文章摘要
李柳阳, 朱青, 刘亚, 廖凯华, 赖晓明.基于气象因子的长三角地区农田站点土壤水分时间序列预测[J].水土保持学报,2021,35(2):131~137
基于气象因子的长三角地区农田站点土壤水分时间序列预测
Time Series Prediction of In-situ Soil Moisture Based on Meteorological Data in the Yangtze River Delta
投稿时间:2020-08-28  
DOI:10.13870/j.cnki.stbcxb.2021.02.018
中文关键词: 土壤水文  时空变化  长三角  遥感反演
英文关键词: soil hydrology  temporal and spatial changes  the Yangtze River Delta  remote sensing inversion
基金项目:国家自然科学基金项目(41971117);中国科学院前沿科学重点研究项目(QYZDB-SSW-DQC038)
作者单位E-mail
李柳阳1,2, 朱青1,2, 刘亚1,2, 廖凯华1,2, 赖晓明1,2 1. 中国科学院南京地理与湖泊研究所, 中国科学院流域地理学重点实验室, 南京 210008

2. 中国科学院大学
, 北京 100049 
qzhu@niglas.ac.cn 
摘要点击次数: 20
全文下载次数: 19
中文摘要:
      以长三角3省1市为研究区,旨在构建长三角地区土壤水分长时间序列,为农业生产和遥感算法提供数据支撑。研究基于空间匹配的站点土壤水分数据和气象数据,利用主成分分析得到4个有效主成分作为线性回归和BP神经网络模型的输入因子,建立土壤水分与气象因子间的定量关系,并评估所构建模型的精度。结果表明,基于全部站点数据建立的单一BP神经网络模型优于单一线性回归模型。单一线性回归模型的R2=0.34,RMSE=0.046 m3/m3,MAE=3.67%;而单一BP神经网络模型的训练、验证和测试3个数据集的R2均在0.64以上,RMSE<0.043 m3/m3,MAE低于3.4%。根据逐个站点分别构建分站点的BP神经网络模型,其总体精度高于基于全部站点数据构建的单一BP神经网络模型。分站点构建的BP神经网络模型的总体精度方面,3个数据集的R2均值在0.75以上,RMSE<0.039 m3/m3,MAE低于3%。通过对逐个站点分别构建BP神经网络模型,获得了精度较高、较稳定的土壤水分拟合结果。
英文摘要:
      In order to construct a long-term series of soil moisture in the Yangtze River Delta, which is frequently used for the validation of various satellite products and agricultural production, the linear regression and BP neural network models were constructed respectively to describe the quantitative relationship between soil moisture and meteorological factors. Based on the paired data of in-situ soil moisture and meteorological observation in the Yangtze River Delta including the city of Shanghai and the provinces of Jiangsu, Anhui and Zhejiang, four effective principal components were obtained by principal component analysis as the input of the established models, and the accuracies of these models were evaluated systematically. The results showed that, although both based on all paired data from soil moisture and meteorological observation, the prediction accuracy of the single BP neural network model (the R2 were all above 0.64, and RMSE and MAE were less than 0.043 m3/m3 and 3.4% separately in the sets of the training, validation and testing data) was better than the single linear regression model (R2, RMSE and MAE were 0.34, 0.046 m3/m3 and 3.67% respectively). However, accuracies of both these models were worse than the BP neural network models based on each paired soil moisture and meteorological observation data, with the average of R2above 0.75, and RMSE and MAE less than 0.039 m3/m3 and 3% respectively. By constructing the BP neural network models based on each paired soil moisture and meteorological observation data, a more accurate and stable soil moisture fitting result was obtained.
查看全文   查看/发表评论  下载PDF阅读器
关闭