Massachusetts Institute of Technology, US, 2009, 232 pp. - ISBN-13
978-1-59693-372-9.
This book is intended to provide a practical, applications-oriented treatment of neural network methodologies for use in atmospheric remote sensing. We focus on the retrieval of atmospheric parameters, such as the Earth's temperature and water vapor profiles and precipitation rate, but the techniques described can be applied to a wide variety of problems where function approximation is required. We use simple, largely theoretical examples to provide the reader with intuition on how performance is affected by basic neural network attributes such as model selection, initialization, and training methodology, and we then build these simple techniques into larger, "real-world" applications that are common throughout the field of atmospheric remote sensing. Many of the examples are accompanied by Matlab™ (www.mathworks.com) software codes (available on the accompanying CD-ROM in the back of the book) that can be used as building blocks for larger and more complex problems. These codes were written using the freely available Netlab Neural Network package and do not require any Matlab™ "addon" toolboxes.
Contents.
Preface.
Introduction.
Physical Background of Atmospheric Remote Sensing.
An Overview of Inversion Problems in Atmospheric Remote Sensing.
Signal Processing and Data Representation.
Introduction to Multilayer Perceptron Neural Networks.
A Practical Guide to Neural Network Training.
Pre- and Post-Processing of Atmospheric Data.
Neural Network Jacobian Analysis.
Neural Network Retrieval of Precipitation from Passive Microwave Observations.
Neural Network Retrieval of Atmospheric Profiles from Microwave and Hyperspectral Infrared Observations.
Discussion of Future Work.
This book is intended to provide a practical, applications-oriented treatment of neural network methodologies for use in atmospheric remote sensing. We focus on the retrieval of atmospheric parameters, such as the Earth's temperature and water vapor profiles and precipitation rate, but the techniques described can be applied to a wide variety of problems where function approximation is required. We use simple, largely theoretical examples to provide the reader with intuition on how performance is affected by basic neural network attributes such as model selection, initialization, and training methodology, and we then build these simple techniques into larger, "real-world" applications that are common throughout the field of atmospheric remote sensing. Many of the examples are accompanied by Matlab™ (www.mathworks.com) software codes (available on the accompanying CD-ROM in the back of the book) that can be used as building blocks for larger and more complex problems. These codes were written using the freely available Netlab Neural Network package and do not require any Matlab™ "addon" toolboxes.
Contents.
Preface.
Introduction.
Physical Background of Atmospheric Remote Sensing.
An Overview of Inversion Problems in Atmospheric Remote Sensing.
Signal Processing and Data Representation.
Introduction to Multilayer Perceptron Neural Networks.
A Practical Guide to Neural Network Training.
Pre- and Post-Processing of Atmospheric Data.
Neural Network Jacobian Analysis.
Neural Network Retrieval of Precipitation from Passive Microwave Observations.
Neural Network Retrieval of Atmospheric Profiles from Microwave and Hyperspectral Infrared Observations.
Discussion of Future Work.