Advancing the satellite remote sensing of heterogeneous clouds through the development of a tomographic technique that uses 3D radiative transfer

Loveridge, Jesse Ray

Permalink

https://hdl.handle.net/2142/121943 Copy

Description

Title

Advancing the satellite remote sensing of heterogeneous clouds through the development of a tomographic technique that uses 3D radiative transfer

Author(s)

Loveridge, Jesse Ray

Issue Date

2023-09-21

Director of Research (if dissertation) or Advisor (if thesis)

Di Girolamo, Larry

Doctoral Committee Chair(s)

Di Girolamo, Larry

Committee Member(s)

• Nesbitt, Stephen
• Waldrop, Lara
• Xu, Feng

Department of Study

Atmospheric Sciences

Discipline

Atmospheric Sciences

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Clouds
• Remote Sensing
• 3D Radiative Transfer
• Tomography
• Atmospheric Radiation

Abstract

Clouds are the largest source of uncertainty in our understanding of climate and its change, due to their strong radiative effect and its dependence on dynamical and microphysical processes at a range of spatio-temporal scales through, for example, Aerosol Cloud Interactions (ACI). Satellite remote sensing techniques are one of the few methods for measuring the optical and microphysical properties that are associated with these processes at a near-global scale to understand their impact on climate. Retrievals of cloud optical depth (τ), droplet effective radius (r_e), Liquid Water Path (LWP), and droplet number concentration (N_d) from satellite remote sensing observations in the solar spectrum are the primary piece of observational evidence used for understanding the global impact of marine boundary layer clouds on the Earth’s radiation budget. All of these remote sensing retrievals utilize two strong assumptions for interpreting the reflected solar radiation. These are the Plane-Parallel Homogeneous Approximation (PPHA) and Independent Pixel Approximation (IPA), which assume, respectively, that clouds form horizontally infinite, homogeneous slabs and that each pixel is radiatively independent of others for the purpose of retrieving cloud optical and microphysical properties. These assumptions are imperfect for all clouds but are still used due to their computational efficiency. Much of our current observational understanding of the global impact of anthropogenic pollution on climate rests on these two assumptions. The work in this thesis has two objectives. The first objective is to test for the first time the impact of the IPA and PPHA on our understanding of the climatology of cloud microphysical properties and the assessment of ACI using satellite remote sensing from reflected solar radiation. The second objective is to develop a tomographic remote sensing technique that is targeted towards the heterogeneous clouds that are ill-served by the IPA and PPHA and will enable more detailed and accurate retrievals of the volumetric optical and microphysical properties of heterogeneous clouds. To achieve the first objective, a wide range of stochastically generated cloud fields were combined with 3D radiative transfer to study retrieval errors in the widely used bispectral technique for retrieving τ and r_e and inferring LWP and N_d. These simulations were used to test, for the first time, the effectiveness of sampling strategies used in studies of ACI that claimed to select only retrievals that were effectively unaffected by retrieval errors due to the PPHA and IPA. These sampling strategies select pixels based on the spatial homogeneity of the radiance field. We show that retrievals of τ, r_e, LWP and N_d are affected by systematic errors that are almost unchanged whether these subsampling strategies are applied or not. Systematic errors vary from ~0% to -55% for τ with increasing cloud heterogeneity, -10% to +50% for r_e, and +30% to -80% for N_d, consistent with other studies. The widely used sampling strategies do preferentially reject retrievals from heterogeneous cloud conditions but the retrievals that are selected are still just as biased as those retrievals that were rejected. Retrieval errors are therefore expected to confound the estimation of the climatology of cloud microphysical properties and the assessment of ACI. To understand the implications of these errors, the results of the simulations were extrapolated to the globe using data from the Moderate Resolution Imaging Spectroradiometer (MODIS). This is done, not to provide a fool-proof estimate of climatological error in retrieved droplet number concentration, but rather to estimate plausible errors due to the use of the PPHA and IPA. Based on the simulations, we parameterized the bias in retrieved droplet number concentration as -50% for retrievals from non-overcast (9 km)2 patches and as 0% for retrievals from (9 km)2 patches and analyzed the impact on MODIS data for different sampling strategies. We found that the implied bias in averages over 1 by 1 regions increases in magnitude linearly from 0% to -50% as the cloud fraction in the region decreases. These results imply a bias in the climatology of N_d that peaks at -40% in the trade cumulus regions of the tropical and subtropical oceans and reaches a minimum of -10% in the subtropical stratocumulus decks. The correlation between cloud fraction, and hence implied retrieval bias, retrieved droplet number concentration and aerosol concentration, in the MODIS data mean that the sensitivity of cloud droplet number concentration to aerosol concentration will plausibly be overestimated by 15% to 20%. The sensitivity of cloud fraction to changes in aerosol will plausibly be overestimated by 50%. With this study, we have demonstrated that the methodologies used to estimate radiative forcing due to ACI in marine boundary layer clouds from satellite remote sensing retrievals using reflected solar radiation can plausibly contain substantial systematic errors. This is the first quantification of such errors at regional scales over the ocean. This thesis also tested the assumptions of a climatological bias in the r_e estimated by Liang et al. (2015) based on the observed inconsistency in the optical depth retrieved across the cloudbow scattering angles, which is very sensitive to errors in the retrieval of r_e using simulations. The climatological bias estimate provides upper and lower bounds on the bias in the r_e retrieved by MODIS using simultaneous measurements in the rainbow scattering angle range from the Multiangle Imaging Spectroradiometer (MISR). The lower bounds range from +10% to +60% regionally, while the upper bounds can reach +230%. The fidelity of these bounds rests on the assumption that the view-zenith angle dependent bias in the retrieval of cloud optical depth due to cloud optical depth is monotonic over the range required for MISR to sample across the range of scattering angles of the cloud bow from different cameras. This assumption was tested for the first time in this thesis using 3D radiative transfer simulations and was confirmed. However, it was also identified in this thesis that undetected thin cirrus were also causing an overestimate in the bias in the r_e derived using the cloudbow technique. With an idealized model of the occurrence of thin cirrus, we estimated that the largest positive biases indicated by the r_e were unlikely, and that the lower bound range of error, with regional variations from +10% to +60%, is a more reasonable estimate of the range of error. These two lines of evidence, both the cloudbow technique and the simulations, demonstrate the presence of large climatological biases in retrievals of cloud microphysical properties from techniques using the PPHA and IPA, that are comparable to differences between climate models or the discrepancies between models and observations. This demonstrates that these techniques are unable to provide further insight into the details of the global distribution of cloud microphysical properties in marine boundary layer clouds. This motivates the development of new observational techniques that relax these assumptions. The second part of this thesis addresses the need for new observational techniques by furthering the development of a tomographic technique that retrieves of volumetric cloud optical and microphysical properties from multi-angle imagery in the solar spectrum using 3D radiative transfer. The implementation of this technique is developed here and made available in the software package Atmospheric Tomography with 3D Radiative Transfer (AT3D). This tomographic technique builds on previous demonstrations of cloud tomography that developed an approximate, computationally efficient, linearization of the 3D radiative transfer model, Spherical Harmonics Discrete Ordinates Method (SHDOM). In this thesis, the theoretical underpinnings of this approximate linearization are derived as an approximation in forward-adjoint perturbation theory and its accuracy is quantified for the first time. We find that for cloud-like media over dark surfaces, the approximate linearization has an error that increases from 2% to 12% as the optical depth of the medium increases. Larger errors occur over brighter surfaces. A linearized analysis of the cloud tomography problem demonstrates that the cloud tomography problem becomes progressively more ill-conditioned as the optical depth of the cloud increases and the closer it approaches plane-parallel geometry. This ill-conditioning is expressed as a loss of sensitivity to regions that are optically far from the sensors and the sun. This loss of sensitivity is found to also be strongly sensitive to the forward scattering peak of the phase function, and its numerical treatment. Stronger forward scattering peaks result in more efficient transmission and therefore reduce the ill-conditioning of the tomography problem. Based on these considerations, we conclude that cloud tomography will be most effective for heterogeneous clouds over ocean, as well as other optically thin media such as aerosol. This prediction is tested with numerical experiments on 40 stochastically generated, isolated cumuliform clouds with a range of optical thicknesses. The 3D volumetric extinction coefficient at 672 nm is retrieved at high (40 m) resolution from synthetic measurements at 9 multi-angle narrowband images at 35 m resolution at a solar zenith angle of 60. The relative root-mean-square error (RMSE) and relative bias of the retrievals are less than 20% and 1%, respectively, for clouds that have maximum cloud optical depths less than 17. This demonstrates that tomography is highly effective for isolated shallow cumulus. The retrieval errors grow with optical depth reaching RMSE of ~70% and bias of -36% as the maximum optical depth of the cloud reaches 88. All tomographic retrievals outperform a retrieval of optical depth using the PPHA and IPA with relative RMSE that are a factor of 2 to 10 better. Retrievals also become sensitive to radiometric noise in this optically thick limit with deviations in the mean extinction coefficient of 18% possible due to the inclusion of noise. Errors in the optically thick clouds are systematic with a decrease in extinction from the illuminated to the shadowed side of the cloud and with distance from cloud edge, which are attributed to the ill-conditioned nature of the tomography in the optically thick limit. These systematic errors and the sensitivity to noise are sensitive to the numerical treatment of the forward scattering peak. This demonstrates the need for additional measurements or priors to constrain retrievals in thick clouds but also the possibility for other preconditioning techniques to improve the convergence of the tomography problem. Even with these limitations, tomography will be effective for a number of climatically relevant atmospheric scatterers such as cumulus, cirrus, and aerosol, overcoming the limitations of existing techniques which use the PPHA and IPA. The further development and deployment of cloud tomography will derive new insight into the microphysical process in marine boundary layer clouds, such as ACI, which are important for our understanding of Earth’s climate.

Graduation Semester

2023-12

Type of Resource

Thesis

Copyright and License Information

Copyright 2023 Jesse Loveridge

Owning Collections