-backing STAC collection.
+and reach the dataset classes. `LAYERS` maps each layer id to its provenance
+and backing source (`LayerSpec.source` is `"stac"`, `"ode"`, or `"mosaic"`).
+A probe may also carry `granules`: experimental, non-map-projected dataset
+classes (see [Raw granules](granules.md)) that sit outside the layer/product
+contract entirely and are not part of `LAYERS`.
 
 ::: astrofetch.moon.layers.MOON
     options:

docs/reference/data.md

@@ -1,7 +1,21 @@
 # Data layer
 
-Body-agnostic building blocks shared by every instrument dataset: STAC search,
-windowed COG reads, the target grid, and the disposable cache.
+Body-agnostic building blocks shared by every instrument dataset: STAC and
+PDS ODE search, windowed raster reads, the target grid, and the disposable
+cache. Three sources back instrument datasets — see
+[Instrument datasets](datasets.md) for which each instrument uses:
+
+- **STAC** (`astrofetch.data.stac`): the USGS Astrogeology Analysis Ready
+  Data catalog, searched via `pystac-client`.
+- **PDS ODE** (`astrofetch.data.ode`): the NASA PDS Orbital Data Explorer
+  REST API, for instruments (LROC, LOLA, M3, ...) the STAC catalog does not
+  carry.
+- **Fixed mosaics**: a handful of instruments are published as one global
+  (or near-global) file rather than many searchable items; these are read
+  directly from a well-known URL in `astrofetch.data.endpoints`, no search.
+
+All three reproject through the same windowed-read path and share the same
+sample-dict contract.
 
 ## Target grid
 
@@ -15,6 +29,10 @@ windowed COG reads, the target grid, and the disposable cache.
 
 ::: astrofetch.data.stac
 
+## PDS ODE search
+
+::: astrofetch.data.ode
+
 ## Cache
 
 ::: astrofetch.data.cache

docs/reference/datasets.md

@@ -2,14 +2,68 @@
 
 One dataset class per instrument, each a map-style `torch` dataset that samples
 coregistered patches. Combine instruments with `&` to stack their channels over
-the overlapping region.
+the overlapping region. Every instrument dataset shares the same sample-dict
+contract regardless of where its products are read from — see
+[Data layer](data.md) for the three sources (STAC, PDS ODE, fixed mosaics).
+
+For raw, non-map-projected data (camera-geometry strips), see the separate,
+experimental [Raw granules](granules.md) page — a deliberately different
+contract, not part of the windowed-dataset family below.
+
+## Kaguya (SELENE)
 
 ::: astrofetch.moon.datasets.KaguyaTC
 
 ::: astrofetch.moon.datasets.KaguyaTCImagery
 
+## Lunar Reconnaissance Orbiter
+
+::: astrofetch.moon.datasets.LROCNACDTM
+
+::: astrofetch.moon.datasets.LROCWACMosaic
+
+::: astrofetch.moon.datasets.LOLA
+
+::: astrofetch.moon.datasets.SLDEM2015
+
+::: astrofetch.moon.datasets.MiniRF
+
+::: astrofetch.moon.datasets.DivinerGDR
+
+::: astrofetch.moon.datasets.WACGLD100
+
+::: astrofetch.moon.datasets.WACTiO2
+
+::: astrofetch.moon.datasets.LROCWACGlobal
+
+::: astrofetch.moon.datasets.LROCWACColor
+
+::: astrofetch.moon.datasets.LROCNACROI
+
+## Korea Pathfinder Lunar Orbiter
+
+::: astrofetch.moon.datasets.ShadowCam
+
+## Clementine
+
+::: astrofetch.moon.datasets.ClementineUVVIS
+
+::: astrofetch.moon.datasets.ClementineNIR
+
+## Base classes
+
 ::: astrofetch.moon.datasets.InstrumentDataset
 
+::: astrofetch.moon.datasets.ODEInstrumentDataset
+
+::: astrofetch.moon.datasets.MosaicDataset
+
 ::: astrofetch.moon.datasets.IntersectionDataset
 
+## Product specs
+
 ::: astrofetch.moon.datasets.Product
+
+::: astrofetch.moon.datasets.ODEAsset
+
+::: astrofetch.moon.datasets.MosaicAsset

docs/reference/granules.md

@@ -0,0 +1,48 @@
+# Raw granules (experimental)
+
+!!! warning "Experimental — a different contract than every other dataset"
+    Unlike the [instrument datasets](datasets.md), raw granule datasets are
+    **not** reprojected onto a common grid: each item is one raw PDS product
+    (an NAC/WAC calibrated strip, an M3 radiance cube, ...), read in its own
+    native camera/instrument geometry. No reprojection, resampling, ISIS, or
+    SPICE processing is applied.
+
+    Consequences:
+
+    - **No bbox windowing.** `__getitem__` returns a whole granule (or a row
+      range, via `rows=`), not a patch cropped to a requested extent.
+    - **Ragged shapes across items.** The default `DataLoader` collation will
+      not work; use `batch_size=None` or a custom `collate_fn`.
+    - **No `&` composition.** There is no shared grid to stack channels onto.
+
+    `len()` is the number of PDS ODE products matching a bbox, fetched once
+    and eagerly in `__init__`, so `len()` never needs a network call.
+
+## Reading large strips
+
+NAC/WAC strips can be gigapixel. Reading a whole granule with no `rows=`
+raises once its pixel count exceeds `max_pixels` (about 512 MiB as float32
+by default), naming the granule's size and suggesting a row range:
+
+```python
+import astrofetch as af
+
+# Read only the first 512 rows of every matching strip instead of the whole
+# multi-gigapixel granule.
+dataset = af.LROCNACRaw(bbox=(-26.3, -50.6, -25.5, -49.7), rows=slice(0, 512))
+sample = dataset[0]
+sample["image"]  # (bands, 512, W) float32, physical values
+sample["mask"]   # same-shaped bool validity
+```
+
+## Datasets
+
+::: astrofetch.moon.granules.LROCNACRaw
+
+::: astrofetch.moon.granules.LROCWACRaw
+
+::: astrofetch.moon.granules.M3
+
+## Base class
+
+::: astrofetch.moon.granules.GranuleDataset

docs/roadmap.md

@@ -7,10 +7,25 @@ this page is the short version.
 | Phase | Deliverable | Status |
 |:-----:|:------------|:------:|
 | 0 | Scaffolding — package, CI, docs, target API | Done |
-| 1 | STAC sampler MVP — bbox to coregistered `(C, H, W)` tensor | In progress |
-| 2 | Datasets and transforms — grid-tile dataset, spatial splits, transforms, LRO WAC | Planned |
+| 1 | STAC sampler MVP — bbox to coregistered `(C, H, W)` tensor | Done |
+| 2 | Datasets and transforms — new data sources, grid-tile dataset, spatial splits, transforms | In progress |
 | 3 | Release and community — PyPI, planetarypy affiliation, paper | Planned |
 
-**Current phase: Phase 1.** `InstrumentDataset.read(bbox)` now fetches real COGs
+**Current phase: Phase 2.** `InstrumentDataset.read(bbox)` fetches real COGs
 from the USGS ARD catalog, reprojects them onto a common geographic grid,
-applies scale/offset, mosaics overlapping items, and caches the result.
+applies scale/offset, mosaics overlapping items, and caches the result — the
+Phase 1 exit criterion. Phase 2 has delivered new data sources beyond STAC:
+the NASA PDS Orbital Data Explorer (`ODEInstrumentDataset`) now backs a wide
+roster of instruments — LROC NAC stereo DTM sites and region-of-interest
+mosaics, Mini-RF S-band radar, Diviner rock abundance and regolith
+temperature, WAC GLD100, TiO2, tiled global morphology, and 7-color
+reflectance, ShadowCam polar mosaics and DTMs, and the Clementine UVVIS and
+NIR basemaps — and fixed-URL mosaics (`MosaicDataset`) add the LRO WAC
+global mosaic and the LOLA and SLDEM2015 global DEMs, all behind the same
+sample-dict contract and `&` composition as the STAC-backed datasets. See
+[Instrument datasets](reference/datasets.md) for the full list. An
+experimental, separately-contracted raw-granule path
+(`astrofetch.moon.granules`) also now exists for camera-geometry NAC/WAC
+strips and M3 radiance cubes; see [Raw granules](reference/granules.md).
+Still open for Phase 2: `GridTileDataset`, spatial-autocorrelation-aware
+train/val/test splitting, transforms, and the WMS/WMTS rendered mode.

mkdocs.yml

@@ -28,6 +28,7 @@ nav:
   - Roadmap: roadmap.md
   - API Reference:
       - Instrument datasets: reference/datasets.md
+      - Raw granules (experimental): reference/granules.md
       - Catalog & registry: reference/catalog.md
       - Data layer: reference/data.md
 

pyproject.toml

@@ -22,6 +22,7 @@ dependencies = [
   "torch>=2.0",
   "pystac-client>=0.7",
   "rasterio>=1.3",
+  "requests>=2.31",
 ]
 
 [project.urls]

src/astrofetch/__init__.py

@@ -2,19 +2,58 @@
 
 One dataset class per instrument; combine them with ``&`` to receive
 coregistered multichannel samples. AstroFetch is a thin layer over existing
-archive tooling (STAC, COGs), never a mirror.
+archive tooling (STAC, COGs, PDS ODE), never a mirror of any archive.
 """
 
 from astrofetch import moon
-from astrofetch.moon import MOON, IntersectionDataset, KaguyaTC, KaguyaTCImagery
+from astrofetch.moon import (
+    LOLA,
+    LROCNACDTM,
+    LROCNACROI,
+    M3,
+    MOON,
+    SLDEM2015,
+    WACGLD100,
+    ClementineNIR,
+    ClementineUVVIS,
+    DivinerGDR,
+    IntersectionDataset,
+    KaguyaTC,
+    KaguyaTCImagery,
+    LROCNACRaw,
+    LROCWACColor,
+    LROCWACGlobal,
+    LROCWACMosaic,
+    LROCWACRaw,
+    MiniRF,
+    ShadowCam,
+    WACTiO2,
+)
 
 __version__ = "0.1.0"
 
 __all__ = [
+    "LOLA",
     "MOON",
+    "M3",
+    "ClementineNIR",
+    "ClementineUVVIS",
+    "DivinerGDR",
     "IntersectionDataset",
     "KaguyaTC",
     "KaguyaTCImagery",
+    "LROCNACDTM",
+    "LROCNACROI",
+    "LROCNACRaw",
+    "LROCWACColor",
+    "LROCWACGlobal",
+    "LROCWACMosaic",
+    "LROCWACRaw",
+    "MiniRF",
+    "SLDEM2015",
+    "ShadowCam",
+    "WACGLD100",
+    "WACTiO2",
     "moon",
     "__version__",
 ]

src/astrofetch/data/endpoints.py

@@ -9,3 +9,36 @@
 
 STAC_API_ROOT = "https://stac.astrogeology.usgs.gov/api/"
 """USGS Astrogeology Analysis Ready Data STAC API root (pystac-client entry)."""
+
+ODE_API_ROOT = "https://oderest.rsl.wustl.edu/live2/"
+"""NASA PDS Orbital Data Explorer (ODE) REST API root, Washington Univ. St.
+Louis. Used to search PDS3/PDS4 products (LROC, LOLA, M3, ...) by instrument
+and bounding box; the USGS ARD STAC catalog does not carry these instruments.
+Last verified 2026-07-20."""
+
+LROC_WAC_MOSAIC_100M_URL = (
+    "https://asc-pds-services.s3.us-west-2.amazonaws.com/mosaic/"
+    "Lunar_LRO_LROC-WAC_Mosaic_global_100m_June2013.tif"
+)
+"""LRO LROC WAC global morphology mosaic, 100 m/px, equirectangular.
+
+Not a Cloud Optimized GeoTIFF (striped, no overviews): windowed reads at
+native resolution (100 m) are efficient; heavily downsampled reads are not.
+Last verified 2026-07-20.
+"""
+
+LOLA_DEM_128_URL = (
+    "https://pds-geosciences.wustl.edu/lro/lro-l-lola-3-rdr-v1/lrolol_1xxx/"
+    "data/lola_gdr/cylindrical/float_img/ldem_128_float.lbl"
+)
+"""LOLA global DEM, 128 px/degree (~237 m/px at the equator), float32 metres
+above the IAU 2015 Moon reference sphere. Detached PDS3 label; GDAL's PDS
+driver resolves the sibling ``.img`` over HTTPS. Last verified 2026-07-20."""
+
+SLDEM2015_URL = (
+    "https://pds-geosciences.wustl.edu/lro/lro-l-lola-3-rdr-v1/lrolol_1xxx/"
+    "data/sldem2015/global/float_img/sldem2015_128_60s_60n_000_360_float.lbl"
+)
+"""SLDEM2015: LOLA + Kaguya Terrain Camera co-registered DEM, 128 px/degree,
+float32 metres. Source coverage is 60S-60N only (not a bug); windows outside
+that band read back with ``mask`` all ``False``. Last verified 2026-07-20."""