三江源高寒退化草原土壤有机质含量的光谱模拟估算

Estimation of soil organic matter content of degradated alpine grassland in Three Rivers Headwater region by spectral simulation

摘要:

以三江源玛多县不同退化程度高寒草原土壤为研究对象，采集 0~30 cm 土层的 90 个土壤样品，测定土壤样品的光谱反射率和有机质(SOM)含量，分析不同光谱数据转换方式与土壤 SOM 含量的相关性，据此挑选 P＜0. 001 水平的显著波段作为特征波段，并与土壤 SOM 含量建立多元逐步回归 (MLSR)、支持向量机(SVM)、决策树(DT)、随机森林(RF)模型。结果表明：1)高寒草原土壤 SOM 含量属中等变异，且与土壤原始反射率呈负相关，与倒数对数呈正相关；2)光谱数据的数学转换扩大了光谱的吸收特征，log(1/R)、R’、[log(1/R)]’与土壤 SOM 含量相关系数绝对值的最大值比 R 分别提高了 0. 099、0. 156、0. 160；3)RF 反演模型精度高于其他反演模型，Log(1/R)‐RF 模型的预测效果较好，其建模组和检验组的决定系数(R² )、均方根误差(RMSE)分别为 0. 949 1、0. 252 69 和 0. 717 23、0. 496 9，可以准确估算高寒草原土壤 SOM 的含量。

Abstract:

The alpine meadow soils with five degradation gradients in Maduo County of Sanjiangyuan were the research subject. A total of 90 soil samples of 0~30 cm below ground were collected. Spectral reflectance and soil or‐ ganic matter (SOM) content in the soil samples were measured. The correlation between different spectral data con‐ version methods and SOM content was analyzed. According to the result of correlation analyses，the significant band with P＜0. 001 level was selected as the characteristic band. Multiple stepwise regression (MLSR)，support vector machine (SVM)，decision tree (DT)，and random forest (RF) models were established with the SOM content. The results showed that，1) The content of SOM in alpine meadow soil was moderately variable and was negatively corre‐ lated with the soil original reflectance and positively correlated with the reciprocal logarithm；2) The spectrum ab‐ sorption characteristics were expanded by mathematical conversion of spectral data，while log(1 /R)，R' and [log(1/ R)]' of correlation coefficient to maximum absolute value of SOM content were 0. 099，0. 156，0. 159 9 higher than R respectively；3) The accuracy of the RF inversion model was higher than other inversions Model，and the Log(1/R)‐ RF model had the batter predictive effect. The coefficient of determination (R² ) and root mean square error (RMSE) of the modeling group and test group were 0. 949 1，0. 252 69 and 0. 717 23，0. 496 9，respectively，which can accu‐ ratelyestimate the SOM content of the alpine grassland soil.