GIS-based urban land use characterization and population modeling with subpixel information measured from remote sensing data

This dissertation provides deeper understanding on the application of Vegetation-Impervious Surface-Soil (V-I-S) model in the urban land use characterization and population modeling, focusing on New Orleans area. Previous research on the V-I-S model used in urban land use classification emphasized on the accuracy improvement while ignoring the discussion of the stability of classifiers. I developed an evaluation framework by using randomization techniques and decision tree method to assess and compare the performance of classifiers and input features. The proposed evaluation framework is applied to demonstrate the superiority of V-I-S fractions and LST for urban land use classification. It could also be applied to the assessment of input features and classifiers for other remote sensing image classification context. An innovative urban land use classification based on the V-I-S model is implemented and tested in this dissertation. Due to the shape of the V-I-S bivariate histogram that resembles topological surfaces, a pattern that honors the Lu-Weng’s urban model, the V-I-S feature space is rasterized into grey-scale image and subsequently partitioned by marker-controlled watershed segmentation, leading to an urban land use classification. This new approach is proven to be insensitive to the selection of initial markers as long as they are positioned around the underlying watershed centers. This dissertation links the population distribution of New Orleans with its physiogeographic conditions indicated by the V-I-S sub-pixel composition and the land use information. It shows that the V-I-S fractions cannot be directly used to model the population distribution. Both the OLS and GWR models produced poor model fit. In contrast, the land use information extracted from the V-I-S information and LST significantly improved regression models. A three-class land use model is fitted adequately. The GWR model reveals the spatial nonstationarity as the relationship between the population distribution and the land use is relatively poor in the city center and becomes stronger towards the city fringe, depicting a classic urban concentric pattern. It highlighted that New Orleans is a complex metropolitan area, and its population distribution cannot be fully modeled with the physiogeographic measurements.

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Social and Behavioral Sciences

Grantor

Geography and Anthropology

Year dc:date.available

2012

Author dc:creator

• Tang, Quan

• evaluation framework
• population modeling
• urban remote sensing
• kappa-error diagram
• GWR
• V-I-S model
• watershed segmentation

Statement dc:rights

• unrestricted
• Release the entire work immediately for access worldwide.

Identifier

etd-06192012-110252
https://repository.lsu.edu/gradschool_dissertations/1282

OAI identifier oai:identifier

oai:repository.lsu.edu:gradschool_dissertations-2281