| 摘要: |
| 【目的】探讨星载高分五号（GF-5）高光谱影像不同光谱波段对SOM 含量预测精度差
异，明确有效光谱波段范围，以便提高SOM 含量高光谱预测精度。【方法】该研究以黑龙江
省建三江农垦区为研究区域，将GF-5 可见短波红外高光谱相机（AHSI）获取的高光谱数据
划分为可见光- 近红外（VNIR）、短波红外（SWIR）和VNIR-SWIR 3 种不同光谱波段，并
将光谱反射率进行了9 种光谱数学变换；分别采用多元逐步回归（MLSR）和偏最小二乘回
归（PLSR）构建SOM 含量预测模型，评价分析了3 种不同光谱波段预测SOM 含量的精度差
异。【结果】在MLSR 模型中，VNIR-SWIR 的对数倒数一阶微分SOM 含量预测精度相对较
高，验证精度决定系数R2
val 为0.383，均方根误差RMSEP 为5.009；在PLSR 模型中，VNIR
反射率的SOM 含量预测精度较高，验证精度R2
val 为0.359，RMSEP 为4.170。【结论】GF-5
AHSI SOM 含量预测精度较高的光谱波段为VNIR 和VNIR-SWIR。卫星数据质量、研究区域
自然条件、数据预处理过程、建模方法选择等因素共同影响SOM 含量预测模型精度，通过
技术和方法改进，GF-5 数据预测SOM 含量的潜力更大。 |
| 关键词: GF-5 高光谱影像 可见短波红外高光谱相机AHSI 土壤有机质 VNIR SWIR 预
测模型 |
| DOI：10.12105/j.issn.1672-0423.20200602 |
| 分类号: |
| 基金项目:高分辨率对地观测系统国家科技重大专项“高分农业遥感监测与评价示范系统（二期）”（09-Y30F01-
9001-20/22）；中国农业科学院科技创新工程（CAAS-2020-IARRP-G202020-2） |
|
| Prediction accuracy comparison of soil organic matter content indifferent spectral bands based on GF-5 AHSI hyperspectral imagery |
|
Yan Xiangzhao1, Yao Yanmin※1, Zhang Xiaoyu1, Liu Junming2
|
|
1.Key Laboratory of Agricultural Remote Sensing，Ministry of Agriculture and Rural Affairs / Institute of Agricultural
Resources and Regional Planning，Chinese Academy of Agricultural Science，Beijing 100081，China;2.College of Land
Science and Technology，China Agricultural University，Beijing 100094，China
|
| Abstract: |
| ［Purpose］The purpose of this study is to explore the difference of SOM content
prediction accuracy in different spectral bands based on GF-5 hyperspectral image，and to
define the effective spectral band range，so as to improve SOM content prediction accuracy.
［Method］Taking Jiansanjiang reclamation area of Heilongjiang Province as the study area，
GF-5 spectral data by the Advanced Hyperspectral Imager（ AHSI） was divided into three different
spectral bands as visible-near infrared（ VNIR），short wave infrared（ SWIR） and VNIRSWIR，
and the spectral reflectance was transformed by nine spectral mathematical methods.
Then，multiple stepwise regression（ MLSR） and partial least squares regression（ PLSR） methods
were used to build SOM content prediction models，and the accuracy differences of SOM content
prediction in three different spectral bands were evaluated and analyzed.［Result］In MLSR
model，VNIR with the first order differential has a relatively high SOM predicting accuracy，the
coefficient of verification accuracy determination R2
val is 0.383，and RMSEP is 5.009. In PLSR
model，SOM content prediction accuracy of VNIR with reflectance（ R） is relative higher，R2
val
is 0.359 and RMSEP is 4.170.［Conclusion］VNIR and VNIR-SWIR by GF-5 AHSI are the
spectral bands with high SOM prediction accuracy. SOM content prediction accuracy is affected by
the quality of satellite data，natural conditions of the study area，data preprocessing process and
modeling method selection. Through improvement of hyperspectral remote sensing technology and
method of SOM content hyperspectral prediction，Gf-5 data has more potential to predict SOM
content. |
| Key words: GF-5 hyperspectral imagery the Advanced Hyperspectral Imager soil organic
matter VNIR SWIR prediction |
