Skip to content

Welcome to the Planetary Data Reader (pdr)

This tool provides a single command---read(‘/path/to/file’)---for ingesting all common planetary data types. It is currently in development. Almost every kind of "primary observational data" product currently archived in the PDS (under PDS3 or PDS4) should be covered eventually. Currently-supported datasets are listed here.

If the software fails while attempting to read from datasets that we have listed as supported, please submit an issue with a link to the file and information about the error (if applicable). There might also be datasets that work but are not listed. We would like to hear about those too. If a dataset is not yet supported that you would like us to consider prioritizing, please fill out this request form.

You can access the source code for pdr on github at:


If you use pdr in your work, please cite us using our Zenodo DOI: DOI


pdr is now on conda and pip. We recommend (and only officially support) installation into a conda environment. You can do this like so:

conda create --name pdrenv
conda activate pdrenv
conda install -c conda-forge pdr

The minimum supported version of Python is 3.9.

Using the conda install will install all dependencies in the environment.yml file (both required and optional) for pdr. If you'd prefer to forego the optional dependencies, please use minimal_environment.yml in your installation. This is not supported through a direct conda install as described above and will reqiore additional steps. Optional dependencies and the added functionality they support are listed below:

  • pvl: allows Data.load("LABEL", as_pvl=True) which will load PDS3 labels as pvl objects rather than plain text
  • astropy: adds support for FITS files
  • jupyter: allows usage of the Example Jupyter Notebook (and other jupyter notebooks you create)
  • pillow: adds support for TIFF files and browse image rendering
  • matplotlib: allows usage of save_sparklines, an experimental browse function


(You can check out our example Notebook on Binder for a quick interactive demo of functionality: Binder)

Just open a python shell and run import pdr and then, where filename is the full path to a data file or a metadata / label file (extensions .LBL, .lbl, or .xml). read() will look for corresponding data or metadata files in the same path, or read metadata directly from the data file if it has an attached label.

read returns a pdr.Data object whose attributes include all of the data and metadata. Data attributes take their names directly from the product's label. They can be accessed either as attributes or using dict-style [] index notation. For example, PDS3 image objects are often named "IMAGE", so you could examine a PDS3 image as an array with:

>>> data ="/path/to/cr0_398560467edr_f0030004ccam02012m1.LBL")
>>> data['IMAGE']
array([[21, 21, 20, ..., 19, 19, 20],
       [21, 21, 21, ..., 19, 20, 20],
       [21, 21, 20, ..., 20, 20, 20],
       [25, 25, 25, ..., 26, 26, 26],
       [25, 25, 25, ..., 27, 26, 26],
       [24, 25, 25, ..., 26, 26, 26]], dtype=int16)

Parsed metadata are stored in a pdr.Metadata object and exposed as the metadata property of a pdr.Data object. You can access metadata values with dict-style [] index notation or the convenience method metaget. For instance:

>>> data.metaget('INSTRUMENT_HOST_NAME')

Some PDS products (like this one) contain multiple data objects. You can look at all the objects associated with a product with .keys():

>>> data.keys()

Output data types

In general: + Image data are presented as NumPy ndarray objects. + Table data are presented as pandas DataFrame objects. + Parsed label contents (metadata fields + values) are presented in a pdr.Metadata object (behaves much like a dict). + Header and label contents are presented as plain text (str objects), bytes, or, for PDS4 labels, pds4_tools.reader.label_objects.Label objects. + Other data are presented as simple python types (str, tuple, dict). + There might be rare exceptions.

Notes and Caveats

Additional processing

Some data, especially calibrated image data, require the application of additional offsets or scale factors to convert the storage units to meaningful physical units. The information on how and when to apply such adjustments is typically stored (as plain text) in the instrument SIS, and the scale factors themselves are often (but not always) stored in the label. Image data also often contain special constants (like missing or invalid data), and these constants are often not explicitly specified in the label. pdr is therefore not guaranteed to correctly apply -- or even know anything about -- these constants.

pdr.Data objects offer a convenience method that attempts to mask invalid values and apply any scaling and offset specified in the label. Use it like: scaled_image = data.get_scaled('IMAGE'). However, we do not perform science validation of these outputs, so do not trust that they are ready for analysis without further processing or validation. Contributions towards making this more effective for specific data product types are very much welcomed.

If you'd like to visualize the outputs that this creates, the dump_browse method creates separate browse files for all currently-loaded objects (as .jpg, .txt., or .csv) in your working directory. Use it like: data.dump_browse(). This uses the get_scaled method for images and will also output browse products for tables and labels.

.FMT files

Some PDS3 table formats are defined in external reference files (usually with a .FMT extension). You can often find these in the LABEL or DOCUMENT subdirectories of data archive volumes. If you place the relevant format files in the same directory as the data files you are trying to read, pdr will be able to use them to interpret the table data. If you attempt to read a table object that requires a format file that is not present, pdr will not be able to open the table object, and will throw a warning that includes the format file name in order to help you go find it. Future functionality may make this process smoother.

Data attribute naming

The observational and metadata attributes (or keys) of pdr.Data objects take their names directly from the metadata files. We believe that maintaining this strong correlation between the representation of the data in-language and the representation of the data in-file is important, even when it causes us to break strict PEP-8 standards for attribute capitalization. There are three exceptions at present: 1. Some table formats include repeated column names. For usability and compatibility, we force these to be unique by suffixing 0-indexed increasing integers. So a table definition with two separate columns named "COLUMN" will return a pandas DataFrame with columns named "COLUMN_0" and "COLUMN_1." 2. PDS3 data object names sometimes contain spaces. pdr replaces the spaces with underscores in order to make them easily usable as Python attributes.

PDS4 products

pdr.Data wraps pds4_tools to read PDS4 products. All valid PDS4 products should be fully supported. pdr modifies some pds4_tools outputs in order to provide interface and behavior consistency. In general, you should be able to use pdr with PDS4 products the same way you do with PDS3 products.

Some PDS data products have both PDS3 and PDS4 labels. Data object names, metadata, and even data field names and format specifications often differ slightly between these labels, so pdr.Data may produce different outputs depending on which label you use to initialize it. This is not a bug. However, in general, if a PDS3 label is available, we recommend initializing the object from the PDS3 label rather than the PDS4 label.

FITS files

pdr.Data wraps to read data from FITS files. pdr converts objects produced byastropy to np.ndarrays (FITS arrays and compressed arrays), pd.DataFrames (FITS ASCII and binary tables), or MultiDicts (FITS headers), so you do not need to change your code simply because a file is in FITS format.

Whenever you load a data object from a FITS file, pdr also places the associated FITS header in a key of your Data object named "$objectname_HEADER" -- for instance, if you load an object named "HK_TABLE", its FITS header will appear in Data.HK_TABLE_HEADER. You can also use that name to directly load the header without loading the entire data object.

pdr's FITS-loading behavior is somewhat different depending on whether a PDS3 label, a PDS4 label, or no PDS label is present:

FITS with no PDS label

pdr has 'first-class' support for FITS files and does not require a PDS label to open them. This gives FITS users access to all of the pdr convenience functions and idioms even if no PDS labels are available.

pdr names data objects in FITS files opened 'directly' by FITS extension names (EXTNAME) when specified, and by extension types ('PRIMARY', 'COMPRESSED_IMAGE', etc.) when not. Like duplicate objects in PDS products, pdr disambiguates duplicate names by appending integers.

Note that any time you pass a FITS file to pdr and there is no PDS label in that file's directory, pdr will default to this behavior. Conversely, if you'd like to force pdr to open a file 'directly' from its FITS headers, but there is a PDS label in its directory, simply specify that the FITS file is also the label file, like:'name_of_data_file.fits', label_fn='name_of_data_file.fits').

With PDS3 labels

pdr prefers the data specification given in FITS headers to the data specification in the PDS3 label, with one exception: it uses object names from the PDS3 label. In our experience, because FITS is more rigorously standardized than PDS3, using the FITS header is more reliable. Note that in some cases, PDS3 and FITS specifications may differ, even when the specification in the PDS3 label is technically valid. For instance, column names might be given differently in the FITS header, or a PDS3 TABLE might be stored as a FITS array HDU.

With PDS4 labels

Because PDS4 is more reliable and rigorously standardized than PDS3, pdr prefers the specification given in PDS4 labels to the specification given in FITS headers. You can always override this by passing the label_fn argument.

Lazy loading

Because many planetary data objects are very large, pdr helps conserve your time and memory by loading them lazily. It loads data objects into memory when they are explicitly referenced, not when pdr.Data is initialized. For example, referencingdata.IMAGE will immediately load the IMAGE object if it has not already been loaded. Alternatively, you can load objects by using the load method, like data.load("IMAGE"). You can also pass the 'all' argument to load all data objects, like data.load("all").

Missing files

If a file referenced by a label is missing, pdr will throw warnings and populate the associated attribute from the portion of the label that mentions that file. You are most likely to encounter this for DESCRIPTION files in document formats (like .TXT). These warnings do not prevent you from using objects loaded from files that are actually present in your filesystem.

Big files (like HiRISE)

pdr currently performs no special memory management, so use caution when attempting to read very large files. We intend to implement memory management in the future.


.jp2 support is not guaranteed for WSL (Windows Subsystem for Linux). It is supported on Windows itself and Linux.



Our testing strategy focuses on end-to-end systems integration testing to ensure consistency, coverage of supported datasets, and (to the extent we can verify it) correctness of output. We prioritize this kind of testing because pdr is intended to work with a wildly diverse body of data. Most of pdr's source code is highly multifunctional, designed to accept format ambiguity and unexpected inputs.

This means that obtaining meaningful coverage from unit tests is extremely difficult. No written standard is reliable: the products are the only reliable representation of the products. Describing their idiosyncracies well enough to test against them requires replicating a representative subset of the data corpus and verifying that pdr behaves consistently when executed on this subset.


This is exactly what our application ix is designed to facilitate. It serves both as a regression testing framework and a development tool. Although it serves as pdr's primary test suite, it is really a distinct application, so it lives in a different repository: pdr-tests. Note that ix is feature-complete and stable, but lacks comprehensive user-facing documentation.

Unit Tests

All this being said, many potential bugs in pdr are detectable by unit testing on simple cases, and although ix is well-optimized, running a comprehensive regression test against >100 GB of data products is simply not a fast affair. For these reasons, pdr also features a small suite of unit tests. You can run them by executing pytest from the repository root.

Funding Acknowledgement

This work is supported by NASA grant No. 80NSSC21K0885.