Land degradation causes serious environmental problems in many regions of the world,and although it can be effectively assessed and monitored using a time series of rainfall and a normalized difference vegetation index(NDVI)from remotely-sensed imagery,dividing human-induced land degradation from vegetation dynamics due to climate change is not a trivial task.This paper presented a multilevel statistical modeling of the NDVI-rainfall relationship to detect human-induced land degradation at local and landscape scales in the Ordos Plateau of Inner Mongolia,China,and recognized that anthropogenic activities result in either positive(land restoration and re-vegetation)or negative(degradation)trends.Linear regressions were used to assess the accuracy of the multilevel statistical model.The results show that:(1)land restoration was the dominant process in the Ordos Plateau between 1998 and 2012;(2)the effect of the statistical removal of precipitation revealed areas of human-induced land degradation and improvement,the latter reflecting successful restoration projects and changes in land management in many parts of the Ordos;(3)compared to a simple linear regression,multilevel statistical modeling could be used to analyze the relationship between the NDVI and rainfall and improve the accuracy of detecting the effect of human activities.Additional factors should be included when analyzing the NDVI-rainfall relationship and detecting human-induced loss of vegetation cover in drylands to improve the accuracy of the approach and eliminate some observed non-significant residual trends.
