Authors: Hsiao-chien Shih*, San Diego State University, Douglas Stow, San Diego State University, Dar Roberts, University of California, Santa Barbara, John Weeks, San Diego State University, Konstadinos Goulias, University of California, Santa Barbara
Topics: Remote Sensing, Land Use and Land Cover Change, Population Geography
Keywords: spectral mixture analysis, time series analysis, machine learning classification, urbanization, urban population growth
Session Type: Paper
Presentation File: No File Uploaded
The relative timing of urban land use change and population growth was explored based on a case study for north Taiwan from 1990 to 2015. An efficient approach to estimating areal coverage of urban land use types in an urbanizing region over time is to generate an urban expansion map labeled with the date that urbanization commenced, and then conduct an overlay analysis with conditional statement on an accurate land use map for the end date of study period. An approach to identifying urban expansion time was developed based on dense time series of Vegetation-Impervious-Soil (V-I-S) maps derived from Landsat imagery. The identified location and time for urbanized lands were accurate, with 80% of urban expansion estimated within ± 2.4 years.
To derive an accurate land use map for the end date of the study period, a random forest (RF) classifier was applied to a 2015 Landsat image. The top 10 most contributing features for the classifier were the seven spectral wavebands, textural homogeneity of V-I-S, and V temporal variation, which yielded the most accurate map compared to other feature inputs. Thus, areal coverages of annual urban land uses were derived.
The relative timing and general relationship between urban land use and population dynamics was explored at the district level. Residential land use and change was most related to population and change. Population growth occurred 2.5 years later than Residential land expansion based on the median time lag.