Classes and Configuration

This page describes the classes that are implemented in MPAS-JEDI. It explains in broad terms what they are responsible for, how they are implemented, and the configuration options that are available for controlling their behavior.

The individual classes are the model-dependent building blocks of the interface to MPAS-Model, which are used in all generic applications that are implemented in OOPS. All JEDI applications are configurable through a top-level application-specific YAML file, the general layout and formatting of which is described in detail within the JEDI Configuration File Format documentation. The run-time behavior of each MPAS-JEDI class instance, or object, is configurable via that application-specific YAML file or in one of the MPAS-Model configuration files, such as namelist.atmosphere and streams.atmosphere.

In the sections that follow, there are context-free examples of how to configure each class in any MPAS-JEDI application, such as would be described in the aforementioned application-specific YAML file. In general, YAML files are hierarchical, such that the small snippets of YAML given as examples here might be indented under some combination of parent YAML keys that are not shown here. It is often helpful to have the clarity of context-specificity, for which it is recommended to follow along with one or more full application YAML files, such as those used in the MPAS-JEDI ctests and described here.

Before continuing, it is important to define MPAS-Model as referring to the common framework that is used for all the various MPAS cores (e.g., Atmosphere, Ocean, Seaice). At this time, MPAS-JEDI only works with MPAS-Atmosphere. Most of the elements that bind MPAS-JEDI and MPAS-Atmosphere are accomplished through the generic MPAS-Model framework, but only atmospheric fields are accessed within the MPAS-JEDI code.

Geometry

The Geometry class is responsible for generating descriptors of the MPAS mesh and for associating an MPI communicator passed down from the generic application with a particular Geometry object. The Geometry class constructor uses the MPAS-Model subroutine mpas_init (located within MPAS-Model/src/driver/mpas_subdriver.F) to initialize pointers to two MPAS-Model derived types, domain_type and core_type, which fully describe the MPAS-Model mesh information and MPAS-Atmosphere fields that are required in an MPAS-JEDI application. The domain_type and its underlying components facilitate the operations described by most other classes in MPAS-JEDI.

Inheritance from MPAS-Model

Every MPAS-JEDI application requires a Geometry object and a call to mpas_init, as described above. The calling structure of that subroutine requires there to be several files present in the run directory, some of which are specific to the Atmosphere core. They are as follows:

  • namelist.atmosphere

  • streams.atmosphere

  • stream_list.atmosphere.output

  • GENPARM.TBL

  • LANDUSE.TBL

  • OZONE_DAT.TBL

  • OZONE_LAT.TBL

  • OZONE_PLEV.TBL

  • SOILPARM.TBL

  • VEGPARM.TBL

  • CAM_ABS_DATA.DBL

  • CAM_AEROPT_DATA.DBL

  • RRTMG_LW_DATA.DBL

  • RRTMG_SW_DATA.DBL

  • config_block_decomp_file_prefix.npe

If Thompson microphysics or the “convection_permitting” physics suite (see the MPAS-Atmosphere User’s Guide) is specified in the nml_file, more files like MP_THOMPSON*BL should be included in the run directory. While the names namelist.atmosphere and streams.atmosphere are hard-coded in MPAS-Atmosphere, the names of those files can be configured at run-time in MPAS-JEDI, as explained later in the nml_file and streams_file configuration element descriptions. That configurability is needed because dual-mesh data assimilation requires that those file names are specified independently for each of the MPAS-Model meshes.

config_block_decomp_file_prefix.npe is the graph partition file for the MPAS-Model mesh. The npe suffix refers to the number of processors over which the MPAS-Model mesh is decomposed. config_block_decomp_file_prefix is set in namelist.atmosphere under &decomposition. So far MPAS-JEDI applications have followed the naming conventions for the MPAS-Atmosphere uniform meshes, x1.init.nCells.graph.info.part., where nCells is the number of horizontal columns (e.g., 40962 for the 120 km mesh).

The number of processors used for the top-level generic MPAS-JEDI application is equal to the number of ranks in the global MPI communicator. Most applications (including 3DVar, 3DEnVar, hybrid 3DEnVar, and HofX) do not split the communicator, in which case all Geometry objects are associated with that same global communicator, and same total number of processors. However, OOPS splits the communicator for some ensemble applications (e.g., EDA, EnsHofX) and 4DEnVar such that the number of processors available to each Geometry object is variable at run-time, depending on how the application is configured.

The Geometry object’s communicator is passed to MPAS-Model’s mpas_init subroutine, which associates it with an instance of a MPAS-Model mesh. See MPAS-Model/src/driver/mpas_subdriver.F for the details. Up to this point, MPAS-JEDI has only been tested in situations where the MPAS-Model mesh uses all available ranks in the Geometry object’s communicator. The user must ensure that the number of processors available to the potentially split communicator corresponds to one of the graph partition files available for the MPAS-Model mesh used in that application. The total number of processors utilized is selected via the -n flag to mpiexec, mpirun, or whatever command is used on a particular compute system. There is no configuration element in the YAML file to select the number of processors utilized, but each selection in the configuration must be consistent with the number of processors selected.

Presently, it is only possible to build MPAS-JEDI for use with double floating point precision. Therefore, static lookup tables such as RRTMG_LW_DATA must be provided in double precision.

Configuration

There are four run-time options available to users in MPAS-JEDI geometry sections of the YAML file. An example for a 240km mesh is given below, although there is no requirement for the names of the nml_file and streams_file to follow any particular format, such as including a date or a mesh-spacing.

geometry:
  nml_file: "./namelist.atmosphere_2018041500_240km"
  streams_file: "./streams.atmosphere_240km"
  deallocate non-da fields: true
  interpolation type: unstructured

nml_file and streams_file

The nml_file and streams_file specify the namelist.atmosphere and streams.atmosphere files, respectively, which are used to initialize the MPAS-A mesh and additional model fields that are used to initialize MPAS-A physical quantities. Some of those physical quantities may not be needed for MPAS-JEDI applications, which will be discussed later in the context of the deallocate non-da fields option. For most MPAS-JEDI applications, it suffices to use the default namelist and streams file names, again namelist.atmosphere and streams.atmosphere, respectively. One specific case where the default names cannot be used is a dual-mesh Variational application.

For Variational applications, the geometry configuration is provided in two places, under cost function and within each of the iterations vector members under variational. The cost function.geometry configuration specifies the fine mesh used for background and analysis states, and the variational.iterations[:].geometry configurations specify the coarse meshes for the inner loop increments and ensemble input states (i.e., for EnVar). With that freedom of configurability, each outer loop iteration of the variational minimization can potentially use a different mesh in the inner loop. Note that for EnVar applications, using unique coarse meshes in each outer iteration requires users to provide ensemble states on those multiple meshes. For a single-mesh Variational application, the cost function.geometry and variational.iterations[:].geometry are identical. As an example, for a dual-mesh application with the above 240-km mesh being used for the outer loop’s fine mesh, the first inner loop might use a 480-km coarse mesh, in which case variational.iterations[0].geometry would look like

variational:
  iterations:
  - geometry:
      nml_file: "./namelist.atmosphere_2018041500_480km"
      streams_file: "./streams.atmosphere_480km"
      deallocate non-da fields: true
      interpolation type: unstructured

In this particular example, the primary difference between streams.atmosphere_240km and streams.atmosphere_480km is the MPAS state file that is used to initialize their respective model meshes. The 240km file would include a section that looks like

<immutable_stream name="restart"
                  type="input;output"
                  filename_template="restart.240km.$Y-$M-$D_$h.$m.$s.nc"
                  input_interval="initial_only"
                  clobber_mode="overwrite" />

where the prefix to filename_template is given as restart.240km.. The filename_template also contains date substitution strings, so that the date in the filename must correspond to the config_start_time option in namelist.atmosphere_2018041500_240km. The respective 480km streams_file would have an entry that looks like

<immutable_stream name="restart"
                  type="input;output"
                  filename_template="restart.480km.$Y-$M-$D_$h.$m.$s.nc"
                  input_interval="initial_only"
                  clobber_mode="overwrite" />

using a filename_template prefix of restart.480km., or whatever prefix the user prefers. There are also differences in the 240-km and 480-km nml_file’s, which are primarily related to settings that are mesh-specific. For example, config_block_decomp_file_prefix, which is desribed earlier in this section of the documentation.

Although the above discussion shows how to handle restart files in an MPAS-JEDI application, using full restart files requires significant disk space in cycling workflows and will likely be deprecated in the future. Another approach, which we call “2-stream”, avoids restart files and is available in both MPAS-JEDI and MPAS-Model, as distributed with MPAS-BUNDLE. For purposes of definition, it is important to understand that MPAS-Model uses the term “stream” to define the flow of data in and out of the model from and to files stored on the hard disk. Each stream can be defined as an input stream, an output stream, or both an input and an output stream. The “restart” stream is conveniently defined as both input and output, meaning that all of the same fields are read and written through that stream.

The 2-stream approach defines two unique input streams, and saves significant disk space in cycling workflows by splitting out time-invariant fields into an input stream called “static” and keeping only time-varying fields in the input stream simply named “input”. The “static” input stream includes the mesh, some of surface input variables (landmask, shdmin, albedo12m, etc.) and parameters for gravity wave drag over orography. Because those fields (“static”) are time-invariant, they are stored in a single file that does not require multiple copies across workflow cycles. It is important to note that this “static” stream does not correspond to the x1.nCells.static.nc (nCells being the number of mesh cells) file that is used, for example, in the MPAS-Model init_atmosphere core. x1.nCells.static.nc include a much smaller set of information that pertains only to the mesh, while the files read through the “static” stream defined in the MPAS-Model atmosphere core for 2-stream input also includes fields contained in cold-start files produced by the init_atmosphere executable. That executable, named init_atmosphere_model in a standalone MPAS-A build, is named mpas_init_atmosphere in an MPAS-BUNDLE build. Therefore, it is recommended to produce a single x1.*.init.nc using either the init_atmosphere_model or mpas_init_atmosphere executables following the instructions in the MPAS-A user guide available for download here.

The “input” input stream reads all of the fields needed for a cold-start forecast. The “input” input stream is efficient for replacing the “restart” stream as the sole input stream in cycling workflows, because the “restart” stream includes physical tendency and other fields needed for a perfect restart to within machine/IO precision that are inconsequential in cycling. In order to use the 2-stream input, one needs to modify the streams_file in both the forecast step and in all MPAS-JEDI applications (e.g., HofX, Variational) as follows, using the quasi-uniform 120km mesh as an example,

<immutable_stream name="static"
                  type="input"
                  filename_template="static.40962.nc"
                  io_type="pnetcdf,cdf5"
                  input_interval="initial_only" />
<immutable_stream name="input"
                  type="input"
                  filename_template="init.40962.$Y-$M-$D_$h.$m.$s.nc"
                  io_type="pnetcdf,cdf5"
                  input_interval="initial_only" />

The “input” input stream is used to read the forecast initial state in MPAS-Model (above) and the fields needed to initialize the MPAS-JEDI Geometry in all MPAS-JEDI applications. However, it is important to note that the fields initialized in the Geometry object are not the same as the background State object that feeds Variational and HofX applications. The fields in the Geometry object are only placeholders or templates for the fields that will eventually be read in the State::read method. The State::read and State::write (used for analysis output) methods use the “output” output stream, whose fields are entirely described at run-time. This is achieved by adding an input/output stream to the streams_file that includes a file name element giving a hard-coded name for a file that lists all fields to be read/written , i.e. stream_list.atmosphere.output in the example below

<stream name="output"
        type="input;output"
        io_type="pnetcdf,cdf5"
        filename_template="history.$Y-$M-$D_$h.$m.$s.nc"
        clobber_mode="overwrite"
        input_interval="initial_only"
        output_interval="none" >
        <file name="stream_list.atmosphere.output"/>
</stream>

Although the above example “output” stream gives a filename_template with the “history.” prefix, the actual filenames for input and output is generated within State::read and State::write methods respectively. Those methods use the filename specified in the YAML for each applicable State object.

In addition to the fields that are used for cold-start forecasts, there are other fields that are needed exclusively in the State objects of MPAS-JEDI applications, including analysis variables and fixed fields needed for CRTM or other observation operators. Because those extra fields can undergo IO through the State class, completely bypassing the hard-coded streams in MPAS-A’s Registry.xml (i.e., MPAS-Model/src/core_atmosphere/Registry.xml), it is useful to codify the fields in a unique output stream so that forecasts preceeding an MPAS-JEDI application will write all the necessary fields. Thus, the MPAS-A code distributed with MPAS-BUNDLE has an additional “da_state” output stream defined in Registry.xml. That stream should be added to the streams.atmosphere file used to configure the forecast as follows

<immutable_stream name="da_state"
                  type="output"
                  clobber_mode="truncate"
                  filename_template="mpasout.$Y-$M-$D_$h.$m.$s.nc"
                  io_type="pnetcdf,cdf5"
                  output_interval="0_06:00:00" />

The output_interval and clobber_mode should be modified to fit the user’s application in terms of the forecast file writing interval and need to overwrite existing files, respspectively. For a working example of using 2-stream input in MPAS-JEDI, users are referred to the JEDI-MPAS HofX tutorial. Step 4 of that tutorial shows how to read a file with an “mpasout.” prefix that was written using the “da_state” output stream. The same file is used both for the “input” input stream and for the State::read method.

Finally, there are three MPAS-A namelist settings of which MPAS-JEDI users ought to be aware:
  1. When running any MPAS-JEDI application, but not when running MPAS-A executables, the assimilation namelist and config_jedi_da option need to be added to the nml_file as

    &assimilation
      config_jedi_da = true
    /
    
  2. When the “restart” stream is used, config_do_restart should be set to true in the nml_file during all MPAS-JEDI applications and during the MPAS-A forecast. For 2-stream input, config_do_restart should be set to false.

  3. When conducting cycled forecast and data assimilation workflows, config_do_DAcycling should be set to true, which forces MPAS-A to re-initialize the coupled prognostic model fields from the MPAS-JEDI analysis fields. The analysis fields are described in more detail in the State / Increment class descriptions.

deallocate non-da fields

The geometry.deallocate non-da fields option is used to reduce physical memory usage in 3D JEDI applications that do not require time-integration of the MPAS-Model, i.e., applications that do not utilize the Model class. This setting controls deallocation of those unused fields, which are allocated in the domain_type object that is created in mpas_init. Refer to the fortran-level Geometry class source code for detailed information about which fields are deallocated.

interpolation type

The geometry.interpolation type setting is optional, and allows flexibility in dual-mesh applications, where interpolation is performed between the two meshes. Valid settings are bump or unstructured, which refer to two different interpolation implementations in SABER and OOPS, respectively. The default is bump interpolation if the option is not specified. GetValues also employs interpolation and the method can be chosen there by setting the identically named key interpolation type under GetValues and LinearGetValues.

State / Increment

The State and Increment classes in MPAS-JEDI have a fair amount of overlap between them. The constructors are largely the same and they share a number of methods, such as read, write, and mathematical operations. In order to simplify the code, MPAS-JEDI contains shared subroutines for both of those c++ classes in a fortran class called mpas_field within mpas-jedi/src/mpas-jedi/mpas_fields_mod.F90.

Inheritance from MPAS-Model

MPAS-JEDI leverages the MPAS-Model mpas_pool_type (i.e., MPAS-Model/src/framework/mpas_pool_types.inc) to manage groupings of model fields. That paradigm eases iterations over model fields. When considering mathematical operations on fields with different dimensions (i.e., 3D vs. 2D), special cases are handled using the mpas_pool_iterator_type%nDims attribute. This approach makes it easy to add new variables to State and Increment objects at run-time without much additional code. New code must be added when converting between previously unimplemented sets of MPAS-Model state variables, MPAS-JEDI analysis variables, and UFO GeoVaLs variables.

State

State objects are defined uniquely by three keys in the yaml file:

state:
  state variables: [temperature, spechum, uReconstructZonal, uReconstructMeridional, surface_pressure,
                    theta, rho, u, qv, pressure, landmask, xice, snowc, skintemp, ivgtyp, isltyp,
                    qc, qi, qr, qs, qg, cldfrac,
                    snowh, vegfra, t2m, q2m, u10, v10, lai, smois, tslb, pressure_p]
  filename: mpasout.2018-04-15_00.00.00.nc
  date: 2018-04-15T00:00:00Z

state variables

The state variables key determine which MPAS-Model fields (e.g., field1DReal, field2DReal, etc…) will comprise the State object’s stored data. This key, which is used during the constructor of the State object, affects all down-stream operations with this State object. All fields that are needed by an MPAS-JEDI application must be listed here and must also be included in stream_list.atmosphere.output. Note that five hydrometeors (qc, qi, qr, qs, and qg) in state variables are listed as scalar.

Although the user specifies which state variables are created and operated on within an MPAS-JEDI application using the yaml file, the fields for which IO is conducted are specified in stream_list.atmosphere.output. The list of variables there does not need to exactly match the list in the yaml. However, the user should be aware that State::read and State::write will attempt to read and write all fields listed in stream_list.atmosphere.output. There are three MPAS-JEDI fields that are derived directly from MPAS-Model fields within the read method, spechum, pressure, and temperature. None of those need to be listed within stream_list.atmosphere.output, but only a warning and not an error will result if they are included.

filename

The filename key determines which file is associated with IO during calls to the read and write methods from within a generic application. The filename may or may not contain any of the date-time placeholders recognized by MPAS-Model (i.e., $Y, $M, $D, $h, $m, $s). For example, '$Y-$M-$D_$h:$m:$s'. All of the date-time placeholders will be replaced with quantities associated with the valid date for this State.

date

The date key has two purposes. During the read method, this ISO8601-formatted date-time is used to tell the generic application the valid date for this State. Secondly, it may be used to subsitute the actual date components into the filename as described above, but only during the State::read method. The State::write method is not subject to this YAML key, because the generic application determines the valid date of the output State.

Increment

The Increment class differs from the State class in that it is primarily used to conduct mathematical operations. Often an Increment object is constructed by taking the difference between two State objects or by copying a subset of fields from a single State. That is why only the fields for which such a difference is calculated need be specified in the YAML. For example, an Increment object with the MPAS-JEDI standard analysis variables is defined with

analysis variables: [temperature, spechum, uReconstructZonal, uReconstructMeridional, surface_pressure]

The correct specification of state variables and analysis variables is application-dependent. Hydrometeor fields are added to State and Increment objects with the following strings: qc, qi, qr, qs, and qg for cloud, ice, rain, snow, and graupel, respectively. However, hydrometeor fields are only updated in an MPAS-JEDI variational application when assimilating observations that are sensitive to them. Users should refer to existing ctests and tutorials as examples.

Variable Changes

Control2Analysis

This linear variable change converts the control variables (which are used in the B matrix) to the analysis variables. For various variational applications of MPAS-JEDI, we have chosen the following set of analysis variables.

analysis variables:
- uReconstructZonal       # zonal wind at cell center [ m / s ]
- uReconstructMeridional  # meridional wind at cell center [ m / s ]
- temperature             # temperature [ K ]
- spechum                 # specific humidity [ kg / kg ]
- surface_pressure        # surface pressure [ Pa ]
- qc                      # mixing ratio for cloud water [ kg / kg ]
- qi                      # mixing ratio for cloud ice [ kg / kg ]
- qr                      # mixing ratio for rain water [ kg / kg ]
- qs                      # mixing ratio for snow [ kg / kg ]
- qg                      # mixing ratio for graupel [ kg / kg ]

The latter five hydrometeor variables are optional. These variables are chosen because fewer variable changes are required for implementing (1) the multivariate background error covariance and (2) the simulation of observation equivalent quantities from analysis variables.

We have chosen stream_function and velocity_potential for the momentum control variables in the B matrix. Thus, the wind transform from stream function and velocity potential to zonal and meridional winds is implemented in control2analysis.

variable change: Control2Analysis
input variables: [stream_function, velocity_potential, temperature, spechum, surface_pressure]        # control variables
output variables: [uReconstructZonal, uReconstructMeridional, temperature, spechum, surface_pressure] # analysis variables

We can also choose the pseudo relative humidity, relhum, as an optional moisture control variable. The pseudo relative humidity is defined as a specific humidity normalized by saturation specific humidity (of background temperature and pressure). For this, variable transform from pseudo relative humidity to specific humidity is implemented in control2analysis.

variable change: Control2Analysis
input variables: [stream_function, velocity_potential, temperature, relhum, surface_pressure]         # control variables
output variables: [uReconstructZonal, uReconstructMeridional, temperature, spechum, surface_pressure] # analysis variables

A YAML example for this linear variable change can be found in CTest mpas-jedi/test/testinput/linvarcha.yaml or mpas-jedi/test/covariance/yamls/3dvar.yaml

Analysis to Model Variable Change

After getting the analysis increment for [uReconstructZonal, uReconstructMeridional, temperature, spechum, surface_pressure] from the minimization of cost function, the full field analysis state is calculated in += (add increment) method of State class. In this method, the variables [qv, theta, rho, u], which are closely related to MPAS prognostic variables, are also updated. The full-field pressure pressure is obtained by integrating the hydrostatic equation from the surface. The dry potential temperature theta is then calculated from temperature and pressure, and dry air density rho is derived using the equation of state. The edge normal wind u is incrementally updated by using subroutine uv_cell_to_edges, originally from MPAS DART.

When performing a forecast from an MPAS-JEDI analysis state, the variables [qv, theta, rho, u] are converted to MPAS-A prognostic variables during the forecast initialization by setting config_do_DAcycling to true in namelist.atmosphere, as described in nml_file and streams_file. One example of the conversions at this stage is the coupling of water vapor, velocities and moist potential temperature with density.

Variable Change from Reading MPAS file to Analysis Variable

As described in “state variables”, State::read includes several variable changes. First, pressure_base and pressure_p (pressure perturbation) are read separately, and then added to the full fields pressure. The water vapor mixing ratio qv is read from the file, and converted to the specific humidity spechum. The dry potential temperature theta is read from the file, and converted to temperature. Because the reconstructed winds at the cell center, uReconstructZonal and uReconstructMeridional, are usually available in the MPAS file, they are directly read from the file.

Nonlinear and Linear Variable Change from Analysis to GeoVaLs Variable

The model interface interacts with UFO through Geophysical Variables at Locations, or GeoVaLs. Translating from model variables on the MPAS mesh to GeoVaLs is a two-step process divided into a variable change between the MPAS variable and the UFO variable, and an interpolation from the MPAS mesh to the observation locations. The Model2GeoVars and LinearModel2GeoVars classes are used to translate from the background state variables and the increment variables in MPAS-JEDI to the UFO variables that occupy GeoVaLs. The variable transforms are conducted across the entire model mesh one time for all observation operators. Oftentimes a UFO variable identically matches an MPAS field variable, in which case the identity transform is applied in Model2GeoVars and/or LinearModel2GeoVars.

There is a list of GeoVars that are available in MPAS-JEDI in code:mpas-jedi/test/testinput/namelists/geovars.yaml. As an example, consider the entry for air_pressure, which is the name of a UFO GeoVars

- field name: air_pressure
  mpas template field: theta
  mpas identity field: pressure

Each such entry within the fields vector must include the fields[i].field_name and the fields[i].mpas template field. The fields[i].mpas template field entry must be the name of an MPAS-Model field that is present in the Geometry’s domain_type member object. It is a field with the same shape as the UFO GeoVars in the sense that it has the same number of vertical levels. Each fields member may also include the fields[i].mpas template field entry, which indicates that the specified MPAS-Model field and the UFO GeoVars are identical, even down to having the same units. For such fields, of which there are many, the Model2GeoVars and/or LinearModel2GeoVars classes utilize an identity transform. For each entry in geovars.yaml that does not have an mpas identity field, those classes must explicitly describe the transformation between MPAS-Model fields that are available in the State and the corresponding UFO GeoVars

GetValues and LinearGetValues

After the model variable fields are converted to GeoVars, the GetValues and LinearGetValues classes are responsible for interpolating the GeoVars from the MPAS unstructured Voronoi mesh to the observation locations requested by the UFO observation operators. Currently OOPS instructs GetValues and LinearGetValues to carry out interpolation for each observation operator independently, in a serial loop. The interpolation weights are calculated only once for each observation operator in an HofX application and only once per outer iteration in a Variational application.

There are two algorithms that can be used for the horizontal interpolation, one from the BUMP library and another from the OOPS repository. The latter is referred to as “unstructured interpolation”, even though both algorithms are technically generalized for unstructured meshes. In MPAS-JEDI, the user can choose the interpolation algorithm via the observations[:].get values.interpolation type configuration element under each observation type in the YAML. For example,

observations:
- get values:
    interpolation type: unstructured
  obs space:
    name: Radiosonde
    obsdatain:
      obsfile: sondes_obs_2018041500_m.nc4
    obsdataout:
      obsfile: obsout_sondes.nc4
    simulated variables: [air_temperature, eastward_wind, northward_wind, specific_humidity]
  obs operator:
    name: VertInterp
  obs error:
    covariance model: diagonal

The valid values of get values.interpolation type in MPAS-JEDI are unstructured and bump, where bump is the default value. Both interpolation algorithms have pros and cons, and are subject to further improvement.

Somewhat different interpolation methods are used depending upon the data-type of the field. Barycentric weights (unstructured) or mesh triangulation (bump) are used for scalar real fields, while a form of nearest neighbor interpolation is used for integer fields. Integer fields are primarily associated with surface quantities, such as vegetation and soil types.

MPAS-JEDI has two ctests that exercise the OOPS interfaces of the GetValues and LinearGetValues classes. The two tests are configured with the getvalues.yaml and lineargetvalues.yaml YAML files described here.