Date of Award

Fall 12-1-2003

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Geography, Geology, and Anthropology

First Advisor

Qihao Weng

Second Advisor

Paul Mausel

Third Advisor

Brian Ceh

Abstract

The world population and associated housing area have experienced a high growth rate in the past century and generated great pressure on Earth resources and environments. Accurate and timely population and housing estimates and distributions become important for many applications, such as regional design and development, policy-making, etc. This research reviewed four categories of methods to estimate population and housing using remotely sensed data in a previous study and explored the potentials of integration of Landsat ETM+ data with census data to estimate population and housing density in Marion County, Indiana. Spectral signatures, principal components, vegetation indices, fraction images, texture, and temperature were tested. Correlation analysis was used to explore the relationships between remote sensing variables and population/housing and stepwise regression analysis was used to develop models for estimating population/housing quantities. Two different sampling schemes (non-stratified and stratified sampling schemes) were compared. It was found that population and housing densities have significantly negative correlation with variables that relate to green vegetation, such as ETM+ band 4, the second principal component, vegetation fraction, and vegetation indices and have positive correlation with surface temperature; population and housing can be estimated using remote sensing variables derived from ETM+ data with reasonable accuracy; incorporation of textures, temperature and spectral response improved the model performance; stratification of population and housing into low, medium and high densities and developing models separately provide better estimation than a non-stratified scheme. The total estimated population for Marion county is 832,792 with accuracy of96.8% using a stratified sampling scheme compared to accuracy of91.8% of non-stratified; the total estimated housing units are 384,415 with accuracy of99.3% using a stratified scheme compared to accuracy of 96.9% using a non-stratified scheme. It is also found that ETM+ data is more suitable to estimate population and housing in medium density areas, not suitable for low and high density areas.

Download

Share

COinS