| 摘要: |
| 【目的】可见光、近红外、短波红外等波段范围对于土壤有机质有着不同的敏感程度和反应特征。将不同波长范围的光谱与土壤有机质建立对应关系并对比精度，从而准确估测土壤有机质的含量，对精准农业、土壤肥力监测和改良等领域意义重大。【方法】该研究选定的研究区域位于黑龙江建三江农垦区中，采用该区域对应的高分五号（GF-5）高光谱数据作为数据源，对GF-5数据按照其设计特性划分为可见光-近红外（VNIR）和短波红外（SWIR）两个部分，将两部分光谱和获取的原始全光谱（VNIR-SWIR）数据进行多元逐步回归（MLSR）和偏最小二乘回归（PLSR）建模，将三种数据的模型精度进行对比分析。【结果】结果表明，在多元逐步回归模型中，精度最高为VNIR-SWIR的对数倒数一阶微分（R2cal=0.538，RMSEC=3.602，R2val=0.383，RMSEP=5.009），偏最小二乘回归模型中，最优结果为VNIR的原始反射率（R2cal=0.339，RMSEC=4.310，R2val=0.359，RMSEP=4.170）。比较来看，MLSR的VNIR-SWIR对数倒数一阶微分模型精度更好，而且这两种模型中，VNIR数据的精度普遍高于SWIR精度，表明该研究中VNIR更适合进行土壤有机质反演。【结论】卫星数据质量、研究区域自然条件等不可控因素以及数据预处理过程、建模方法的选择等人为因素共同决定了最后的模型精度，因此要基于这些因素进行更深入的研究，从而在更深层面探讨GF-5数据和土壤有机质的细微关系，并提高GF-5数据预测土壤有机质含量的潜力。 |
| 关键词: 高光谱 土壤有机质 GF-5 VNIR SWIR 反演模型 |
| DOI： |
| 分类号:P283.8 |
| 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) |
|
| Comparison of Prediction Accuracy of Soil Organic Matter Content in Different Spectral Bands of GF-5 AHSI Hyperspectral Imagery |
|
yanxiangzhao1, yaoyanmin1, zhangxiaoyu1, liujunming2
|
|
1.Institute of Agricultural Resources and Regional Planning;2.College of Land Science and Technology, China Agricultural University
|
| Abstract: |
| [Purpose]Visible light, near infrared and short wave infrared have different sensitivity and response characteristics to soil organic matter. It is of great significance for precision agriculture, soil fertility monitoring and improvement to accurately estimate the content of soil organic matter by establishing the corresponding relationship between different wavelength range spectra and soil organic matter and comparing the accuracy. [Method] The research area selected in this study is located in Jiansanjiang Reclamation Area of Heilongjiang Province. GF-5 hyperspectral data corresponding to this area was used as the data source. According to its design characteristics, GF-5 data was divided into visible-near infrared (VNIR) and short wave infrared (SWIR) parts, and the two parts of spectrum and the original full spectrum (VNIR-SWIR) data were analyzed step by multiple linear stepwise regression(MLSR) and partial least squares regression(PLSR) were used to build the model, and the model accuracy of the three kinds of data was compared and analyzed. [Result] The results show that in the multiple stepwise regression models, the highest accuracy is the logarithmic reciprocal first-order differential of VNIR-SWIR (R2cal=0.538，RMSEC=3.602，R2val=0.383，RMSEP=5.009), and in the partial least squares regression models, the best result is the original reflectance of VNIR (R2cal=0.339，RMSEC=4.310，R2val=0.359，RMSEP=4.170). In comparison, the accuracy of VNIR-SWIR log reciprocal first-order differential model of MLSR is better, and the accuracy of VNIR data in these two models is generally higher than that of SWIR, indicating that VNIR is more suitable for soil organic matter inversion in this study. [Conclusion] The uncontrollable factors such as the quality of satellite data, the natural conditions of the research area and the artificial factors such as the data preprocessing process and the selection of modeling methods jointly determine the final model accuracy. Therefore, it is necessary to conduct more in-depth research based on these factors, so as to explore the subtle relationship between GF-5 data and soil organic matter in a deeper level, and improve the potential of GF-5 data to predict soil organic matter content. |
| Key words: hyperspectral soil organic matter GF-5 VNIR SWIR inversion model |