多源遥感数据在土地覆盖变化监测中的应用

Application of multiple satellite data to land cover change detection

基于MODIS、Landsat TM、Quick Bird数据,对山西芦芽山2000年、2005年和2010年植被变化作了定量和定性分析。结果显示:3年间,部分植被MODIS NDVI年际波动可达5%~10%。1年之内,多数植被MODIS NDVI季节性明显,Landsat TM季节差对应的NDVI差可达0.2。在自然地带的过渡区,因难以区分人类活动引起的变化和季节变化,这种季节差使变化检测面临不确定性。利用2002年和2008年Quick Bird数据对森林采伐迹地对比发现,只有局部地区的植被类型开始草丛向灌木林的演替,说明植被恢复缓慢。由此可以认为,自然植被过渡带植被固有的分类模糊性,人类活动反复干扰—恢复后形成植被的破碎性,以及Landsat数据可能存在的季节不一致性,都会影响土地覆盖分类精度和变化检测可靠性。因此,需要使用多源数据对自然过渡带进行重点分析。

Land cover change detection includes two categories: conversion(qualitative) and modification(quantitative).Both of them have close relation with temporal and spatial scales.The present paper explored land cover change during 2000,2005,and 2010,by means of top-down method,using 250 m-16day MODIS time serials,30 m-1year TM,and0.6 m-6 year Quick-Bird data in Mt.Luya,Shanxi province,China.The results show:(1) Inter-annual MODIS-NDVI variation is estimated at 5%-10%,and intra-year MODIS-NDVI time serial has clear seasonality for most land cover types.The inter-annual change may be due to mixed results of climate variation and human activities.Therefore,it is necessary to de-couple the effect of nature from that of human beings in order to assess the efficiency of ecological project.(2) The intra-year change(seasonality) reflected by MODIS-NDVI means that inconsistency of TM acquisition time may lead to uncertainty for change detection since corresponding NDVI could be as much as 0.2.In transitional eco-zones,it is especially true because it is not easy to differentiate between actual change and seasonal change.To deal with this problem,normalization is recommended to minimize the difference resulted from seasonal variations.(3) Land cover type changes were found at clear-cut deforestation sites,some slow transition from grass to bush can be detected by 0.6 m Quick Bird data,but not by 30 m TM and 250 m MODIS.In conclusion: Landsat data have been successfully used in many operational database developments in the last 40 years; however,this triumph should not obscure its shortfalls in some specific fields.This study is such a case where vegetations have experienced a long history of disturbance-recovery,thus they may be not a typical land cover type in a 30-m scale.For this naturally transitional eco-zone,plus frequent human disturbance,Landsat data only is far from enough,let alone its inconsistent acquisition time.So,multi-source data should be applied to handle the uncertainties.