耦合植被物候的新疆绿洲土壤盐分信息挖掘  

Soil Salinization Mining in Xinjiang Based on Vegetation Phenology

作　　者：蒙莉娜 丁建丽[1,2] 张振华 MENG Lina;DING Jianli;ZHANG Zhenhua(College of Resources and Environmental Sciences,Xinjiang University,Urumqi 830046,China;Laboratory of Oasis Ecology Under Ministry of Education,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学资源与环境科学学院,乌鲁木齐830046 [2]新疆大学绿洲生态教育部重点实验室,乌鲁木齐830046

出　　处：《土壤》2022年第3期629-636,共8页Soils

基　　金：国家自然科学基金项目(41961059、41261090、41771470)资助。

摘　　要：随着土壤环境问题涉及的尺度日趋增大,小区域斑块化盐渍化信息的提取难以了解土壤环境总体的变化趋势。本文以野外监测的南北疆典型绿洲区域——渭库绿洲和艾比湖流域为分析靶区,通过实测数据建立土壤–环境关系,并通过MODIS EVI数据反演得到植被物候特征,耦合植被物候、植被指数、盐度指数、地表温度和地形参数作为随机森林(random forest,RF)模型的输入因子,预测新疆绿洲区域土壤盐分含量信息并绘制土壤盐分空间分布图。结果表明:通过深入挖掘植被物候信息,物候参数在预测土壤盐分方面具有较高的相对重要性,代表生物积累量的LSI和SSI参数表征土壤盐渍化的能力较强,优于其他几个物候参数。耦合物候参数后土壤盐分信息预测精度明显提高,决定系数R^(2)从0.53提升到0.61。经模型反复迭代进一步筛选出适合研究区的23个环境参数,大幅提升了预测精度(R^(2)=0.73,RMSE=5.19,MAE=3.59)。从得到的盐渍化空间分布特征来看,新疆绿洲大部分区域分布的是非盐渍化土和轻盐渍化土,且普遍分布在绿洲内部,中度及以上盐渍化土多分布在绿洲外围,总体盐渍化水平依次为:伊犁平原<北疆绿洲<南疆绿洲<东疆绿洲。With the increasing scale of soil environmental problems,it is difficult to understand the overall changing trends of soil environment by extracting the information on patchy salinization in small areas.In this study,the Weigan-Kuqa River delta oasis and Ebinur River basin were taken as the research area,topsoil samples(0–10 cm)were collected from 209 of typical sites based on the representative grade sampling method,the electric conductivity(EC)and salt contents of the samples were determined,and then the relationship between soil EC and various environment variables were established.The characteristics of vegetation phenology were inversed by MODIS EVI data and were coupled with vegetation index,salinity index,surface temperature and topographic parameters as input factors of the random forest(RF)model,and then the information of soil salt content in the oasis regions in Xinjiang was deduced and the spatial distribution maps of soil salt content was drawn.The results show that digging deeper into the information of vegetation phenology promote significantly the importance of the phenological parameters in predicting soil salinity,the large seasonal integral(LSI)and small seasonal integral(SSI)representing bioaccumulation are better in characterizing soil salinization than other phenological parameters.The accuracy of saltness prediction is significantly improved after coupling with the phenological parameters with the coefficient of determination R^(2) increased from 0.53 to 0.61.23 environmental parameters suitable for the study area are screened out after the iterative selection,which significantly improve the prediction accuracy(R^(2)=0.73,RMSE=5.19,MAE=3.59).According to the spatial distribution of salinization,non-salinized and lightly salinized soils are distributed in most areas of the Xinjiang oases,and they are generally distributed in the interior of the oases,while the moderate and above salinized soils are mostly distributed in the periphery of the oases.The total salinization level is in order of

关 键 词：土壤盐渍化 植被物候 机器学习 数字土壤制图 

分 类 号：S153[农业科学—土壤学]