Scattering graph method for 3D radiative transfer

Antti Mikkonen 1,*, Anssi Koskinen 1,2, Johanna Tamminen 1 and Hannakaisa Lindqvist 3 · 2025

Read Scattering graph method for 3D radiative transfer by Antti Mikkonen 1,*, Anssi Koskinen 1,2, Johanna Tamminen 1 and Hannakaisa Lindqvist 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16037.txt.

Introduction

Mikkonen and three co-authors gave a new method for computing how light scatters in three-dimensional scenes in twenty twenty-five. Radiative transfer models track sunlight bouncing through clouds and air. They are essential for reading satellite images. Full three-dimensional calculations are accurate but slow.

Their scattering graph method replaces detailed physics between pixels with a heuristic coupling function, organized as a graph. The implementation, called TURSCA, runs on graphics processors. Against two established models it agreed within three and six percent. Test cases included simulating a satellite view of an emission plume. The authors say the method opens new avenues for satellite remote sensing of atmospheres. It trades a little physical detail for speed and freedom to adapt.

Catalog record

CREATORAntti Mikkonen 1,*, Anssi Koskinen 1,2, Johanna Tamminen 1 and Hannakaisa Lindqvist 3
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16037

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Scattering graph method for 3D radiative transfer Antti Mikkonen 1,*, Anssi Koskinen 1,2, Johanna Tamminen 1 and Hannakaisa Lindqvist 3 Abstract A novel method for monochromatic scalar 3D radiative transfer, designed primarily for modeling remote sensing imaging, is presented. For simulating an observation of an imaging satellite instrument, the method uses a heuristic scattering coupling function to model the inter-pixel scattering of radiation, which is represented with a graph. The GPU-capable code implementation of the method, TURSCA, was validated against two established 3D RT models, Siro and SHDOM with relative agreement at 3% and 6%, respectively. The capabilities of TURSCA in modeling a satellite observation of an emission plume are examined. The presented method opens up new avenues of research, especially in satellite-based remote sensing of atmospheres. Journal: opticajournal Publication type: Research Article 1 Introduction Atmospheric remote sensing is the process of inferring information about the planetary atmosphere based on observations of electromagnetic radiation which it has interacted with, and this information is extracted from the observations with a retrieval algorithm. At the core of atmospheric remote sensing retrieval algorithms is a radiative transfer (RT) model, which simulates the phenomena associated with the propagation of radiation in a medium, such as absorption, scattering and emission, in the atmosphere. In the context of inferring trace gas concentrations from space-based spectral measurements, for example, the Atmospheric CO 2 Observations from Space (ACOS) algorithm [ 1 ] , used in retrieving total column of carbon dioxide (CO 2 ) from NASA’s Orbiting Carbon Observatory -2 (OCO-2) satellite observations, utilizes a modified version of LInearized Discrete Ordinate Radiative Transfer (LIDORT) [ 2 ] and the methane column retrieval algorithm [ 3 ] employs LINTRAN [ 4 ] to process TROPOspheric Monitoring Instrument (TROPOMI) observations. In both of these RT models, the atmosphere is assumed to be one-dimensional, with the only variation being altitudal. This is a good assumption when the ground pixel sizes are in the scale of several km 2 as is the case with OCO-2 and TROPOMI. In addition, OCO-2 and TROPOMI are both in a solar synchronous orbit with their equatorial crossing times close to local noon, which cause the observed solar zenith angles (SZA) be relatively small, further supporting the assumption of 1D atmosphere and yielding less atmospheric scattering compared to larger SZAs, which can complicate the retrieval. In the vicinity of clouds, however, the RT modeling in a 3D atmosphere is required for OCO-2 and TROPOMI [ 5 , 6 ] . A nearby cloud may cast a shadow onto the pixel to be retrieved, causing a deviation from the assumed 1D atmosphere, resulting into biases in the retrieved trace gas column concentrations. These cloud 3D effects are the most prominent, but airborne aerosols and gas concentration differences could cause 3D effects as well. These 3D effects likely become more prominent when retrieving atmospheric information from high-resolution imaging spectrometer satellite instruments, such as GHGSat [ 7 ] , EMIT [ 8 ] and PRISMA [ 9 ] . In nadir-viewing passive remote sensing, when the ground pixel sizes are in the scale of thousands of m 2 , the incident solar radiation will be affected by atmospheric columns in several adjacent pixels due to their small size. Atmospheric scattering further increases the radiative interaction between the columns. This effect between the pixels is also increased with the increasing SZA. In high latitudes, where SZAs are consistently large, these adjacent pixel interactions arise even in the km 2 size scale. Additionally, large viewing zenith angles (VZA) strengthen the radiative interaction with neighbouring pixels. A case with large SZAs and VZAs could be Sentinel-4 imaging spectrometer observations of the high latitudes from the geostationary orbit [ 10 ] , OCO-2 observations of the sunglint reflected from the Arctic ocean, or continuous observation of a single point on the Earth’s surface throughout the whole satellite overpass (i.e. target mode observation). Therefore, to be as realistic as possible, the RT problem in these cases should not be considered as individual pixels with 1D RT, but instead a full 3D RT should be calculated. There are two main approaches for 3D RT: discrete ordinates and Monte Carlo. These two different approaches have been verified to give consistent results [ 11 , 12 ] . Theoretically these methods are quite general, but their computational implementations can be based on widely different assumptions. These 3D RT codes are thoroughly intercompared by the International Intercomparison of 3D Radiation Codes (I3RC) project [ 13 ] . With discrete ordinates the radiation field is solved at each volume element into discrete zenith and azimuthal angular grid. This systematic approach is conceptually simple, but sets constraints on medium geometries and it can be prohibitively expensive to compute if a small subset of the radiation field is of interest, as is the case in satellite remote sensing. The discrete ordinates approaches can be used conjointly with other numerical methods to tackle the 3D RT problem. The Spherical Harmonic Discrete Ordinates Method (SHDOM) combines modeling of source and scattering functions using spherical harmonics with spatial integration using discrete ordinates [ 14 ] . The Discrete Anisotropic Radiative Transfer (DART) model employs ray-tracing and discrete ordinates [ 15 ] . In Monte Carlo RT (e.g. [ 16 ] ), photon packets are traced through the medium and are scattered into different directions sampled from the scattering phase functions. It is straightforward to incorporate different phenomena into the computation, such as polarization or refraction. However, Monte Carlo requires simulating millions of photon packets, which can be slow to compute and is non-deterministic due to stochastic nature of scattering. Code implementations for 3D Monte Carlo RT are, for example, I3RC community Monte Carlo [ 11 ] , LargE-Scale Simulation framework (LESS) [ 17 ] and MONKI (Monte Carlo KNMI) [ 5 ] . Due to the heavy computational load present in the RT calculations, graphical processing units (GPU) have been of increasing use. GPUs have been primarily designed for rapidly visualizing computer graphics, but with the advent of CUDA [ 18 ] and OpenCL [ 19 ] , they can be used for more general computation also. The main difference between GPUs and CPUs are that GPUs control thousands of computation threads compared to CPU’s tens of threads, but the GPU threads operate on single-instruction-multiple-data (SIMD) paradigm, where maximum computation efficiency is attained when homogenous calculations are carried out on heterogenous data. This makes the GPUs especially suitable for Monte Carlo RT models [ 20 , 21 , 22 , 23 ] , of which there are several implementations, but also the discrete ordinates method benefits from the GPU implementation [ 24 ] . In this work a novel methodology for 3D RT is presented. The scattering graph method is a way to model monochromatic scalar radiation fields observed by a remote sensing instrument in an absorbing and scattering medium, which is as flexibly defined as in Monte Carlo models, while being faster and deterministic. The method aims to enable new approaches on satellite-based atmospheric remote sensing retrieval algorithm development. The software implementation of the method, Transmittance calcUlator with Radiative Scattering Coupling Approach (TURSCA), can be used on both CPUs and GPUs for simulating remote

Read the complete text