A 3D Monte Carlo scattering code to keep track of individual parcels in a plane-parallel homogenous slab
The theory behind the model, many of its parameters, and model validation have been described by Petrich et al. (2012), https://doi.org/10.1016/j.coldregions.2011.12.004
The Monte Carlo code itself is implemented in C for speed and integrated with a Python interface for convenience.
License: Apache 2.0.
- Python 2.7, 3.5, 3.6, 3.7. Should work on Linux, MacOS, and Windows.
- Numpy
- When compiling from source, a C-compiler, and typically setuptools and wheel.
Run pip install mcmodel
. Done.
Later, upgrade with pip install -U mcmodel
, or uninstall with
pip uninstall mcmodel
.
Get the latest source code from https://github.com/cpetrich/MCmodel
Run python setup.py build
and copy the contents of the built/lib.[...]
directory into the same directory as your own modeling code.
Get the latest source code from https://github.com/cpetrich/MCmodel
A wheel will be built locally and installed with pip.
Build the module and create a wheel file
with python setup.py bdist_wheel
, cd dist
,
pip install mcmodel-[...].whl
. To uninstall the package later run
pip uninstall mcmodel
.
To test the result of the build process, a rather incomplete test suite
can be run with python setup.py test
.
The code is currently not reentrant since it uses global state for the random number generator, the phase function, and buffers that hold detailed information about the most recent particle track. I.e., only one instance of the model per process.
The special case - in a scattering model - of "no scattering" (w0 == 0
) is
implemented only for k == 0
(i.e. no absorption).
The usual mode of operation is:
- seed the random number generator
- define a scattering phase function
- call the scattering code with slab optical properties, and position and direction of one parcel of light
- store entrance and exit positions and angles of the parcel and information about plane crossings inside the medium
- if desired, store complete track of the light parcel
- continue with (2) a couple of thousand or million times.
- read the simulation results from disk, and determine parcel intensity due to absorption and Fresnel reflection
- analyze result as desired
An example of seeding the random number generator of the Monte Carlo model is
import os
import mcmodel
mcmmodel.set_seed(int.from_bytes(os.urandom(4),'little'))
A helper function exists to generate common scattering phase functions, e.g.
import mcmodel_util
pf = mcmodel_util.make_phase_function('Henyey-Greenstein', (0.98, 0))
The phase function is transfered to the model code like
mcmodel.define_phase_function(pf['lookup_cdf'],pf['lookup_phi'])
A random number generator is usually used to generate an incoming distribution (that random number generator is separate from the one used by the Monte Carlo model and should also be seeded).
import random
random.seed()
A sample from the angular distribution of a Lambert emitter (i.e., overcast over snow-covered ground) is
import math
elevation_angle = math.arcsin(random.uniform(-1., 0))
azimuth = random.uniform(-math.pi, math.pi)
Note that the Monte Carlo code works in elevation angles. A normally incident beam at the upper surface of the slab has an elevation angle of -pi/2.
The domain parameters and incident angle are passed on to the Monte Carlo code in a dictionary, e.g.
in = {'angle': elevation_angle,
'azimuth': azimuth,
'thickness': 1, # slab thickness
'k': 1} # densiy of microscopic interactions used by the model
with additional optional input parameters described below. Many of the output data are obtained through updates to a dictionary that is passed to the scattering simulation. To calculate the track of a single parcel call
out = {}
mcmodel.simulate(in, out)
where exit position and angles are
exit_position = (out['x'], out['y'], out['z'])
exit_elevation_angle = out['angle']
exit_azimuth = out['azimuth']
path_length = out['path_length']
Attenuation of the parcel inside the medium is accounted for by decreasing its amplitude. The precise method of this depends on the definition and units of thickness and microscopic extinction coefficient passed into the model above. However, the intensity of the emerging parcel will be of a form similar to
exit_intensity = math.exp(-kappa * path_length)
where kappa is the absorption coefficient.
Intensity reduction due to Fresnel reflection (other than total reflection) would be accounted for based on entrance and exit angles and relative refractive indices. To do this in hindsight is not trivial and requires the simulations to be run with a relative refractive index of 1 and appropriate scaling of the input angular distribution to account for non-trivial refractive indices. This is outside the scope of this overview. [TODO: in a future version, add an option to have Monte Carlo model take care of this by selecting a path stochastically.]
The C code exports currently four functions (see Usage above):
mcmodel.set_seed(...)
mcmodel.define_phase_function(..., ...)
mcmodel.simulate(..., ...)
mcmodel.get_last_particle_track()
In addion, there is one helper function defined in Python
mcmodel_util.make_phase_function(...)
The model domain is a horizontal slab of thickness thickness
, extending from
z=0
at the upper surface to z=-thickness
at the bottom. Parcles are injected
at depth z=z0
(-thickness <= z0 <= 0
) and x=0
and y=0
. Injection at
z0 = -thickness
and z0 = 0
are on the outside of the slab at the lower and
upper interface, respectively (i.e. still subject to refraction).
If an anisotropic scattering coefficient is used, i.e. sigma_anisotropy != 0
,
then the vertical and horizontal scattering coefficients are
sigma_v = k * w0
and sigma_h = k * w0 * (sigma_anisotropy+1)
, respectively.
Name | Default | Meaning |
---|---|---|
angle |
none | elevation angle of incident parcel [-pi; pi] |
azimuth |
none | azimuth angle of incident parcel [-pi; pi) |
z0 |
0 |
vertial coordinate of parcel injection |
thickness |
none | slab thickness |
k |
none | microscopic extinction coefficient |
w0 |
1 |
single scattering albedo |
n_surface |
1 |
refractive index n_slab / n(z>0) |
n_bottom |
1 |
refractive index n_slab / n(z<-thickness) |
sigma_anisotropy |
0 |
anisotropy of scattering coefficient (cf. Trodahl et al.), sigma_anisotropy = sigma_h/sigma_v - 1 |
do_record_track |
False |
log coordinates of scattering the parcel. Retrieve with mcmodel.get_last_particle_track() |
do_record_plane_crossings |
False |
log position and angle of parcel every time it crosses specified planes |
record_planes |
[] |
list of z coordinates to record parcel crossings at |
Name | Meaning |
---|---|
angle_in |
copy of angle in input dictionary |
azimuth_in |
copy of azimuth in input dictionary |
angle |
exit eleation angle of parcel |
azimuth |
exit azimuth of parcel |
x |
lateral exit coordinate, same unit as thickness in input |
y |
lateral exit coordinate, same unit as thickness in input |
z |
vertical exit coordinate, same unit as thickness in input |
path_length |
total path length of parcel in the medium, same unit as thickness in input |
exit_direction |
+1 : exit at z=0 , -1 : exit at z=-thickness |
min_z |
lowest z coordinate along the track |
max_R |
largest value of sqrt( x^2 + y^2 ) along the track |
n_collisions |
number of scattering events along the path (NB: not total reflection) |
n_missed_collisions |
number of absorption events along the path, all of which have been ignored (count is slightly too high if w0>0 ) |
n_refractions |
number of refractions experienced along the path |
n_total_reflections |
number of total reflections along the path |
n_plane_crossings |
number of plane crossing events along the path |
plane_crossings |
2D numpy array containing 6 entries (columns) per crossing event (see below) |
The six column entries for each plane crossing are:
- elevation angle
- azimuth angle
- x-coordinate at the point of crossing
- y-coordinate at the point of crossing
- z-coordinate at the point of crossing
- parcel path length so far