Abstract
Abstract. Global, timely, accurate and cost-effective cropland monitoring with a fine spatial resolution will dramatically improve our understanding of the effects of agriculture on greenhouse gases emissions, food safety, and human health. Time-series remote sensing imagery have been shown particularly potential to describe land cover dynamics. The traditional change detection techniques are often not capable of detecting land cover changes within time series that are severely influenced by seasonal difference, which are more likely to generate pseuso changes. Here,we introduced and tested LTSM ( Landsat time-series stacks model), an improved Continuous Change Detection and Classification (CCDC) proposed previously approach to extract spectral trajectories of land surface change using a dense Landsat time-series stacks (LTS). The method is expected to eliminate pseudo changes caused by phenology driven by seasonal patterns. The main idea of the method is that using all available Landsat 8 images within a year, LTSM consisting of two term harmonic function are estimated iteratively for each pixel in each spectral band .LTSM can defines change area by differencing the predicted and observed Landsat images. The LTSM approach was compared with change vector analysis (CVA) method. The results indicated that the LTSM method correctly detected the “true change” without overestimating the “false” one, while CVA pointed out “true change” pixels with a large number of “false changes”. The detection of change areas achieved an overall accuracy of 92.37 %, with a kappa coefficient of 0.676.
Highlights
Land use/cover change is among the main driving factor of global environmental change, affecting ecosystem services and biodiversity (Dixon et al, 1994; Foley et al, 2005; Wulder et al, 2008)
Based on the change detection and classification (CCDC) approach, we developed an improved change detection algorithm called Landsat time-series stacks model (LTSM) for detecting cropland change, which uses all available Landsat images to remove “flase” one. We differ from their method by: (1)using more targeted harmonic model with different frequencies and removing coefficients for inter-annual change to describe cropland change trajectory (Rayner, 1971)(2)using LevenburgMarquardt fitting algorithms instead of Partial Least Squares (PLS) because it is faster and more accurate (Xue et al, 2014) (3) using an interative method based on the ExpectationMaximization (EM) algorithmto determine threshold for defining change (Bruzzone et al, 2000).The main objectives of the paper is test that LTSM is able to compress the pseudo change signal more effectively compared to existing change detection methods
Assuming no land cover change has occurred, in order to test whether the predicted image obtained by time series model can eliminating the pseudo ones caused by cultivated land phenology
Summary
Land use/cover change is among the main driving factor of global environmental change, affecting ecosystem services and biodiversity (Dixon et al, 1994; Foley et al, 2005; Wulder et al, 2008). To minimize phenonlogy differences and make multitemporal image differencing possible, all the image used should be within the same season and at the same time they should be almost cloud and snow free Recognizing these limitation, several approaches have been proposed for analyzing image time series (i.e. trajectory-based change detection methods), such as the Continuous Monitoring of Forest Disturbance Algorithm(CMFDA), the Vegetation Change Tracker(VCT) and the Landsat-based detection of Trends in Disturbance and Recovery (LandTrendr) made full use of the Landsat time-series stacks (LTS) to reconstruct forest disturbance histories (Zhu et al, 2012; Huang et al, 2010; Kennedy et al, 2010). We differ from their method by: (1)using more targeted harmonic model with different frequencies and removing coefficients for inter-annual change to describe cropland change trajectory (Rayner, 1971)(2)using LevenburgMarquardt fitting algorithms instead of Partial Least Squares (PLS) because it is faster and more accurate (Xue et al, 2014) (3) using an interative method based on the ExpectationMaximization (EM) algorithmto determine threshold for defining change (Bruzzone et al, 2000).The main objectives of the paper is test that LTSM is able to compress the pseudo change signal more effectively compared to existing change detection methods
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
