基于高光谱和机器学习模型的冬小麦土壤含水率监测研究  

Monitoring of Soil Moisture Content of Winter Wheat Based on Hyperspectral and Machine Learning Models

作　　者：唐子竣 张威 向友珍[1,2] 李志军[1,2] 张富仓[1,2] 陈俊英[1,2] TANG Zijun;ZHANG Wei;XIANG Youzhen;LI Zhijun;ZHANG Fucang;CHEN Junying(Key Laboratory of Agricultural Soil and Water Engineering in Arid and Semiarid Areas,Ministry of Education,Northwest A&F University,Yangling,Shaanxi 712100,China;Institute of Water-saving Agriculture in Arid Areas of China,Northwest A&F University,Yangling,Shaanxi 712100,China)

机构地区：[1]西北农林科技大学旱区农业水土工程教育部重点实验室,陕西杨凌712100 [2]西北农林科技大学中国旱区节水农业研究院,陕西杨凌712100

出　　处：《农业机械学报》2023年第12期350-358,共9页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(52179045)。

摘　　要：为及时获取大田作物根区土壤含水率(Soil moisture content, SMC),实现精准灌溉,运用高光谱技术,通过连续2年(2019—2020年)田间试验采集了冬小麦拔节期不同土层深度SMC及高光谱数据,构建了3类植被指数(蓝、黄和红边面积等三边光谱参数,与冬小麦根区SMC相关性最高的任意两波段植被指数和前人研究与作物参数相关性较好的经验植被指数)并筛选与各土层深度SMC相关系数最高的植被指数,随后将筛选后的植被指数作为模型输入,分别采用随机森林(Random forest, RF)、反向神经网络(Back propagation neural network, BPNN)和极限学习机(Extreme learning machine, ELM)构建冬小麦拔节期不同土层深度SMC估算模型。结果表明,绝大部分三边参数、任意两波段植被指数和经验植被指数在深度0~20 cm土层的SMC相关系数较20~40 cm和40~60 cm更高,在深度0~20 cm土层两波段组合构建的光谱指数与SMC的相关系数最高,均超过0.8,其中RI与SMC的相关系数最高,为0.851,其波长组合为675 nm和695 nm。RF模型是SMC的最佳建模方法,其中深度0~20 cm土层的模型精度最高,估算模型验证集的决定系数R2达0.909,均方根误差(RMSE)为0.008,平均相对误差(MRE)为3.949%。本研究结果可为高光谱监测冬小麦根区SMC提供依据,为快速评估水分胁迫下的作物生长提供应用参考。Aiming to promptly obtain soil moisture content(SMC)in the root zone of field crops for precise irrigation,hyperspectral technology was utilized.Over a 2-year period spanning from 2019 to 2020,during the winter wheat jointing stage,SMC data at varying soil depths and hyperspectral data were collected.Three categories of vegetation indices were created,comprising'trilateral'spectral parameters related to blue,yellow,and red-edge areas,any two-band vegetation indices with the highest correlation to winter wheat root zone SMC,and empirical vegetation indices showing good correlation with crop parameters from previous studies.The vegetation indices exhibited the highest correlation with SMC at different soil depths were selected.Subsequently,random forest(RF),back propagation neural network(BPNN),and extreme learning machine(ELM)were employed to construct SMC estimation models,using the selected vegetation indices as model inputs.The results revealed that a majority of the‘trilateral'spectral parameters spectral indices,any two-band vegetation indices,and empirical vegetation indices displayed stronger correlations with SMC in the O~20 cm soil layer in comparison with the 20~40 cm and 40~60 cm layers.The two-band combinations in the 0~20 cm layer exhibited the highest correlations with SMC,all surpassing O.8.Among which,RI showed the highest correlation with SMC at 0.851,utilizing a wavelength combination of 675 nm and 695 nm.The RF model emerged as the most effective modeling method for SMC,with the highest accuracy observed in the O~20 cm soil layer.The coefficient of determination(R^(2))for the validation set of the estimation model in the O~20 cm layer reached 0.909,and the root mean square error(RMSE)was 0.008,while the mean relative error(MRE)was 3.949%.The outcomes can serve as a foundation for hyperspectral monitoring of winter wheat root zone SMC and provide valuable insights for the rapid assessment of crop growth under water stress.

关 键 词：冬小麦 土壤含水率 高光谱 机器学习 

分 类 号：S512.1[农业科学—作物学] S127