基于高光谱遥感和机器学习技术的黑龙江友谊县土壤有机质含量反演

Retrieval of Soil Organic Matter Contents in Black Soil Region Based on Hyperspectral Remote Sensing and Machine Learning Techniques

  • 摘要:
    目的 探究星载高光谱影像对黑土区土壤有机质的预测能力,构建有机质高光谱反演模型。
    方法 以东北典型黑土区友谊县为研究区,基于资源一号02D卫星高光谱遥感数据,结合9种不同光谱预处理方法,分别构建随机森林(Random Forest,RF)、支持向量机(Support Vector Machine,SVM)、偏最小二乘回归(Partial Least Squares Regression,PLSR)三种模型,对友谊县土壤有机质含量进行估测。
    结果 RF和SVM两种非线性机器学习模型对研究区土壤有机质的预测效果明显优于线性模型PLSR。其中,应用Savitzky-Golay滤波光谱预处理技术的RF模型精度最高(R2 = 0.52、RMSE = 7.25 g kg−1),能较好地估测友谊县土壤有机质含量。
    结论 星载高光谱影像能够较好的预测黑土区土壤有机质含量分布,且研究区土壤有机质含量整体呈现东北高、西南低的空间分布格局。本文为黑土区土壤有机质空间分布研究提供相关方法支撑,并为其他利用高光谱数据预测土壤属性含量的研究提供科学依据和可靠思路。

     

    Abstract:
    Objective The aims were to explore the predictive ability of spaceborne hyperspectral images for soil organic matter (SOM) in black soil areas, the SOM inversion model was developed based hyperspectral and machine learning techniques.
    Method This study selected Youyi County, a typical black soil area in Northeast China, as the research area. Based on the high-resolution remote sensing data of the ZY-1 02D satellite, nine different spectral preprocessing methods were combined and three models were constructed: Random Forest (RF), Support Vector Machine (SVM), and Partial Least Squares Regression (PLSR), to estimate SOM content in Youyi County.
    Result The predictive performance of two nonlinear machine learning models (i.e., RF and SVM) for SOM in the study area was significantly higher than that of the linear model of PLSR. Among them, the RF model with SG spectral preprocessing technique had the highest accuracy (R2 = 0.52, RMSE = 7.25 g kg−1), and estimated the SOM content in Youyi County well.
    Conclusion The overall trend of SOM changes in the area showed a spatial distribution pattern of high values in the northeast and low values in the southwest. This study provides not only relevant method support for the spatial distribution of SOM in black soil areas, but also scientific basis and reliable ideas for other studies with hyperspectral data for predicting soil attribute content.

     

/

返回文章
返回