Graduate Thesis Or Dissertation
 

Mining Landsat Time-series Imagery to Understand Spatiotemporal Dynamics of the Urban Landscape: A Perspective from Local Climate Zone Mapping

Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/7h149x711

Descriptions

Attribute NameValues
Creator
Abstract
  • We live on an Urban Planet. The current unprecedented urbanization is accompanied by intensive land cover transition and demographic shifts from rural to urban livelihoods. Cities serve as cultural, economic, political centers that facilitate wealth creation and innovation of the society, sustaining human from multi-dimensions with large ecological footprints far beyond the boundaries of the human settlement. To date, the spatial and temporal dynamics of the urban extent have generally been examined through the lens of binary built-up / non-built-up or impervious surface masks derived from remotely sensed data. Although the binary urban map is necessary for presenting urban extension, it fails to reveal the spatial structure, function, and change of different urban zone types within the cities. To address the urgent need for urban landscape mapping that goes beyond urban footprint analysis, the local climate zone (LCZ) scheme has been increasingly used to reveal the urban forms and functions across the globe. As a generic and culturally-neutral classification scheme, the LCZ map has the potential to reveal the urban dynamics in a relatively long-term period. Through three manuscript-based chapters, the objective of this study is to develop a workflow to map urban dynamics into multi-year LCZ maps and to characterize how the urban landscape evolves in terms of its composition and configuration. The manuscript comprising Chapter 2 established best practices for building broad-scale LCZ maps. This study focused on the critical issue of how training data impact broad-scale LCZ mapping. The collection and application of LCZ training areas brings with it two challenges that may affect mapping success. First, because digitizing training areas is a time-consuming task, there is a broad effort in the LCZ mapping community to create a crowdsourced data collection among different experts. However, this strategy likely leads to inconsistencies in labels that could weaken models. Second, the LCZ labeling process typically involves the delineation of large zones from which multiple training samples are drawn, but those samples are likely spatially autocorrelated and lead to overly optimistic estimates of model accuracy. By applying Random Forests and ResNets for LCZ mapping, we found that: (1) The inconsistency in crowdsourced training areas weakened both models’ performance; (2) Spatial autocorrelation in training data led to underestimations of both models’ predictive errors; (3) interplay of these issues results in erroneous interpretation of LCZ maps. We suggested that broad-scale LCZ mapping results should be interpreted with caution. This paper has been subject to peer review and has been published. Chapter 3’s manuscript developed a workflow for multiple-year LCZ mapping. A key challenge in multiple-year mapping is linking temporally-limited training data to dense time-series image stacks to create consistent maps across time. The contributions of this manuscript include: (1) a labeling workflow to collect historical ground-truth land surface labels on recent years’ high-resolution imagery from Google Earth; (2) an evaluation of classification modeling strategy to best capture temporarily-stabilized training data across time; and (3) an approach to evaluate each classification model’s temporal consistency and robustness in land cover prediction across time. Additionally, we assessed the effectiveness of temporal stabilization of time-series Landsat on long-term LCZ mapping. The results showed that a classification model trained in a single year showed poor performance in predicting historical labels, implying that the model should not be directly extended to predict time-series image stack. Instead, models that included historical training data improved LCZ prediction across years. We also found that temporal stabilization improved mapping consistency across time and should be used to pre-process annual composites for time-series mapping. This study can provide insight into how to model urban dynamics as well as other long-term land cover mapping tasks with Landsat time series. In Chapter 4’s manuscript, we utilized the LCZ maps to quantify spatial pattern and temporal change of urban form for eight major metropolitan areas in the United States for the years 1995, 2000, 2005, 2010, 2015, and 2018. With the map products from Chapter 3, we characterized urban dynamics in both horizontal and vertical dimensions using concentric ring analysis and direction analysis. This approach improves on the more traditional methods of urban footprint analysis, in allowing us to evaluate not only urban expansion, but also within-urban transitions. The results show that: (1) The dominant transition at the periphery of urban areas is from natural zones to urban zones comprised of open low-rise types; (2) Although increase in urban area measured by absolute extent was relatively modest, a larger proportion of urban zones changed over the study period compared to the natural zones, with an average annual growth rate of 3.1% (urban zones) versus -1.7% (natural zones); (3) Evaluated from urban core to periphery, the density of different urban types showed different trends: high-rise and compact zones decrease monotonically with increasing the distance from the urban core, whereas the urban densities of the midrise, low-rise, and open urban zones often showed increasing then decreasing densities with increasing the distance from the urban core; (4) Low-rise and open urban types are more spatially aggregated than denser urban types, but the aggregation level of high-rise, midrise, and compact types often was shown to increase over time; (5) Finally, there are considerable disparities in the urban structure for MSAs, and unbalanced urban development can be found in different directions.
Contributor
License
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Rights Statement
Publisher
Peer Reviewed
Language
Embargo reason
  • Pending Publication
Embargo date range
  • 2021-12-21 to 2024-01-22

Relationships

Parents:

This work has no parents.

In Collection:

Items

Thumbnail Title Date Uploaded Visibility Actions
file details XuChunxue2021.pdf 2021-12-21 Public Download