Sasktran-Disco¶

Sasktran-Disco is a discrete-ordinates radiative transfer model. The engine is based of the well known DISORT and LIDORT models and has been verified to agree with DISORT v2.0 to 1e-7.

Features

Scalar radiative transfer computations
Scalar weighting function computations
BRDF and Lambertian surfaces

Limitations

Nadir-like geometry’s only (observer must be looking at the ground)
Plane parallel model (atmosphere is discretized into homogeneous slabs)
No vector calculations, delta-M scaling, thermal emissions, two-stream approximations, single scatter solutions

Strengths

Fast
Low-memory usage
Easy to use
Simultaneous weighting functions (multiple species)

Usage¶

Simple Example¶

The following is a simple example of how to (1) setup a basic radiative transfer calculation using the Sasktran framework, and (2) default configure a Disco engine.

In this example we’ll setup a single line of sight. The observer will be TEMPO and our single line of sight will be looking at the ARG office.

import numpy as np
import sasktran as sk

""" Geometry configuration.
First we need to configure our sk.Geometry object. This object specifies the
line of sight vectors as well as the solar position (based on mean mjd).
"""
geometry = sk.NadirGeometry()

# make the look vector from TEMPO to the ARG office
tempo = sk.Geodetic()
tempo.from_lat_lon_alt(0, -100, 35786000)
geometry.from_lat_lon(mjd=57906.843, observer=tempo, lats=52.131638, lons=-106.633873, elevations=0)

""" Atmosphere configuration.
Next we need to configure our sk.Atmosphere object. This object specifies all
atmospheric and surface properties to the engine.
"""
atmosphere = sk.Atmosphere()

# add our species
atmosphere['rayleigh'] = sk.Species(sk.Rayleigh(), sk.MSIS90())
atmosphere['o3'] = sk.Species(sk.O3OSIRISRes(), sk.Labow())
atmosphere['no2'] = sk.Species(sk.NO2OSIRISRes(), sk.Pratmo())

# add our surface properties
atmosphere.brdf = sk.Kokhanovsky()


""" Radiance calculation.
We are now ready to perform the radiance calculation. Note that the engine can
be default constructed (with properties set after construction), or the
properties can be passed to the constructor.
"""
engine = sk.EngineDO(geometry=geometry, atmosphere=atmosphere, wavelengths=[350, 700])
rad = engine.calculate_radiance()

# rad now hold the calculated radiances, shape: (num_wav, num_los)

Weighting Functions¶

Sasktran-Disco support the same weighting function interface as Sasktran-HR (weighting functions are in the same form as well). Disco also supports multiple-species weighting functions. Below is an overview of the weighting function interface.

# Standard atmosphere configuration
atmosphere = sk.Atmosphere()
atmosphere['rayleigh'] = sk.Species(sk.Rayleigh(), sk.MSIS90())
atmosphere['o3'] = sk.Species(sk.O3OSIRISRes(), sk.Labow())
atmosphere['no2'] = sk.Species(sk.NO2OSIRISRes(), sk.Pratmo())

atmosphere.wf_species = 'o3'            # single species syntax

atmosphere.wf_species = ['no2', 'o3']   # multiple species syntax

# calculate radiance and weighting functions
rad, wf = engine.calculate_radiance()

Configuring Weighting Functions¶

Weighting functions are calculated by perturbing the interesting species at specific altitudes. The altitudes are set with the wfaltitudes engine option. Around each altitude in wfaltitudes the engine will linearly interpolate the perturbation for a given region. This region is speficied using the wfwidths engine option. The weighting function widths are the height above and below the weighting function altitude over which the perturbation will be linearly interpolate. Weighting function widths must be specified for each weighting function altitude. The user must take care when setting wfaltitudes and wfwidths so that the upper bound and lower bound of the perturbations are within the defined atmosphere (ie. if the widths are all 500m then the minimum, maximum weighting function altitudes are 500m and 99500m respectively assuming a TOA altitude of 100km).

Shape of `wf`¶

Length definitions for this section: lwav = len(engine.wavelengths) llos = len(geometry.lines_of_sight) lwfs = len(atmosphere.wf_species) lwfa = len(weighting function altitudes)

The shape of wf depends on the syntax you used to specify the weighting functions. If you used the single species syntax to specify atmosphere.wf_species then the shape of wf is (lwav, llos, lwfa). If you used the multiple species syntax then the shape is (lwav, llos, lwfs, lwfa).

Advanced Configuration¶

Sasktran-Disco has four configuration parameters which can be adjusted. The following are the default values of each parameter.

# Sasktran-Disco defaults
engine.num_streams = 16
engine.num_layers = 50
engine.num_phasef_quadrature_terms = 64
engine.num_brdf_quadrature_terms = 64

Below the significance of each parameter is explained and demonstrated.

Number of Streams¶

In the discrete-ordinates algorithm phase functions and the BRDF are expanded into series of Legendre polynomials. The value of NSTR is the number of polynomials used in these expansions.

engine.num_streams = 16         # default value
Symbol

NSTR

Complexity

\(O(n^3)\)

Valid range

NSTR = {4, 6, 8, …, 40}

Default

16

Logic

Increase with the complexity of the phase function (and potentially BRDF).

Note

This parameter will have the most drastic effect on runtime (and potentially neglibile effects on precision).

Phase Functions and the BRDF Expansions

The following is how these expansions are computed.

\[p_l(\vec{x}) = \dfrac{1}{2} \int_{-\pi}^{\pi} L_l(\theta) f(\vec{x}, \theta) \ \mathrm{d} \theta \qquad l = 0, \ldots, 2N -1\]

where

\(p_l\) The \(l\)’th polynomial in the Legendre expansion of the function being expanded.

\(\vec{x}\) Some position in the atmosphere.

\(L_l\) The \(l\)’th Legendre polynomial.

\(f\) The function being expanded (phase functions or surface BRDF).

Azimuth Separation

In the discrete-ordinates algorithm the azimuth component of the resultant intensity is expanded into a cosine Fourier series. During the computation, solutions to the radiative transfer equation for different azimuth components are lazily evaluated until the solution has converged. The value of NSTR is also the maximum number of azimuth components allowed before the computation is said to have failed.

Solutions to the radiative transfer equation are reconstructed as follows.

\[I(h, \mu, \phi) = \sum_{m = 0}^{C}I^m (\tau, \mu) \cos{m(\phi_0 - \phi)} \qquad C < \mathrm{NSTR}\]

where

\(I\) Intensity.

\(h\) Altitude.

\(\mu\) Line of sight zenith angle.

\(\phi\) Line of sight azimuth angle.

\(\phi_0\) Solar azimuth angle.

Number of Layers¶

In the discrete-ordinates method, the atmosphere is discretized into NLYR homogeneous slabs. These slabs are spaced in optical depth and atmospheric properties are averaged within each slab (constant through slab).

engine.num_layers = 50          # default value
Symbol

NLYR

Complexity

\(O(n)\)

Valid range

0 < NLYR

Default

50

Logic

Increase with optical depth.

Note

Reasonable values may vary from 10 to hundreds.

Number of Quadrature Terms¶

In the Number of Streams section we saw the following integral.

\[\int_{-\pi}^{\pi} L_l(\theta) f(\vec{x}, \theta) \ \mathrm{d} \theta\]

Internally this integral is computed as a quadrature sum. The num_brdf_quadrature_terms and num_phasef_quadrature_terms parameters set the number of quadrature terms used evaluated this integral numerically.

engine.num_phasef_quadrature_terms = 64         # default value
engine.num_brdf_quadrature_terms = 64           # default value
Symbol

NPHASEF_QT, NBRDF_QT

Complexity

\(O(n)\)

Valid range

NPHASEF_QT, NBRDF_QT = {64, 128, 256, 512, 1024}

Default

64

Logic

Increase with complexity of function, \(f\).

Note

The integral is evaluated seperately over the intervals [-1, 0] and [0, 1].

Engine Options¶

The following are settable options in Sasktran-Disco. Properties are lowercase strings with spaces removed. These are set your engines sasktran.Engine.options.

Keyword	Type	Description
`sun`	`np.array`	Three element unit vector pointing to the sun in Geodetic coordinates.
`altitudegrid`	`np.array`	Array of altitudes (in meters) which define the range of altitudes where Sasktran-Disco will calculate radiance. These are also the altitudes where climatologies will be polled. The default is `np.linspace(0, 100e3, 201)` which means poll every 500m from the ground to 100km.
`radiancetoa`	`np.array`	Two element array defining radiance at the top of the atmosphere. The defaults are `np.array([1, 0])`. The first element is the direct radiance and the second element is the diffuse radiance (at the top of the atmosphere).
`wfaltitudes`	`np.array`	Altitudes where weighting functions are calculated at (in meters).
`wfwidths`	`np.array`	Defines the region in which the perturbation occurs. If this is set to 500 (meters) then the perturbations will occur 500m above and 500m below each `wfaltitudes`. A width must be specified for each `wfaltitudes`.
`wfeps`	`np.array`	The percentage by which the species will be perturbed when calculating the weighting function (eg. a value of 0.01 means the size of the perturbation will be 1%). An epsilon must be specified for each `wfaltitudes`.
`numstreams`	`int`	Sets the number of streams.
`numlayers`	`int`	Sets the number of layers the atmosphere is discretized into.
`numphasefexpansionterms`	`int`	Sets the number of quadrature terms used in the expansions of the phase function.
`numbrdfexpansionterms`	`int`	Sets the number of quadrature terms used in the expansion of the BRDF.
`diagnostics`	`bool`	Tells the engine to cache all diagnostics. This is useful for when you want to query various parameters from the engine afterwards.
`losdiagnostics`	`bool`	Tells the engine to cache diagnostics related to the lines of sight (ie. geometry).
`rtsdiagnostics`	`bool`	Tells the engine to cache diagnostics related to the solution of the radiative transfer equation and discrete-atmosphere.
`wfform`	`int`	Specifies the form that you would like the weighting functions returned in. A value of 0 return weighting function in the form \((I' - I)/\mathrm{wfeps}\). A value of 1 returns weighting in the form \((I' - I)/(\mathrm{wfeps} \times \mathrm{nd})\) where \(\mathrm{nd}\) is the species unperturbed number density (this essentially gives you contribution per molecule).
`convergencecriteria`	`double`	Specifies convergence criteria. Once consecutive (2) iteration of the azimuth expansion yield contributions less than this value the solution is said to have converged. If a solution does not converge its value (radiance) will be set to `nan` and a warning will be shown.
`forcenumazimuthterms`	`int`	Specifies the exact number of terms used in the azimuth expansion. The convergence criteria is ignored if this value is set.
`ssadither`	`double`	Singularities occur for the case where the SSA (scattering albedo) is exactly 0 or 1 (which are unphysical but nevertheless useful). If one of thse cases occurs the value of the SSA will be dithered by this amount.
`averagereferencepoint`	`bool`	If true the engine will use a single reference point for all the lines of sight (the average of all ground-intersections), otherwise the ground-intersection of each line of sight will be used as the reference point (i.e. different reference point for each line of sight).

About These Figures¶

Configuration¶

Geometry

Observer: Geostationary satellite similar to TEMPO (location: 0N,-100E, altitude: 35,786 km)
Line of sight: Single line of sight looking at the ARG office in the Physics building at the University of Saskatchewan (location: 52.131638N,-106.633873E)
Time: 3:14PM June 2, 2017

Atmosphere

Species: Rayleigh, O3, NO2, and an aerosol
Surface: Snow (Kokhanovsky, default settings)
Wavelengths: 2 ozone retrieval wavelengths for OSIRIS, 322.5nm and 350.31nm are optically relatively thick and thin respectively. 1 aerosol retrieval wavelength, 749.94 where the influence of the aerosol is relatively significant.

Equations¶

Someone that is familiar with the discrete-ordinates method might notice that some of the equations presented above omit some important details. This was done for the sake of clarity.