|国家科技期刊平台
首页|期刊导航|农业机械学报|基于Landsat 8的节水改造背景下盐渍化土壤含盐量反演

基于Landsat 8的节水改造背景下盐渍化土壤含盐量反演OA北大核心CSTPCD

Inversion of Salt Content in Salinized Soil under Background of Water-saving Transformation Based on Landsat 8

中文摘要英文摘要

为探明沈乌灌域节水改造后因渠道衬砌、引排水量减少引起的土壤含盐量时空分布特征及变化规律,采用区域土壤信息定点监测,结合经典统计学、空间插值以及机器学习建模反演等技术手段,利用Landsat 8卫星获取光谱数据,通过对实测土壤含盐量、光谱指数及波段反射率进行处理,运用Adaboost回归、BP神经网络回归、梯度提升树回归、KNN回归、决策树回归、随机森林回归方法构建了沈乌灌域土壤含盐量空间反演模型.采用最优反演模型对沈乌灌域土壤含盐量空间分布特征进行了遥感反演.结果表明:通过全变量单一回归法筛选出相关系数大于0.55的9个光谱因子,使用SPSS PRO软件构建6种机器学习反演模型,对比6种反演模型精度,验证集决定系数R2由大到小依次为随机森林回归、梯度提升树回归、Adaboost回归、KNN回归、决策树回归、BP神经网络回归.其中随机森林回归模型的拟合精度最佳,训练集与验证集的决定系数R2分别为0.834和0.86,说明随机森林回归模型的反演效果较好.反演结果表明:节水改造后非盐渍土面积增加391.7 km2,占灌域总面积的21%,中度盐渍土面积、重度盐渍土面积、盐土面积分别减少95.61、63.37、45.7 km2,分别占灌域总面积的5%、3%、2%.综上所述,节水改造工程完成后,沈乌灌域土壤盐渍化程度减轻,作物生长安全区面积增加,但由于渠道衬砌以及引排水量减少,土壤盐分淋洗效果减弱,土壤盐分在灌域内部运移,整体土壤环境得到改善,局部地区出现盐分聚集.

In order to explore the spatial-temporal distribution characteristics and variation rules of soil salt content caused by the reduction of channel lining and drainage water after water-saving transformation in Shenwu irrigated district,fixed-point monitoring of regional soil information was adopted,combined with classical statistics,spatial interpolation and machine learning modeling and inversion,and spectral data was obtained by Landsat 8 satellite.By processing the measured soil salt content,spectral index and band reflectance,using Adaboost regression,BP neural network regression,gradient lifting tree regression,KNN regression,decision tree regression and random forest regression,the spatial inversion model of soil salt content in Shenwu irrigated district was constructed.The optimal inversion model was used to invert the spatial distribution characteristics of soil salt content in Shenwu irrigated district.The results showed that the correlation coefficient was screened by the whole variable single regression method.With nine spectral factors of 0.55,six inversion models of machine learning were constructed using SPSS PRO software,and the accuracy of the six inversion models was compared.The verification set R2 from high to low was random forest regression,gradient lifting tree regression,Adaboost regression,KNN regression,decision tree regression,and BP neural network regression.The random forest regression model had the best fitting accuracy,and the R2 of training set and verification set were 0.834 and 0.86,respectively.It was showed that random forest regression model had better inversion effect.The inversion results showed that the non-salinization soil increased by 391.7 km2,accounting for 21%of the total irrigation area,while the moderate salinization soil,severe salinization soil and saline soid reduced the square by 95.61 km2,63.37 km2 and 45.7 km2,accounting for 5%,3%and 2%of the total irrigation area,respectively.In summary,after the completion of the water-saving transformation project,the degree of soil salinization in Shenwu irrigated district was reduced,and the area of crop growth security area was increased.However,due to the reduction of channel lining and drainage water,the effect of soil salt leaching was weakened,and soil salt migrated within the irrigation area,the overall soil environment was improved,and salt accumulation occurred in some areas.

刘伟;史海滨;苗庆丰;刘晓志;段倢;王禹森

内蒙古农业大学水利与土木建筑工程学院,呼和浩特 010018||内蒙古生态环境大数据有限公司,呼和浩特 010010内蒙古农业大学水利与土木建筑工程学院,呼和浩特 010018中国水利水电科学研究院流域水循环模拟与调控国家重点实验室,北京 100038

农业科学

节水改造土壤含盐量遥感反演Landsat 8

water-saving transformationsoil salt contentremote sensing inversionLandsat 8

《农业机械学报》 2024 (001)

294-304 / 11

国家重点研发计划项目(2021YFC3201202-05)和国家自然科学基金项目(52269014)

10.6041/j.issn.1000-1298.2024.01.028

评论