植被波谱空间尺度效应及尺度转换方法初步研究

Preliminary Research on Scale Effect and Scaling-up of the Vegetation Spectrum

基于遥感图像的地物识别需要大量地物波谱数据的支持,而现有地物波谱库中收集到的同一地物的波谱,由于其测量尺度和方法的差异,波谱也存在很大差异。以冬小麦为例,首先介绍了材料波谱、端元波谱和像元波谱这3种不同尺度波谱的概念,并以实验数据分析了不同测量尺度下波谱的差别,以此说明波谱尺度转换的必要性。然后利用物理模型和统计模型建立不同测量尺度下的波谱转换关系。分别验证了SAILH模型和线性光谱混合模型在波谱转换中的精度。研究表明,在大尺度上采用统计模型,在小尺度上采用非线性的物理模型可以解释不同尺度观测植被波谱之间的差异。

Remote sensing imagery provides vast information about the land surface,and the spatial distribution of land cover types by classification could be obtained.Moreover,the spectrum of land surface objects is useful to improve the accuracy of image classification.However,the spectrum of the same object may be different when they are measured at different measuring scale and with different method.For example,the spectrum of winter-wheat extracted from Landsat TM and measured in the field are different.So it is important to study the scale effect and scaling method on the spectrum at different measuring scale.In this paper,we took the winter wheat as example,and selected Shunyi region in Beijing as our study area.Firstly the definition of three-scale spectrum was explained,then we analyzed the discrimination using the measuring data to highlight the importance of the scale transformation of spectrum.The collected data included: field measured spectrum of leaf and canopy and the hyper-spectrum high-resolution remote sensed imagery OMIS,Landsat TM and MODIS data.We compared the winter-wheat spectrums and calculated the slope of 'red-edge' at different measuring scales.We also studied the scaling-up method of the spectrum,and the physical model(SAILH) and statistical model(Linear mixing model) were used to describe the relationship between the spectrum at different measuring scales.SAILH is a typical radiation transfer model,which can be used to simulate the canopy spectrum by taking the leaf spectrum,some structural parameters and environmental variables as inputs.In this experiment,the input parameters were acquired with high accuracy,so the error of simulation result is very small: 8.45%.Linear mixing model was used to describe the relationship between endmember spectrum and pixel spectrum.The resolution of MODIS imagery(visible and infrared bands) is 250m,which was taken as pixel spectrum,and the endmember one can be got by multiple methods,here we adopt two methods: Broadman method from MODIS imagery and aggregation one from TM imagery.The unmixing results were compared and analyzed,and the linear mixing model was validated.Through the spectrum data in the study area,the winter-wheat spectrum of leaf,canopy and OMIS imagery is different and the character reflecting the plant growing status is also varied.As for the scaling method,we found statistical models and physical models are fit on the three research scales respectively.However,the endmember selecting method from the imagery also needs more improvement,and more statistical models or coupling physical models should be explored in the further work.