Date of Award

Spring 5-1-2001

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Geography, Geology, and Anthropology

First Advisor

Paul Mausel

Second Advisor

Brian Ceh

Third Advisor

Eduardo Brondizio

Abstract

This research focuses on measurement of forest stand parameters using remote sensing which is applied to classification and change detection of tropical successional and mature forests in selected areas in the Brazilian Amazon Basin. Three study areas, Altamira, Bragantina, and Pedras, each with different biophysical environments, were selected for this research. Previous research has indicated that relationships between selected stand parameters and remotely sensed data are not clearly understood, especially in tropical areas. Rarely has remote sensing research been successfully conducted in quantitative estimation of forest stand parameters (e.g. biomass) and identification of vegetation growth stages. In this research, atmospherically corrected multi-temporal Thematic Mapper (TM) images and field sampled vegetation inventory data were integrated. Twenty-three vegetation indices and four texture measures with five different window sizes were implemented in analysis. Pearson's correlation coefficient, stepwise regression analysis, and multiple regression models were employed to analyze relationships between forest stand parameters and remotely sensed data. Ratio of tree biomass to total biomass (RTB) value thresholding was used to differentiate vegetation growth stages and post-classification comparison was conducted to detect forest cover changes. Biomass change detection was also explored. This study concludes that different biophysical environments influence relationships between forest stand parameters and remotely sensed data. In a study area with rapid vegetation growth rates, TM textural information has stronger correlations with forest stand parameters (correlation coefficients up to 0.84) than do TM-based spectral signatures (correlation coefficients are less than 0.63). However, in a study area with slow vegetation growth rates, the correlations of forest stand parameters are relatively weak with TM textural information (correlation coefficients are less than 0.65) but strong with TM spectral signatures (correlation coefficients up to 0.88). A middle infrared wavelength (TM 5) and linear combinations of TM bands such as PCl (the first component of principal component analysis), KTI (the first component of Tasseled Cap transform), and Albedo are strongly related to forest stand parameters, which are not greatly influenced by different environmental conditions. Multiple regression models using spectral and textural TM data improved estimation accuracy of forest stand parameters in all study areas. As an example, the estimation accuracy of biomass in Altamira using spectral signatures is 85%, while using textural information it is 78%; however, using a combination of spectral and textural information accuracy reaches 93%. These values vary depending on the study area and parameters used, but most of the resulting accuracies derived from the multiple regression models using spectral and textural signatures exceed 80% for successional and mature forests, and exceed 98% when excluding initial secondary succession (SSl) sites. The models developed in this research have been used to estimate forest stand parameters (e.g. biomass) in Altamira and Bragantina. The established models have been determined to be suitable to extrapolate estimation on the same scene of TM image, but if they are transferred to multiple dates of TM images for biomass change detection, additional calibration of the results is frequently necessary because of the impacts of different climate conditions between different image acquisition dates.

Download

Share

COinS