基于长短期记忆网络与植被指数的春小麦产量估测研究  

Yield Estimation of Spring Wheat Based on Long Short-Term Memory Network and Vegetation Index

作　　者：苟曜 董莉霞[1] 李广[2] 燕振刚[1] 逯玉兰 GOU Yao;DONG Lixia;LI Guang;YAN Zhengang;LU Yulan(College of Information Science and Technology,Gansu Agricultural University,Lanzhou,Gansu 730070,China;College of Forestry,Gansu Agricultural University,Lanzhou,Gansu 730070,China)

机构地区：[1]甘肃农业大学信息科学技术学院,甘肃兰州730070 [2]甘肃农业大学林学院,甘肃兰州730070

出　　处：《作物研究》2024年第5期343-350,共8页Crop Research

基　　金：国家自然科学基金(32360438);甘肃省重点研究发展计划(22YF7FA116);甘肃省财政专项(GSCZZ20160909);甘肃省高等学校产业支撑项目(2022CYZC-41)。

摘　　要：春小麦作为我国北方地区重要的粮食作物,其产量预测和空间分布特征分析对于保障国家粮食安全、推动农业区域的均衡发展以及优化农业资源配置具有极其重要的意义。本研究选取甘肃省兰州市和武威市作为研究区,采用中分辨率成像光谱仪(Moderate Resolution Imaging Spectroradiometer,MODIS)卫星遥感获取2001—2022年的遥感数据,对遥感图像进行预处理,并提取包括归一化差异水分指数(Normalized Difference Water Index,NDWI)在内的7种遥感植被指数。对实测产量数据进行趋势分析,结合长短期记忆循环网络(Long Short-Term Memory,LSTM)构建产量估算模型,对兰州市和武威市的春小麦产量进行预测研究,并与传统估产模型反向传播神经网络(Back Propagation Neural Network,BP)模型对比,研究不同模型的估产性能。结果表明:NDWI在所有植被指数中对产量估算模型的贡献最为显著,其决定系数达到0.69,并且在去除趋势影响后,该指数的决定系数进一步提升至0.77。此外,采用五折交叉验证法对模型进行验证,进一步确保了模型的稳健性和泛化能力。本研究可为市级尺度的春小麦产量估算提供参考。As an important grain crop in Northern China,the yield prediction and spatial distribution characteristics of spring wheat are of great significance for ensuring national food security,promoting the balance of agricultural regional development and optimizing the allocation of agricultural resources.In this study,Lanzhou City and Wuwei City in Gansu Province were selected as case areas,and the remote sensing data from 2001 to 2022 were obtained using MODIS(MOD09A1)satellite remote sensing data,and the remote sensing images were preprocessed,and seven remote sensing vegetation indices including the Normalized Difference Water Index(NDWI)were extracted.By analyzing the trend of the measured yield data and constructing a yield estimation model combined with the Long Short-Term Memory Recurrent Network(LSTM),the spring wheat yield prediction in Lanzhou City and Wuwei City was studied,and the yield estimation performance of different models was studied by comparing with the traditional yield estimation model BP neural network model.The results showed that NDWI contributed the most significantly to the yield estimation model among all vegetation indices,with a coefficient of determination as high as 0.69,and the coefficient of determination of the index was further increased to 0.77 after removing the influence of the trend.In addition,the model was verified by the five-fold cross-validation method,which further ensured the robustness and generalization ability of the model.The results of this study provide an important scientific reference for estimating spring wheat yield at the municipal scale.

关 键 词：春小麦 产量预测 LSTM 植被指数 数据预处理 

分 类 号：S512.12[农业科学—作物学]