摘要
为探索基于光学卫星遥感数据的冬小麦地上生物量估算方法,该研究通过3 a田间试验,获取冬小麦4个关键生育期(拔节期、抽穗期、开花期和灌浆期)和3种施氮水平下的地上生物量以及对应的近地冠层高光谱反射率数据。通过将高光谱数据重采样为具有红边波段的RapidEye、Sentinel-2和WorldView-2卫星波段反射率数据,构建任意两波段归一化植被指数。同时,将卫星波段反射率数据与6种机器学习和深度学习算法相结合,构建冬小麦生物量估算模型。研究结果表明:任意两波段构建的最佳植被指数在冬小麦开花期对生物量的敏感性最强(决定系数R^(2)为0.50~0.56)。在不同施氮水平条件下,高施氮水平增强了植被指数对生物量的敏感性。Sentinel-2波段数据所构建的植被指数优于其他两颗卫星波段数据。对6种机器学习和深度学习算法,总的来说,基于深度神经网络(Deep Neural Networks,DNN)算法所构建的模型要优于其他算法。在单一生育期中,在拔节期(R^(2)为0.69~0.78,归一化均方根误差为26%~31%)和开花期(R^(2)为0.69~0.70,归一化均方根误差为24%~25%)的估算精度最高。Sentinel-2波段数据与DNN算法结合的估算精度最高,在全生育期中预测精度R^(2)为0.70。施氮水平的提高同样增强了DNN模型的估算精度,3颗卫星波段数据在300 kg/hm^(2)施氮条件下的预测精度R^(2)都在0.71及以上,均方根误差小于219 g/m^(2)。研究结果揭示了光学卫星遥感数据在不同生育期和施氮条件下估算冬小麦生物量的潜力。
Crops biomass has been one of the most important indicators to predict the plant growth status and crop yield.This study aims to estimate the dry biomass of the winter wheat aboveground using optical satellite remote sensing.The winter wheat biomass was acquired at four growth stages(jointing,heading,flowering,and filling stage),and three nitrogen treatments(N1,N2,and N3,application rates of 0,150 and 300 kg/hm^(2))in 2011,2012,and 2014 at the agricultural meteorological experiment station,Nanjing University of Information Science and Technology,Nanjing,China.Simultaneously,the narrow-band spectral reflectance of canopy was also collected from the winter wheat using Analytical Spectral Device(ASD).Afterwards,the hyperspectral remote sensing data was resampled into the broad band reflectance of RapidEye,Sentinel-2,and WorldView-2 satellites with the red edge bands using the satellite spectral response functions.All possible band combinations of Normalized Difference Vegetation Index like(NDVI-like)were validated in the different growth stages and nitrogen treatments.Meanwhile,the satellite broad-band reflectance was used as the inputs for the six machine and deep learning for the biomass estimation,including the Random Forest(RF),Support Vector Regression(SVR),Partial Least Squares Regression(PLSR),Deep Neural Network(DNN),Long-and Short-Term Memory recursive neural network(LSTM),and one-dimensional Convolutional Neural Network(1D-CNN).Finally,the models were developed using Leave-One-Out cross validation under different growth stages and nitrogen treatments.The results showed that the optimal NDVI-like vegetation indices that derived from the two arbitrary bands presented the highest sensitivity to the winter wheat biomass at the flowering stage(coefficient of determination,R^(2)=0.50-0.56).It was also difficult to accurately estimate the biomass of winter wheat using only one vegetation index in the whole growth period.The nitrogen treatments were dominated the correlation between the vegetation indices and winter wheat biomass.Specifically,the high nitrogen treatment was enhanced the sensitivity of vegetation index to the winter wheat biomass.The vegetation index with the Sentinel-2 bands performed better than that with the rest.The R^(2) was over 0.50 between the vegetation index and biomass at the jointing and flowering stages.The best performance was achieved in the estimation model of winter wheat biomass using DNN among the six models.The performance of DNN-based model was also depended on the growth stages and nitrogen treatments.In the single growth stage,the highest estimation accuracy was observed at the jointing stage(R^(2)=0.69-0.78 and the normalized root mean square error(NRMSE)=0.26-0.31),and at the flowering stage(R^(2)=0.69-0.70 and NRMSE=0.24-0.35).The highest estimation accuracy was obtained in the DNN-based model with the Sentinel-2 bands as the inputs,indicating the R^(2) of 0.70 in the whole growth period.The high nitrogen treatment was also enhanced the estimation accuracy of DNN model,where the R^(2) was not lower than 0.71 and RMSE was within 219 g/m^(2) at the N3 condition for all the three satellite bands.Therefore,the optical satellite remote sensing data can be expected to estimate the winter wheat biomass under the different growth stages and nitrogen treatment conditions.
作者
兰仕浩
李映雪
吴芳
邹晓晨
Lan Shihao;Li Yingxue;Wu Fang;Zou Xiaochen(School of Remote Sensing and Geomatics Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China;Schoole of Applied Meteorology,Nanjing University of Information Science and Technology,Nanjing 210044,China;Xinghua Meteorological Bureau,Xinghua 225700,China)
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2022年第24期118-128,共11页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家自然科学基金项目(41801243)。
