|
Abstract:
|
In geological science, knowing when the urbanization starts or ends would be helpful to understand the urbanization process and investigate further environmental impact. Motivated by change detection for mapping annual urban dynamics in 134 locations, we developed a sparse functional approach of detecting changes in mean function by using three indicators representing vegetation, water and bare land in Landsat data. We modeled the three indicators as a multivariate functional data and calculated multiple annual times series of PC scores by the newly developed multivariate sparse FPCA approach. Estimations based on CUSUM for starting and ending years of gradual change were conducted separately for each component after transformation by an appropriate sigmoid function, and then integrated together to make the final decision. The proposed detection procedure performs well and gets 90% accuracy within one year tolerance, which is better than the latest linear regression approach. Furthermore, the statistic and estimator hold nice asymptotic properties under sparse functional data settings. A simulation study was also conducted to evaluate our method under different circumstances.
|
