Locally adaptive change-point detection (LACPD) with applications to environmental changes

Abstract We propose an adaptive-sliding-window approach (LACPD) for the problem of change-point detection in a set time-ordered observations. The proposed method is combined with sub-sampling techniques to compensate lack enough data near time series’ tails. Through simulation study, we analyse its behaviour presence early/middle/late mean, and compare performance some frequently used recently developed methods terms power, type I error probability, area under ROC curves (AUC), absolute bias, variance, root-mean-square (RMSE). conclude that LACPD outperforms other by maintaining low probability. Unlike methods, does not depend on index change-points, it generally has lower bias than alternative methods. Moreover, variance RMSE, when change-points are close tails, whereas shows similar (sometimes slightly poorer) as middle series. Finally, apply our proposal two sets real data: well-known example annual flow Nile river Awsan, Egypt, from 1871 1970, novel remote sensing application consisting 34-year time-series satellite images Normalised Difference Vegetation Index Wadi As-Sirham valley, Saudi Arabia, 1986 2019. good detecting change well magnitude conditions.