Skip to content

Validation and verification

This page summarizes how JURASSIC has been validated and how users can verify correct installation and numerical behavior for their own applications. Validation is a critical aspect of radiative transfer and retrieval modelling and underpins the scientific credibility of the results produced with JURASSIC.

Validation philosophy

JURASSIC validation follows three complementary approaches:

  1. Intercomparison with reference models
  2. Regression testing and example workflows
  3. Scientific application and peer-reviewed publications

Together, these approaches ensure that the model is both numerically correct and scientifically reliable across a wide range of use cases.

Intercomparison with reference models

The core radiative transfer algorithms implemented in JURASSIC have been extensively benchmarked against established, high-accuracy reference models, including:

  • the Karlsruhe Optimized and Precise Radiative Transfer Algorithm (KOPRA),
  • the Reference Forward Model (RFM),
  • the Stand-alone AIRS Radiative Transfer Algorithm (SARTA).

These intercomparisons cover:

  • limb and nadir viewing geometries,
  • temperature and trace gas sensitivities,
  • clear-sky and simplified aerosol conditions,
  • a broad range of atmospheric states.

Results demonstrate that JURASSIC reproduces reference-model radiances, Jacobians, and retrieval-relevant quantities within the accuracy expected from its spectral approximations.

Published validation and application studies are listed on the References page, including comparisons with KOPRA and retrieval studies using satellite observations.

Validation of spectral approximations

Particular attention has been given to validating the:

  • Emissivity Growth Approximation (EGA),
  • Curtis–Godson Approximation (CGA),
  • band-averaged emissivity lookup table approach.

Comparisons against line-by-line calculations show that these approximations provide a good balance between accuracy and performance for typical infrared remote sensing applications, especially at moderate spectral resolution.

Residual differences relative to line-by-line models are well characterized and documented in the literature.

Retrieval validation

The optimal estimation retrieval framework in JURASSIC has been validated through:

  • synthetic retrieval experiments,
  • comparison with independent retrieval systems,
  • application to real satellite measurements.

Key aspects of retrieval validation include:

  • convergence behavior,
  • consistency of Jacobians and averaging kernels,
  • realism of retrieved error estimates,
  • physical plausibility of retrieved atmospheric states.

Retrieval results obtained with JURASSIC have been published in numerous peer-reviewed studies.

Example projects and regression tests

The JURASSIC distribution includes example projects (e.g. limb, nadir, and zenith configurations) that serve as both tutorials and regression tests.

These examples:

  • generate forward-model output for known configurations,
  • compare results against reference data,
  • produce diagnostic plots for visual inspection.

Running these examples after installation is the recommended way to verify that JURASSIC is functioning correctly on a given system.

From the repository root, example projects are typically executed via:

cd projects/limb
./run.sh

and similarly for other configurations.

Automated test suite

In addition to example projects, the build system provides a test target. From the src/ directory, run:

make check

This target runs a predefined set of tests and reports pass/fail status. These tests are intended to catch regressions caused by code changes or build-system issues.

The default make check target currently runs the following suites from tests/:

  • cli_test - command-line behavior and help output for all executables
  • tools_test - small utility programs such as time conversion and Planck/brightness-temperature tools
  • atm_test - atmospheric-profile generation, interpolation, hydrostatic adjustment, and format conversion
  • mat_test - matrix input/output and format conversion
  • obs_test - observation geometry, ray tracing, and observation format conversion
  • tbl_test - filter functions, lookup tables, and table format conversion
  • formod_test - limb, nadir, and zenith forward-model calculations and kernels
  • ret_test - retrieval workflow checks, including shared netCDF input and output

Most regression tests generate fresh output files and compare them with checked-in reference files under the corresponding data.ref/ directory. This makes the test suite useful for detecting unintended changes in numerical output, file-format behavior, and command-line interfaces.

An additional mpi_test suite is available for MPI-enabled retrieval checks, but it is not part of the default make check target.

Numerical reproducibility

Due to floating-point arithmetic and parallel execution, JURASSIC does not guarantee bitwise-identical results across:

  • different compilers,
  • different MPI/OpenMP configurations,
  • different hardware architectures.

The regression tests use exact file comparisons for controlled test setups and supported reference outputs. For broader user workflows, small numerical differences may occur across compilers, hardware, and parallel configurations. Validation should therefore focus on physical and statistical consistency rather than bitwise identity alone.

User-level validation recommendations

Users are encouraged to perform application-specific validation by:

  • comparing selected results against reference models or datasets,
  • testing sensitivity to configuration parameters,
  • inspecting Jacobians, averaging kernels, and residuals,
  • documenting configuration choices and assumptions.

Such validation is especially important when introducing new lookup tables, instrument configurations, or retrieval setups.

Summary

JURASSIC has undergone extensive validation through model intercomparisons, regression testing, and scientific application. These efforts demonstrate that the model provides reliable and accurate results within the scope of its documented assumptions and approximations.

Users are encouraged to make use of the provided example projects and tests to verify correct installation and to perform additional application-specific validation as needed.