基于加权算法的空-天遥感升尺度土壤含盐量监测模型

利用无人机-卫星遥感升尺度转换方法可以有效提高土壤含盐量监测精度。以内蒙古河套灌区沙壕渠灌域为研究区，4月裸土期表层土壤为研究对象，分别采用主导变异权重法、局部平均法和最邻近法将试验区无人机4波段影像(0.1m)升尺度至与GF-1卫星(16m)同一尺度，引入3种变量组合作为模型输入变量并利用多元线性回归模型（Multivariable linear regression，MLR）和BP神经网络模型（Back propagation neural networks，BPNN）构建不同数据源关于土壤含盐量的定量监测模型。在此基础上，采用波段比值均值法对GF-1卫星数据进行修正，实现基于卫星因子的研究区土壤盐分升尺度反演。结果表明，经统计指标评价后得出主导变异权重法在4块试验区针对4波段影像的尺度转换效果总体上优于其他2种转换方法；3种无人机-卫星遥感升尺度转换方法中，主导变异权重法监测效果最佳，局部平均法次之，最邻近法效果最差；对筛选得到的2个模型进行升尺度修正，得到验证效果最佳的监测模型为基于混合变量组的多元线性回归模型，其R2v为0.420，RMSEv为0.219%，比直接采用GF-1卫星数据得到的混合变量组多元线性回归模型R2v高0.217，RMSEv低0.013个百分点。本文研究结果可为卫星、无人机多光谱遥感一体化监测裸土期农田土壤含盐量提供参考。

UAV-Satellite Remote Sensing Scale-up Monitoring Model of Soil Salinity Based on Dominant Class Variability-weighted Method

UAV-satellite remote sensing scale-up transformation method can effectively improve the monitoring accuracy of soil salt content. Sand trench canal irrigation area in Hetao Irrigation Area of Inner Mongolia was taken as the study area, the surface soil in bare soil period in April was taken as the research object. The dominant class variability-weighted method, local average method and nearest neighbor method were used to scale up the quadruple-band image (0.1 m) of UAV in the experimental area to the same scale as GF-1 satellite (16m). Subsequently, three combinations of variables were introduced as the input variables of the model for the UAV dataset and GF-1 satellite dataset, and the quantitative monitoring model of soil salt content was constructed by using multivariable linear regression (MLR) and back propagation neural networks (BPNN). On this basis, the GF-1 satellite data was modified by the mean band ratio method, and the scale-up inversion of soil salinity in the study area based on satellite factors was realized. The results showed that the dominant class variability-weighted method had the best monitoring effect, followed by the local average method. The nearest neighbor method had the worst monitoring effect among the three UAV-satellite remote sensing scale-up transformation methods;after comparing the four statistical evaluation indexes of mean value, standard deviation, information entropy and average gradient with the original UAV image, it was found that the quadruple-band UAV image pushed by the three methods had scale differences with the original image data to different degrees;by comparing R2 and RMSE of three variable combinations based on different data sources, it was found that the accuracy of the model constructed by the dominant class variability-weighted method was better than that of the other three data sources as a whole, and the scale-up dataset of the dominant class variability-weighted method based on mixed variable groups achieved the best monitoring effect in MLR model and BPNN model;the monitoring model with the best validation effect was multivariate linear regression model, its validation R2 was 0.420, RMSE was 0.219%. The research results can provide reference for integrated monitoring of farmland soil salt content in bare soil period by multi-spectral remote sensing of satellite and unmanned aerial vehicle.