| 李港,孔亚聪,代元帅,吕新.基于野外原位光谱的棉田耕层土壤全氮含量监测模型研究[J].干旱地区农业研究,2023,(6):273~280 |
| 基于野外原位光谱的棉田耕层土壤全氮含量监测模型研究 |
| Study on monitoring model for total nitrogen content in plow layer of cotton field based on field in\|situ spectroscopy |
| |
| DOI：10.7606/j.issn.1000-7601.2023.06.29 |
| 中文关键词: 土壤全氮 原位光谱 全氮监测 土壤质地 土壤耕层 |
| 英文关键词:soil total nitrogen in\|situ spectroscopy soil total nitrogen monitoring soil texture plow layer |
| 基金项目:新疆生产建设兵团重点领域创新团队项目(2018CB004)；石河子大学创新发展专项(CXFZ201903) |
|
| 摘要点击次数: 445 |
| 全文下载次数: 429 |
| 中文摘要: |
| 针对可见光和近红外光穿透力不强、难以监测到耕层土壤全氮的问题，以石河子垦区播前棉田土壤为研究对象，获取3种不同土壤质地类型的原位光谱及不同耕层的土壤全氮含量信息，通过Savitzky-Golay平滑处理及最大归一化处理对土壤原位光谱进行预处理，并利用广义传播神经网络(GRNN)、随机森林回归(RFR)、支持向量机回归(SVR)、最小二乘回归等4种建模方法，建立和筛选出不同土壤质地类型的棉田耕层土壤含量的最佳监测模型。结果表明：(1)不同建模方法在各耕层监测精度不同，在浅、中、深耕层表现最佳的监测模型均为GRNN模型（R2分别为0.72、0.68、0.63）。(2)优化后的NGRO-GRANN模型监测精度高于GRNN模型，在浅、中、深三个耕层的预测精度提高16.2%~30.2%。(3)基于原位光谱建立不同耕层的土壤全氮监测模型具有较好的监测效果，且大量简化了室内光谱处理的繁琐步骤。该研究为野外原位光谱快速获取棉田播前土壤各耕层养分信息提供了理论基础与技术支撑，具有一定的可行性与鲁棒性。 |
| 英文摘要: |
| In response to the challenge of weak penetration of visible and near\|infrared light, which hinders the detection of total nitrogen in the plow layer soil, this study focuses on pre\|planting cotton soil in the Shihezi Karez area. It aims to obtain in\|situ spectral data of three different soil texture types and the total nitrogen content of different plow layers. The soil in\|situ spectra are preprocessed using Savitzky-Golay smoothing and maximum normalization. Four modeling methods, namely Generalized Regression Neural Network (GRNN), Random Forest Regression (RFR), Support Vector Machine Regression (SVR), and Least Squares Regression, are employed to establish and select the optimal monitoring models for cotton field soil content based on different soil texture types. The results showed that: (1) Different modeling methods exhibit varying monitoring accuracies across the plow layers, with GRNN models consistently delivering the best performance in the shallow, medium, and deep layers, achieving accuracies R2 of 0.72, 0.68, and 0.63, respectively. (2) The optimized NGRO-GRANN model outperforms the GRNN model, with R2 values increasing by 16.2%~30.2% in the shallow, medium, and deep layers. (3) The monitoring models for soil total nitrogen in different plow layers, established based on in\|situ spectral data, demonstrate R2 values greater than 0.6, indicating excellent monitoring performance and significant savings in the cumbersome steps of indoor spectral processing. This study provides a theoretical basis and technical support for the rapid acquisition of nutrient information in different plow layers of pre\|planting cotton soil using in\|situ spectroscopy, demonstrating feasibility and robustness. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |