Use this URL to cite or link to this record in EThOS:
Title: A fast forward model for the assimilation of radiances from the EOS-MLS
Author: Scorgie, Donald
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2006
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
In this thesis the idea of using neural networks as a forward model for the EOS-MLS (Earth Observation System – Microwave Limb Sounder) is considered for a direct assimilation scheme. Neural networks are a type of non-linear regression technique that can provide fast, accurate results and are used extensively in many different fields. Here a neural network is constructed to act as a forward model for the EOS-MLS. The neural network uses a temperature profile and tangent pressure levels as inputs and produces the corresponding radiance profile for one channel of the EOS-MLS. The work here primarily concentrates on one band of the EOS-MLS that is centred on an oxygen line and whose radiances are affected only by temperature for the majority of the channels. It shows that a neural network can function as a forward model in this case, producing radiances that are within instrument noise and for most channels, within half the instrument noise. Adding ozone to the forward model affects the radiances in only two channels of this band, increasing the radiances in some minor frames by around ~10K. It was found that this difference could be accounted for in the neural network forward model by adding ozone to the inputs. A second band, which is centred on an ozone line, is briefly considered. It was found that above 150hPa the radiances from this band could be modelled well using a neural network. Below this height, the neural network produced large errors in radiance (of around 1.5K – four times the instrument noise). This is thought to be due to the effects of water vapour. A problem specific to limb sounders that must be faced when doing direct assimilation is determining the tangent pressures of the radiances. During retrieval, these tangent pressures are normally retrieved as part of the state vector and discarded. For an assimilation process, these tangent pressures may be unavailable and have to be deduced in some way. Here, a neural network is used to retrieve tangent pressures outside the assimilation process. These retrieved tangent pressures can then be used by the forward model and assumed to be correct. It was found that tangent pressures could be retrieved with an accuracy of around 50m, much better than required for a forward model. The final problem faced within this thesis is the creation of the Jacobian of the instrument forward model. This is the derivative of the radiances with respect to the state vector and is used by the assimilation process to update the model fields during the assimilation process.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available