Identifier

etd-03302006-122704

Degree

Doctor of Philosophy (PhD)

Department

Oceanography and Coastal Sciences

Document Type

Dissertation

Abstract

Mapping coastal marshes is an important component in the management of coastal environments. Classification of marshes using remote sensing data has traditionally been performed by employing either parametric supervised classification algorithms or unsupervised classification algorithms. The implementation of these conversional classification methods is based on the underlying distributions concerning the probability density functions (PDF). Neural networks provide a practical approach to this classification because they are essentially non-parametric data transformations that are not restricted by any underlying assumptions. The major objective of this study was to evaluate the ability of neural networks using Moderate Resolution Imaging Spectro-radiometer (MODIS) data to classify coastal marshes based on the phenelogical stages of plants. The first stage of the study was to develop a neural network model. The analysis has shown that six day images with eight input variables each are required to perform the classification. The variables are: MODIS bands - the near infrared and the near infrared composite bands, biophysical variables – the leaf area index (LAI) and the fraction of photosynthetically active radiation (fPAR). Other variables are vegetation indices – the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), and the wetness index (WI), and, the day time land surface temperature. The near infrared and the wetness index were found to be the strongest predictor variables in the classification. Six hidden neurons and one output neuron were required in the neural network model for the output of six classes. The second stage of the dissertation was the model application. Images from four years: 2001, 2002, 2003, and 2004 were classified using the model. Accuracy assessment of the classification indicated that neural network techniques using MODIS data could achieve an accuracy of over 80% (at 0.95 confidence level). Using the classified images change detection was performed to determine the loss and gain of four marsh types; saline marsh, brackish marsh, intermediate marsh, and, fresh water marsh found in the south eastern coastal areas of Louisiana. The greatest gain was in the intermediate marsh, 3.0% of the study area, and the greatest loss was in the saline marsh, 3.8% of the study area.

Date

2006

Document Availability at the Time of Submission

Release the entire work immediately for access worldwide.

Committee Chair

Lawrence Rouse

Share

COinS