论文部分内容阅读
叶面积指数LAI是众多气象、环境、农业等模型的关键输入参数。尽管具有多个传感器的全球LAI产品已经相继发布,但是由于受反演方法的局限性以及反射率产品质量的影响,这些由单一传感器数据得到的LAI产品在时间上表现出一定的不连续性,这与自然生长植被的LAI变化规律不能一致。而神经网络在对复杂的、非线性数据的模式识别能力方面具有出色的表现。如在3层神经网络中,只要对隐层采用非线性递增映射函数,输出层采用线性映射函数,就可以用于对任意连续函数进行逼近。对于具有相同植被覆盖类型的同一地点多年的LAI数据,在无自然灾害和人为破坏的前提下,可以构成一个非线性的、连续的时间序列。通过融合MODIS和VEGETATION两种传感器产品,在利用相同植被类型的LAI时间序列来建立自回归神经网络,即NARX神经网络的同时,引入红、近红外和短波红外3个波段上时间序列的反射率以及相应的太阳天顶角、观测天顶角和相对方位角作为NARX神经网络的外部输入变量,并最终达到估算时间序列LAI的目的。验证结果表明,NARX神经网络非常适用于时间序列的LAI估算,并且其预测的LAI比原始的MODIS LAI在时间序列上表现的更连续和平滑。因此,该方法在改进典型植被类型的LAI遥感数据产品质量方面具有一定的应用价值。
LAI is a key input parameter for many meteorological, environmental and agricultural models. Although global LAI products with multiple sensors have been released one after another, these LAI products obtained from single sensor data show some discontinuities in time due to the limitations of inversion methods and the influence of reflectivity product quality. However, This can not be consistent with the change law of LAI of naturally growing vegetation. Neural networks perform well in pattern recognition of complex, non-linear data. For example, in the 3-layer neural network, as long as the hidden layer adopts the non-linear incremental mapping function and the output layer adopts the linear mapping function, it can be used to approximate any continuous function. For many years of LAI data of the same place with the same vegetation cover type, a non-linear, continuous time series can be constructed without natural disasters and man-made damage. By combining MODIS and VEGETATION sensor products, the LAI time series of the same vegetation type is used to establish the autoregressive neural network, NARX neural network, and the time series reflectivity of red, near infrared and shortwave infrared And the corresponding solar zenith angle, observed zenith angle and relative azimuth as the external input variables of NARX neural network, and finally achieve the purpose of estimating the time series LAI. The verification results show that the NARX neural network is very suitable for time series LAI estimation, and the predicted LAI is more continuous and smooth than the original MODIS LAI in time series. Therefore, this method has certain application value in improving the quality of LAI remote sensing data of typical vegetation types.