基于神经网络的玛纳斯河流域植被地上生物量反演

Inversion of Vegetable Aboveground Biomass in the Manas River Basin Based on Neural Network

植被生物量反映了生态系统获取能量的能力,分析其分布特征对了解生态系统结构和功能具有十分重要的意义。传统的反演植被地上生物量的方法往往由于样本的缺少,以及影响因子的不确定性而导致预估精度不高。本文选用ELM对105块实测样本的遥感因子(TM影像灰度值和植被因子等10个因子)进行训练,用余下34块样地进行验证,结果表明:ELM反演植被地上生物量,可以获得较高的精度,模型预测结果与实测结果的曲线拟合决定系数R^2达0. 89。此外,对2010-2015年玛纳斯河流域的植被地上生物量进行反演,认为流域内上游山区生物量大部分较为稳定,中游平原区生物量呈现增加趋势,下游荒漠区生物量则呈现退化趋势。

Aboveground biomass reflects the capability of ecosystems to obtain energy. Analysis on the spatial distribution pattern is of great significance for understanding the structure and function of ecosystems. The accuracy of inverting aboveground biomass with the conventional approach is low due to the lack of samples and the uncertainty of impact factors. In this study,Extreme Learning Machine (ELM) was used to train the remote sensing factors of 105 samples which included seven-band pixel values of TM image and vegetation factors,and the remaining 34 samples were used for verification. The results confirmed that ELM approach could invert vegetable aboveground biomass with a higher accuracy,and its determination coefficient of curve fitting reached 0. 89. In addition,the inversion of vegetation aboveground biomass in the Manas River Basin from 2010 to 2015 found that the biomass was relatively stable in the upper area of the Manas River Basin,was in an increase trend in the middle plains and was in a deterioration trend in the downstream desert.