Multi-Level Classification Based on Trajectory Features of Time Series for Monitoring Impervious Surface Expansions

Beibei Wang,A-Xing Zhu,Changqing Xu,Zhenjie Chen,Yuzhu Hao

doi:10.3390/rs11060640

Abstract

As urbanization has profound effects on global environmental changes, quick and accurate monitoring of the dynamic changes in impervious surfaces is of great significance for environmental protection. The increased spatiotemporal resolution of imagery makes it possible to construct time series to obtain long-time-period and high-accuracy information about impervious surface expansion. In this study, a three-step monitoring method based on time series trajectory segmentation was developed to extract impervious surface expansion using Landsat time series and was applied to the Xinbei District, Changzhou, China, from 2005 to 2017. Firstly, the original time series was segmented and fitted to remove the noise caused by clouds, shadows, and interannual differences, leaving only the trend information. Secondly, the time series trajectory features of impervious surface expansion were described using three phases and four types with nine parameters by analyzing the trajectory characteristics. Thirdly, a multi-level classification method was used to determine the scope of impervious surface expansion, and the expansion time was superimposed to obtain a spatiotemporal distribution map. The proposed method yielded an overall accuracy of 90.58% and a Kappa coefficient of 0.90, demonstrating that Landsat time series remote sensing images could be used effectively in this approach to monitor the spatiotemporal expansion of impervious surfaces.

Highlights

Over the past few decades, human activities have changed the global environment at an unprecedented rate and scale [1]
The proposed method yielded an overall accuracy of 90.58% and a Kappa coefficient of 0.90, demonstrating that Landsat time series remote sensing images could be used effectively in this approach to monitor the spatiotemporal expansion of impervious surfaces
Considering the shortcomings mentioned above, we proposed a multi-level classification approach combining support vector machine (SVM) and decision tree classification based on the trajectory features of the time series

Summary

Introduction

Over the past few decades, human activities have changed the global environment at an unprecedented rate and scale [1]. Urbanization has rapidly increased the impervious surface area and has degraded large swaths of high-quality farmland and ecological settings [2,3,4]. As this impervious surface expansion profoundly affects food security and the ecosystems around the world [5,6], monitoring this trend is important for global sustainable development. The corresponding studies have been focused on finding spectral differences between impervious surfaces and other land covers using single or multiple images. Time series images can better solve the spectral confusion problem between similar land cover types [13,14,15,16] and provide clearer distinctions between vegetation changes caused by phenological changes, interannual variations, and vegetation changes caused by impervious surface changes [17,18]

Methods

Results

Conclusion