基于SARIMA-BP神经网络组合方法的MODIS叶面积指数时间序列建模与预测

发布时间：2018-02-28 22:11

本文关键词： SARIMA BP神经网络 LAI SARIMA-BP神经网络组合方法 LAI时间序列建模与预测　出处：《光谱学与光谱分析》2017年01期 　论文类型：期刊论文

【摘要】：植被叶面积指数(LAI)时间序列的建模及预测是陆面过程模型和遥感数据同化方法的重要组成部分。MODIS数据产品MOD15A2是目前应用最为广泛的LAI数据源之一,然而MODIS LAI时间序列产品包含了一些低质量的数据,例如由于云层、气溶胶等的影响,该产品在时间和空间上缺乏连续性。MODIS LAI时间序列包含线性部分和外在干扰产生的非线性部分,单一的线性方法或非线性方法都不能对其精确建模和预测。首先利用Savitzky-Golay(SG)滤波和线性插值平滑受到干扰的LAI时间序列,然后采用季节自回归积分滑动平均(SARIMA)方法、BP神经网络方法及二者的组合方法(SARIMA-BP)对MODIS LAI时间序列进行建模及预测。在SARIMA-BP神经网络组合方法中,各自在线性与非线性建模的优势得以充分发挥,其中SARIMA方法用于建模及预测LAI时间序列中的线性部分,BP神经网络方法用于对非线性残差部分进行建模及预测。实验结果显示:SG滤波和线性插值后的LAI时间序列比原LAI时间序列更平滑;SARIMA-BP神经网络组合方法的决定系数为0.981,比SARIMA和BP神经网络的0.941和0.884更接近于1;SARIMA-BP神经网络组合方法的预测值同观测值之间的相关系数为0.991,高于SARIMA(0.971)和BP神经网络(0.942)的相关系数。由此得出结论:SARIMA-BP神经网络组合方法对MODIS LAI时间序列具有更好的适应性,其建模和预测准确性高于SARIMA方法或BP神经网络方法。
[Abstract]:Modeling and prediction of vegetation leaf area index (Lai) time series is an important part of land surface process model and remote sensing data assimilation method. MODIS data product MOD15A2 is one of the most widely used LAI data sources at present. However, MODIS LAI time series products contain some low-quality data, such as clouds, aerosols, etc. The product lacks continuity in time and space. The MODIS LAI time series consists of linear parts and nonlinear parts produced by external disturbances. Neither a single linear method nor a nonlinear method can accurately model and predict it. Firstly, Savitzky-Golayn filtering and linear interpolation are used to smooth the disturbed LAI time series. Then the MODIS LAI time series is modeled and predicted by using the seasonal autoregressive integral moving average (SARIMA) method and the combined method of BP neural network and SARIMA-BP. in the SARIMA-BP neural network combination method, the BP neural network method is used to model and predict the MODIS LAI time series. Their respective advantages in linear and nonlinear modeling have been brought into full play, The SARIMA method is used to model and predict the linear part of the LAI time series and the BP neural network method is used to model and predict the nonlinear residual part. The experimental results show that the LAI time series ratio after being filtered by the w / SG filter and the linear interpolation is obtained. The determination coefficient of the original LAI time series smoother SARIMA-BP neural network combination method is 0.981, which is closer to the correlation coefficient between the predicted value and the observed value of the SARIMA and BP neural network combination method and the SARIMA-BP neural network combination method, which is higher than that of SARIMA0.971. It is concluded that the combined method of the 1: SARIMA-BP neural network has better adaptability to the MODIS LAI time series. The accuracy of modeling and prediction is higher than that of SARIMA or BP neural network.
【作者单位】：中国科学院东北地理与农业生态研究所;中国科学院大学;Lancaster
【基金】：国家自然科学基金项目(41271196) 中国科学院重点部署项目(KZZD-EW-07-02)资助
【分类号】：Q948;TP79

【相似文献】