中文摘要：

MODIS等全球叶面积指数（LAI）产品空间分辨率偏低（250m～7km）,不能满足高空间分辨率遥感应用的需求。为获取大区域高空间分辨率LAI,有必要对物理模型用于高空间分辨率遥感影像LAI反演的可行性进行探讨。本文基于4-scale模型LAI反演算法,以甘肃省张掖为研究区,利用TM影像实现研究区人工林LAI反演。反演算法考虑了反射率入射-观测角度对LAI与植被指数关系的影响和植被冠层尺度的集聚程度。利用地面实测LAI数据对反演结果进行验证与分析,并与NDVI经验模型进行对比,同时分析LAI反演结果对波段反射率敏感性。结果表明：4-scale模型LAI反演结果与实测LAI一致性良好（R2=0.67,RMSE=0.50）,且优于NDVI经验模型（R2=0.59,RMSE=0.67）;当LAI大于2时,4-scale模型LAI反演算法误差小于NDVI经验模型,能有效避免植被指数的饱和现象;红光波段反射率减小时,引起4-scale模型LAI反演结果的变化幅度比其增大时更高,且影响程度大于近红外波段反射率。研究表明,4-scale模型LAI反演算法可用于TM数据反演人工林LAI,模型应用普适性较强。

英文摘要：

Global leaf area index（LAI） products such as MODIS LAI product have relatively low spatial resolution（250 m-7 km） and thus can not meet the needs of high spatial resolution remote sensing applications.It is necessary to explore the feasibility of the algorithm based on physical model for LAI retrieval using high spatial resolution remote sensing imagery.This study utilizes the algorithm based on 4-scale model to retrieve LAI in planted forest from TM imagery.The bidirectional reflectance distribution function（BRDF） and clumping representation at canopy scale are both considered in the algorithm.A validation study is conducted with in-situ measurements of LAI in planted forest from Zhangye City,Gansu Province.For comparison,the empirical model using NDVI as predicted variable is also considered for LAI estimation.The results show that better fit was found between the LAI produced by the algorithm based on 4-scale model and measured LAI（R2=0.67,RMSE=0.50） than that between LAI predicted by NDVI and measured LAI（R2=0.59,RMSE=0.67）.The accuracy of the algorithm based on 4-scale model is evidently better than that of empirical model when LAI2.Better resistance to saturation limits of vegetation index is observed for the algorithm.Moreover,the sensitivity analysis of inversed LAI to band reflectance is carried out.For red band,LAI produced from the algorithm is more influenced by the decreasing reflectance than by the increasing condition.And LAI was more sensitive to reflectance at red band（ρred） than that at near infrared band（ρnir）,with uncertainty value of reflectance range from-10% to-30%.This study proved the effectiveness of the algorithm based on 4-scale model in LAI estimation from TM imagery in planted forest and will be helpful in further development of physical models for high spatial resolution LAI retrieval.