基于相空间重构与RBF神经网络模型的面上干旱预测研究

田 苗<sup>1</sup>; 王鹏新<sup>1*</sup>; 侯姗姗<sup>2</sup>; 韩 萍<sup>3</sup>

田苗,王鹏新，侯姗姗，韩萍.基于相空间重构与RBF神经网络模型的面上干旱预测研究[J].干旱地区农业研究,2013,31(6):164~168

基于相空间重构与RBF神经网络模型的面上干旱预测研究

Research of regional drought forecasting based on phase space reconstruction and RBF neural network model

DOI：10.7606/j.issn.1000-7601.2013.06.029

中文关键词: 条件植被温度指数干旱预测相空间重构神经网络 RBF

英文关键词:vegetation temperature condition index drought forecasting phase space reconstruction neural network RBF

基金项目:国家自然科学基金项目(41071235，40871159)；高等学校博士学科点专项科研基金项目（20100008110031）

作者	单位
田苗¹,王鹏新^1*，侯姗姗²，韩萍³	（1.中国农业大学信息与电气工程学院，北京 100083； 2.中国科学院对地观测与数字地球科学中心，北京 100094； 3.中国农业大学理学院，北京 100193）

摘要点击次数: 2179

全文下载次数: 1423

中文摘要:

条件植被温度指数(VTCI)是一种适合关中平原的近实时定量化的干旱监测方法，在前期基于以旬为单位的VTCI样本点上相空间重构与RBF神经网络干旱预测研究的基础上，进一步进行了VTCI遥感面上的干旱预测研究。通过分析样本点VTCI时间序列的延迟时间和重构维数，确定整个面上VTCI时间序列相空间维数为7，从而对面上VTCI数据进行了相空间重构。对重构后的VTCI数据应用RBF神经网络模型预测得到了2009年4月上旬到5月中旬的VTCI预测结果。结果表明，多旬预测结果都较好地反映了监测结果的特征，各旬预测结果的绝对误差频数分布主要集中在-0.2到0.2之间。应用Kappa系数评价预测结果与监测结果的一致性程度：5月中旬为显著，4月上旬和中旬为中度，4月下旬和5月上旬的一致性为弱，但阳性一致率较高。该模型的面上预测精度较好，适合关中平原的干旱预测研究。

英文摘要:

The vegetation temperature condition index (VTCI) was a real time and quantification drought monitoring method which was suitable to Guanzhong Plain. Based on the earlier research of the drought forecasting of the phase space reconstruction on the VTCI samples in a period of ten days and R BF neural network, further carried out the drought forecasting research of the VTCI by the regional remote sensing. Through analysis of the delay time and reconstruction dimension of the sample VTCI time series, has determined the phase space dimension in whole region VTCI time series was 7. Thereby has carried out the phase space reconstruction for the regional VTCI data. Applied the neural netwo rk model on the reconstructed VTCI data to do forecast and obtained the forecasting results from early April to middle May of 2009. The result shown that: The multi-period forecasting results can be welll reflected the feature of the monitoring result, and the absolute error frequency in each ten days period was mainly distributed between -0.2 to 0.2. Applied the Kappa Coefficient to evaluate the consistence of the forecasting result with monitoring results: In middle of May was significant, in first and middle of April was moderate, and in late of April and early of May, the consistence was weak, but the positive consistence was high. These results indicated that this forecasting model can be suitable to the drought forecasting in Guanzhong Plain.

查看全文查看/发表评论下载PDF阅读器