论文部分内容阅读
叶面积指数(LAI)是估算作物生长的关键参数。基于物理模型的LAI反演,被认为是当前最为可靠的方法,但其反演复杂。本文提出了将物理模型和神经网络相结合,从地表反射率反演叶面积指数的算法,利用MOD IS地表反射率和4-scale模型反演作物LAI。(1)利用4-scale模型模拟不同LAI与地表反射率的关系,生成训练数据;(2)利用模型模拟的LAI训练神经网络;(3)以MOD IS地表反射率输入训练后的神经网络,反演LAI。估算的LAI与其他LAI产品进行了比较,结果表明,估算的作物LAI和MOD IS及CYCLOPES LAI产品空间和时间分布一致,均方根误差分别为0.4994和0.6558。以2004年衡水的作物LAI地面观测数据进行了直接验证,估算的LAI与研究区地表植被分布一致,但是,三种卫星LAI产品都小于地表测量,故需针对华北平原浓密作物设计模型参数化方案。
Leaf Area Index (LAI) is a key parameter for estimating crop growth. LAI inversion based on physical model is considered as the most reliable method at present, but its inversion is complicated. In this paper, an algorithm of combining physical model with neural network to retrieve leaf area index from surface reflectance is proposed. Crop LAI is retrieved using MODIS surface reflectance and 4-scale model. (1) using 4-scale model to simulate the relationship between different LAI and surface reflectivity, and generating training data; (2) using LAI training neural network model simulation; (3) MODIS surface reflectivity input training neural network, Inversion LAI. The estimated LAI is compared with other LAI products and the results show that the spatial and temporal distributions of the estimated crop LAI and MOD IS and CYCLOPES LAI products are consistent with rms errors of 0.4994 and 0.6558, respectively. Direct LAI field observations of Hengshui crop in 2004 were conducted. The estimated LAI is consistent with the distribution of surface vegetation in the study area. However, the LAI products of the three satellites are all less than the surface measurements. Therefore, the parameterization scheme of the North China Plain thick-crop design model .