EGI OpenIR
基于多源遥感数据的中亚咸海流域农田土壤水盐反演
Alternative TitleInversion of Soil Salinity Content and Soil Moisture Content in The farmland in the Aral Sea Basin Based on Multi-source Remotely Sensed Data
王浩
Subtype硕士
Thesis Advisor罗格平 ; 郑宏伟
2019-06-30
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline理学硕士
Keyword土壤盐分含量 土壤水分含量 随机森林 多源遥感 GEE 中亚咸海流域 Soil Salinity Content soil moisture content Random Forest multi-source remotely sensed data GEE Aral Sea Basin
Abstract中亚咸海流域(锡尔河流域与阿姆河流域) 绿洲农田盐渍化问题突出。土壤盐渍化破坏农业安全与生态环境, 制约中亚咸海流域国家的可持续发展。 “盐随水动”; 遥感技术是大面积精确估计土壤盐分含量(Soil Salinity Content,SSC) 与土壤水分含量(Soil Moisture Content, SMC) 的关键手段。 将擅长处理变量关系的机器学习方法与多源遥感数据结合,构造反演模型,反演 SSC 和SMC,具有显著优势。 但基于机器学习方法的 SSC/SMC 反演因子,通常包括主动微波雷达的后向散射系数,和基于可见光近红外数据的植被因子如 NDVI以及这两类数据的组合; 较少考虑温度、蒸散、高程等重要因子对 SSC/SMC的影响,这在一定程度上增加 SSC/SMC 反演结果的不确定性。 因此,本文研究以锡尔河流域和阿姆河流域的绿洲农田为研究区,以谷歌地球引擎(GoogleEarth Engine, GEE) 为平台获取微波遥感物理量、植被、温度、蒸散发、地形、盐分指数等 SSC/SMC 的影响因子,基于相关分析挑选与 SSC/SMC 显著相关的影响因子作为反演因子。 将随机森林(Random Forest, RF)、支持向量回归(Support Vector Regression, SVR)、 BP 神经网络(Back Propagation NeuralNetwork, BPNN) 3 种机器学习方法与 SSC/SMC 反演因子相结合, 建立SSC/SMC 反演模型, 定量反演 2017 与 2018 年生长季中亚咸海流域绿洲农田SSC 与 SMC 的时空变化。结论如下:(1)基于 GEE 平台, 获取微波遥感物理参量、植被、温度、 蒸散发、地形、下垫面反射条件 6 大类共 32 个 SMC 影响因子;在 SMC 影响因子的基础上增加盐分指数,即 7 大类共 44 个 SSC 影响因子。相关分析结果中, 在锡尔河流域,基于 2017 年数据挑选出 24 个 SSC 反演因子,基于 2018 年数据挑选出 22 个SMC 反演因子和 12 个 SSC 反演因子。 在阿姆河流域,基于 2018 年数据挑选出16 个 SMC 反演因子和 8 个 SSC 反演因子。(2) 将 SSC/SMC 反演因子分为 3 组与 3 种机器学习方法结合,精度验证表明随机森林 RF 与全部因子(多因子)结合在训练与验证阶段具有最高的精度。在锡尔河流域,基于 2017 年的数据,SSC 反演验证阶段结果 RMSE 为 2.17/g/kg;基于 2018 年数据, SSC 反演验证阶段结果 RMSE 为 2.31/g/kg, SMC 反演验证阶段结果 RMSE 为 0.033m³/m³。 在阿姆河流域, 基于 2018 年数据, SSC 反演验证阶段结果 RMSE 为 2.28/g/kg, SMC 反演验证阶段结果 0.041m³/m³(3) 采用最小显著性差异法比较生长季的不同季节(春、夏、秋)内锡尔河流域或阿姆河流域不同子区域(上、中、下游段)之间的绿洲农田 SSC/SMC 差异,并采用该方法分别比较热量条件(LST)、植被状况(NDVI)、土壤质地,以分析 3 者的季节空间差异对 SSC/SMC 时空格局差异造成的影响。 锡尔河流域2017、 2018 生长季与阿姆河流域 2018 生长季中不同季节的宏观主控因素为热量条件、 植被状况、 土壤质地, 3 者单独地或者共同作用影响 SSC、 SMC 时空分布。基于多源遥感数据尽可能全面获取 SSC/SMC 的影响因子,在此基础上采用机器学习模型,可显著提高 SSC/SMC 反演的精度,这在一定程度上克服了因考虑 SSC/SMC 因子不足获取更高 SMC 精度的限制。 作为干旱区盐渍化评估和农田干旱监测补充研究,为干旱区农业安全与生态保护提供一定支持。
Other AbstractSalinization of oasis farmland in the Aral Sea Basin of Central Asia (Syr DaryaBasin and Amu Darya Basin) is a prominent problem. Soil salinization destroysagricultural security and ecological environment, and restricts the sustainabledevelopment of the countries in Central Asia. "Salt moves with water". Remote sensingtechnology is the key means to accurately estimate soil salt content (SSC) and soilmoisture content (SMC) in large area. The use of machine learning method to estimateSSC and SMC from multi-source remotely sensed data is a hot topic in the SMCinversion research has significant advantages. However, the input parameters ofSSC/SMC of machine learning method, which usually include backscatteringcoefficient based on active microwave radar, vegetation factors such as NDVI based onvisible and near-infrared data and the combination of such two kinds of data. Theinfluence of temperature, evapotranspiration and elevation on SSC/SMC is seldomtaken account, which increases the uncertainty of SSC/SMC inversion results.Therefore, this paper chooses oasis farmland in the Syr Darya Basin and Amu DaryaBasin as the study area, and obtain the factors of SSC/SMC such as backscatteringcoefficient of active microwave radar, vegetation factors, temperature factors,evapotranspiration factors, topography factors and salinity index from Google EarthEngine (GEE). Based on correlation analysis, the significant factors of SSC/SMC areselected as input parameters. 3 machine learning methods, Random Forest (RF),Support Vector Regression (SVR) and Back Propagation Neural Network (BPNN),were test with different SSC/SMC input parameters groups for establishing SSC/SMCinversion model. Spatial and temporal variations of SSC and SMC in oasis farmland ofSyr Darya Basin and Amu Darya Basin during the growing season of 2017 and 2018were estimated quantitatively. The conclusion is as follows:(1) Based on GEE platform, 32 SMC factors of 6 categories that backscatteringcoefficient of active microwave radar, vegetation index, temperature, evaporation,topography and underlying surface reflection conditions were obtained. On the basis of SMC factors, the salinity index factor was added, and there are 44 SSC factors of 7categories. Referring to the correlation analysis result, in Syr Darya Basin, 24 SSC inputparameters were selected based on 2017 data, 22 SMC input parameters and 12 SSCinput parameters were selected based on 2018 data. In Amu River Basin, 16 SMC inputparameters and 8 SSC input parameters were selected based on 2018 data.(2) Test different SSC/SMC input parameters groups with 3 machine learningmethods. The Random forest with the group that all input parameters (multi-parameters)showed the best estimation accuracy in training and verification. In Syr Darya Basin,RMSE of SSC inversion validation is 2.17g/kg based on the 2017 data, and the RMSEof SSC inversion validation is 2.31g/kg and that of SMC inversion validation is 0.033m³/m³based on the 2018 data. In Amu Darya Basin, RMSE of SSC inversion validationis 2.28g/kg and that of SMC inversion validation is 0.041m³/m³based on 2018 data.(3) The SSC/SMC differences in oasis farmland in different seasons (spring,summer and autumn) of different subregions of Syr Darya Basin and Amu Darya Basin(upper, middle and lower reaches) during growing season were compared by the leastsignificant difference method, and the heat condition (LST), vegetation condition(NDVI) and soil texture were compared by this method to analyze the influence of theirspatial and temporal differences on SSC/SMC. The main dominant variable in thegrowth season of the Syr Darya Basin in 2017 and 2018 and in the Amu Darya Basinin 2018 are heat condition, vegetation condition and soil texture. Temporal and spatialdistribution of SSC and SMC were affected by heat condition, vegetation condition andsoil texture that work together or alone.Based on multi-source remotely sensed data, the SSC/SMC factors were obtained.On the basis, the machine learning model can significantly improve the inversionaccuracy of SSC/SMC to a large extent for overcoming the limitation of obtaininghigher SMC accuracy due to taking no account of the important SSC/SMC factors. Asa supplementary study on salinization assessment and farmland drought monitoring inarid areas, it serves agricultural safety and ecological protection in arid areas.
Subject Area地图学与地理信息系统
Language中文
Document Type学位论文
Identifierhttp://ir.xjlas.org/handle/365004/15358
Collection中国科学院新疆生态与地理研究所
研究系统
Affiliation中国科学院新疆生态与地理研究所
First Author Affilication中国科学院新疆生态与地理研究所
Recommended Citation
GB/T 7714
王浩. 基于多源遥感数据的中亚咸海流域农田土壤水盐反演[D]. 北京. 中国科学院大学,2019.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王浩]'s Articles
Baidu academic
Similar articles in Baidu academic
[王浩]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王浩]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.