叶面积指数遥感尺度效应与尺度纠正

Simulation and correction of spatialscaling effects for leaf area index

由于地表空间异质性的普遍存在,遥感反演模型的非线性必然会导致不同分辨率观测的遥感结果不一致,从而产生遥感产品尺度效应。本文研究了遥感产品尺度效应概念、模拟方法和定量计算模型,并利用锡林浩特草原研究区的实测数据,对尺度效应模型和方法进行了定量计算与验证分析。首先,基于不同升尺度方法与多尺度遥感成像机理之间的机理联系,通过"先反演再平均"与"先平均再反演"之间的差异,可计算"高"分辨率与"低"分辨率之间的遥感产品尺度差异。其次,分别以红光、近红外两波段反射率和归一化植被指数(NDVI)为自变量,对叶面积指数(LAI)非线性遥感模型进行泰勒展开,研究了模型非线性、遥感数据空间异质性对LAI遥感产品尺度差异的影响,发现高阶项可忽略,利用二阶导数项和遥感数据方差项可定量计算遥感产品尺度差异,经过二阶导数项纠正后的尺度差异相对偏差从5.6%分别降低到0.78%和1.45%。最后,分析了LAI遥感产品尺度效应的特征规律,得出以下结论:随着植被覆盖的增大,同等遥感空间异质性的LAI遥感产品尺度差异越大,且红光波段比近红外波段的尺度差异敏感性高近2个数量级;对于绝大部分陆地植被区域,存在"低分辨率低估"尺度效应,且遥感产品尺度差异的主导要素为LAI模型非线性,NDVI变量自身非线性对尺度效应贡献占23.5%;对于湿地类植被与水体混合情形,NDVI变量非线性的贡献为主导贡献,出现"低分辨率高估"尺度效应,必须利用红光、近红外两波段的二阶导数项非线性尺度差异,才能解释这一类型的LAI遥感产品尺度效应。本文建立了具有一定普适意义的遥感产品尺度效应定量模拟与尺度纠正方法,对推动定量遥感的尺度问题研究有一定参考价值。

Various remote sensing sensors observe the Earth＇s surface at different spatial resolutions. Due to the spatial heterogeneivy and model＇s nonlinearity, there would be some scale difference among different remote sensing surface parameter （such as leaf area index, LAI） derived fi＇om remote sensing images with different resolution. In this paper, the spatial scale effects and transformation methods are studied using both experiment at Xilinhaote steppe region and theoretic models. Firstly, different upscaling methods were presented to simulate the scale effects between fine resolution and coarse resolution. Secondly, Taylor expansion was conducted for both NDVI model and reflectance model for LAI estimation, and the nonlinearity can be well explained by the second derivatives. The scaling difference was reduced from 5.6% to 1.45% and 0.78%, respectively, if the contributions of the second derivatives were corrected for LAI models based on NDVI and reflectances of red and NIR bands. Finally, the effects of the nonlinearity and heterogeneity on scaling are quantified. It can be observed： （1） the scaling error increases with the vegetation coverage under same spatial heterogeneity; （2） the heterogeneity in red band is about 100 times sensitive to scale error than it in near-infrared band for high NDVI conditions; （3） for terrestrial vegetation region, the LAI would be underestimated at coarse reso- lution. The nonlinearity of the exponent LAI model based on NDVI is the primary factor, and the nonlinearity of NDVI variable contributes about 23.5% scaling difference; （4） for wetland region （mixed by vegetation and water）, the LAI would be overestima- ted at coarse resolution. The nonlinearity of NDVI variable becomes the dominant factor, and the scaling difference can still be corrected by the contribution of the second derivates of the LAI model based on reflectances of red and NIR bands. Therefore, we developed a series methods and models to quantify the scale effect of LAI, and the scaling error was consistent with contributions of the second derivates by Taylor expansion, which can also be applied to other surface parameters.