基于多时相遥感影像的农安县土壤有机质含量估算OA北大核心CSTPCD
Estimation of Soil Organic Matter Content in Nongan County Based on Multi-temporal Remote Sensing Images
精准获取农田土壤碳储量对确保国家粮食安全及应对气候变化具有重要意义.土壤有机质(Soil Organic Matter,SOM)作为土壤碳库的重要组成部分,采用遥感影像估算SOM的精度受单时相、多时相影像的影响程度,及究竟选取多少景影像进行SOM的估算较为合适尚不清楚.本研究以吉林省长春市农安县为研究区,以2013-2018年Landsat 8遥感影像作为数据源,选取归一化植被指数(Normalized Difference Vegetation Index,NDVI)、比值植被指数(Ratio Vegetation Index,RVI)、有机碳指数(Soil Organic Carbon Index,SOCI)等光谱指数,并提取高程、坡度、坡向、地形湿度指数(Topographic Wetness Index,TWI)等地形因子,及粮食产量、化肥施用量等农田管理措施,采用随机森林算法筛选后的指示因子作为模型的输入变量,构建SOM估算模型.结果表明:相较于使用单时相卫星影像提取的指示因子,使用多时相卫星影像能更好的利用影像多时相信息,对SOM的估算精度较高.随着使用影像年份的增加,模型反演精度会达到上限,本研究中使用数据年限为3年,后续增加使用影像的数量,反演精度基本不变.使用3年影像进行SOM反演的模型验证集精度R2为0.682,RMSE为3.152 g/kg.本研究成果不仅对应用多时相遥感影像进行SOM精准估算具有重要意义,也对农田土壤质量的提升及农业可持续发展具有重要意义.
Accurate acquisition of farmland soil carbon storage is of great significance for ensuring national food security and addressing climate change.Soil Organic Matter(SOM)is an important part of soil carbon pool.The accuracy of SOM estimation using remote sensing images is affected by single-phase and multi-temporal images,and it is not clear that how many images would be optimal to achieve the best performance for SOM estimation.This study took Nongan County,Changchun City,Jilin Province as the research area,and used Landsat 8 remote sensing images from 2013 to 2018 as the data source.Normalized Difference Vegetation Index(NDVI),Ratio Vegetation Index(RVI)and Soil Organic Carbon index(SOCI)were selected as indicators,and other spectral indices,topographic factors such as elevation,slope,slope direction and Topographic Wetness Index(TWI)were extracted.The farmland management measures such as grain production and fertilizer application were also employed for SOM estimation.SOM estimation model was constructed by using the random forest algorithm with a variety of indicators as input variables.The results show that compared with the indicators extracted from single phase satellite images,multi-temporal satellite images can make better use of multi-temporal image information,and the accuracy of SOM estimation is higher.With the increase of the employed image numbers,the improvement of model accuracy was marginal.In this study,the period of data used is 3 years,and the inversion accuracy will remain basically unchanged with the subsequent increase in the number of images used.The accuracy of model validation set R2 and RMSE for SOM inversion using 3-year image is 0.682 and 3.152 g/kg respectively.The results of this study not only have important significance for applying multi-temporal remote sensing image to SOM accurate estimation,but also have important significance for improving farmland soil quality and agricultural sustainable development.
李华森;张继真;郝航;张月
吉林农业大学资源与环境学院,吉林 长春 130118||吉林省商品粮基地土壤资源可持续利用重点实验室,吉林长春 130118松辽水利委员会 松辽流域水土保持监测中心站,吉林 长春 130021吉林农业大学资源与环境学院,吉林 长春 130118||吉林省商品粮基地土壤资源可持续利用重点实验室,吉林长春 130118||秸秆综合利用与黑土地保护教育部重点实验室,吉林 长春 130118
农业科学
土壤有机质多光谱影像多时相随机森林贡献率
Soil organic mattermultispectral imagemulti-temporalrandom forestcontribution rate
《山东农业大学学报(自然科学版)》 2024 (003)
304-313 / 10
国家自然科学基金青年基金项目(42301074);吉林省自然科学基金面上项目(20240101041JC);国家重点研发计划项目(2021YFD1500800);中国科学院战略性先导科技专项课题(XDA28080500)
评论