Neural network satellite retrievals of nocturnal stratocumulus cloud properties
I investigate the feasibility of retrieving cloud top droplet effective radius, optical thickness and cloud top temperature of nocturnal marine stratocumulus clouds by inverting infrared satellite measurements using an artificial neural network. For my study, I use the information contained in the t...
Main Author: | |
---|---|
Language: | English |
Published: |
University of British Columbia
2011
|
Online Access: | http://hdl.handle.net/2429/32189 |
id |
ndltd-UBC-oai-circle.library.ubc.ca-2429-32189 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-UBC-oai-circle.library.ubc.ca-2429-321892018-01-05T17:46:28Z Neural network satellite retrievals of nocturnal stratocumulus cloud properties Rautenhaus, Marc I investigate the feasibility of retrieving cloud top droplet effective radius, optical thickness and cloud top temperature of nocturnal marine stratocumulus clouds by inverting infrared satellite measurements using an artificial neural network. For my study, I use the information contained in the three infrared channels centred at 3.7, 11.0 and 12.0 μm of the Moderate Resolution Imaging Spectroradiometer (MODIS) on-board NASA ' s Terra satellite, as well as sea surface temperature. A database of simulated top-of-atmosphere brightness temperatures of a range of cloud parameters is computed using a correlated-k parameterisation which I have embedded in the radiative transfer package libRadtran. The database is used to train feed-forward neural networks of different architecture to perform the inversion of the satellite measurements for the cloud properties. I investigate the application of Bayesian methods to estimate the retrieval uncertainties, and analyse the Jacobian of the networks in order to gain information about the functional dependence of the retrieved parameters on the inputs. A high variability in the Jacobian indicates that the nocturnal retrieval problem is ill-posed. My experiments show that because the problem is ill-conditioned, it is very difficult to find a network that approximates the database of simulated brightness temperatures well. Sea surface temperature proves to be a necessary input. I compare the retrievals of a selected network architecture with in-situ cloud measurements taken during the second Dynamics and Chemistry of Marine Stratocumulus experiment. The results show general agreement between retrievals and in-situ observations, although no collocated comparsions are possible because of a time lag of five hours between both measurements. I establish that the uncertainty estimates are prone to numerical problems and their results are questionable. I show that the Jacobian is a valuable tool in evaluating the retrieval networks. Science, Faculty of Earth, Ocean and Atmospheric Sciences, Department of Graduate 2011-03-09T00:43:46Z 2011-03-09T00:43:46Z 2007 Text Thesis/Dissertation http://hdl.handle.net/2429/32189 eng For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use. University of British Columbia |
collection |
NDLTD |
language |
English |
sources |
NDLTD |
description |
I investigate the feasibility of retrieving cloud top droplet effective radius, optical thickness and cloud top temperature of nocturnal marine stratocumulus clouds by inverting infrared satellite measurements using an artificial neural network. For my study, I use the information contained in the three infrared channels centred at 3.7, 11.0 and 12.0 μm of the Moderate Resolution Imaging Spectroradiometer (MODIS) on-board NASA ' s Terra satellite, as well as sea surface temperature. A database of simulated top-of-atmosphere brightness temperatures of a range of cloud parameters is computed using a correlated-k parameterisation which I have embedded in the radiative transfer package libRadtran. The database is used to train feed-forward neural networks of different architecture to perform the inversion of the satellite measurements for the cloud properties. I investigate the application of Bayesian methods to estimate the retrieval uncertainties, and analyse the Jacobian of the networks in order to gain information about
the functional dependence of the retrieved parameters on the inputs. A high variability in the Jacobian indicates that the nocturnal retrieval problem is ill-posed. My experiments show that because the problem is ill-conditioned, it is very difficult to find a network that approximates the database of simulated brightness temperatures well. Sea surface temperature proves to be a necessary input. I compare the retrievals of a selected network architecture with in-situ cloud measurements taken during the second
Dynamics and Chemistry of Marine Stratocumulus experiment. The results show general agreement between retrievals and in-situ observations, although no collocated comparsions are possible because of a time lag of five hours between both measurements. I establish that the uncertainty estimates are prone to numerical problems and their results are questionable. I show that the Jacobian is a valuable tool in evaluating the retrieval networks. === Science, Faculty of === Earth, Ocean and Atmospheric Sciences, Department of === Graduate |
author |
Rautenhaus, Marc |
spellingShingle |
Rautenhaus, Marc Neural network satellite retrievals of nocturnal stratocumulus cloud properties |
author_facet |
Rautenhaus, Marc |
author_sort |
Rautenhaus, Marc |
title |
Neural network satellite retrievals of nocturnal stratocumulus cloud properties |
title_short |
Neural network satellite retrievals of nocturnal stratocumulus cloud properties |
title_full |
Neural network satellite retrievals of nocturnal stratocumulus cloud properties |
title_fullStr |
Neural network satellite retrievals of nocturnal stratocumulus cloud properties |
title_full_unstemmed |
Neural network satellite retrievals of nocturnal stratocumulus cloud properties |
title_sort |
neural network satellite retrievals of nocturnal stratocumulus cloud properties |
publisher |
University of British Columbia |
publishDate |
2011 |
url |
http://hdl.handle.net/2429/32189 |
work_keys_str_mv |
AT rautenhausmarc neuralnetworksatelliteretrievalsofnocturnalstratocumuluscloudproperties |
_version_ |
1718594668836421632 |