基于多源数据融合的盐分遥感反演与季节差异性研究

为提高多光谱盐分遥感反演的精度,利用实测高光谱与多光谱进行数据融合,并分析了不同季节盐分遥感的差异性。以河套灌区永济灌域为研究区域,以实测光谱仪测定的土壤高光谱数据和Landsat-8 OLI多光谱数据为基础,通过光谱变换和多元逐步回归方法筛选特征波段和特征光谱指数,构建了春、秋两季土壤盐分多光谱、高光谱反演模型,并利用特征光谱指数的线性回归构建了高-多光谱数据融合反演模型。结果表明:高光谱的反射率总体比多光谱高36.83%,春季反射率比秋季平均高23.78%。利用模型中最优变量特征光谱指数对多光谱模型与高光谱模型进行融合,高多光谱融合反演模型训练集和验证集R2平均值分别为0.651和0.635,RMSE平均值分别为2.44 g/kg和2.49 g/kg,精度明显高于对应的多光谱反演模型,其中训练集、验证集的R2平均值分别提高了36.19%和35.64%,RMSE平均值分别降低了34.28%和41.72%。春季多光谱、高光谱和融合反演模型的精度均高于秋季,其中训练集R2平均值比秋季模型分别提高了6.03%、6.05%和4.40%,验证集R2平均值分别提高了19.07%、12.21%和1.75%。构建的高多光谱融合模型反演灌域春秋两季平均盐分含量分别为6.05、5.97 g/kg,平均相对误差分别为9.65%和10.68%,总体上该区域春季土壤主要为重盐化土,秋季土壤主要为中盐化土。

Remote Sensing Inversion of Soil Salinity and Seasonal Difference Analysis Based on Multi-source Data Fusion

The fusion technology based on measured hyperspectral and multispectral data was used to remote sensing inversion of soil salinity to improve the multispectral model precision,and the difference for different seasons was analyzed.Yongji of Hetao Irrigation District,a typical salinization region,was chosen as the study region for establishing hyper-multispectral inversion model of spring and autumn,respectively.The optimal spectral transformation and multiple stepwise regressions were used to get the characteristic bands and spectral indices by using the measured data of the hyperspectral inversion model and Landsat-8 OLI multispectral inversion model.Additionally,the fusion model was established with measured hyperspectral and multispectral data by multiple stepwise regression based on characteristic spectral indices.The results showed that the reflectivity of hyperspectral was 36.83%higher than that of the multispectral,and the average reflectivity in spring was 23.78%higher than that in autumn.The R2 of the training set and validation set of the hyper-multispectral inversion model with characteristic spectral indices were 0.651 and 0.635 on average,the RMSE were 2.44 g/kg and 2.49 g/kg on average,respectively,R2 were 36.19%and 35.64%higher than those of training set and validation set of the multispectral inversion model,and the RMSE were 34.28%and 41.72%lower than that,respectively.In addition,The accuracy of multispectral,hyperspectral and fusion inversion models in spring was higher than that in autumn,the R2 of the training set was improved by 6.03%,6.05%and 4.40%on average,and the verification set was improved by 19.07%,12.21%and 1.75%on average.The average salinity of the spring and autumn was 6.05 g/kg and 5.97 g/kg which used the hyper-multispectral fusion model inversed,respectively,and the average relative errors with the measured salinity were 9.65%and 10.68%,respectively.On the whole,the soil of this region was mainly highly salinization in spring and moderate salinization in autumn.