Contents
Tables
AER-maintained Models
Configuration File Fields
Output netCDF Dimensions
Output netCDF Variables
Assuming the user has installed all dependencies, is using the gfortran compiler, and edited the configuration file to their liking (or is OK with using the default settings), the steps for a complete, out-of-the-box run of this library are:
git clone --recursive [email protected]:ReFRACtor/ABSCO.git; cd ABSCOconda env create -n absco -f environment.ymlconda activate absco./build_models.py -c gfortran -i ABSCO_config.inirun_LBLRTM_ABSCO.py -e2e
Software that can generate a set of ABSCO tables that span the thermal IR to UV spectral range.
Assuming the user has set the user.name and user.email Git configuration variables and has setup their Github account SSH keys, the repository can be cloned by using:
git clone --recursive [email protected]:ReFRACtor/ABSCO.git
Note the --recursive keyword -- it will force the clone to copy the necessary subroutines. Otherwise, the user will have to do (assuming the repository is just checked out with the default ABSCO name in the working directory):
cd ABSCO
git submodule update --init --recursive
This library depends on a number of standard Python libraries (os, sys, configparser, subprocess, glob, argparse), some widely distributed third-party libraries (see "Python Packages" subsection), and some ad hoc subroutines that are available as a GitHub repository in the AER-RC account (utils, RC_utils, lblTools). The latter modules are located in the common subdirectory, which is its own repository but also a submodule that is cloned along with the ABSCO repository if submodules are updated (e.g., --recursive clone). Additionally, some AER-maintained models and line parameters are required to run the ABSCO software.
The following libraries were installed with miniconda for Python 3.6.3:
- numpy (v1.13.3)
- scipy (v1.0.0)
- pandas (v0.23.0)
- netCDF4 (v1.3.1)
- xarray (v0.10.8)
More recent versions of the software gather the line file from a Zenodo archive with zenodo_get (v1.3.0). This library can be installed with pip, so it has been added to requirements.txt with the other packages that were miniconda installs.
All are used in this ABSCO library. The software is optimized for Python 3 usage -- Python 2 has not been tested and is not recommended.
LNFL (LiNe FiLe) FORTRAN code that converts ASCII text line parameter files to the binary files that LBLRTM expects (TAPE3) is located in its own repository. Because LNFL has been declared a submodule of the ABSCO library, using the --recursive keyword in the clone of this ABSCO repository will also clone the LNFL source code the is necessary. The source code is fetched and saved under the LNFL subdirectory.
LBLRTM (Line-By-Line Radiative Transfer Model) FORTRAN code also has its own Git repository and is a declared ABSCO submodule. It is stored under the LBLRTM subdirectory. LBLRTM in the context of this software simply calculates optical depth at specified pressures, temperatures, and spectral ranges.
The AER line parameter database (LPD) is distributed as a set of ASCII text files in the AER_Line_File directory. Initially, Verison 3.6 of the LPD was available on the AER external Git server and linked as a submodule of this ABSCO repository (so a --recursive clone would take care of this dependency as well). As of LPD v3.7, the dataset should no longer be retrieved from Git. Rather, it is archived as a dataset in Zenodo -- archiving this way makes the ABSCO repository completely open and allows for easier implementation of updated line file releases. The Zenodo request is handled internally in build_models.py, which will be described shortly.
Periodically, the models and LPD will be updated to reflect new line parameters, a new continuum, or bug fixes. These revisions can have significant effects on the model output. For reference, the model and LPD version numbers associated with the initial release of the ABSCO software are listed in Table 1.
AER-maintained models that are linked as ABSCO submodules
| Model | Version (GitHub Tag/Release) |
|---|---|
| LNFL | v3.2 |
| LBLRTM | v12.9 |
| LPD | v3.7 |
Currently, there are no plans on updating these three repositories.
LNFL and LBLRTM can be built with the build_models.py script:
% ./build_models.py -c gfortran -i ABSCO_config.ini
This script call specifies a gfortran compiler (-c) and replaces the paths to the executables in ABSCO_config.ini with the paths of the newly-built executables. Other valid compilers are ifort and pgf90. Use the -h option with the script for more options. Path replacement also occurs with the line file-associated paths (tape1_path, tape2_path, extra_params, xs_path, and fscdxs). If -i is not set, no path replacement occurs even though the executable are compiled. The two executables follow the naming convention model_version_os_compiler_precision, e.g., lblrtm_v12.9_linux_intel_dbl. It is recommended that build_models.py with the -i option be run after the initial clone of this repository to minimize any path issues.
For the line file, build_models.py utilizes the zenodo_get Python library to first download a list of Zenodo archive files, check if the archive needs to be downloaded, then extract the .tar.gz in which the dataset is packaged. If the user already has an AER_Line_File directory in the working directory, it is not overwritten (so for updates, one should consider either moving AER_Line_File or removing it recursively).
With the exception of the --run_lnfl, --run_lbl, and --end_to_end (alternatively -lnfl, -lbl, or -e2e) keywords that dictate which executable will be run, ABSCO_config.ini contains all of the inputs expected from the user. The parameters in Table 2 are expected in the file:
ABSCO configuration file (ABSCO_config.ini by default) items
| Field | Parent Directory | Notes |
|---|---|---|
| pfile | PT_Grid | text file with 1 pressure level in millibars per line. these will be the pressures on which the ABSCOs are calculated. this needs to be a relative path with respect to the directory in which this repo is cloned |
| ptfile | PT_Grid | for every pressure level, there are different allowed temperatures. this file contains a set of pressures and their permitted temperatures. this needs to be a relative path with respect to the directory in which this repo is cloned |
| vmrfile | PT_Grid | CSV files generated with VMR/standard_atm_profiles.py that provide interpolated/extrapolated volume mixing ratios (VMRs) for entire user-specified profile. this needs to be a relative path with respect to the directory in which this repo is cloned |
| xs_lines | Repository top-level directory | this file contains the species names for when XS and line parameters exist and line parameter usage is recommended by HITRAN. this should be a full path and can be assigned in build_models.py |
| wn1, wn2 | N/A | starting and ending spectral points for every desired band. can be in wavenumbers, microns, or nanometers. 200 cm-1 should be the minimum extent of the window -- broader windows increase the accuracy of the calculations |
| lblres, outres | N/A | spectral resolution at which LBLRTM is run and spectral resolution of the output (after spectral degradation). for now, this should be in wavenumbers |
| units | N/A | spectral units ("cm-1", "um", and "nm" are allowed) |
| wv_vmr | N/A | water vapor VMR (in ppmv) that will be used for all levels in a given profile for H2O, CO2, O2, and N2, since their continua are dependent on water vapor amounts |
| molnames | N/A | HITRAN molecule names, space delimited, case-insensitive. really should only be one molecule per run |
| scale | N/A | multiplicative factors used for continuum or extinction scaling (separate factors for H2O self continuum, H2O foreign continuum, CO2 continuum, O3 continuum, O2 continuum, N2 continuum, and Rayleigh extinction) |
| tape5_dir | LNFL_Runs and LBL_Runs | Directory underneath both lnfl_run_dir and lbl_run_dir to which TAPE5 files will be written (should just be a single string) |
| lnfl_run_dir | intdir |
Path to directory where LNFL runs will occur. Additional subdirectories (one for each molecule) will be created underneath this directory. this needs to be a relative path with respect to the directory in which this repo is cloned |
| tape1_path | AER_Line_File/line_file | Full path to the full TAPE1 ASCII line file that should be used in LNFL runs. assignment can be automated with build_models.py (aer_v_3.6) |
| tape2_path | AER_Line_File/line_file | Full path to the full TAPE2 ASCII line coupling file that should be used in LNFL runs with O2, CO2, and CH4. assignment can be automated with build_models.py |
| lnfl_path | LNFL | Full path to LNFL executable to be run. assignment can be automated with build_models.py |
| extra_params | AER_Line_File | directory with CO2-CO2, CO2-H2O, O2-H2O, O2-O2, and H2O-CO2 broadening parameter specifications. assignment can be automated with build_models.py |
| tape3_dir | intdir |
directory relative to the intdir to which LNFL output (binary line files) will be written |
| lbl_path | LBLRTM | Full path to LBLRTM executable to be run. assignment can be automated with build_models.py |
| xs_path | AER_Line_File | Full path to LBLRTM cross section file directories. assignment can be automated with build_models.py |
| fscdxs | AER_Line_File | Full path to cross section "lookup" file used with xs_path. assignment can be automated with build_models.py |
| lbl_run_dir | intdir |
Path to directory where LBL runs will occur. Additional subdirectories (one for each molecule) will be created underneath this directory. this needs to be a relative path with respect to the directory in which this repo is cloned |
| intdir | N/A | Directory where intermediate and output directories (lnfl_run_dir, lbl_run_dir, tape3_dir, and outdir) will be written. |
| outdir | intdir |
directory relative to intdir where the netCDFs will be written. |
| sw_ver | N/A | used in the output netCDF file to specify the software version number |
| out_file_desc | N/A | used in the output netCDF file. allows use to document run more specifically |
| nc_compress | N/A | compression level for netCDF output |
ABSCO_config.ini can be named something else, but that will have to be specified at the command line (otherwise it's assumed that ABSCO_config.ini is the configuration file to use):
run_LBLRTM_ABSCO.py -i your_ini_file
Some molecules have both line parameters and XS parameters. HITRAN makes recommendations on the preferred parameters given the species and the band, and these are taken into account in the error checking that the ABSCO_preprocess.py module does. Molecules where line parameters are recommended, the associated bands, and the flag (0: only XS exist, 1: both exist, use line params, 2: both exist, use XS) are recorded in FSCDXS_line_params.csv, which was generated with a separate script not in version control. For now, if there is any overlap of the user-specified region and a HITRAN-recommended XS region, the XS parameters are used.
Two modules exist in this library -- ABSCO_preprocess.py and ABSCO_compute.py. All that needs to be done to calculate cross sections for a given layer is an optical depth (OD) calculation, then the OD is divided by a layer amount (molecule number density), which amounts to an LNFL and LBLRTM run, both of which are done in ABSCO_compute.py. However, a number of things need to be determined before we get to the line file generation and subsequent OD computation:
- Configuration file read
- Verify existence of necessary paths
- Channels check
- are channel inputs consistent (limits, input and output resolution have the same number of elements)?
- are units defined and valid (wavenumber, micron, or nanometer)?
- is the expected ratio of the output resolution to the LBLRTM resolution an integer exponent of 2?
- are band limits in the correct order?
- conversion to wavenumber fo use in LBLRTM
- is the band no larger than 2000 cm-1? Break up into separate bins if it is
- Verify that input molecules are allowed
- Ensure that only two H2O values are provided
- Confirm that each of the remaining configuration attributes has only one value assigned to it
- Verify all necessary attributes have been found in the configuration file
- Check that the input pressures are monotonic, then force them to be in descending order (surface to TOA)
- Ensure that the same number of pressures are in the arrays from the input VMR (
vmrfile) and P (pfile) files (but here is no check that the pressures are equal) - Find the molecules that are radiatively active in the given spectral regions and ask user if they want to include omitted or extra species (this provision allows the
molnamesfield in the configuration file to be empty, but if it is, all active molecules will be processed) - Determine when to use cross section parameters as opposed to the line parameter database (as specified by
FSCDXS_line_params.csv) - Calculate the kernel and weights for spectral degradation
- Ascertain the source (AER LPD v3.6 or HITRAN 2012) for each molecule and each band
- Compute the amount of memory needed for the calculation and output file
These items are part of a configure object that is required input for the makeABSCO class.
Note that the objects contain the information for all user specified species. In the driver, we loop over each molecule so that a single netCDF output file is made for each molecule. This allows for parallelization of the processing (e.g., using a single core per molecule), but the code to facilitate it has not yet been written (partially because of concerns of the amount of HD space and RAM is needed for a single molecule).
Additionally, this driver handles H2O-affected molecules (CO2, N2, O2, H2O) and O2 by looping over VMRs (in ppmv) for both species, and then generating ABSCO objects at these VMRs. The water vapor VMRs are provided in the configuration file, while the oxygen VMRs are hard set at 1.9 * 105 and 2.3 * 105. For this processing, the H2 and O2 profiles are held constant at a given VMR while others (N2 and CO2) follow the profiles given by vmrfile (see Table 2).
The user must provide a spectral range in the configuration file. If multiple bands are provided, each band must have assigned to it an associated starting value, ending value, LBLRTM resolution, and output (netCDF) resolution.
Species specification is optional, but nothing is done if not molecule is provided. If molnames is not assigned anything, the code will check to see what molecules are radiatively active in the given band, then provide them to the user in a standard output prompt.
In the VMR subdirectory, standard_atm_profiles.py should be run if the user ever wants to use a different profile (rather than the default 1976 United States standard atmopshere provided in the repository). This module utilizes a standard atmosphere (the different kinds of standard atmospheres computed by LBLRTM are documented in the constructor of the vmrProfiles class) and the pressures expected by the user and performs an interpolation of the associated volume mixing ratios onto the user-specified grid. The interpolated profile is then used as a user-input atmosphere in the TAPE5 files that are generated for LBLRTM. Whatever the user chooses to be the output file name of standard_atm_profiles.py should be entered into the vmrfile field in ABSCO_config.ini (see Table 2).
As documented in Table 2, pressures are extracted from pfile, and each pressure has an associated range of acceptable temperature values listed in ptfile. Both of these files are underneath the PT_Grid. By default, the pressures that are used are based on the AIRS instrument (PT_Grid/AIRS_P_air.txt). build_temp_array.txt is the associated ptfile.
While LBLRTM handles a number of molecules via line parameters and cross sections, the allowed molecules for ABSCO processing is more limited. From ABSCO_preprocess.py, where we verify that the user-provided molecule can be processed:
allowed = ['H2O', 'CO2', 'O3', 'N2O', 'CO', 'CH4', 'O2', \
'NO', 'SO2', 'NO2', 'NH3', 'HNO3', 'OCS', 'H2CO', 'N2', \
'HCN', 'C2H2', 'HCOOH', 'C2H4', 'CH3OH', 'CCL4', 'CF4', \
'F11', 'F12', 'F22', 'ISOP', 'PAN', 'HDO', 'BRO', 'O2-O2']
The molecule names match those in the HITRAN database and are case-insensitive.
Line files need to be generated for every molecule and spectral range. Depending on the range and the number of lines for a given species in the range, this can be a time consuming process. However, the TAPE3 files likely only need to be generated once and can be saved to disk, which can be done by setting the -lnfl keyword:
run_LBLRTM_ABSCO.py -lnfl
In the call, we assume ABSCO_config.ini to be the configuration file, which contains the molecule name, spectral range, and output TAPE3 subdirectory, all of which are incorporated into the file naming convention of the TAPE3 files: working_dir/TAPE3_dir/molecule/TAPE3_molname_wn1-wn2. In the examples that follow, We will assume ABSCO_config.ini is populated with its default values (i.e., is unchanged from its original checkout).
LNFL runs are performed inside an LNFL_Runs directory (also defined in ABSCO_config.ini). Links to the LNFL executable and necessary input files (line coupling TAPE2, broadening parameters, full ASCII TAPE1 line file), and TAPE5 files that direct LNFL on what to do are also automatically generated and saved in the LNFL_Runs/TAPE5_dir subdirectory by default.
For the absorption coefficient calculation, LBLRTM must be run to compute a) optical depths, and b) layer amounts (done with the LBLATM subroutine). Once TAPE3 files are generated for the specified molecule and bands, LBLRTM TAPE5s (LBLRTM input file with state specifications and radiative transfer parameters) are written for every specified pressure level, temperature, and band combination. In each iteration, the optical depth spectrum is converted to absorption coefficients (k values) by dividing them by the layer amount for the given molecule. This k spectrum is then degraded to lessen the amount of RAM and hard drive space needed for the output.
The LBLRTM runs are followed by array manipulation (i.e., tranposing arrays to the dimensions that were agreed upon, adding fill values, etc.) and writing the necessary arrays to an output netCDF for the given species. LBLRTM is run inside the directory specified by lbl_run_dir in Table 2.
To run this LBLRTM process, use the driver script with the -lbl keyword.
run_LBLRTM_ABSCO.py -lbl
The process is repeated over both water vapor VMRs for species whose continua are affected by water vapor (CO2, O2, and N2). Separate objects for each VMR are instantiated, then their ABSCO arrays are combined.
Initially, it may be best to just run LNFL and LBLRTM in series, i.e., the end-to-end run. This can be done in the driver with the e2e keyord:
run_LBLRTM_ABSCO.py -e2e
Separating the LNFL and LBL runs may be useful after the user has generated all of the TAPE3 files that they need, but it will not be detrimental to run the end-to-end mode everytime. The only bottlenecks are the LNFL runs and the loop over all LBL cases. The latter will happen whenever ABSCOs are computed, and the former will not take a noticeable amount of time because LNFL will not be run if the expected TAPE3 exists.
Output from this software is relatively simple but potentially very large (when the software is run, a prompt will warn the user of the potential size of the file and memory footprint of the code). This is primarily because of the many dimensions of the cross section arrays ([nP x nT x nSpec]), with the nSpec dimension being the biggest contributor. The number of spectral points can be manipulated by the input bandwidth, resolution, and spectral degradation. The output are stored in a netCDF in the outdir directory listed in Table 2 (nc_ABSCO by default). A single netCDF is generated for each molecule.
Cross section array dimensions are also dependent on the molecule, because species whose continuum is dependent on water vapor amount contain an extra dimension for the water vapor VMR. Absorption coefficients are calculated at two different H2O VMRs at all applicable pressures and temperatures, so another dimension is necessary to store all of the calculations. All possible dimensions are given in Table 3.
| Dimension Name | Description |
|---|---|
| nfreq | Number of spectral points (wavenumbers, microns, nanometers) |
| nlev | Number of pressure levels |
| nlay | Number of pressure layers (nlev-1) |
| ntemp | Number of temperatures. This is always 15, regardless of the input pressures; and any invalid temperature (for a given pressure) will just be populated with fill values for k |
| nranges | Number of spectral ranges (AKA bands, channels) |
| nranges_lims | Number of limits for spectral ranges. This is always two because there is a single minimum and single maximum spectral point associated with each range |
| nvmr | Number of H2O VMR values. This will always be 2 |
Since their continua are dependent on water vapor content, netCDF variables for H2O, CO2, O2, and N2 will include all of the dimensions listed in Table 3. Table 4 contains the names, dimensions, units, valid ranges, and descriptions of variables for these four molecules.
Output Arrays for Water Vapor-affected Species
| Array Name | Dimensions | Units | Range | Description |
|---|---|---|---|---|
| P_level | (nlev) | mbar | [0, 1050] | User-provided layer boundary pressures |
| P_layer | (nlay x ntemp) | mbar | [0, 1050] | LBLRTM-calculated layer pressures |
| Cross_Section | (nfreq x ntemp x nlay x nvmr) | cm-2 | [0, 10-20] | Absorption coefficients k calculated from LBLRTM optical depths and layer amounts |
| Spectral_Grid | (nfreq) | cm-1 | [0, 50000] | Spectral points corresponding to ABSCOs in a single layer for a single temperature and in a given spectral range |
| Temperature | (nlev x ntemp) | K | [180, 320] | Applicable temperatures associated with each layer boundary pressure |
| Extent_Ranges | (nranges x nranges_lims) | cm-1 | [0, 50000] | Starting and ending spectral points for each input channel |
| Extent_Indices | (nranges x nranges_lims) | unitless | [0, "infinity"] | Pairs of indices defining the start and end indices of the Cross_Section frequency dimension for non-continuous calculation regions |
| H2O_VMR | (nvmr) | ppmv | [0, 50000] | Water vapor volume mixing ratio |
| O2_VMR | (nvmr) | ppmv | [0, 250000] | Oxygen volume mixing ratio. Only added to oxygen k array |
All other allowed molecules will conform to a similar convention as Table 4, only without the H2O VMR dimension. There is thus no H2O_VMR array for these species.
In the global attributes, there is a "source" field. There are only two sources -- HITRAN 2012 and AER LPD v3.6 -- and they can vary by band. The code accounts for the by-band differences in source. All fill values in the netCDF files are NaN (not-a-number).
We also designed a module to read the ABSCO_compute.py output. The following script reads in a file like the ones in the Output netCDF section and prints out the cross section value at a given coordinate (it also prints out the array coordinates -- i.e., zero-offset coordinates -- of the absorption coefficient):
% read_ABSCO_tables.py -h
usage: read_ABSCO_tables.py [-h] [-p IN_PRESSURE] [-T IN_TEMP]
[-s IN_SPECTRAL IN_SPECTRAL] [-wv IN_H2O]
[-tol TOLERANCE]
ncFile
Read in netCDF generated with run_LBLRTM_ABSCO.py module and print out absorption
coefficient (k) at a given pressure, temperature, spectral point, and water
vapor amount (if molecule continuum is affected by water vapor).
positional arguments:
ncFile Output netCDF generated by run_LBLRTM_ABSCO.py.
optional arguments:
-h, --help show this help message and exit
-p IN_PRESSURE, --in_pressure IN_PRESSURE
Reference pressure level [mbar] for which k is
retrieved. There are two pressure boundaries for a
given layer, and this is the lower bound (ie, closest
to surface). (default: 1050.0)
-T IN_TEMP, --in_temp IN_TEMP
Reference temperature [K] for which k is retrieved.
(default: 230.0)
-s IN_SPECTRAL IN_SPECTRAL, --in_spectral IN_SPECTRAL IN_SPECTRAL
Reference spectral point AND units [cm-1, um, or nm]
for which k is retrieved. (default: [500, 'cm-1'])
-wv IN_H2O, --in_h2o IN_H2O
Reference water vapor VMR (ppmv) for which k is
retrieved IF the specified molecule is H2O, CO2, O2,
or N2. (default: 10000.0)
-o2 IN_O2, --in_o2 IN_O2
Reference oxygen VMR (ppmv) for which k is retrieved
IF the specified molecule is O2. (default: 190000.0)
-tol TOLERANCE, --tolerance TOLERANCE
Tolerance used when searching for floating point
matches in *each* of the dimensions. This should be a
relative tolerance (e.g. 0.01 would mean P from netCDF
is within 1% of in_pressure). (default: None)