Abstract: Remote sensing is a very fast and effective way to monitor the grassland biomass,the previous studies are mostly based on the correlation of vegetation index(VI) and biomass.In this paper,the method of partial least squares regression(PLSR) was used to set up the regression and prediction models between grassland biomass and normalized difference vegetation index(NDVI) based on the multi-year average annual maximum normalized difference vegetation index(NDVI) combined with annual rainfall,annual accumulated temperature,and other meteorological materials in Tibet.In addition,this method was also compared with the stepwise regression of ordinary least squares(OLS) method.The results show that there was a strong correlation between grass biomass and the annual maximum NDVI value and the annual rainfall.PLSR achieved better effects of fitting and prediction than the stepwise regression of OLS and the correlation coefficient was 0.895,meanwhile reliable results were obtained.PLSR would provide a new way for data processing in vegetation biomass monitoring by remote sensing because it is particularly effective in the case of more predictor variables,less samples,and existent multicollinearity among variables.

Key words: Partial least squares regression(PLSR), Ordinary least squares regression(OLSR), Biomass, Normalized difference vegetation index(NDVI), Rainfall

摘要： 在多年平均年最大归一化植被指数(NDVI)的基础上,结合西藏地区年降雨量、年积温等气象资料,利用偏最小二乘(partial least squares, PLS)回归方法对数据进行分析并建立西藏地区草地生物量与归一化植被指数、降雨量等解释变量的回归估测模型.并和一般最小二乘法(ordinary least squares, OLS)中的逐步回归法(Stepwise)相比较.结果表明:草地生物量与年最大NDVI值和年降雨量有很强的相关性,偏最小二乘回归在拟合及估测效果上均优于一般最小二乘的逐步回归法,回归方程的相关系数为0.89,取得了较为可靠的结果.偏最小二乘回归在解释变量多、样本个数少、变量间存在多重共线性时尤为有效,为遥感监测植被生物量时的数据处理提供了新的途径.

关键词: 偏最小二乘回归, 一般最小二乘回归, 生物量, NDVI, 降雨量