基于组分光谱识别的黄河三角洲盐碱土盐分定量估测研究

The extraction of soil salinization information from hyperspectra has increasingly become a hot research because of the economic and efficient advantages. However, the recognition wavebands of saline-alkali soil salt composition should be definited and put into practice, as the real-time feature and precision of salinity quantitative estimation needs to be promoted. In this research, the Kenli county in the Yellow River Delta will be as the core study area, the content of soil salinity, its compositon and the visible-near infrared hyperspectra will be obtained through sampling analysis and ground remote sensing method in the fields. Aiming at the major salt composition (NaCl and CaCl2) of the research area, based on the dry soil samples of different salt composition content in laboratory, the response and recognition wavebands will be proved up. Then, the evaporation and leaching process of saline-alkali soil will be simulated outdoors, the different content of soil moisture and salinity will be combined with the spectral changes, the influence of soil moisture and measurement conditions of outdoors on spectral response will be analyzed and removed, then the quantitative spectral recognition model of salt composition will be built. Afterwards the spectral response will be integrated with statistical analysis, the quantitative estimation model of soil salinity will be built based on the recognition wavebands of salt compositon in situ, then the quantitative, quick, accurate obtaining of soil salinity will be realized. Through this research, not only the theoretical and technical reference will be provided to the quantitative remote sening extraction and the real-time monitoring in situ of soil salinization, but also the data and decision-making support will be offered for the treatment and utilization of saline-alkali soil, and the construcion of the national efficient and ecological economy zone in the Yellow River Delta.

基于高光谱提取盐碱土信息，因经济、高效的独特优势日益成为研究热点，但盐碱土盐分组成的识别谱系有待明确并实用化，盐分定量估测的实时性和准确性有待提升。本项目以黄河三角洲垦利县为研究区，通过实地采样分析和地面遥感手段，获取盐碱土盐分及其组分含量特征与可见-近红外光谱数据；针对研究区主要盐分组成（NaCl和CaCl2），室内基于盐分组成不同含量的干土，探明盐分组成的光谱响应和识别谱系；室外模拟盐碱土蒸发和淋洗，将不同水盐含量与光谱变化相耦合，摸清并消除土壤水分、室外测定条件对组分光谱响应的影响，构建盐分组成定量光谱识别模型；进而将光谱响应与统计分析相融合，构建基于组分光谱识别的盐分野外原位定量估测模型，实现盐碱土盐分定量、快速、准确光谱获取。通过研究，不仅为土壤盐碱化定量遥感提取和实时原位监测提供理论和技术参考，而且为黄河三角洲盐碱地的治理、利用和国家级高效生态经济区建设提供数据和决策支撑。

土壤盐分定量光谱估测的实时性和准确性有待提升，本项目以黄河三角洲垦利县为研究区，将室内实验与室外模拟、水盐含量与光谱变化、光谱响应与统计分析相结合，从室内干土、室外模拟土到野外原状土，逐渐增加光谱复杂度，逐步探明了主要盐分组成（NaCl和CaCl2）的光谱响应，探索了去除土壤水分影响因素的方法，摸清了盐分组成的室外定量光谱识别谱系，进而基于光谱识别构建了土壤盐分含量的原位光谱估测模型。研究表明，盐分组成在550 nm尤其是1000 nm之后有较为明显的光谱响应，NaCl的识别谱系范围为1 440-1 550、2 100-2 150和2 300-2 370 nm，CaCl2的识别谱系范围为1 440-1 580、1 870-1 920和2 270-2 300 nm；盐分的室外光谱特征受土壤水分的影响较为明显，采用奇异值分解和非负矩阵分解都可以一定程度去除土壤水分影响，提高土壤盐分的估测精度；基于组分光谱识别的土壤盐分估测模型，精度明显优于单纯基于统计分析的模型，尤其是以1 445和1 920 nm反射率一阶导数为变量，基于非负矩阵分解去除土壤水分影响后的土壤盐分随机森林模型精度最高，建模R2为0.95，RMSE为0.22，验证R2为0.78，RMSE为0.43，RPD为2.16，从机理上实现了土壤盐分的快速、原位、准确估测。通过研究，不仅为土壤盐碱化定量遥感提取和实时原位监测提供理论和技术参考，促进土壤定量遥感的发展，而且可为黄河三角洲盐碱地的治理、利用和国家级高效生态经济区建设提供数据和决策支撑。项目资助发表相关论文6篇，其中SCI收录1篇，EI收录1篇，注册软件著作权1项。协助培养博士1人，硕士3人。项目投入经费26万，支出13.89万，各项支出基本与预算相符，结余12.11万，剩余经费计划用于本项目研究的后续支出。