How to HITRAN line-by-line

This is a basic introductory guide on how to write a line-by-line calculation routine using the HITRAN database which is widely used for the calculation of atmospheric and molecular absorption spectra of the most abundant atmospheric gases (see list here) for given input parameters of the desired temperature, total pressure and molecular concentrations.   This guide is a practical supplement to the information provided by the HITRAN group at this link which contains a list of the HITRAN database parameters, their detailed description and the most relevant calculation formulas.

The line-by-line calculation procedure involves computation of absorption for individual spectral line data records provided in the HITRAN database over a desired spectral range and summing these calculated single spectral line absorption profiles together to form the total absorption spectrum as a function of wavenumber (or wavelength).

What is supplied by HITRAN?

The HITRAN database supplies all data needed to perform line-by-line calculations except for the atmospheric model profiles which are described in more detail further.   The supplied data include the
(1)   HITRAN database data file (traditionally an ASCII file with the PAR extension) which may now be obtained from HITRANonline,
(2)   the molparam.txt file which contains reference data such as the molecular mass etc. for each isotope in the HITRAN compilation, and
(3)   the files with the Total Internal Partition Sums (TIPS) for different temperatures in the "q_Global_Isotopologue_Id.txt" format available here.

Alternatively the Total Internal Partition Sums from (3) above may also be calculated using the FORTRAN and Python routines provided at the HITRAN website here.

HITRAN database records (legacy 160-char ASCII format)

The 2004 - 2012 year ASCII edition of the HITRAN database consists of 160 character records for each single spectral line.   Below is the listing of the fields for each HITRAN database spectral line record with their corresponding ASCII character field lengths.   This table is based on that provided at HITRAN.org as well as Table 2 in the HITRAN 2004 edition paper.   Not all data supplied in each HITRAN database record is required to run line-by-line calculations.   The parameters required to run the line-by-line calculation using the Voigt profile are indicated with a tick mark.

- parameter used in calculations, - parameter not used in calculations.

      Symbol Length Units Description
  $M$ 2 - Molecule number
  $I$ 1 - Isotope number
  $\nu_{ij}$ 12 $cm^{-1}$ Vacuum wavenumber
  $S_{ij}$ 10 $cm^{-1} / (molecule \cdot cm^{-2})$ Line intensity
  $A$ 10 $s^{-1}$ Einstein A-coefficient
  $\gamma_{air}$ 5 $cm^{-1} / atm$ Air-broadened half width at half maximum (HWHM)
  $\gamma_{self}$ 5 $cm^{-1} / atm$ Self-broadened half width at half maximum (HWHM)
  $E^{ \prime\prime}$ 10 $cm^{-1}$ Lower-state energy
  $n_{air}$ 4 $cm^{-1}$ Temperature dependence exponent for $\gamma_{air}$
  $\delta_{air}$ 8 $cm^{-1} / atm$ Air pressure-induced line shift
  $V^{ \prime}$ 15 - Upper-state "global" quanta
  $V^{ \prime\prime}$ 15 - Lower-state "global" quanta
  $Q^{ \prime}$ 15 - Upper-state "local" quanta
  $Q^{ \prime\prime}$ 15 - Lower-state "local" quanta
  $I_{err}$ 6 - Uncertainty indices
  $I_{ref}$ 12 - Reference indices
  $*$ 1 - Line mixing flag
  $g^{\prime}$ 7 - Statistical weight of the upper state
  $g^{\prime\prime}$ 7 - Statistical weight of the lower state
HITRANonline database records

With the transition to the online version of the HITRAN database (HITRANonline, at HITRAN.org) additional spectral paramters (see listing here) are being added to the HITRAN database records to enbale modeling using lineshape profiles more advanced and more accurate that Voigt.   HITRANonline still permits requesting data in the old 160-char HITRAN database format but contains many more parameters allowing usage of advanced spectral lineshapes such as the Hartmann-Tran and the Speed-dependent Voigt.   A listing of the HITRANonline database record parameters is available at the following link.

The molparam.txt file

The molparam.txt file contains basic data for each molecule and isotope needed to perform the line-by-line calculations.   These include the molecule number $M$, the Isotope code, terrestrial abundance $B_{T}$, total internal partition sum value at 296K denoted as $Q(296K)$, ground state degeneracy factor $gj$, and the molar mass $W_{g}$ in grams per mole.   The molecule numbers $M$ are indicated in brackets in the molparam.txt after each molecule name followed by the isotopes arranged in ascending order by their isotope number $I$.   Please note that the Isotope code has nothing to do with the isotope number $I$ field in the HITRAN database records.   The second field called $I$ used for isotope numbering in the HITRAN database is simply an index of each isotope starting from 1 in the increasing order of their relative abundance in the Earth atmosphere for a given molecule $M$.   The only exceptions are isotopes 10, 11 and 12 of CO2 which have indexes of 0 (number zero), A, and B respectively.

Global Isotopologue ID

With the switch to HITRANOnline a new field called Global isotopologue ID was instroduced into the HITRAN database as can be seen in the first column of this list.  The "Global isotopologue IDs" are unique identifiers for each isotope (unique across all molecules and isotopes in the HITRAN database) and as such are different from the Isotope number $I$ and the Isotope code.  The "Global isotopologue IDs" are not currently present in the molparam.txt file and are not available in the standard legacy 160-ASCII *.par HITRAN database format but may be retrieved for each spectral line record from the HITRANOnline database which uses the new extended database format.

The TIPS files

The Total Internal Partition Sums (TIPS) files are available here.   Each TIPS file contains 2 columns.   The first column contains the temperature in Kelvin, and the second one the corresponding Total Internal Partition Sum (TIPS) for that temperature $Q_{T}$.   The value of the total internal partition sum at $296K$ denoted as $Q_{296K}$ may be taken from the TIPS files or the molparam.txt file.   The TIPS values are used to recalculate the spectral line intensity from the reference temperature of 296K into the user-specified calculation temperature.

The "pre-lineshape" calculations for a single spectral line

Several parameters need to be calculated for a single spectral line prior to running the lineshape calculation for it.   First of all, the center of the spectral line as indicated in the HITRAN database will change due to pressure.   As such the pressure induced spectral line shift needs to be taken into account.   The shifted spectral line center $ \nu_{ij}^{*}$ due to pressure compared to the unperturbed value specified in the HITRAN database is

$$ \nu_{ij}^{*} = \nu_{ij} + \delta_{air} \cdot P_{total} $$

where $P_{total}$ is the total atmospheric pressure in atmospheres (atm).   The intensity of the spectral line also needs to be recalculated for the user-specified temperature

$$ S_{ij}^{*} = S_{ij} \cdot \left( \frac{Q_{296K}}{Q_{T}} \right) \cdot \frac{exp\left(-c_{2} \cdot E^{\prime\prime} / \, T\right)}{exp\left(-c_{2} \cdot E^{\prime\prime} / \, 296\right)} \cdot \frac{1 - exp\left(-c_{2} \cdot \nu_{ij} \, / \, T\right)}{1 - exp\left(-c_{2} \cdot \nu_{ij} \, / \, 296\right)} $$

where $T$ is the calculation temperature in Kelvin, $Q_{T}$ is the total internal partition sum value for the calculation temperature, $Q_{296K}$ is the total internal partition sum for the temperature of 296K, and $c_{2} = c \cdot h / k $ = 1.43880285 (cm · K) is the second black body radiation constant.   After the above calculation, we need to factor in the pathlength, molecular concentration and the molecule isotopic abundance into the calculated line intensity.

$$ S_{L} = S_{ij}^{*} \cdot \frac{B}{B_{T}} \cdot \frac{296}{T} \cdot N_{L} \cdot P_{mol} \cdot L $$

where $B$ is the isotope abundance in the atmosphere being modelled, $B_{T}$ is the isotope abundance in the terrestrial atmosphere, $T$ is the calculation temperature in degree Kelvin, $N_{L}$= 2.47937196e+19 is the Loschmidt's number (in $cm^{-3}$) at a pressure of 1 atm and recalculated for a temperature of 296K,   $P_{mol}$ is the partial pressure in atmospheres (atm) for the molecule with index $M$ being calculated, and $L$ is the pathlength in centimeters.   For a more detailed description of this and several other derivation steps, please refer to Appendix A1.1 (pages 124 - 132) in this manual.

Since the line intensities provided in the HITRAN database factor in the natural isotope abundances in the terrestrial atmosphere the $B / B_{T}$ term will always be equal to $1$ when modeling terrestrial atmosphere because $B = B_{T}$ for all isotopes in that case.   If however the calculations are performed for non-terrestrial atmospheres (i.e. Mars, Titan etc.) or absorption cell experiments, the isotope abundaces will then likely be different from those encountered in the terrestrial atmosphere ($B$ will no longer be equal to $B_{T}$).   This is also explained in the last paragraph of Appendix A1.1 on page 125 of this manual.

At the next step the Doppler and Lorentzian lineshape half widths at half maximum (HWHM) in 1/cm need to be found.   The Doppler HWHM is calculated as follows:

$$ \alpha_{doppler} = \frac{\nu_{ij}}{c} \cdot \sqrt{ \frac{2 \cdot N_{A} \cdot k \cdot T \cdot ln(2)}{10^{-3} \cdot W_{g}} } $$

where $T$ is the calculation temperature, $N_{A}$ = 6.02214086e+23 is the Avogadro number (in 1/mol) , $k$ = 1.3806503e-23 is the Boltzman's constant (in $m^{2} \cdot kg \cdot s^{-2} \cdot K^{-1}$),   $c$ is the speed of light (in m/s), $W_{g}$ is the molar mass of the molecule (in g/mol, located in the molparam.txt) for a given molecule $M$.

The Lorentzian HWHM (in 1/cm) is given by the following formula:

$$ \gamma = \left(\dfrac{296}{T}\right)^{n_{air}} \cdot \Big( \gamma_{air} \cdot \left( P_{total} - P_{mol} \right) + \gamma_{self} \cdot P_{mol} \Big) $$

where $\gamma$ is the half width at half maximum for a given line, $P_{total}$ is the total atmospheric pressure in atmospheres (atm), $P_{mol}$ is the partial pressure in atmospheres (atm) corresponding to the molecule being calculated.

Lineshape calculation for a single spectral line

The data in the HITRAN database is supplied in wavenumbers, as such the calculation is typically performed in wavenumbers (cm-1).   Usually one of the first steps in the line-by-line calculation would involve forming a wavenumber grid over the entire calculation range to run the calculation.   The easiest solution is to use an equidistant grid with a fixed wavenumber step.   The calculation for each individual spectral line HITRAN database record is then carried out on a subset of points within the wavenumber grid.   Some spectral lines with weak line intensities (or the pre-calculated peak absorption) may be left out in the line-by-line calculation to speed up the computation.

The wavelength calculation span for the individual spectral lines is specified either as an absolute value (typically in the units of 1/centimeter) or in terms of the number of saturated halfwidths (multiples of values of $\alpha_{doppler}$ or $\gamma$ described above) and should be selected wide enough to provide sufficient accuracy while also resulting in acceptable calculation speeds.   For example, the default values used for in the HAPI source code is the larger of 50 saturated halfwidths and the manually specified absolute value in 1/cm which is not used by default (according to lines 4632 - 4644, 18262 and 18213 in the HAPI version source code), and the HITRAN-PC program default value is the lesser of 12 saturated half widths and 25 1/cm.

The section below lists equations for the calculation of the Dopper, Lorentzian or Voigt profile spectral lines.   These are the basic general purpose profiles.

(1)   If the Doppler lineshape profile is desired, it is calculated by varying $\nu$ over the wavenumber span of a single spectral line as follows

$$ f_{doppler}(\nu) = \sqrt{ \frac{ln(2)}{\pi \cdot \alpha_{doppler}^{2}}} \cdot exp \left( - \frac{ \left( \nu - \nu_{ij}^{*} \right)^{2} \cdot ln(2) }{ \alpha_{doppler}^{2} } \right) $$

(2)   When Lorentz lineshape profile is used, it is calculated over the wavenumber range of a single spectral line $ \nu $ using the following equation

$$ f_{lorentz}(\nu) = \frac{1}{\pi} \cdot \frac{\gamma}{\gamma^2 + \left( \nu - \nu_{ij}^{*} \right)^{2} } $$

(3)   For the calculation of the Voigt profile (see this presentation and this thesis) approximations are typically used.   The Voigt function $K(x,y)$ is the real part of the complex probability function $W(x,y)$

$$ W(x,y) = K(x,y) + iL(x,y) $$ $$ Voigt(x,y) = K(x,y) = \frac{y}{\pi} \cdot \int_{-\infty}^{\infty} \frac{exp(-t^2)}{y^2 + (x - t)^2} dt $$ $$ L(x,y) = \frac{1}{\pi} \cdot \int_{-\infty}^{\infty} \frac{(x - t) \cdot exp(-t^2)}{y^2 + (x - t)^2} dt $$


$$ x = \frac { \sqrt{ln(2)} \cdot (\nu - \nu_{ij}^{*}) }{ \alpha_{doppler} } ; \qquad y = \frac { \sqrt{ln(2)} \cdot \gamma }{ \alpha_{doppler} } $$

and the Voigt spectral line shape $f_{voigt}(\nu)$ is computed as follows (see this presentation, slides 17-22, please note that our $\alpha_{doppler}$ and $\gamma$ values are HWHM and those in the link are given in full width at half maximum (FWHM)):

$$ f_{voigt}(\nu) = \frac { \sqrt{ln(2)} } { \sqrt{\pi} \cdot \alpha_{doppler} } \cdot Voigt(x,y) $$

The spectral lineshape is now combined together with the calculated line intensity as follows to get the final attenuation value due to a single line in the units of optical depth $O_{d}(\nu)$ as a function of wavenumber which may be further converted into other units:

$$ O_{d}(\nu) = f_{voigt}(\nu) \cdot S_{L} $$

Please note that parameter $S_{L}$ in this formula has the units of 1/cm because it already includes the optical pathlength $L$ as shown earlier.   If we substitute $S_{L}$ into the last formula for $O_{d}(\nu)$ it will look as follows

$$ O_{d}(\nu) = f_{voigt}(\nu) \cdot S_{ij}^{*} \cdot \frac{B}{B_{T}} \cdot \frac{296}{T} \cdot N_{L} \cdot P_{mol} \cdot L $$

which is equivalent to the following equation (see this link, as well as Appendices 1.1 (page 124) and 9 (page 176) in this manual):

$$ O_{d}(\nu) = -ln(\tau_{\nu}) = k_{\nu} \cdot \frac{296}{T} \cdot N_{L} \cdot P_{mol} \cdot L$$

where $k_{\nu}$ is the absorption cross-section in $cm^{2}/molecule$, and $\tau_{\nu}$ is the transmittance (on a scale of 0 to 1) at a given wavenumber $\nu$.

The Voigt function $Voigt(x,y)$ may be calculated using a number of algorithms.   The current version of the bytran software uses the rational approximation algorithm described in Abrarov, S., and Quine, B. "A rational approximation for efficient computation of the Voigt function in quantitative spectroscopy". Journal of Mathematics Research, 7(2), p 163, (2015) available here which also contains the algorithm source code in the Matlab language.   This algorithm is accurate and very easy to implement by substituting values of $x$ and $y$ as described above.  The usage of the Voigt profile is sufficient for many applications, however more advanced lineshape models may be needed where higher accuracy is required.

After the calculation for all spectral lines within a selected spectral range is complete, the individual spectral line contributions are added together to obtain the total spectrum by taking into account individual spectral line center positions and spans along the wavenumber grid.

For a recent short survey of several other types of precise Voigt profile algorithms currently suggested for radiative transfer applications the Discussion Section 4 in this paper by Schreier may be useful.  The Introduction Section 1 in this same paper as well as another publication by Schreier also list several low accuracy approximations of the Voigt profile.   The low accuracy algorithms such as that by Kielkopf also implemented in bytran for survey purposes may be useful for fast calculations where high accuracy is not required.

Radiance calculation

The radiance calculation involves the application of a simple formula after the high resolution spectrum has been calculated which can be found in a procedure called "radianceSpectrum" of the HAPI source code or else in Appendix 4 (page 141) of this manual which presents a better explanation of the physics involved.

The Radiance and the Instrument function calculations in bytran are a direct conversion of the HAPI Python code into the Qt/C++ with some modifications to take into account the specifics of the Qt/C++ environment and program structure.   Currently the only notable improvement in bytran over HAPI's Radiance code is the "separate molecule" calculations which make it possible to view contributing molecular spectral bands in the Radiance spectrum.

Instrument functions

The instrument function routines used in bytran are a direct conversion of those introduced in the HAPI source code under the "SPECTRAL CONVOLUTION" section.   The explanations below outline the steps and formulas used in the algorithm.   Before the spectral convolution can be performed, a wavenumber array corresponding to the instrument function to be used needs to be generated over the wavenumber range $x$ spanning from $-x_{0}$ to $x_{0}$ with the spectral resolution equivalent to that of the calculated transmission spectrum to be smoothed with the instrument function.   Here $n$ is the number of points in the instrument function array, and $x_{0}$ is equal to the instrumental function wing parameter which has to be larger than the instrument resolution parameter $g$.   The instrument function shape ("y-axis") values $B(x)$ are then generated over the array grid of $x$ values (from $-x_{0}$ to $x_{0}$) using one of the formulas presented below depending on the desired instrument function.

Rectangular   (where $g$ is a slit width or the instrumental resolution):

$$ B(x) = \newcommand\ddfrac[2]{\frac{\displaystyle #1}{\displaystyle #2}} \begin{equation} \begin{cases} \ddfrac {1}{g}, \enspace when \enspace |x| \leqslant \ddfrac {g}{2} \\ \, 0, \enspace when \enspace |x| > \ddfrac {g}{2} \\ \end{cases} \end{equation} $$

Triangular   (where $g$ is the line width equal to the half base of the triangle):

$$ B(x) = \begin{equation} \begin{cases} \ddfrac {1}{g} \cdot \left( 1 - \ddfrac {|x|}{g} \right) , \enspace when \enspace |x| \leqslant g \\ \, 0, \enspace when \enspace |x| > g \\ \end{cases} \end{equation} $$

Gaussian   (where $g/2$ is a Gaussian half-width at half-maximum):

$$ B(x) = \frac {\sqrt{ \ddfrac {ln(2)}{\pi} }} { \ddfrac{g}{2} } \cdot exp \left(-ln(2) \cdot \left( \frac {x}{(g/2)} \right)^2 \right) $$

Dispersion (Lorentz)   (where $g/2$ is a Lorentzian half-width at half-maximum):

$$ B(x) = \frac{g}{2 \pi} \cdot \frac{1} {x^{2} + \left( \ddfrac {g}{2} \right) ^{2}} $$


$$ B(x) = \begin{equation} \begin{cases} \ddfrac{1}{g} \cdot \ddfrac { \left( sin \left( \ddfrac {\pi \cdot x}{g} \right) \right) ^2} { \left( \ddfrac {\pi \cdot x}{g} \right) ^2 }, \enspace when \enspace x \neq 0 \\ 1, \enspace when \enspace x = 0\\ \end{cases} \end{equation} $$

Michelson   (where $1/g$ is the maximum optical path difference):

$$ B(x) = \begin{equation} \begin{cases} \ddfrac{2}{g} \cdot \ddfrac { sin \left( \ddfrac {2\pi \cdot x}{g} \right)} { \left( \ddfrac{2\pi \cdot x}{g} \right) }, \enspace when \enspace x \neq 0 \\ 1, \enspace when \enspace x = 0\\ \end{cases} \end{equation} $$

The instrument function $B(x)$ is then normalized before the convolution procedure

$$ B_{n}(x) = \frac{B(x_{i})}{\sum\limits_{i=1}^n B(x_{i})} $$

The application of the instrument function smoothing to the generated high resolution spectrum is carried out by convolving the generated instrument function array $B_{n}(x)$ with the high resolution spectrum in a "scanning" manner along the wavenumber axis.   For an easy to understand introductory explanation of the convolution procedure these lecture course notes at Cornell University and University of Cape Town may be consulted.

Molecular concentrations and the Atmospheric models

The concentrations of molecules may be specified manually for line-by-line calculations.   However, the standard way to model the atmospheric transmission spectra is by using the molecular concentration profiles, the total pressures and temperatures from atmospheric models.   It is very common to use the US Standard 1986 Atmospheric model described in this document even though other model profiles are widely available.   There are 6 models available in the 1986 US Standard: Tropical, Midlatitude summer, Midlatitude winter, Subarctic summer, Subarctic winter, and the US Standard.   Even though these standard profiles do not include data for all molecules currently available in the HITRAN database, they provide values for the most abundant molecules which is sufficient for many applications. For more exotic calculation scenarios, the unavailable values need to be supplemented from other sources.

www.bytran.org -|- Page created in 2016, last edited November 23, 2020
   Email:  bytran@bytran.org