Adding a Custom Reader to Satpy

In order to add a reader to satpy, you will need to create two files:

  • a YAML file for describing the files to read and the datasets that are available

  • a python file implementing the actual reading of the datasets and metadata

Satpy implements readers by defining a single “reader” object that pulls information from one or more file handler objects. The base reader class provided by Satpy is enough for most cases and does not need to be modified. The individual file handler classes do need to be created due to the small differences between file formats.

The below documentation will walk through each part of making a reader in detail. To do this we will implement a reader for the EUMETSAT NetCDF format for SEVIRI data.

Naming your reader

Satpy tries to follow a standard scheme for naming its readers. These names are used in filenames, but are also used by users so it is important that the name be recognizable and clear. Although some special cases exist, most fit in to the following naming scheme:

<sensor>[_<processing level>[_<level detail>]][_<file format>]

All components of the name should be lowercase and use underscores as the main separator between fields. Hyphens should be used as an intra-field separator if needed (ex. goes-imager).

sensor:

The first component of the name represents the sensor or instrument that observed the data stored in the files being read. If the files are the output of a specific processing software or a certain algorithm implementation that supports multiple sensors then a lowercase version of that software’s name should be used (e.g. clavrx for CLAVR-x, nucaps for NUCAPS). The sensor field is the only required field of the naming scheme. If it is actually an instrument name then the reader name should include one of the other optional fields. If sensor is a software package then that may be enough without any additional information to uniquely identify the reader.

processing level:

This field marks the specific level of processing or calibration that has been performed to produce the data in the files being read. Common values of this field include: sdr for Sensor Data Record (SDR), edr for Environmental Data Record (EDR), l1b for Level 1B, and l2 for Level 2.

level detail:

In cases where the processing level is not enough to completely define the reader this field can be used to provide a little more context. For example, some VIIRS EDR products are specific to a particular field of study or type of scientific event, like a flood or cloud product. In these cases the detail field can be added to produce a name like viirs_edr_flood. This field shouldn’t be used unless processing level is also specified.

file format:

If the file format of the files is informative to the user or can distinguish one reader from another then this field should be specified. Common format names should be abbreviated following existing abbreviations like nc for NetCDF3 or NetCDF4, hdf for HDF4, h5 for HDF5.

The existing reader’s table can be used for reference. When in doubt, reader names can be discussed in the GitHub pull request when this reader is added to Satpy, or in a GitHub issue.

The YAML file

If your reader is going to be part of Satpy, the YAML file should be located in the satpy/etc/readers directory, along with the YAML files for all other readers. If you are developing a reader for internal purposes (such as for unpublished data), the YAML file should be located in any directory in $SATPY_CONFIG_PATH within the subdirectory readers/ (see Configuration).

The YAML file is composed of three sections:

  • the reader section, that provides basic parameters for the reader

  • the file_types section, that gives the patterns of the files this reader can handle

  • the datasets section, that describes the datasets available from this reader

The reader section

The reader section provides basic parameters for the overall reader.

The parameters to provide in this section are:

name

This is the name of the reader, it should be the same as the filename (without the .yaml extension). The naming convention for this is described above in the Naming your reader section above. short_name (optional): Human-readable version of the reader ‘name’. If not provided, applications using this can default to taking the ‘name’, replacing _ with spaces and uppercasing every letter.

long_name

Human-readable title for the reader. This may be used as a section title on a website or in GUI applications using Satpy. Default naming scheme is <space program> <sensor> Level <level> [<format>]. For example, for the abi_l1b reader this is "GOES-R ABI Level 1b" where “GOES-R” is the name of the program and not the name of the platform/satellite. This scheme may not work for all readers, but in general should be followed. See existing readers for more examples.

description

General description of the reader. This may include any restructuredtext formatted text like links to PDFs or sites with more information on the file format. This can be multiline if formatted properly in YAML (see example below).

status

The status of the reader (one of: Nominal, Beta, Alpha, Defunct; see Status Description for more details).

supports_fsspec

If the reader supports reading data via fsspec (either true or false).

sensors

The list of sensors this reader will support. This must be all lowercase letters for full support throughout in Satpy.

reader

The main python reader class to use, in most cases the FileYAMLReader is a good choice.

reader:
  name: seviri_l1b_nc
  short_name: SEVIRI L1b NetCDF4
  long_name: MSG SEVIRI Level 1b (NetCDF4)
  description: >
    NetCDF4 reader for EUMETSAT MSG SEVIRI Level 1b files.
  sensors: [seviri]
  reader: !!python/name:satpy.readers.yaml_reader.FileYAMLReader

Optionally, if you need to customize the DataID for this reader, you can provide the relevant keys with a data_identification_keys item here. See the Satpy internal workings: having a look under the hood section for more information.

The file_types section

Each file type needs to provide:

  • file_reader, the class that will handle the files for this reader, that you will implement in the corresponding python file. See the The python file section below.

  • file_patterns, the patterns to match to find files this reader can handle. The syntax to use is basically the same as format with the addition of time. See the trollsift package documentation for more details.

  • Optionally, a file type can have a requires field: it is a list of file types that the current file types needs to function. For example, the HRIT MSG format segment files each need a prologue and epilogue file to be read properly, hence in this case we have added requires: [HRIT_PRO, HRIT_EPI] to the file type definition.

file_types:
    nc_seviri_l1b:
        file_reader: !!python/name:satpy.readers.nc_seviri_l1b.NCSEVIRIFileHandler
        file_patterns: ['W_XX-EUMETSAT-Darmstadt,VIS+IR+IMAGERY,{satid:4s}+SEVIRI_C_EUMG_{processing_time:%Y%m%d%H%M%S}.nc']
    nc_seviri_l1b_hrv:
        file_reader: !!python/name:satpy.readers.nc_seviri_l1b.NCSEVIRIHRVFileHandler
        file_patterns: ['W_XX-EUMETSAT-Darmstadt,HRV+IMAGERY,{satid:4s}+SEVIRI_C_EUMG_{processing_time:%Y%m%d%H%M%S}.nc']

The datasets section

The datasets section describes each dataset available in the files. The parameters provided are made available to the methods of the implemented python class.

If your input files contain all the necessary metadata or you have a lot of datasets to configure look at the Dynamic Dataset Configuration section below. Implementing this will save you from having to write a lot of configuration in the YAML files.

Parameters you can define for example are:

  • name

  • sensor

  • resolution

  • wavelength

  • polarization

  • standard_name: The CF standard name for the dataset that will be used to determine the type of data. See existing readers for common standard names in Satpy or the CF standard name documentation for other available names or how to define your own. Satpy does not currently have a hard requirement on these names being completely CF compliant, but consistency across readers is important.

  • units: The units of the data when returned by the file handler. Although not technically a requirement, it is common for Satpy datasets to use “%” for reflectance fields and “K” for brightness temperature fields.

  • modifiers: The modification(s) that have already been applied to the data when it is returned by the file handler. Only a few of these have been standardized across Satpy, but are based on the names of the modifiers configured in the “composites” YAML files. Examples include sunz_corrected or rayleigh_corrected. See the metadata wiki for more information.

  • file_type: Name of file type (see above).

  • coordinates: An optional two-element list with the names of the longitude and latitude datasets describing the location of this dataset. This is optional if the data being read is gridded already. Swath data, from example data from some polar-orbiting satellites, should have these defined or no geolocation information will be available when the data are loaded. For gridded datasets a get_area_def function will be implemented in python (see below) to define geolocation information.

  • Any other field that is relevant for the reader or could be useful metadata provided to the user.

This section can be copied and adapted simply from existing seviri readers, like for example the msg_native reader.

datasets:
  HRV:
    name: HRV
    resolution: 1000.134348869
    wavelength: [0.5, 0.7, 0.9]
    calibration:
      reflectance:
        standard_name: toa_bidirectional_reflectance
        units: "%"
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b_hrv

  IR_016:
    name: IR_016
    resolution: 3000.403165817
    wavelength: [1.5, 1.64, 1.78]
    calibration:
      reflectance:
        standard_name: toa_bidirectional_reflectance
        units: "%"
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b
    nc_key: 'ch3'

  IR_039:
    name: IR_039
    resolution: 3000.403165817
    wavelength: [3.48, 3.92, 4.36]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: K
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b
    nc_key: 'ch4'

  IR_087:
    name: IR_087
    resolution: 3000.403165817
    wavelength: [8.3, 8.7, 9.1]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: K
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  IR_097:
    name: IR_097
    resolution: 3000.403165817
    wavelength: [9.38, 9.66, 9.94]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: K
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  IR_108:
    name: IR_108
    resolution: 3000.403165817
    wavelength: [9.8, 10.8, 11.8]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: K
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  IR_120:
    name: IR_120
    resolution: 3000.403165817
    wavelength: [11.0, 12.0, 13.0]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: K
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  IR_134:
    name: IR_134
    resolution: 3000.403165817
    wavelength: [12.4, 13.4, 14.4]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: K
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  VIS006:
    name: VIS006
    resolution: 3000.403165817
    wavelength: [0.56, 0.635, 0.71]
    calibration:
      reflectance:
        standard_name: toa_bidirectional_reflectance
        units: "%"
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  VIS008:
    name: VIS008
    resolution: 3000.403165817
    wavelength: [0.74, 0.81, 0.88]
    calibration:
      reflectance:
        standard_name: toa_bidirectional_reflectance
        units: "%"
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  WV_062:
    name: WV_062
    resolution: 3000.403165817
    wavelength: [5.35, 6.25, 7.15]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: "K"
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

  WV_073:
    name: WV_073
    resolution: 3000.403165817
    wavelength: [6.85, 7.35, 7.85]
    calibration:
      brightness_temperature:
        standard_name: toa_brightness_temperature
        units: "K"
      radiance:
        standard_name: toa_outgoing_radiance_per_unit_wavelength
        units: W m-2 um-1 sr-1
      counts:
        standard_name: counts
        units: count
    file_type: nc_seviri_l1b

The YAML file is now ready and you can move on to writing your python code.

Dynamic Dataset Configuration

The above “datasets” section for reader configuration is the most explicit method for specifying metadata about possible data that can be loaded from input files. It is also the easiest way for people with little python experience to customize or add new datasets to a reader. However, some file formats may have 10s or even 100s of datasets or variations of datasets. Writing the metadata and access information for every one of these datasets can easily become a problem. To help in these cases the available_datasets() file handler interface can be used.

This method, if needed, should be implemented in your reader’s file handler classes. The best information for what this method does and how to use it is available in the API documentation. This method is good when you want to:

  1. Define datasets dynamically without needing to define them in the YAML.

  2. Supplement metadata from the YAML file with information from the file content (ex. resolution).

  3. Determine if a dataset is available by the file contents. This differs from the default behavior of a dataset being considered loadable if its “file_type” is loaded.

Note that this is considered an advanced interface and involves more advanced Python concepts like generators. If you need help with anything feel free to ask questions in your pull request or on the Pytroll Slack.

The python file

The python files needs to implement a file handler class for each file type that we want to read. Such a class needs to implement a few methods:

  • the __init__ method, that takes as arguments

    • the filename (string)

    • the filename info (dict) that we get by parsing the filename using the pattern defined in the yaml file

    • the filetype info that we get from the filetype definition in the yaml file

    This method can also receive other file handler instances as parameter if the filetype at hand has requirements. (See the explanation in the YAML file filetype section above)

  • the get_dataset method, which takes as arguments

    • the dataset ID of the dataset to load

    • the dataset info that is the description of the channel in the YAML file

    This method has to return an xarray.DataArray instance if the loading is successful, containing the data and metadata of the loaded dataset, or return None if the loading was unsuccessful.

    The DataArray should at least have a y dimension. For data covering a 2D region on the Earth, their should be at least a y and x dimension. This applies to non-gridded data like that of a polar-orbiting satellite instrument. The latitude dimension is typically named y and longitude named x. This may require renaming dimensions from the file, see for the xarray.DataArray.rename() method for more information and its use in the example below.

    If the reader should be compatible with opening remote files see Adding remote file support to a reader.

  • the get_area_def method, that takes as single argument the DataID for which we want the area. It should return a AreaDefinition object. For data that cannot be geolocated with an area definition, the pixel coordinates will be loaded using the get_dataset method for the resulting scene to be navigated. The names of the datasets to be loaded should be specified as a special coordinates attribute in the YAML file. For example, by specifying coordinates: [longitude_dataset, latitude_dataset] in the YAML, Satpy will call get_dataset twice, once to load the dataset named longitude_dataset and once to load latitude_dataset. Satpy will then create a SwathDefinition with this coordinate information and assign it to the dataset’s .attrs['area'] attribute.

  • Optionally, the get_bounding_box method can be implemented if filtering files by area is desirable for this data type

On top of that, two attributes need to be defined: start_time and end_time, that define the start and end times of the sensing. See the Time Metadata section for a description of the different times that Satpy readers typically use and what times should be used for the start_time and end_time. Note that these properties will be assigned to the start_time and end_time metadata of any DataArrays returned by get_dataset, any existing values will be overwritten.

If you are writing a file handler for more common formats like HDF4, HDF5, or NetCDF4 you may want to consider using the utility base classes for each: satpy.readers.hdf4_utils.HDF4FileHandler, satpy.readers.hdf5_utils.HDF5FileHandler, and satpy.readers.netcdf_utils.NetCDF4FileHandler. These were added as a convenience and are not required to read these formats. In many cases using the xarray.open_dataset() function in a custom file handler is a much better idea.

Note

Be careful about the data types of the DataArray attributes (.attrs) your reader is returning. Satpy or other tools may attempt to serialize these attributes (ex. hashing for cache keys). For example, Numpy types don’t serialize into JSON and should therefore be cast to basic Python types (float, int, etc) before being assigned to the attributes.

Note

Be careful about the types of the data your reader is returning. It is easy to let the data be coerced into double precision floats (np.float64). At the moment, satellite instruments are rarely measuring in a resolution greater than what can be encoded in 16 bits. As such, to preserve processing power, please consider carefully what data type you should scale or calibrate your data to.

Single precision floats (np.float32) is a good compromise, as it has 23 significant bits (mantissa) and can thus represent 16 bit integers exactly, as well as keeping the memory footprint half of a double precision float.

One commonly used method in readers is xarray.DataArray.where() (to mask invalid data) which can be coercing the data to np.float64. To ensure for example that integer data is coerced to np.float32 when xarray.DataArray.where() is used, you can do:

my_float_dataarray = my_int_dataarray.where(some_condition, np.float32(np.nan))

One way of implementing a file handler is shown below:

# this is seviri_l1b_nc.py
from satpy.readers.file_handlers import BaseFileHandler
from pyresample.geometry import AreaDefinition

class NCSEVIRIFileHandler(BaseFileHandler):
    def __init__(self, filename, filename_info, filetype_info):
        super(NCSEVIRIFileHandler, self).__init__(filename, filename_info, filetype_info)
        self.nc = None

    def get_dataset(self, dataset_id, dataset_info):
        if dataset_id['calibration'] != 'radiance':
            # TODO: implement calibration to reflectance or brightness temperature
            return
        if self.nc is None:
            self.nc = xr.open_dataset(self.filename,
                                      decode_cf=True,
                                      mask_and_scale=True,
                                      chunks={'num_columns_vis_ir': "auto",
                                              'num_rows_vis_ir': "auto"})
            self.nc = self.nc.rename({'num_columns_vir_ir': 'x', 'num_rows_vir_ir': 'y'})
        dataset = self.nc[dataset_info['nc_key']]
        dataset.attrs.update(dataset_info)
        return dataset

    def get_area_def(self, dataset_id):
        return pyresample.geometry.AreaDefinition(
            "some_area_name",
            "on-the-fly area",
            "geos",
            "+a=6378169.0 +h=35785831.0 +b=6356583.8 +lon_0=0 +proj=geos",
            3636,
            3636,
            [-5456233.41938636, -5453233.01608472, 5453233.01608472, 5456233.41938636])

class NCSEVIRIHRVFileHandler():
  # left as an exercise to the reader :)

If you have any questions, please contact the Satpy developers.

Auxiliary File Download

If your reader needs additional data files to do calibrations, corrections, or anything else see the Auxiliary Data Download document for more information on how to download and cache these files without including them in the Satpy python package.