Forward Models

A forward model in skretrieval is something that simulates the observation data, given the state vector. This is accomplished through the skretrieval.retrieval.ForwardModel class and the calculate_radiance() method.

Available Forward Models

skretrieval includes some pre-built forward models that are suitable for general instruments. We recommend determining if your problem is suited to one of these pre-built objects before creating your own forward model.

skretrieval.forwardmodel.IdealViewingSpectrograph(...)

A forward model for the retrieval that uses an ideal viewing geometry and a spectrometer

The Default Forward Model

If no options are specified when construcing the Retrieval object, the default forward model, skretrieval.forwardmodel.IdealViewingSpectrograph object, is created. This behaviour can be explicitly set by passing

forward_model_kwargs = {"class": skretrieval.forwardmodel.IdealViewingSpectrograph}

to the retrieval.

The ideal viewing component indicates that no integration is performed over the viewing dimension, i.e., the spatial point spread function is assumed to be a delta function. For the spectral dimension, the default is also to assume delta function behaviour, but this can be changed by specifying extra parameters, for example to switch to a Gaussian line shape,

from skretrieval.retrieval.core.lineshape import Gaussian

forward_model_kwargs = {"class": skretrieval.forwardmodel.IdealViewingSpectrograph,
                        "line_shape_fn": lambda w: Gaussian(fwhm=1.5),
                        "model_res_nm": 0.1
                        }

Since the lineshape may be a function of wavelength, we have to specify a function that takes in wavelength and returns back the lineshape rather than just a singular lineshape. In addition we pass in the model_res_nm parameter which indicates the resolution the forward model should simulate the radiance at.

Configuring the Radiative Transfer Model

Every forward model is passed a sasktran2.Config object that may be used to configure various radiative transfer settings. These settings are configured through the model_kwargs parameter that is passed to the skretrieval.Retrieval object. For example,

import sasktran2 as sk
import skretrieval as skr

model_kwargs = {
    "num_threads": 8,
    "multiple_scatter_source": sk.MultipleScatterSource.DiscreteOrdinates
}

ret = skr.Retrieval(..., model_kwargs=model_kwargs)

This is equivalent to creating a sasktran2.Config object with the properties

config = sk.Config()

config.num_threads = 8
config.multiple_scatter_source = sk.MultipleScatterSource.DiscreteOrdinates

Setting the radiative transfer settings is something that is almost almost problem dependent, and therefore the defaults that skretrieval uses are likely not suitable for your specific problem. Therefore we always recommend