摘要
监测土壤有机质含量状况，可为土壤肥力诊断及土壤资源的合理开发利用提供科学依据。本研究通过对南疆191个水稻土样品的反射率数据进行连续统去除处理后，构建了有机质连续统去除光谱指数并提取了850 - 1 380、1 380 - 1 550、1 730 - 2 150、2 150 - 2 380 nm 4个波段的吸收特征参数，据此建立了多种定量反演模型。结果表明：经连续统去除后，有机质的吸收特征得到了有效放大，不同有机质含量的连续统去除曲线在850 - 1 380 nm，其有机质含量与连续统去除值呈正相关，与吸收面积呈负相关，而在1 730 - 2 150 nm波段则呈现相反的规律。反射率连续统去除值与有机质含量的相关性要优于反射率与之的相关性，而反射率一阶微分与连续统去除一阶微分与有机质的相关性差异不明显。不同有机质光谱指数模型之间的建模参数与预测能力差异不大，但均只具备初略估测有机质的能力。吸收特征参数模型中，仅有850 - 1 380 nm 波段的面积归一化最大吸收深度（NMAD850-1380nm）所建模型具有较好的定量预测能力。以反射率、反射率连续统去除、反射率一阶微分、反射率连续统去除一阶微分所建的PLSR模型均具有较好的预测能力，相对分析误差均大于2.00。所有模型中，连续统去除一阶微分（CR′）模型的决定系数与相对分析误差最高，分别为0.91、2.58，均方根误差最低，其值为5.62，具有最好的预测能力。
Monitoring soil organic matter（SOM） content can provide scientific basis for soil fertility diagnosis and rational utilization of soil resources. Through continuum removal process on reflectance curves of 191 soil samples from southern Xinjiang, SOM continuum removal spectral indexes were constructed and four absorption parameters were identified in accordance with 850–1380, 1380–1550, 1730–2150 and 2150–2380 nm, respectively. Subsequently, a number of quantitative inversion models were established to illustrate SOM contents. The results showed that absorption characteristics of SOM were amplified owing to the continuum removal reflectance, and SOM content was positively correlated with continuum removal values and negatively correlated with absorption area in 850–1380 nm, however, inverse correlations were observed in 1730– 2150 nm. Additionally, the correlation between SOM content and values of continuum removal reflectance was better than that between SOM content and reflectance, meanwhile, the correlation between SOM content and first derivative of continuum removal reflectance was not significantly different from that between SOM content and first derivative of reflectance. As a whole, there were no significant differences in predictive accuracies when using exponential models based on SOM spectral indexes, and these models could estimate roughly SOM content. The model generated by NMAD850-1380nm （normalization maximum absorption depth area in 850–1380 nm） could predict SOM content better among all models based on absorption characteristic parameters. All of the PLSR models, which were established based on reflectances（R）, continuum removal（CR）, first derivative of reflectance （R′） or first derivative of continuum removal reflectance （CR′） values, respectively, could predict SOM content with all RPD values higher than 2.00, while the CR′ model was the best with highest R2 （0.91） and RPD （2.58） but lowest RMSE （5.62）.
出处
《土壤》
CAS
CSCD
北大核心
2016年第2期389-394,共6页
Soils
基金
国家自然科学基金项目(41271234
41061031
41261083
41361048)资助
关键词
连续统去除
高光谱
水稻土
有机质含量
预测
Continuum removal
Hyperspectral
Paddy soil
Soil organic matter content
Predicting
作者简介
向红英（1980-），女，湖南常德人，助理研究员，主要从事干旱区资源与环境的遥感监测与评价方面的研究。E-mail：hongyingxiang@163．com
通讯作者（pjzky@163．com）
