| 乔娟峰,熊黑钢,王小平,郑曼迪,刘靖朝,李荣荣.新疆阜康荒地土壤有机质高光谱特征及其反演模型研究[J].干旱地区农业研究,2018,36(5):207~214 |
| 新疆阜康荒地土壤有机质高光谱特征及其反演模型研究 |
| Wasteland soil organic matter hyperspectral characteristics and inversion model research in Fukang, Xinjiang |
| |
| DOI：10.7606/j.issn.1000-7601.2018.05.30 |
| 中文关键词: 荒地 土壤有机质 高光谱 显著性波段 偏最小二乘回归 |
| 英文关键词:wasteland soil organic matter hyperspectral significant band partial least squares regression method（PLSR） |
| 基金项目:国家自然科学基金“干旱区人类活动胁迫下绿洲水盐时空变化规律研究”(41671198) |
|
| 摘要点击次数: 1072 |
| 全文下载次数: 497 |
| 中文摘要: |
| 针对干旱区荒地土壤贫瘠且有机质含量少，难以快速、准确测定的问题，以阜康中部荒地土壤为研究对象，对64个样点野外光谱进行测定和室内土壤样品农化分析，在原始反射率（R）基础上，利用ENVI5.1软件提取光谱反射率一阶微分(R′)、倒数的对数(lg(1/A))、倒数的对数一阶微分(lg(1/A)′)、去包络线(CR)等4种光谱反射率，分析了5种光谱反射率的变换形式与土壤有机质含量的相关性，基于全波段（450～2 350 nm）和显著性波段（相关系数通过P=0.01水平检验），利用偏最小二乘法回归（PLSR）建立土壤有机质含量的高光谱预测模型。结果表明：（1）对不同有机质含量的土壤光谱去包络线后，光谱曲线吸收特征差异更加显著，且土壤有机质含量越多，土壤光谱反射率越低。（2）土壤反射率经过数学变换后提高了与有机质含量的相关系数。（3）在全波段的PLSR中，CR、R′和lg(1/R)′模型的RPD均大于2.0，表明预测能力极好。其中以CR的预测精度最为突出，其模型R2和RMSE分别为0.79、4.12，RPD为2.18。在显著性波段的PLSR中，虽然R′和CR的模型RPD均大于2.0，可以准确预测有机质含量，但CR的R2，RPD更高；基于全波段PLSR模型精度均略优于显著性波段，但其使用数据量大，增加了计算量。同时，其CR模型的RPD仅比显著性波段模型的高0.03。因此，选择显著性波段CR模型作为估测该荒地土壤有机质含量的模型更为简洁、科学、可行。 |
| 英文摘要: |
| It is difficult to determine concentration of soil nutrition and organic matter accurately and quickly for poverty of soil with less organic matter in waste land, arid region. Concentration of organic matter and field spectra of 64 soil samples from barren land in central Fukang, were analyzed. Based on the original reflectance (R), the software ENVI5.1 was used to extract the first derivative reflectance (R′), logarithmic reciprocal (lg(1/A)), the logarithm of reciprocal derivative (lg(1/A)′),to envelope (CR) and other 4 kinds of spectral reflectance. The correlation between the 5 spectral reflectance transformation form and concentration of soil organic matter was analyzed. Based on the full band (450~2 350 nm) and significant band (correlation coefficient by P=0.01 level test), prediction models of soil organic matter conenttation were regressed by partial least squares regression (PLSR) hyperspectral. The results showed that:(1)The spectral curve absorption characteristics of soil with different organic matter concentration were more significant.Concentration of soil organic matter, correlated inversely to the spectral reflectance of soil.(2)The correlation coefficient between soil reflectance and the concentration of organic matter was enhanced by mathematical transformation.(3)In the whole band of PLSR, the RPD of CR, R′ and lg(1/R) model was greater than 2, which indicated that the prediction ability was excellent. Among them, the prediction accuracy of CR was the most prominent, and the model R2 and RMSE were 0.79 and 4.12, respectively, and RPD was 2.18. In a significant band of PLSR, while RPD R′ and CR model were greater than 2, could accurately predict the concentation of organic matter. However, R2 and RPD of the CR were higher precision, full spectrum PLSR model was slightly better than significant band based and increased the amount of calculation for its use of large amount of data. At the same time, the RPD of the CR model was only 0.03 higher than which of the significant band model. Therefore, it is more concise, scientific and feasible to choose the significant band CR model as a model to estimate the content of organic matter in the uncultivated land. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |