光谱学与光谱分析, 2013, 33 (12): 3354, 网络出版: 2014-01-09
光谱测定黑河上游土壤有机质的预测模型
Prediction Models of Soil Organic Matter Based on Spectral Curve in the Upstream of Heihe Basin
有机质含量 高光谱 反射率 逐步线性回归 Soil organic matter Hyperspectral Reflectance Stepwise linear regression analysis
摘要
地面高光谱遥感光谱分辨率高, 能详细地反映地物波谱特征; 多光谱遥感时域宽, 覆盖范围广, 对较大时空区域的地物特征反演具有更大的优势。 探求以不同反射率指标的土壤有机质含量预测模型, 及其敏感波段, 可以结合两种光谱数据的优点, 为研究土壤有机质含量的时空变化规律提供新途径。 本研究选取黑河上游223个土壤样品测定其有机质含量和高光谱曲线, 应用原始光谱曲线反射率(λ)、 倒数(REC)、 倒数之对数(LR)、 归一化(CR)和一阶微分(FRD)五种指标, 采用逐步线性回归分析方法建立预测模型。 通过统计检验, 结果表明, 以反射率指标为自变量的模型预测效果最佳, 其相关系数(r)和均方根误差(RMSE)分别为: 0.863和4.79。 最优模型中得出的敏感波段有TM1内的474 nm、 TM3内的636 nm和TM5内的1 632 nm。 研究结果可为使用TM遥感数据反演黑河上游土壤有机质含量提供参考。
Abstract
Benefiting from the high spectral resolution, ground hyperspectral remote sensing technology can express the ground surface feature in detail, meanwhile, multispectral remote sensing has more advantages in studying the features in a large space-time region, because of its long time-series images and wide coverage. Investigating the prediction models between the soil organic matter (SOM) content and the hyperspectral data and the sensitive bands based on different indices mathematically obtained from reflectance could combine the advantages of both kinds of spectral data, and provide a new method to search the spatio-temporal characteristics of SOM. Two hundred twenty three soil samples were chosen from the upper reaches of Heihe Basin to measure the SOM content and hyperspectral curve. Taking 181 of them, the stepwise linear regression methods were used to establish models between the SOM and five indices, including reflectance (λ), reciprocal (REC), logarithm of the reciprocal (LR), continuum-removal (CR) and the first derivative reflectance (FDR). After then, the left 42 samples were used for model validation: firstly, the best model of the same index was chosen by the values of Pearson correlation coefficient (r) and Root mean squared error (RMSE) between the measured value and predicted value; secondly, the best models of different indices were compared. As a result, the model built by reflectance has a better estimation of SOM with the r: 0.863 and RMSE: 4.79. And the sensitive bands of the reflectance model contain 474 nm during TM1, 636 nm during TM3 and 1 632 nm during TM5. This result could be a reference for the retrieval of SOM content of the upper reaches by using the TM remote sensing data.
刘娇, 李毅, 刘世宾. 光谱测定黑河上游土壤有机质的预测模型[J]. 光谱学与光谱分析, 2013, 33(12): 3354. LIU Jiao, LI Yi, LIU Shi-bin. Prediction Models of Soil Organic Matter Based on Spectral Curve in the Upstream of Heihe Basin[J]. Spectroscopy and Spectral Analysis, 2013, 33(12): 3354.