Abstract
Frequent human activities resulted by rapid urbanization lead to a variety of urban-related environmental and socio-economic issues. Therefore, for effective environmental management and urban planning, monitoring urban growth and detecting its resulting land use and land cover (LULC) change is very important. Most of the previous studies focused on bi-temporal or coarsely multi-temporal change detection to extract stationary change information over a time span. However, higher-order change information, for instance, acceleration or deceleration of urban growth, which would not be observed by bi-temporal method, is more meaningful information for policy makers to understand the urbanization process. With the free access to the USGS Landsat archive and development of remote sensing techniques, detecting urban growth pattern (intensification or sprawl) and LULC change dynamics with temporally high frequent datasets become possible. In this study, bi-temporal, multi-temporal and long-term annual change detection were applied to the Region of Waterloo, Ontario, Canada, to identify the urban growth pattern and LULC change dynamics. Classification was performed for each scene to extract LULC information from 1984 to 2013. This study demonstrates that machine learning classifiers, such as support vector machine (SVM), random forest (RF) and artificial neural network (ANN), perform better than classical maximum likelihood classifier (MLC), among which SVM performs the best. Total urban built-up area of the Region of Water increased from 30% in 1984 to 55% in 2013, replacing large area of vegetated area (agricultural lands and grassland. Outward (sprawl) and inward (intensification) growth patterns were detected both spatially and temporally. Within this time iv span, built-up area experienced a relatively accelerating growth in 1990s and in early 2000s. In terms of long-term record, Kitchener had the fastest growing rate of low-density built-up (residential) area. The coverage of high-density built-up (commercial and industrial) area in Cambridge increased most dramatically. And the built-up area of Waterloo experienced the lowest growth rate. These important findings indicate that using long-term temporal-dense Landsat datasets enables monitoring of urban growth and LULC change dynamics. Such valuable long-term results can be used for better analysis of urban growth and LULC change patterns for planners and policy makers to comprehensively understand the urbanization process in the past and make better planning in the future.