基于Sentinel-2卫星数据的灌区农田土壤水盐协同反演

Synergistic inversion of water and salt in irrigated agricultural soils based on Sentinel-2

土壤含水量(soil water content, SWC)和土壤含盐量(soil salt content, SSC)是影响作物生长和农业生产力的重要因素。光学卫星图像已成为SWC和SSC估计的主要数据源。然而,在SWC或SSC变化较大地区,土壤水分和盐分会影响对方对光谱反射率的响应,使得SSC和SWC的反演精度较差。对此,该研究提出了一个半解析性的反射率模型—RVS模型,来模拟植被光谱反射率(R_(v))对作物根区土壤含水量和含盐量的响应;并通过构建的RVS模型,对植被覆盖区域的土壤含水量和土壤含盐量进行同步监测。研究表明:RVS模型在反演研究区土壤含盐量和含水量时,精度较为可靠(水分:决定系数R^(2)为0.63~0.74,均方根误差为0.017~0.028;盐分:决定系数R^(2)为0.68~0.75,均方根误差为0.052 5~0.061 7)。在作物生长过程中,植被光谱反射率对深层土壤的含水量和含盐量的响应比对浅层土壤的含水量和含盐量的响应更加明显,而且随着作物的生长,影响光谱反射率的主导因素从土壤水分慢慢转向土壤盐分和水盐相互作用。该研究在一定程度上揭示了土壤水分、盐分、水盐交互作用对作物光谱反射率的干扰过程,实现土壤水分和盐分的同步监测,对实现区域尺度上土壤含盐量和含水量的精准监测具有一定的意义。

Soil water and salt content have been two of the most important influencing factors on crop growth and agricultural productivity all over the world nowadays.Inappropriate soil water and content can pose a potential threat to crop quality and yield.Fortunately,satellite remote sensing can be expected to monitor the soil salt content and water content,due to the largescale coverage,time-and labor-saving.Among them,the optical satellite has been the major data source for the estimation of the soil water and salt content,particularly with the high temporal return cycle,high spatial resolution,and rich spectral range.However,the soil water and salt can affect each other on the response to the spectral reflectance in the areas with their large variations,similar to their regular change in optical satellite band reflectance.There is a high demand to improve the inversion accuracy of soil water and salt content when the soil salt or water content cannot be maintained consistently.This study aims to improve the estimation accuracy of soil salt and water,and then expand the application scope of optical satellite remote sensing.A semi-analytical reflectance model-RVS model was proposed to simulate the response of vegetation spectral reflectance to the soil water and salt content in the vegetation soil.The influence of soil salt-water interaction was also considered when estimating the soil water and salt content.The RVS model was then used to simultaneously estimate the soil water and salt content in the vegetation-covered soil.The results show that the RVS model performed the better estimation accuracy in the study area in both soil water content(coefficient of determination:R^(2)=0.63-0.74,root mean square error:RMSE=0.017-0.028)and soil salt content(coefficient of determination:R^(2)=0.68-0.75,root mean square error:RMSE=0.053-0.062).There was a more outstanding response of vegetation spectral reflectance to the soil water and salt content in the deep soil during the whole stage of crop growth,compared with the shallow soil.The dominant influencing factors on the vegetation spectral reflectance were slowly shifting from the soil water content to the soil salt content and soil water-salt interactions,as the crop grew.The RVS model was then used to continuously estimate the soil water and salt content in the period of crop growth and irrigation.As such,the salt accumulation was attributed to the insufficient water that was caused by both crop growth depletion and excessive irrigation.The estimation revealed the interference of soil water content,soil salt content,and water-salt interaction on the crop spectral reflectance,considering the influence of soil water(soil salt)when estimating soil salt content(soil water content)under the concept of soil salt-water interaction.The simultaneous monitoring of soil water content and soil salt content was realized to construct the estimation models using Sentinel-2 and machine learning.The finding can provide a strong reference for the accurate monitoring of soil salt content and soil water content on the regional scale.