摘要
为对比同一背景下不同人类干扰程度的荒漠土壤有机质含量的预测模型,以天山北麓阜康市的土壤为研究对象,通过对无人干扰区、人为干扰区全样本和剔除有机质质量分数大于2%的样点的原始光谱反射率进行6种光谱变换,分析不同变换形式与有机质含量的相关性,以相关系数通过P=0.01和0.05水平上显著性检验的敏感波段为自变量,运用多元逐步回归、偏最小二乘回归以及主成分回归法分别建立了无人干扰区、人为干扰区土壤有机质高光谱的预测模型,并选择精度最高的为最优模型。结果表明:(1)无人干扰区与人为干扰区的原始光谱所有波段与有机质含量的相关性都没有通过0.01水平的显著性检验。将有机质质量分数大于2%的样点剔除后,有机质含量与原始光谱反射率的相关系数都大于全样本且有部分波段通过了0.01水平的显著性检验。(2)不论采用何种方法建立的全样本无人干扰区和人为干扰区的预测有机质模型的RPD均小于1.4,不具有预测有机质含量的能力。其中全样本无人干扰区一阶微分、人为干扰区倒数一阶微分多元逐步回归模型是其所有模型中,建模精度最高的,R^2分别为0.652、0.512,但是其RPD仅分别是0.662、0.655,表明模型的预测能力很差。(3)剔除有机质质量分数大于2%的样点之后,预测效果最好的是无人干扰区一阶微分多元逐步回归模型,R^2达到0.776,RMSE为1.408,RPD为2.136;而人为干扰区的二阶微分模型预测效果最优,R^2为0.542,RMSE为2.261,RPD为2.087。
In order to compare the estimation models about the organic matter in desert soil at different levels of human disturbance under the same background, this study took the soil in Fukang City, which is at the north Xin- jiang, of Tianshan Mountains, as the study object and established the hyperspectral-based soil organic matter （SOM） estimation models for soils in the undisturbed area and in the human-disturbed area respectively using the multiple stepwise regression, partial least squares regression and principal component regression method after 6 spectral trans- formations of the original spectral reflectance. The correlation between the transformation and the SOM was analyzed using the sensitive bands as the independent variables which had passed the P =0.01 and 0.05 level of significance test. The results indicated as follows： （1） All the correlation coefficients between the original spectral bands and the SOM in both the undisturbed area and human-disturbed area did not pass the 0.01 level of significance test. Af- ter removing the samples with the mass fraction being greater than 2%, the correlation coefficients were bigger than those concerning all samples and a portion of the bands had passed the 0.01 level of significance test. （2） Regard- less of the method in establishing the SOM estimation model for the full sample size in undisturbed area or in hu- man-disturbed area, the RPD of the models were less than 1.4 which indicated they were not capable of predicting organic content. The first order differential model for all samples in the undisturbed area, and the first order differen- tial multiple stepwise regression model of the reciprocals in the human-disturbed area had the highest precision a- mong all the models with R2 being 0.652 and 0. 512 respectively. （3） After removing the samples whose mass frac- tion were greater than 2% ,for the undisturbed area the best estimation model was the first order differential multiple stepwise regression model with R2 being 0.776 ,the RMSE being
作者
郑曼迪
熊黑钢
乔娟峰
刘靖朝
ZHENG Man-di;XIONG Hei-gang;QIAO Juan-feng;LIU Jing-chao(College of Resource and Environment Science ,Xinjiang University, Urumqi 830046 ,Xinjiang, China;College of Art and Science,Beijing Union University,Beijing 100083 ,China)
出处
《干旱区地理》
CSCD
北大核心
2018年第2期384-392,共9页
Arid Land Geography
基金
国家自然科学基金(41671198)
关键词
干扰程度
荒漠土
有机质
估算模型
level of human disturbance
desert soil
organic matter
estimation models
