黑土地硒含量的航空高光谱综合反演模型

Comprehensive inversion model of Se content in black soil by aerial hyperspectral method

针对单纯利用光谱信息建立土壤硒含量反演模型时,通常会产生模型精度受限、模型参数难以解释的问题,该文基于硒的空间分布影响因素开展综合建模。以黑龙江省海伦典型黑土区为研究区,利用CASI/SASI航空高光谱数据、土壤理化性质和地形因子,在筛选模型变量的基础上采用随机森林法建立综合模型。结果表明:硒与土壤理化性质的关系非常密切,其中与有机质、全氮、全磷、Al2O3、Fe2O3、MgO、CaO、pH呈极显著正相关,与SiO2、Na2O呈极显著负相关;硒与地形因子的关系相对较弱,其中与起伏度和地形粗糙指数呈显著正相关。在光谱特征上,硒与光谱反射率的相关性主要体现了有机质和铁锰氧化物的光谱吸收特征。将筛选的相关指标按照不同组合方式进行建模和对比讨论,结果显示光谱、理化性质和地形因子均对提升模型精度有贡献,整体上光谱和理化性质占主导地位。当模型自变量为全部3类数据时,模型的建模和验证精度均为最高,表明综合模型不仅提升了硒元素的建模精度,而且改善了模型的稳健性。

Simply using spectral information to establish an inversion model of soil selenium(Se) content usually results in problems of limited model accuracy and difficulty in interpreting model parameters. Therefore, a new comprehensive modeling method based on the spatial distribution of Se elements was proposed. The typical black soil area of Hailun in Heilongjiang province was selected as the study area, based on the CASI/SASI aerial hyperspectral data, the soil physical and chemical properties and terrain factor data were introduced to conduct a comprehensive analysis and select model variables, and the random forest method was used to establish a comprehensive model. The results showed that Se had a very close relationship with the physical and chemical properties of soil. Among them, Se had a very significant positive correlation with organic matter,total nitrogen,total phosphorus,Al2 O3,Fe2 O3,MgO,CaO,and pH,and had a significant negative correlation with SiO2 and Na2 O.The relationship between Se and terrain factors was relatively weak,and there was a significant positive correlation between Se and relief and terrain roughness index.On the whole spectral characteristics,the correlation between Se and spectral reflectance mainly reflected the spectral absorption characteristics of organic matter and iron manganese oxide.The relevant indicators were modeled and compared according to different combinations.It was shown that the spectrum,physical and chemical properties,and terrain factors all contributed to improving the accuracy of the model.As a whole,spectrum and physical and chemical properties were dominant.Finally,the modeling accuracy and verification accuracy of the comprehensive model based on all three types of data were the highest.The comprehensive model improved the modeling accuracy of Se and the robustness of the model.