基于Sentinel卫星及无人机多光谱的滨海冬小麦种植区土壤盐分反演研究——以黄三角垦利区为例

删除或更新信息，请邮件至freekaoyan#163.com(#换成@)

本站小编 Free考研考试/2021-12-26

奚雪^,¹, 赵庚星^,¹, 高鹏¹, 崔昆¹, 李涛²

¹山东农业大学资源与环境学院/土肥资源高效利用国家工程实验室,山东泰安 271018

²山东省土壤肥料总站,济南250100

Inversion of Soil Salinity in Coastal Winter Wheat Growing Area Based on Sentinel Satellite and Unmanned Aerial Vehicle Multi-Spectrum— A Case Study in Kenli District of the Yellow River Delta

XI Xue^,¹, ZHAO GengXing^,¹, GAO Peng¹, CUI Kun¹, LI Tao²

¹College of Resources and Environment, Shandong Agricultural University/National Engineering Laboratory for Efficient Utilization of Soil and Fertilizer Resources, Tai’an 271018, Shandong

²Soil and Fertilizer Working Station of Shandong, Jinan 250100

通讯作者: 赵庚星,E-mail： zhaogx@sdau.edu.cn

责任编辑: 杨鑫浩
收稿日期:2020-03-7接受日期:2020-05-5网络出版日期:2020-12-16

基金资助:

国家自然科学基金.41877003
山东省重大科技创新工程项目.2019JZZY010724
山东省“双一流”奖补资金.SYL2017XTTD02

Received:2020-03-7Accepted:2020-05-5Online:2020-12-16
作者简介 About authors
奚雪,E-mail： 1349637259@qq.com。

摘要
【目的】探究黄河三角洲麦田土壤盐分准确高效的遥感提取方法,掌握土壤盐渍化程度与分布。【方法】以垦利区为研究区,均匀布设冬小麦种植区样点77个,同时设置代表性试验区2个,网格布设样点99个,实测采集麦田土壤表层盐分数据及试验区无人机多光谱图像。筛选红、绿、红边、近红4个波段及SI、NDVI、DVI、RVI、GRVI 5个光谱指数中的敏感光谱参量,采用逐步回归、偏最小二乘法、BP神经网络及SVM支持向量机4种方法建立土壤盐分估测模型,使用波段比值均值法得到Sentinel-2A卫星影像相应波段的修正系数,进而将筛选的土壤盐分估测模型转换为基于卫星影像的反演模型,经麦区实测样点数据验证,得到最佳的麦区土壤盐分反演模型,实现试验区和研究区2个尺度的麦田土壤盐分反演。【结果】无人机4个波段及光谱指数NVDI、RVI、SI与土壤盐分含量相关性显著,4种建模方法的13个模型中,以NDVI、RVI、SI建立的4个指数模型的建模及验证R²均优于其他模型;对4个模型进行升尺度修正及验证,效果最佳的反演模型为偏最小二乘法光谱指数模型：Y=-9.4774×NDVI₁+0.4794×RVI₁+3.0747×SI₁+5.0604,验证R²为0.513,RMSE为1.379;利用该模型反演得到了试验区及整个研究区麦田土壤盐分等级分布图,结合实测插值及调查结果,证明反演模型及空间分布结果准确、可靠。【结论】本研究构建了卫星、无人机一体化的滨海麦区土壤盐分反演模型,对滨海盐渍区农作物的生产管理有积极参考价值。
关键词： 冬小麦;无人机;Sentinel-2A卫星;土壤盐分;反演模型

Abstract

【Objective】 The purpose of this paper was to explore an accurate and efficient remote sensing method for soil salinity extraction of wheat field in the Yellow River Delta, and obtain the degree and distribution of soil salinization of wheat fields.【Method】This study took Kenli District as the research area, and set 77 sample points in winter wheat growing area evenly. At the same time, two representative test areas and 99 grid sample points were set, and the surface soil salinity data in wheat field and the multi-spectral images of UAV in the test area were collected. The sensitive spectral parameters were screened from four spectral bands (red, green, red edge, and near-infrared) and five spectral indexes (SI, NDVI, DVI, RVI, and GRVI). Stepwise regression, partial least squares, BP neural network and support vector machine methods were used to establish models for predicting the soil salinity, and the band ratio mean method was used to obtain the correction coefficient of the corresponding band of sentinel-2A satellite image. And then the selected soil salinity estimation model was converted into an inversion model based on satellite image. After using the data from the wheat field sample points to verify the models, the best soil salinity inversion model in wheat field was selected, and two scales of soil salinity inversion are realized in the test areas and the research area.【Result】The results showed that the four bands of UAV and the spectral indexes NVDI, RVI and SI were significantly correlated with soil salinity. Among the 13 models of the four modeling methods, the four index models established by NDVI, RVI and SI were better than the other models in modeling and verifying R²; The best inversion model was the spectral index model obtained by partial least square method: Y=-9.4774×NDVI₁+ 0.4794×RVI₁+ 3.0747×SI₁+ 5.0604, and the accuracy R² was 0.513 and RMSE was 1.379. By using this model, the soil salinity distribution map of the test area and the whole wheat area was obtained. Combined with the measured interpolation and the survey, the inversion model and spatial distribution results were proved to be accurate and reliable. 【Conclusion】In this study, the soil salinity inversion model of the coastal wheat area based on the integration of satellite and UAV was constructed, which had positive reference value for the production and management of crops in the coastal saline area.

Keywords：winter wheat;unmanned aerial vehicle;sentinel-2A satellite;soil salinity;inversion model

PDF (2671KB)元数据多维度评价相关文章导出EndNote|Ris|Bibtex 收藏本文
本文引用格式
奚雪, 赵庚星, 高鹏, 崔昆, 李涛. 基于Sentinel卫星及无人机多光谱的滨海冬小麦种植区土壤盐分反演研究——以黄三角垦利区为例[J]. 中国农业科学, 2020, 53(24): 5005-5016 doi:10.3864/j.issn.0578-1752.2020.24.004
XI Xue, ZHAO GengXing, GAO Peng, CUI Kun, LI Tao. Inversion of Soil Salinity in Coastal Winter Wheat Growing Area Based on Sentinel Satellite and Unmanned Aerial Vehicle Multi-Spectrum— A Case Study in Kenli District of the Yellow River Delta[J]. Scientia Acricultura Sinica, 2020, 53(24): 5005-5016 doi:10.3864/j.issn.0578-1752.2020.24.004

开放科学（资源服务）标识码（OSID）：

0 引言

【研究意义】土壤盐渍化是国内外广泛关注的生态环境问题,由于自然和人为因素的影响,我国北方干旱、半干旱和半湿润地区盐渍化土壤广泛分布,滨海区域尤为突出,其中黄河三角洲自20世纪70年代开始,水沙减少、海水倒灌导致陆地水盐失衡,严重的盐渍化导致土壤退化、板结,地力降低限制了农作物的生长,严重阻碍了区域农业经济的可持续发展^[1]。【前人研究进展】传统的土壤盐分监测方法为实地采样化验分析法,土壤采样样点数量受限且费时费力;电磁感应技术数据获取均匀但处理较为复杂;而遥感图像反演技术已经在土壤盐分监测上取得了较高的应用效果,且数据来源丰富、精度高、获取方便快捷^[2,3]。多数****基于多源卫星影像,结合光谱及地面数据建模,实现了土壤盐分含量的大范围定量反演^[4,5,6,7,8];陈俊英等^[9]和ZHANG等^[10]结合不同空间尺度,使用星机地一体的多平台遥感技术手段进行土壤盐分反演,均取得较好结果;SONG等^[11]选取Landsat TM图像的数字波段、植被指数和地形指数作为变量,利用广义加性模型实现了土壤盐分的定量估测。在植被生长与土壤盐分含量关系遥感检测方面,多数****通过研究土壤盐分与植被覆盖、长势等的时空变化,寻找两者之间的定量关系,如贾吉超等^[12]通过多时相影像叠加分析,研究了黄河三角洲典型区域麦棉种植面积的变化与土壤盐分含量的关系,并建立了研究区土壤盐分与小麦的长势模型;张同瑞等^[13]采用多光谱数据,筛选敏感光谱植被指数,建立并优选出土壤盐分含量监测的最佳模型;ZHANG等^[14]通过对7种典型盐敏植物的高光谱数据与其根区土壤样品的盐分含量进行分析,探讨了黄河三角洲地区植被光谱与土壤盐度的关系;张天举等^[15]和张雪妮等^[16]依据多种植物群落的空间变化,研究与之对应的土壤盐分分布特征。在建模过程中,选择敏感光谱参数与最佳建模类型和方法是提高模型精度的重要环节,如安德玉等^[17]采用实测的高光谱数据进行多种模型的盐分反演,结合HICO波段进行模型修正,从而得到了可运用于大尺度的反演模型。对于不同来源、不同尺度的光谱数据,采用敏感光谱波段组合、光谱指数筛选、改进、微分等方法进行处理,建立多类土壤盐分估测模型,可以有效提升模型的土壤盐分反演效果^{[18,19,20,21,22,23,24,25]}。在模型类型的选择上,多数****使用线性回归方法和机器学习算法建立反演土壤盐分模型,比较了不同方法得到的反演模型的精度^{[26,27,28,29]}。【本研究切入点】在黄河三角洲滨海盐渍土典型区,采用星机一体化的技术方法,专门针对冬小麦种植区域的土壤盐分反演研究较为薄弱。【拟解决的关键问题】本文通过实地调查获取土壤盐分数据,筛选无人机多光谱波段并构建光谱参数,建立小麦光谱与土壤盐分间的多种关系模型,进而根据无人机图像与卫星影像关系进行模型修正,实现麦田土壤盐分估测模型的升尺度反演,最终得到研究区冬小麦种植区域的土壤盐分分布。

1 材料与方法

1.1 研究区概况

研究区选择山东省东营市垦利区,位于东经117°90′—119°82′,北纬37°01′—37°99′,是黄河三角洲滨海盐渍土的典型区域,地处暖温带半湿润大陆性季风气候区,光照充足降水少,蒸发大,旱涝不均,季节性干湿交替明显。研究区地势低平,自西南至东北降低,地表水来源为自然降水及黄河水,地下水矿化度高、埋深浅,主要土壤类型为盐土和潮土,盐化程度高,盐渍土分布广泛。种植的主要农作物为小麦、玉米和棉花,主要的种植区分布在区内西南部的地势较高区和东北部的黄河沿岸区2个部分,其土壤盐渍化变异明显,是本研究的理想区域。

综合垦利区域调查、地貌土壤及作物分布情况,分别在研究区西南部和东北部2个作物集中分布区,选择A、B 2个试验区,均处于麦田集中种植区域,耕作方式一致,灌排条件良好,小麦长势存在空间差异性,具有较好的代表性。其中A试验区为100 m×50 m的长方形区域,其南侧分布着灌溉沟渠及农用道路,B试验区为50 m×50 m的四边形区域,其西侧有灌溉沟渠及乡道,其余相邻区域均分布着麦田（图1）。

图1

新窗口打开|下载原图ZIP|生成PPT
图1研究区及试验区分布图

Fig. 1Distribution of study area and test area

1.2 地面数据采集与处理

4月份垦利降雨较少,地表盐化明显、稳定,各地类与农作物覆被差异显著,其他主要农作物均未播种,冬小麦处于返青拔节期,便于提取光谱特征^[30]。于2019年4月10日至16日进行了研究区野外调查,为确保布点均匀,研究区每5 km×5 km的网格内预布设3个样点,最终采集了位于麦区的77个样点。针对2个试验区,分别在A、B试验区的测量区域外围边界,每10 m放置一个插地牌,使用测绳连接后形成10 m×10 m的样点格网,以其格网交叉点作为采样点,最终共采集99个样点数据,其中A试验区63个,B试验区36个。

使用TrimbleGeo7手持式差分GPS,测量试验区四至点坐标作为无人机图像空间校正的地面控制点。将EC110便携式盐分计进行电导率温度校正后,对每个样点植株下方0—10 cm土层厚度的土壤电导率（EC）进行多次测量,待稳定后记录,取测量值均值作为各样点EC值,dS·m^-1。根据前期研究结果^[31],利用公式SS= 2.18×EC+0.727,将实测EC数据转换为土壤盐分含量（SS）,g·kg^-1。

1.3 无人机图像数据采集及预处理

使用Matrice600PRO六旋翼无人机搭载Sequoia多光谱相机采集多光谱数据,相机带有4个120万单色传感器：绿（b_G）、红（b_R）、红边（b_REG）、近红（b_NIR）光谱波段,中心波长分别为550、660、735、790 nm。阳光传感器可进行自校准,且带有GPS模块,每幅相片均存储了地理标记。飞行时,天气晴朗微风,飞行时间为10：00—14：00,飞行高度50 m,速度5 m·s^-1,地面分辨率2—3 cm。

使用Pix4D软件进行无人机多幅图像拼接及辐射校正,ENVI5.3进行波段叠加与图像裁剪,在ArcGIS10.1中使用控制点进行空间校正后展点,提取99个样点在4个波段的像元反射率。

1.4 卫星图像获取及预处理

Sentinel-2卫星是具有高分辨率、重访率及更新率的多光谱成像卫星,包含A、B 2颗小卫星,重访周期5 d,主要载荷为MSI多光谱成像仪,覆盖0.4—2.4 μm光谱范围,包含10 m（4个波段）、20 m（6个波段）、60 m（3个波段）的地面分辨率,可监测陆地植被生长、覆盖、健康状况,获取农作物种植、土地利用变化等信息。本文下载2019年4月17日的Sentinel-2A卫星影像（https://scihub.copernicus.eu/）,选择的Level-2A级产品是经辐射定标和大气校正的大气底层反射率数据。

选择与无人机光谱波段中心波长相近的Sentinel卫星波段：b₃、b₄、b₆、b₇,中心波长依次为560、665、740、783 nm,空间分辨率依次为10、10、20、20 m,在ENVI5.3中为影像统一投影坐标,进行重采样、边界裁剪、图层叠加,输出多波段影像。

1.5 基于无人机的土壤盐分估测模型构建与验证

根据植被冠层在可见光强吸收及近红外强反射的特性,使用无人机采集的多波段光谱计算归一化植被指数NDVI、差值植被指数DVI、比值植被指数RVI、绿波段比值植被指数GRVI和盐分指数SI（表1）。

Table 1
表1
表1光谱指数计算公式
Table 1Calculation formula of spectral index

光谱指数 Spectral index	计算公式 Computational formula	参考文献 Reference
归一化植被指数NDVI	(b_NIR-b_R)/( b_NIR+b_R)	[32]
比值植被指数RVI	b_NIR/b_R	[32]
差值植被指数DVI	b_NIR-b_R	[32]
绿波段比值植被指数GRVI	b_NIR/b_G	[32]
盐分指数SI	$\sqrt{bG\times bR}$	[33]

新窗口打开|下载CSV

将99个样本随机划分为66个建模样本和33个验证样本,对66个样点土壤盐分含量数据与无人机图像4个波段、5种光谱指数进行相关性分析,利用公式VIF=1/(1-r×r)（r为光谱参量间的相关系数）计算光谱参量间的方差膨胀因子（VIF）^[19],排除相关性较低或VIF>10,即无法通过共线性诊断的参数。对敏感波段及光谱指数,分别在IBM SPSS19和Matlab R2012b软件中使用线性逐步回归方法、偏最小二乘法,建立统计分析模型,在Matlab R2012b软件中使用BP神经网络、支持向量机方法建立机器学习模型,其中BP神经网络设置训练的迭代次数1 000,精度0.003,学习速率为0.01;支持向量机方法设置类型为v-SVR,选用高斯核函数（RBF）,通过网络搜索和交叉验证选择最佳惩罚因子C与核函数参数gamma,训练出支持向量机模型。

用33个样本验证所得模型,以决定系数R²为主要评价依据,结合均方根误差RMSE评价建模及验证精度,R²在0.50—0.65,0.66—0.81,0.82—0.90,0.91—1.00的闭区间内的预测能力依次为一般、中等、良好、优秀^[10]。分别筛选出4种方法中拥有较高R²及相对较低RMSE的土壤盐分估测模型,同时,基于试验区卫星影像数据,建立4个同方法、同光谱自变量的估测模型。通过比较两者的模型精度,以验证采用无人机多光谱建模是否提高了反演精度,确认无人机盐分估测模型进行升尺度修正及应用的必要性,其中BP神经网络模型根据其验证精度进行筛选和比较。

1.6 基于卫星影像的土壤盐分反演模型修正与应用

1.6.1 卫星影像的反射率修正及反演模型由于卫星数据及无人机图像数据存在光谱分辨率的差异,需要对卫星光谱波段进行修正,以期达到一体化的目的。在试验区每个样点对应的卫星影像单个像元范围内,提取无人机图像各波段的平均反射率b_{RⅴGⅴREGⅴNIR}及对应波段的卫星单一像元反射率b_{3ⅴ4ⅴ6ⅴ7},参照前人研究^[10],选用比值均值法确定卫星4个波段的修正系数C,公式如下：

（1）$C=\frac{\sum\nolimits_{i=1}^{n}{bR\vee G\vee REG\vee NIR}/b3\vee 4\vee 6\vee 7}{n}$

式中,n为参与计算的样点数量（n=99）,以各波段修正系数乘以对应卫星波段参量,对筛选出的模型中的光谱自变量进行修正后,得到升尺度土壤盐分反演模型。

1.6.2 反演模型验证导出卫星影像上研究区77个样点的各波段像元值,计算修正后的光谱参数。升尺度土壤盐分反演的统计分析模型代入修正后的光谱参数,得到的结果与冬小麦种植区实测盐分数据进行相关性分析;机器学习模型则直接将修正的光谱参数、实测盐分作为验证集放入BP神经网络及支持向量机模型中进行验证,根据R²、RMSE判断模型修正效果。

1.6.3 研究区冬小麦分布信息提取及土壤盐分反演为得到研究区中的麦区空间分布,使用ENVI对Sentinel-2A多波段影像划分小麦、居民点、水体、滩涂、荒地5类感兴趣区,统计不同地物在各个波段的感兴趣区像元值均值,计算NDVI均值,绘制折线以确定分类的阈值,建立决策树模型进行地物分类,结合目视解译提取研究区小麦种植区域。

综合试验区不同方法盐分估测模型精度及研究区升尺度修正反演模型的验证结果,筛选出2个尺度土壤盐分反演的最佳模型,使用该模型进行土壤盐分反演,得到试验区及冬小麦种植区盐分反演图,使用反距离加权（inverse distance weighted,IDW）方法进行土壤盐分实测值插值,得到2个尺度盐分等级分布图。统计反演图与实测插值图的等级面积比例,综合2个图的空间分布及不同等级占比变化趋势的一致性,结合实际调查分析模型反演效果。土壤盐渍化程度按照相关分级标准划分为5个等级（表2）^[34]。

Table 2
表2
表2垦利区土壤盐渍化程度分级标准
Table 2Grade standard of soil salinization degree in Kenli district

土壤盐渍化等级 Soil salinity grade	非盐渍化 Non-salinized	轻度盐渍化 Mild salinized	中度盐渍化 Moderate salinized	重度盐渍化 Severe salinized	盐土 Salinized soil
分级标准 Grade standard (g·kg^-1)	SS<1	1≤SS<2	2≤SS<4	4≤SS<6	SS>6
等级 Grade	1	2	3	4	5

新窗口打开|下载CSV

2 结果

2.1 光谱参数筛选

4个无人机光谱波段及5种光谱指数与试验区地面盐分数据进行的相关性分析结果如表3—4所示。在相关性矩阵中,4个波段与土壤盐分相关性均较好,光谱指数中的GRVI与土壤盐分相关性较低,DVI与NDVI间的r为0.953,VIF值大于10,存在较强的多重共线性,故选择4个波段和光谱指数NDVI、RVI、SI作为自变量进行建模。

Table 3
表3
表3试验区样点各波段与实测土壤盐分含量相关性
Table 3Correlation between the bands of sample points and the measured soil salinity content in test area

变量 Variable	SS	G	R	REG	NIR
SS	1
G	0.677**	1
R	0.667**	0.506**	1
REG	0.616**	0.431**	0.578**	1
NIR	-0.671**	-0.561**	-0.449**	-0.486**	1

** 表示在 0.01水平（双侧）上显著相关。下同
** indicates the significant correlation at 0.01 level(bilateral). The same as below

新窗口打开|下载CSV

Table 4
表4
表4试验区样点光谱指数与实测土壤盐分含量相关性
Table 4Correlation between the spectral indexes of sample points and the measured soil salinity content in test area

变量 Variable	SS	SI	NDVI	DVI	RVI	GRVI
SS	1
SI	0.770**	1
NDVI	-0.787**	-0.808**	1
DVI	-0.780**	-0.768**	0.953**	1
RVI	-0.560**	-0.696**	0.890**	0.844**	1
GRVI	-0.432**	-0.690**	0.468**	0.557**	0.398**	1

新窗口打开|下载CSV

2.2 土壤盐分估测模型的构建与验证

采用逐步回归、偏最小二乘、BP神经网络及SVM支持向量机方法,基于无人机多光谱图像,以66个建模样本的4个波段和3个光谱指数为自变量,土壤盐分含量为因变量建立土壤盐分估测模型,并使用33个验证样本对模型进行验证（表5）。可以看出,4种建模方法的13个模型中,由NDVI、RVI、SI建立的4个光谱指数模型的建模和验证精度均为最佳,均高于光谱波段模型。其中,建模精度最高的是支持向量机模型,R²为0.835,但其验证精度相比其他3个模型较低,说明该模型稳定性较差。从逐步回归模型可以看出,叠加敏感波段项、指数项可以逐步提高建模精度。使用4种建模方法,基于卫星影像,以指数NDVI、RVI、SI为自变量建立土壤盐分估测模型（表6）。通过比较发现,4个无人机光谱指数模型精度均高于同一方法的卫星光谱指数模型,表明采用无人机光谱建模可以提高土壤盐分的反演精度。因此,选择由NDVI、RVI、SI建立的4个无人机光谱指数模型进行下一步的升尺度模型修正与验证。

Table 5
表5
表5基于无人机多光谱的土壤盐分估测模型
Table 5Soil salinity estimation model based on multi-spectral of UAV

建模方法 Modeling approach	光谱参量 Spectral parameter	估测模型 Estimating model	建模精度 Modeling accuracy		验证精度 Verification accuracy
建模方法 Modeling approach	光谱参量 Spectral parameter	估测模型 Estimating model	R²	RMSE	R²	RMSE
逐步回归 Stepwise regression	b_G	Y=18.609×b_G+0.169	0.458	1.267	0.401	0.882
	b_G,b_R	Y=12.546×b_G+19.044×b_R-1.120	0.599	1.089	0.600	0.729
	b_G,b_R,b_NIR	Y=8.360×b_G+15.775×b_R-9.912×b_NIR +3.116	0.673	0.983	0.650	0.847
	b_G,b_R,b_NIR,b_REG	Y=7.988×b_G+12.282×b_R-8.525×b_NIR+6.979×b_REG+2.101	0.694	0.951	0.648	0.771
	NDVI	Y=-7.507×NDVI+6.308	0.620	1.061	0.668	0.685
	NDVI,RVI	Y=-13.261×NDVI+0.69×RVI+6.734	0.715	0.918	0.692	0.897
	NDVI,RVI,SI	Y=-10.287×NDVI+0.651×RVI+13.486×SI+3.843	0.756	0.850	0.710	0.907
偏最小二乘法 Partial least squares	b_G,b_R,b_NIR,b_REG	Y=6.021×b_G+6.5986×b_R+6.2650×b_NIR-4.1737×b_REG+1.3260	0.689	1.114	0.719	1.177
偏最小二乘法 Partial least squares	NDVI,RVI,SI	Y=-9.4774×NDVI+0.4794×RVI+3.0747×SI+5.0604	0.734	0.954	0.784	0.769
BP神经网络 The BP neural network	b_G,b_R,b_NIR,b_REG	—	—	—	0.714	0.893
BP神经网络 The BP neural network	NDVI,RVI,SI	—	—	—	0.753	0.993
支持向量机 Support vector machine	b_G,b_R,b_NIR,b_REG	—	0.804	0.590	0.467	0.473
支持向量机 Support vector machine	NDVI,RVI,SI	—	0.835	0.353	0.640	0.512

新窗口打开|下载CSV

Table 6
表6
表6基于卫星影像光谱指数的土壤盐分估测模型
Table 6Soil salinity estimation model based on spectral index of satellite image

建模方法 Modeling approach	估测模型 Estimating model	建模精度 Modeling accuracy		验证精度 Verification accuracy
建模方法 Modeling approach	估测模型 Estimating model	R²	RMSE	R²	RMSE
逐步回归Stepwise regression	Y=-97.012×NDVI+22.298×RVI-13.905×SI-8.489	0.509	1.190	0.434	0.874
偏最小二乘法Partial least squares	Y=-57.4889×NDVI+13.3418×RVI+21.9667×SI-9.0323	0.555	0.940	0.414	0.964
BP神经网络The BP neural network	—	—	—	0.602	0.900
支持向量机Support vector machine	—	0.612	1.166	0.438	0.432

新窗口打开|下载CSV

2.3 反演模型的升尺度修正与验证

2.3.2 反演模型修正在试验区每个样点对应卫星单个像元范围内,统计无人机各波段图像多个小像元的平均反射率,提取样点的卫星影像反射率,通过比值均值法,计算出卫星数据每个波段的修正系数为0.82483（b₃）、0.63767（b₄）、0.51249（b₆）和1.15859（b₇）,将4个模型的自变量NDVI、RVI、SI使用卫星波段乘以修正系数进行修正,从而将基于无人机平台的土壤盐分估测模型转换为基于卫星影像的反演模型。修正后的基于卫星图像的各模型自变量为：

（2）$NDVI1=\frac{1.15859\times b7-0.63767\times b4}{1.15859\times b7+0.63767\times b4}$

（3）$RVI1=1.81691\frac{b7}{b4}$

（4）$SI1=0.72524\sqrt{b3\times b4}$

2.3.3 反演模型验证根据研究区中77个麦区样点4个波段反射率,计算已修正的指数NDVI₁、RVI₁、SI₁,分别代入4个修正后的反演模型,结合实测土壤盐分含量进行验证。其中逐步回归反演模型Y=-10.287×NDVI₁+0.651×RVI₁+13.486×SI₁+3.843的验证R²为0.485,RMSE为1.339;偏最小二乘模型Y=-9.4774×NDVI₁+0.4794×RVI₁+3.0747×SI₁+5.0604的验证R²为0.513,RMSE为1.379。同样地,将指数NDVI₁、RVI₁、SI₁作为测试集放入已构建好的BP神经网络和支持向量机模型进行验证,结果R²分别为0.436、0.387,RMSE分别为1.297、1.006。机器学习模型的验证RMSE相对较低,而统计分析模型验证R²均高于机器学习模型,其中仅偏最小二乘法构建的指数模型R²大于0.5。

2.4 模型应用与盐分反演

综合建模及2个尺度的验证结果,4个模型中偏最小二乘法建立的光谱指数具有更好的适用性和稳定性,选用该模型进行2个尺度的土壤盐分反演。

2.4.1 试验区土壤盐分反演基于试验区土壤盐分实测值和最佳模型,得到试验区土壤盐分等级分布（图2）,并对其进行等级面积统计（表7）,总体看,麦田土壤盐分的反演结果与实测结果基本一致,各等级面积占比的变化趋势大致相同,而比较看,反演结果对土壤盐分空间分布的反映则更为精细。试验区A总体盐渍化程度较高,无非盐渍化及轻度盐渍化土壤分布,中、重度盐渍化土壤面积占83.72%,盐土面积占16.28%。土壤盐分含量呈现南部高、北部低趋势,其中盐土集中分布于西南部区域。试验区B总体盐渍化程度偏低,其中非盐化和轻度盐渍化土壤面积比例为39.11%,集中在小麦长势良好的北部,而试验区中部和南部重度盐渍化土壤面积达50.63%,且空间分布差异不大。

图2

新窗口打开|下载原图ZIP|生成PPT
图2试验区土壤盐分实测插值图（左）及反演图（右）

Fig. 2Interpolation diagram (left) and inversion diagram (right) of soil salinity in the test area

Table 7
表7
表7试验区土壤盐分等级面积统计
Table 7Grade-area statistics of soil salinity in test area (%)

土壤盐分等级分布图 Distribution diagram of soil salinity grade	试验区A Test area A					试验区B Test area B
土壤盐分等级分布图 Distribution diagram of soil salinity grade	1	2	3	4	5	1	2	3	4	5
实测图Interpolation diagram	0	0	47.63	36.16	16.21	32.74	0	9.81	57.45	0
反演图Inversion diagram	0	0	47.34	36.38	16.28	31.27	7.84	8.10	50.63	2.16

新窗口打开|下载CSV

2.4.2 研究区麦田面积与分布对研究区典型地物光谱曲线进行分析（图3）,在b₇波段各地类出现清晰反射率区间,水体、滩涂平均反射率小于0.2,首先被去除,NDVI均值大于0.2的地物为冬小麦,由此建立决策树模型,提取冬小麦种植区域（图4）。结果显示,研究区麦田集中分布在垦利西南部及东北部黄河沿岸区域,提取结果符合冬小麦实际分布。

图3

新窗口打开|下载原图ZIP|生成PPT
图3研究区典型地物光谱曲线（左）及冬小麦决策树提取模型（右）

Fig. 3Spectral curve of typical features in the study area (left) and extraction model of winter wheat decision tree (right)

图4

新窗口打开|下载原图ZIP|生成PPT
图4麦区土壤盐分实测插值图（左）及升尺度反演图（右）

Fig. 4Interpolation diagram (left) and upscaling inversion diagram (right) of soil salinity in wheat area

2.4.3 研究区麦田土壤盐分升尺度反演对比研究区麦区土壤盐分反演及实测插值结果（图4）,结合其等级面积统计情况（表8）,可以看出,土壤盐分反演结果与空间插值结果一致,面积占比均呈现出随着盐渍化程度升高而减少的变化趋势,与调查情况相符,表明麦区升尺度反演效果较好。研究区小麦种植区无非盐渍化土壤分布,轻度盐渍化土壤分布广泛,面积占比达73.09%,集中在地势相对较高的西南部和黄河淡水影响的东北部区域;中度盐渍化土壤面积次之,占比14.01%,且在麦区零散分布;重度盐渍化土壤及盐土面积较小,集中分布于研究区中西部沿黄河滩区。

Table 8
表8
表8麦区土壤盐分等级面积统计
Table 8Grade-area statistics of soil salinity in wheat area (%)

土壤盐分等级分布图 Distribution diagram of soil salinity grade	1	2	3	4	5
实测图Interpolation diagram	0	63.07	18.28	14.45	4.20
反演图Inversion diagram	0	73.09	14.01	10.08	2.82

新窗口打开|下载CSV

3 讨论

在星机一体化的土壤盐分反演研究中,针对农作物种植区域的土壤盐分反演研究相对匮乏,需要深入细致的研究探索。本研究为提高采集的农作物光谱精度,使用格网交点采集地面数据,保障样点的均匀及位置准确,无人机飞行高度较低,搭载 Sequoia多光谱相机的分辨率高达2—3 cm,保证了建模精度,因此,反演结果能更精确地反映小麦种植区的盐分分布状况。

植被覆盖差异与土壤盐分空间变化息息相关,特征光谱参数的选择是土壤盐分准确反演的重要基础。本研究中,2个试验区冬小麦长势、覆盖度差异较明显,样点间可见光与近红外光谱差异增强,选用光谱指数进行土壤盐分反演,增强了原本与土壤盐分含量相关性接近的4个无人机光谱波段的敏感程度,提高了模型的精度及空间普适性,与前人研究结果相符^{[28, 35-36]}。光谱指数中,GRVI及RVI与土壤盐分相关性不高,这与前人研究中波段比值与土壤盐分相关性较低的结果一致^[22]。

本研究通过评价模型的精度、不同尺度的稳定性和适应性筛选最佳土壤盐分反演模型。选择的评价指标决定系数R²衡量模型的拟合程度,RMSE反映实测值与预测值的偏差,考虑前者性能度量在0—1之间,而后者对于数据没有准确的度量范围,故模型精度评价以R²作为主要依据。本文建立的统计分析模型的建模R²小于支持向量机模型,说明机器学习方法的数据拟合效率较高,这与前人机器学习算法土壤盐分估测效果优于多元线性回归方法的研究结果一致^[29]。但模型升尺度验证时,4个模型精度均降低,可能与无人机和卫星波段的中心波长存在5—13 nm的差异且大尺度农作物光谱状况较为复杂有关。同时,由于BP神经网络存在数据过拟合现象,使其预测能力与训练能力出现矛盾,而支持向量机模型对新输入样本的泛化能力低^[37,38] ,可能导致了机器学习模型的稳定性较低,精度降幅较大。

在试验时期的考察选点尽量做到了均匀且具有代表性,使其插值结果较好地反映土壤盐分的空间分布情况,但插值方法以距离权重进行赋值,在部分小区域内盐分出现较大变化时,会在一定程度上影响插值准确性。试验区的土壤盐分反演结果与土壤盐分的实测插值图比较时,2个图北部的盐土区域分布存在一定的差异,此外,在试验区和研究区需要减少对插值图定量结果的依赖,而是要结合空间分布情况、各等级占比的相对变化趋势及实际调查结果综合验证反演模型的应用效果。

本研究针对滨海冬小麦作物种植区,实现了多尺度、多类建模方法的土壤盐分遥感反演,在此基础上,后续需考虑区域中玉米、水稻等其他农作物类型,筛选适宜的光谱参数及高效反演模型,开展对农作物种植区的土壤盐分系统定量研究;另外,为提高升尺度反演模型精度,升尺度反演中尺度间数据的修正方法也有待进一步的优化。

4 结论

（1）基于冬小麦试验区无人机图像的多光谱波段,构建光谱指数,4个波段与土壤盐分含量的相关性均大于0.6,5个光谱指数中筛选出了用于建模的NDVI、RVI、SI,使用4种方法构建了9个统计分析模型和4个机器学习模型,不同方法中由NDVI、RVI、SI建立的光谱指数模型综合精度均优于波段模型和基于卫星影像的光谱指数模型。

（2）红、绿、红边、近红波段进行升尺度修正的修正系数分别为0.82483、0.63767、0.51249、1.15859,对4个试验区估测模型进行修正,构建了基于卫星波段的升尺度土壤盐分反演模型,经验证筛选出麦区最佳反演模型Y=-9.4774×NDVI₁+0.4794×RVI₁+3.0747× SI₁+ 5.0604,验证R²为0.513。

（3）由最佳模型反演得到了试验区及研究区2个尺度的麦区土壤盐分等级分布图,反演效果较好,表明模型有良好的预测能力和适用性。研究区小麦种植区土壤以轻度盐渍化为主,面积占73.09%,集中在研究区地势相对较高的西南部和黄河淡水影响的东北部区域,中度盐渍化土壤在麦区零散分布,重度盐渍化及盐土则分布于中西部沿黄滩区。

本研究综合卫星、无人机及地面数据,采用统计分析和机器学习的建模方法,得到了星机一体化的麦区土壤盐分反演最佳模型,为把握麦区土壤盐分等级分布状况、指导研究区农业生产提供了科学依据。

参考文献原文顺序
文献年度倒序
文中引用次数倒序
被引期刊影响因子

[1]

王建华, 胡鹏, 龚家国. 实施黄河口大保护推动黄河流域生态文明建设
人民黄河, 2019,41(10):7-10.

[本文引用: 1]

WANG J

, HU

, GONG J

. Taking the great protection of the Yellow River estuary as the grasp to promote the construction of ecological civilization in the Yellow River basin
Yellow River, 2019,41(10):7-10. (in Chinese)

[本文引用: 1]

[2]

刘丽娟, 李小玉. 干旱区土壤盐分积累过程研究进展
生态学杂志, 2019,38(3):891-898.

[本文引用: 1]

LIU L

, LI X

. Progress in the study of soil salt accumulation in arid region
Chinese Journal of Ecology, 2019,38(3):891-898. (in Chinese)

[本文引用: 1]

[3]

张顺, 贾永刚, 连胜利, 付腾飞, 潘玉英. 电导率法在土壤盐渍化中的改进和应用进展
土壤通报, 2014,45(3):754-759.

[本文引用: 1]

ZHANG

, JIA Y

, LIAN S

, FU T

, PAN Y

. Application and improvement of electrical conductivity measurements in soil salinity
Chinese Journal of Soil Science, 2014,45(3):754-759. (in Chinese)

[本文引用: 1]

[4]

阿尔达克·克里木, 塔西甫拉提·特依拜, 张东, 依力亚斯江·努尔麦麦提. 基于高光谱的ASTER影像土壤盐分模型校正及验证
农业工程学报, 2016,32(12):144-150.

URL [本文引用: 1]

快速准确地获取土壤盐分信息是监测和治理土壤盐渍化现象的重要前提。该文以新疆维吾尔自治区典型盐渍化区域--艾比湖流域为研究区，analytical spectral devices（ASD）光谱仪采集的土壤高光谱数据和advanced space borne thermal emission and reflection radiometer（ASTER）影像为数据源，结合实测土壤盐分含量信息，对遥感定量反演土壤盐渍化现象进行研究。再经过光谱反射率数学变换后，结合相关性分析，利用多元回归方法分别建立基于重采样后的高光谱和影像光谱的土壤含盐量估算模型，对遥感影像光谱盐分估算模型进行校正，以提高遥感定量监测盐渍化土壤的精度。结果表明：ASTER影像光谱反射率二阶导数变换和ASD重采样光谱的对数的二阶导数变换所建立的盐分估算模型最佳，决定系数R2分别为0.59和0.82。经ASD重采样光谱模型校正后的ASTER影像光谱的盐分估算模型精度R2为0.91，有效地提高大尺度条件下土壤盐渍化反演精度。研究为大尺度土壤盐分定量遥感监测提供了一种有效方法。

ARDAK K L

, TAXIPULAT T Y

, ZHANG

, ILYASJ

N R M M T

. Calibration and validation of soil salinity estimation model based on measured hyperspectral and Aster image
Transactions of the Chinese Society of Agricultural Engineering, 2016,32(12):144-150. (in Chinese)

URL [本文引用: 1]

吴亚坤, 刘广明, 苏里坦, 杨劲松. 多源数据的区域土壤盐渍化精确评估
光谱学与光谱分析, 2018,38(11):3528-3533.

[本文引用: 1]

WU Y

, LIU G

, SU L

, YANG J

. Accurate evaluation of regional soil salinization using multi-source data
Spectroscopy and Spectral Analysis, 2018,38(11):3528-3533. (in Chinese)

[本文引用: 1]

[6]

胡盈盈, 王瑞燕, 陈红艳, 李玉萍, 刘燕群. 黄河三角洲春秋两季土壤盐分遥感反演及时空变异研究
测绘与空间地理信息, 2018,41(8):78-81.

[本文引用: 1]

HU Y

, WANG R

, CHEN H

, LI Y

, LIU Y

. Research on remote sensing retrieval and temporal variation of soil salinity in the spring and autumn of the Yellow River Delta
Geomatics & Spatial Information Technology, 2018,41(8):78-81. (in Chinese)

[本文引用: 1]

[7]

SCUDIERO

, SKAGGS T

, CORWIN D

. Regional-scale soil salinity assessment using Landsat ETM+ canopy reflectance
Remote Sensing of Environment, 2015,169:335-343.

DOI:10.1016/j.rse.2015.08.026 URL [本文引用: 1]

[8]

梁静, 丁建丽, 王敬哲, 王飞. 基于反射光谱与 Landsat 8 OLI 多光谱数据的艾比湖湿地土壤盐分估算
土壤学报, 2019,56(2):320-330.

[本文引用: 1]

LIANG

, DING J

, WANG J

, WANG

. Quantitative estimationand mapping of soil salinity in the Ebinur Lake wetland based on Vis-NIR reflectance and Landsat 8 OLI data
Acta Pedologica Sinica, 2019,56(2):320-330. (in Chinese)

[本文引用: 1]

[9]

陈俊英, 王新涛, 张智韬, 韩佳, 姚志华, 魏广飞. 基于无人机-卫星遥感升尺度的土壤盐渍化监测方法
农业机械学报, 2019,50(12):161-169.

[本文引用: 1]

CHEN J

, WANG X

, ZHANG Z

, HAN

, YAO Z

, WEI G

. Soil salinization monitoring method based on UAV-satellite remote sensing scale-up
Transactions of the Chinese Society for Agricultural Machinery, 2019,50(12):161-169. (in Chinese)

[本文引用: 1]

[10]

ZHANG S

, ZHAO G

. A harmonious satellite-unmanned aerial vehicle-ground measurement inversion method for monitoring salinity in coastal saline soil
Remote Sensing, 2019,11(14):1700.

DOI:10.3390/rs11141700 URL [本文引用: 3]

[11]

SONG C

, REN H

, HUANG

. Estimating soil salinity in the Yellow River Delta, Eastern China—An integrated approach using spectral and terrain indices with the generalized additive model
Pedosphere, 2016, 26(5):626-635.

URL [本文引用: 1]

Soil salinity is one of the most severe environmental problems worldwide. It is necessary to develop a soil-salinity-estimation model to project the spatial distribution of soil salinity. The aims of this study were to use remote sensed images and digital elevation model (DEM) to develop quantitative models for estimating soil salinity and to investigate the influence of vegetation on soil salinity estimation. Digital bands of Landsat Thematic Mapper (TM) images, vegetation indices, and terrain indices were selected as predictive variables for the estimation. The generalized additive model (GAM) was used to analyze the quantitative relationship between soil salt content, spectral properties, and terrain indices. Akaike’s information criterion (AIC) was used to select relevant predictive variables for fitted GAMs. A correlation analysis and root mean square error between predicted and observed soil salt contents were used to validate the fitted GAMs. A high ratio of explained deviance suggests that an integrated approach using spectral and terrain indices with GAM was practical and efficient for estimating soil salinity. The performance of the fitted GAMs varied with changes in vegetation cover. Salinity in sparsely vegetated areas was estimated better than in densely vegetated areas. Red, near-infrared, and mid-infrared bands, and the second and third components of the tasseled cap transformation were the most important spectral variables for the estimation. Variable combinations in the fitted GAMs and their contribution varied with changes in vegetation cover. The contribution of terrain indices was smaller than that of spectral indices, possibly due to the low spatial resolution of DEM. This research may provide some beneficial references for regional soil salinity estimation.

[12]

贾吉超, 赵庚星, 高明秀, 王卓然, 常春艳, 姜曙千, 李晋. 黄河三角洲典型区域冬小麦播种面积变化与土壤盐分关系研究
植物营养与肥料学报, 2015,21(5):1200-1208.

DOI:10.11674/zwyf.2015.0513 URL [本文引用: 1]

【目的】将土壤盐分含量与冬小麦分布变化结合,分析两者之间的时空关系,旨在探索土壤盐碱化对冬小麦种植的影响,为冬小麦生产决策提供科学依据。【方法】以黄河三角洲垦利县为研究区,采用2003年4月、2008年4月和2013年3月三期ETM影像,通过分析典型地物光谱曲线生成决策树模型,提取冬小麦分布信息,将各时相冬小麦种植分布提取结果做空间叠加,分析了近10年来冬小麦面积与分布的变化规律;并结合实地土壤盐分调查分析数据,分析了冬小麦种植面积变化与土壤盐分的关系。一方面,将垦利县冬小麦分布图分别与相应时相的土壤含盐量分布图进行空间叠加分析,并对叠加图的属性进行统计,对比分析冬小麦分布与土壤含盐量分布的关系。另一方面,通过叠加2008和2013年土壤盐分含量分布图,将盐分变化分为盐分升高区和盐分降低区,将其与同时段的冬小麦种植范围变化图进行叠加,分析土壤盐分含量变化对冬小麦分布变化的影响。【结果】1)垦利县冬小麦的分布具有明显的空间特征,主要分布在垦利县域西南部和东北部黄河沿岸两个区域,与土壤低含盐量区具有一致的空间分布特征。2)垦利县冬小麦种植面积呈现2003~2008时段大幅减少和2008~2013时段的少许增加趋势。3)冬小麦种植范围变化与土壤盐分含量的相关性极高,冬小麦种植无变化区域土壤含盐量都集中在1.5~2.5 g/kg之间,冬小麦种植增加区域的土壤盐分含量集中在2~3 g/kg,而冬小麦种植减少区的土壤盐分含量都在3 g/kg以上,即超过3 g/kg的土壤含盐量已不再适合冬小麦的生长。4)2008~2013年垦利县冬小麦分布区域变化显著受到土壤含盐量的变化。在土壤含盐量降低的小麦区域中,冬小麦种植增加区和不变区的面积占98.07%,而在土壤含盐量升高的小麦区域中,冬小麦种植减少的面积占84.54%。【结论】冬小麦种植范围及其变化显著受到土壤盐分状况及其变化的影响,冬小麦种植减少区的土壤盐分含量都在3 g/kg以上,且随着土壤含盐量的升高冬小麦种植面积骤减,3 g/kg的土壤含盐量是适合冬小麦生长的上限,土壤含盐量调控是维持和扩大冬小麦种植范围的关键手段。

JIA J

, ZHAO G

, GAO M

, WANG Z

, CHANG C

, JIANG S

, LI

. Study on the relationship between winter wheat sowing area changes and soil salinity in the typical area of the Yellow River Delta
Journal of Plant Nutrition and Fertilizers, 2015,21(5):1200-1208. (in Chinese)

DOI:10.11674/zwyf.2015.0513 URL [本文引用: 1]

张同瑞, 赵庚星, 高明秀, 王卓然, 贾吉超, 李萍, 安德玉. 基于近地面多光谱的黄河三角洲典型地区土壤含盐量估算研究
光谱学与光谱分析, 2016,36(1):248-253.

URLPMID:27228776 [本文引用: 1]

This study chooses the core demonstration area of 'Bohai Barn' project as the study area, which is located in Wudi, Shandong Province. We first collected near-ground and multispectral images and surface soil salinity data using ADC portable multispectral camera and EC110 portable salinometer. Then three vegetation indices, namely NDVI, SAVI and GNDVI, were used to build 18 models respectively with the actual measured soil salinity. These models include linear function, exponential function, logarithmic function, exponentiation function, quadratic function and cubic function, from which the best estimation model for soil salinity estimation was selected and used for inverting and analyzing soil salinity status of the study area. Results indicated that all models mentioned above could effectively estimate soil salinity and models using SAVI as the dependent variable were more effective than the others. Among SAVI models, the linear model (Y = -0.524x + 0.663, n = 70) is the best, under which the test value of F is the highest as 141.347 at significance test level, estimated R2 0.797 with a 93.36% accuracy. Soil salinity of the study area is mainly around 2.5 per thousand - 3.5 per thousand, which gradually increases from southwest to northeast. The study has probed into soil salinity estimation methods based on near-ground and multispectral data, and will provide a quick and effective technical soil salinity estimation approach for coastal saline soil of the study area and the whole Yellow River Delta.

ZHANG T

, ZHAO G

, GAO M

, WANG Z

, JIA J

, LI

, AN D

. Soil salinity estimation based on near-ground multispectral imagery in typical area of the Yellow River Delta
Spectroscopy and Spectral Analysis, 2016,36(1):248-253. (in Chinese)

URLPMID:27228776 [本文引用: 1]

ZHANG T

, ZENG S

, GAO

, QUYANG Z

, LI

, FANG C

, ZHAO

. Using hyperspectral vegetation indices as a proxy to monitor soil salinity
Ecological Indicators, 2011,11(6):1552-1562.

DOI:10.1016/j.ecolind.2011.03.025 URL [本文引用: 1]

The spectral bands most sensitive to salt-stress across diverse plants have not yet been defined; therefore, the predictive ability of previous vegetation indices (VIs) may not be satisfied for salinization monitoring. The hyperspectra of seven typical salt-sensitive/halophyte species and their root-zone soil samples were collected to investigate the relationship between vegetation spectra and soil salinity in the Yellow River Delta (YRD) of China. Several VIs were derived from the recorded hyperspectra and their predictive power for salinity was examined. Next, a univariate linear correlogram as well as multivariate partial least square (PLS) regression was employed to investigate the sensitive bands. VIs examination and band investigation confirmed that the responses of the vegetation differed from species to species, which explained the vibrations of the VIs in many study cases. These differences were primarily between salt-sensitive and halophyte plants, with the former consistently having higher sensitivity than the latter. With the exception of soil adjusted vegetation index (SAVI), most VIs were found to have weak relationships with soil salinity (with average R-2 of 0.28) and some were not sensitive to all species [e.g. photochemical reflectance index (PRI) and red edge position (REP)I, which verified that most currently available VIs are not adequate indicators of salinity for various species. PLS was validated as a more useful tool than linear correlogram for identification of sensitive bands due to well dealing with multicollinear spectral variables. From PLS, wavelengths at 395-410, 483-507, 632-697, 731-762, 812-868, 884-909, and 918-930 nm were determined to be the most sensitive bands. By combining the most sensitive bands in a SAVI form, we finally proposed four soil adjusted salinity indices (SASIs) for all species. Satisfactory relationships were observed between ECe and four SASIs for all species, with largely improved R-2 values ranging from 0.50 to 0.58. Our findings indicate the potential to monitor soil salinity with the hyperspectra of salt-sensitive and halophyte plants. (C) 2011 Elsevier Ltd.

[15]

张天举, 陈永金, 刘加珍. 黄河三角洲湿地不同植物群落土壤盐分分布特征
浙江农业学报, 2018,30(11):1915-1924.

[本文引用: 1]

ZHANG T

, CHEN Y

, LIU J

. Characteristics of spatial distribution of soil salinity under different plant communities on wetland in Yellow River Delta
Acta Agriculturae Zhejiangensis, 2018,30(11):1915-1924. (in Chinese)

[本文引用: 1]

[16]

张雪妮, 吕光辉, 杨晓东, 贡璐, 秦璐, 何学敏, 刘昊奇. 基于盐分梯度的荒漠植物多样性与群落、种间联接响应
生态学报, 2013,33(18):5714-5722.

DOI:10.5846/stxb201306071403 URL [本文引用: 1]

土壤盐分是影响干旱区荒漠植物群落动态的决定因素之一。基于样方调查和不同土壤盐分梯度下植物多样性指数及群落与种间关联的计算结果,分析干旱区荒漠群落植物多样性、群落联接性和种间关联对土壤盐分梯度的响应动态。结果表明,在土壤盐含量为0.03%-0.55% (S1)、0.61%-1.24% (S2)和1.41%-1.79% (S3)的盐分梯度上,(1)随土壤盐含量升高,群落生活型结构改变,草本比例减少,乔木比例增加;(2)植物多样性指数随土壤盐分增加而下降,低盐分梯度下,二者极显著正相关(P<0.01),中盐梯度下二者间呈部分显著负相关(P<0.05),高盐梯度下则转为以正相关为主(P>0.05);(3)群落联接性随土壤盐分梯度转变,在0.61%-1.24%的中度盐含量时正联接性最强(VR=1.89),高盐度含量下群落转为不显著的负联接,稳定性降低;(4)沿盐分升高梯度,种间的负相关种对数增加,正相关种对数减少,种间关联(相关)强度提高,正负相关种对比(PNR)与多样性指数呈显著正相关(P<0.05)。综上可知,干旱区荒漠植物多样性在土壤盐含量达到1.41%-1.79%水平时总体显著降低;土壤盐分水平显著影响植物群落和种间联接性,种间互利性随盐分梯度增加下降,物种趋于独立分布,并最终导致荒漠植物多样性降低。

ZHANG X

, Lü G

, YANG X

, GONG

, QIN

, HE X

, LIU H

. Responses of desert plant diversity, community and interspecific association to soil salinity gradient
Acta Ecologica Sinica, 2013,33(18):5714-5722. (in Chinese)

DOI:10.5846/stxb201306071403 URL [本文引用: 1]

安德玉, 邢前国, 赵庚星. 基于HICO波段的滨海土壤盐分遥感反演研究
海洋学报, 2018,40(6):51-59.

[本文引用: 1]

AN D

, XING Q

, ZHAO G

. Hyperspectral remote sensing of soil salinity for coastal saline soil in the Yellow River Delta based on HICO bands
Acta Oceanologica Sinica, 2018,40(6):51-59. (in Chinese)

[本文引用: 1]

[18]

张素铭, 赵庚星, 王卓然, 肖杨, 郎坤. 滨海盐渍区土壤盐分遥感反演及动态监测
农业资源与环境学报, 2018,35(4):349-358.

[本文引用: 1]

ZHANG S

, ZHAO G

, WANG Z

, XIAO

, LANG

. Remote sensing inversion and dynamic monitoring of soil salt in coastal saline area
Journal of Agricultural Resources and Environment, 2018, 35(4):349-358. (in Chinese)

[本文引用: 1]

[19]

陈红艳, 赵庚星, 陈敬春, 王瑞燕, 高明秀. 基于改进植被指数的黄河口区盐渍土盐分遥感反演
农业工程学报, 2015,31(5):107-114.

URL [本文引用: 2]

快速获取土壤盐分的含量、特征及空间分布信息是盐渍土治理、利用的客观需求。该文针对黄河三角洲盐渍土，以垦利县为例，基于Landsat 8 OLI 多光谱影像，在传统植被指数的基础上引入短波红外波段进行扩展，提出了改进植被指数；然后基于改进前后对应的植被指数，分别采用多元逐步回归（multivariable linear regression，MLR）、反向传播神经网络（back propagation neural networks，BPNN）和支持向量机（support vector machine，SVM）方法构建土壤盐分含量的遥感反演模型，并进行模型验证、对比和优选；最后基于最佳模型进行研究区土壤盐分含量的空间分布反演和分析。结果显示：相对传统植被指数，扩展后植被指数可增强与土壤盐分的相关性，大幅降低指数间的多重共线性；采用上述3种方法建模，改进后模型的精度比改进前都有提高，验证集决定系数R2提高0.04～0.10，均方根误差RMSE降低0.13～0.73，相对分析误差RPD提高0.25～0.34，改进后模型RPD均大于2.0，普遍达到性能良好；对比3种建模方法，SVM建模精度最高，BPNN模型次之，MLR分析精度最低，最佳模型为基于改进植被指数的土壤盐分含量支持向量机反演模型，建模集R2和RMSE为0.75、3.48，验证集R2、RMSE和RPD为0.78、3.02和2.56，模型较为准确、可靠；基于该模型反演的研究区土壤盐分含量整体较高，盐渍化程度空间分布表现为自西南部农业生产区至东北沿海区域逐渐加重，与实地调查一致。研究表明基于Landsat 8 OLI多光谱影像，引入第7波段对植被指数进行改进，从而构建土壤盐分含量的支持向量机模型，可获得较好的土壤盐分空间分布反演结果。

CHEN H

, ZHAO G

, CHEN J

, WANG R

, GAO M

. Remote sensing inversion of saline soil salinity based on modified vegetation index in estuary area of Yellow River
Transactions of the Chinese Society of Agricultural Engineering, 2015,31(5):107-114. (in Chinese)

URL [本文引用: 2]

周晓红, 张飞, 张海威, 张贤龙, 袁婕. 艾比湖湿地自然保护区土壤盐分多光谱遥感反演模型
光谱学与光谱分析, 2019,39(4):12 29-1235.

[本文引用: 1]

ZHOU X

, ZHANG

, ZHANG H

, ZHANG X

, YUAN

. A study of soil salinity inversion based on multispectral remote sensing index in Ebinur Lake Wetland Nature Reserve
Spectroscopy and Spectral Analysis, 2019,39(4):1229-1235. (in Chinese)

[本文引用: 1]

[21]

张贤龙, 张飞, 张海威, 李哲, 海清, 陈丽华. 基于光谱变换的高光谱指数土壤盐分反演模型优选
农业工程学报, 2018,34(1):110-117.

URL [本文引用: 1]

该文探索基于光谱变换建立光谱指数，进而建立土壤盐分反演模型的可行性。运用倒数、导数、对数等15种光谱变换对土壤含盐量进行反演，并利用原始光谱的波段反射率构造光谱指数对土壤盐分进行建模。在15种高光谱变换中，一阶微分R?和一阶对倒数（log1/R?）变换下土壤盐分估算模型的精度较高。但总体而言，基于单一光谱变换和光谱指数的模型模拟精度均较低。采用光谱变换建立光谱指数，并进一步建立土壤盐分反演模型，结果表明，基于（log1/R?）光谱变换构建归一化植被指数，然后建立的土壤盐分精度最高，经验证，其R2为0.89，均方根误差为3.34 g/kg，高于单一方法构建的模型，可为半干旱地区土壤盐分反演提供参考。

ZHANG X

, ZHANG

, ZHANG H

, LI

, HAI

, CHEN L

. Optimization of soil salt inversion model based on spectral transformation from hyperspectral index
Transactions of the Chinese Society of Agricultural Engineering, 2018,34(1):110-117. (in Chinese)

URL [本文引用: 1]

王丹阳, 陈红艳, 王桂峰, 丛津桥, 王向锋, 魏学文. 无人机多光谱反演黄河口重度盐渍土盐分的研究
中国农业科学, 2019,52(10):1698-1709.

DOI:10.3864/j.issn.0578-1752.2019.10.004 URL [本文引用: 2]

【Objective】The purpose of this paper was to improve the extraction accuracy of soil salinity information based on remote sensing and understand accurately the degree and distribution of soil salinization. 【Method】Firstly, the severe and concentrated saline soil area of Huanghekou town, Kenli district, was selected as the experimental area, and the unmanned aerial vehicle (UAV) equipped with Sequoia multispectral camera was adopted to acquire the near earth remote sensing image from April 26th to 28th, 2018, then the image preprocessing, including image splicing, radiation correction, orthorectification and geometric correction, was performed. Secondly, the sensitive bands of soil salinity were screened by correlation analysis and grey correlation analysis, respectively, and the spectral parameters were constructed and screened. Thirdly, the soil salinity quantitative analysis models were built by multivariate linear regression (MLR), support vector machine (SVM) and partial least square (PLS) method, then the models’ accuracy was evaluated and the best one was selected. Finally, the best model was applied to the inversion and analysis of soil salinity distribution in the experimental area, and the inversion accuracy was compared with the interpolation result by inverse distance weighting (IDW) method. 【Result】The results showed that the accuracy and significance of the estimation model based on gray correlation analysis were improved by compared with the correlation analysis; Compared the three modeling methods, the prediction ability of the SVM was the best, followed by the PLS, the MLR models’ precision was the lowest, with the calibration R ² and RMSE of 0.820 and 3.626, the validation R ², RMSE and RPD of 0.773, 4.960 and 2.200, and the SVM model of soil salinity based on screened variables by grey correlation analysis was selected the best one; Based on the best model, the soil salinity content in this region was between 0.323 and 21.210 g·kg ^-1 with the average of 6.871 g·kg ^-1 and the severe salinity accounted for 58.094%, which was consistent with the result of the field investigation; The 80% of the error between the inversion result and the interpolation result by the IDW method was controlled within 20% of the sample salt content average, which showed that the two kind of result were similar. 【Conclusion】It could be concluded that the accurate extraction of severe soil salinity information could be achieved on the UAV multi-spectra.

WANG D

, CHEN H

, WANG G

, CONG J

, WANG X

, WEI X

. Salinity inversion of severe saline soil in the Yellow River estuary based on UAV multi-spectra
Scientia Agricultura Sinica, 2019,52(10):1698-1709. (in Chinese)

DOI:10.3864/j.issn.0578-1752.2019.10.004 URL [本文引用: 2]

HAMZEH

, NASERI A

, ALAVIPANAH S

, MOJARADI

, BARTHOLOMEUS H

, CLEVERS

J G P W

, BEHZAD

. Estimating salinity stress in sugarcane fields with spaceborne hyperspectral vegetation indices
International Journal of Applied Earth Observation & Geoinformation, 2013,21:282-290.

DOI:10.1016/j.jag.2012.07.002 URL [本文引用: 1]

[24]

朱赟, 申广荣, 项巧巧, 吴裕. 基于不同光谱变换的土壤盐含量光谱特征分析
土壤通报, 2017,48(3):560-568.

[本文引用: 1]

ZHU

, SHEN G

, XIANG Q

, WU

. Spectral characteristics of soil salinity based on different pre-processing methods
Chinese Journal of Soil Science, 2017,48(3):560-568. (in Chinese)

[本文引用: 1]

[25]

姚远, 丁建丽, 张芳, 赵振亮, 江红南. 基于高光谱指数和电磁感应技术的区域土壤盐渍化监测模型
光谱学与光谱分析, 2013,33(6):1658-1664.

[本文引用: 1]

YAO

, DING J

, ZHANG

, ZHAO Z

, JIANG H

. Research on model of soil salinization monitoring based on hyperspectral index and EM38
Spectroscopy and Spectral Analysis, 2013,33(6):1658-1664. (in Chinese)

[本文引用: 1]

[26]

厉彦玲, 赵庚星, 常春艳, 王卓然, 王凌, 郑佳荣. OLI与HSI影像融合的土壤盐分反演模型
农业工程学报, 2017,33(21):173-180.

URL [本文引用: 1]

土壤盐渍化问题是黄河三角洲地区主要的土地退化问题，借助遥感技术快速、准确地掌握土壤盐渍化信息，对农业可持续发展具有重要意义。该文以黄河三角洲垦利县为研究区，利用超球体色彩空间变换算法，将环境一号卫星HSI高光谱影像与Landsat 8 OLI多光谱影像进行融合，选择土壤盐分的特征波段，结合土壤盐分的实测数据，建立统计分析模型（多元线性回归、偏最小二乘回归）和机器学习模型（BP神经网络、支持向量机和随机森林），对土壤盐分进行遥感反演。结果表明：OLI影像的统计分析模型和机器学习模型精度均较低，精度最高的随机森林模型相关系数仅为0.570；HSI影像的反演模型精度高于OLI，BP神经网络模型相关系数为0.607；融合影像反演模型精度明显高于HSI影像和OLI影像，土壤盐分含量的实测值与机器学习模型预测值具有良好的相关性，BP神经网络模型、支持向量机模型和随机森林模型的决定系数R2分别达到0.966、0.821和0.926，模型反演精度较高。研究表明，多光谱和高光谱影像融合能显著提高土壤盐分遥感反演精度，机器学习模型的反演效果明显优于统计分析模型。研究结果对黄河三角洲典型地区的土壤盐分反演具有积极的理论和实践意义。

LI Y

, ZHAO G

, CHENG C

, WANG Z

, WANG

, ZHENG J

. Soil salinity retrieval model based on OLI and HSI image fusion
Transactions of the Chinese Society of Agricultural Engineering, 2017,33(21):173-180. (in Chinese)

URL [本文引用: 1]

冯娟, 丁建丽, 杨爱霞, 蔡亮红. 干旱区土壤盐渍化信息遥感建模
干旱地区农业研究, 2018,36(1):266-273.

[本文引用: 1]

FENG

, DING J

, YANG A

, CAI L

. Remote sensing modeling of soil salinization information in arid areas
Agricultural Research in the Arid Areas, 2018,36(1):266-273. (in Chinese)

[本文引用: 1]

[28]

曹肖奕, 丁建丽, 葛翔宇, 王敬哲. 基于光谱指数与机器学习算法的土壤电导率估算研究
土壤学报, 2020,56(4):867-877.

[本文引用: 2]

CAO X

, DING J

, GE X

, WANG J

. Estimation of soil electrical conductivity based on spectral index and machine learning algorithm
Acta Pedologica Sinica, 2020,56(4):867-877. (in Chinese)

[本文引用: 2]

[29]

张智韬, 魏广飞, 姚志华, 谭丞轩, 王新涛, 韩佳. 基于无人机多光谱遥感的土壤含盐量反演模型研究
农业机械学报, 2019,50(12):151-160.

[本文引用: 2]

ZHANG Z

, WEI G

, YAO Z

, TAN C

, WANG X

, HAN

. Soil salt inversion model based on UAV multispectral remote sensing
Transactions of the Chinese Society for Agricultural Machinery, 2019,50(12):151-160. (in Chinese)

[本文引用: 2]

[30]

林芬, 赵庚星, 常春艳, 王卓然, 李慧. 基于相邻轨道图像的冬小麦面积提取及长势分析
农业资源与环境学报, 2016,33(4):384-389.

[本文引用: 1]

LIN

, ZHAO G

, CHANG C

, WANG Z

, LI

. Extraction of winter wheat area and growth analysis based on remote sensing imagery of adjacent tracks
Journal of Agricultural Resources and Environment, 2016,33(4):384-389. (in Chinese)

[本文引用: 1]

[31]

王卓然, 赵庚星, 高明秀, 常春艳, 姜曙千, 贾吉超, 李晋. 黄河三角洲垦利县夏季土壤水盐空间变异及土壤盐分微域特征
生态学报, 2016,36(4):1040-1049.

DOI:10.5846/stxb201406231296 URL [本文引用: 1]

黄河三角洲作为我国重要的后备土地资源区,土壤盐渍化问题突出,切实掌握季节性土壤水盐状况及其微域特征是该区土壤盐渍化防控和土地资源高效利用的重要基础。选择黄河三角洲垦利县,通过野外调查实测与室内化验分析获取土壤水盐含量数据,利用统计分析、GIS空间插值、实地观测与数据分析对比等方法,分析了研究区夏季土壤水盐状况及其微域变异规律。结果显示:研究区夏季土壤水盐含量总体较高,含盐量以中度盐渍化为主,随着土层深度的增加含盐量呈上升趋势,且各层土壤含盐量呈显著正相关性;含盐量较高的地区主要分布在该区东北部和中东部,含盐量较低的地区主要分布在西南部和中部;土壤含盐量从大到小的植被类型依次为光板地→碱蓬→高粱→芦苇→茅草→水稻→棉花→玉米;土壤盐分微域变化特征明显,含盐量受距路边远近、不同耕作措施、地形部位、植被群落等因素影响较大,表现出微域规律性和复杂性。该研究基本摸清了研究区夏季时相的土壤水盐状况及其微域特征,为黄河三角洲农作物栽培管理及土壤资源可持续利用提供了科学依据。

WANG Z

, ZHAO G

, GAO M

, CHANG C

, JIANG S

, JIA J

, LI

. Spatial variation of soil water and salt and microscopic variation of soil salinity in summer in typical area of the Yellow River Delta in Kenli county
Acta Ecologica Sinica, 2016,36(4):1040-1049. (in Chinese)

DOI:10.5846/stxb201406231296 URL [本文引用: 1]

方孝荣, 高俊峰, 谢传奇, 朱逢乐, 黄凌霞, 何勇. 农作物冠层光谱信息检测技术及方法综述
光谱学与光谱分析, 2015,35(7):1949-1955.

URLPMID:26717758

Compared with the traditional chemical methods and the subjective visual ways for measuring plant physiology information indicators, the assessments of crop canopy information through spectral radiometer are more simple, rapid and accurate. The applications of different types of spectral radiometer, especially for international general used Cropscan multispectral radiometer, for predicting crop canopy leaf area index under different growth stage, biomass, nitrogen, chlorophyll and yield, and monitoring plant diseases and insect pests were summarized based on crop group information acquisition methods in recent years. The varity of vegetation indices (VIs) were concluded after comparing regression coefficients of related models among different crops. In general, the correlation coefficients of mathematical models were high and it can realize the crop detection of various kinds of physiological information. Besides, the combination of multispectral radiometer and other sensors can provide useful information to evaluate the status of crops growth, which is very important in practice.

FANG X

, GAO J

, XIE C

, ZHU F

, HUANG L

, HE

. Review of crop canopy spectral information detection technology and methods
Spectroscopy and Spectral Analysis, 2015,35(7):1949-1955 . (in Chinese)

URLPMID:26717758

DOUAOUI A E

, NICOLAS

, WALTER

. Detecting salinity hazards within a semiarid context by means of combining soil and remote-sensing data
Geoderma, 2006,134(1/2):217-230.

[34]

鲍士旦. 土壤农化分析. 北京: 中国农业出版社, 2000.

[本文引用: 1]

BAO S

. Soil and Agricultural Chemistry Analysis. Beijing: China Agricultural Press, 2000. (in Chinese)

[本文引用: 1]

[35]

NGUYEN K

, LIOU Y

, TRAN

, HOANG P

, NGUYEN T

. Soil salinity assessment by using nearinfrared channel and vegetation soil salinity index derived from Landsat 8 OLI data: A case study in the Tra Vinh province, Mekong Delta, Vietnam
Progress Earth and Planetary Science, 2020,7(1).

[本文引用: 1]

[36]

ALLBED

, KUMAR

, ALDAKHEEL

Y Y

. Assessing soil salinity using soil salinity and vegetation indices derived from IKONOS high-spatial resolution imageries: Applications in a date palm dominated region
Geoderma, 2014,230/231:1-8.

DOI:10.1016/j.geoderma.2014.03.025 URL [本文引用: 1]

[37]

葛哲学, 孙志强. 神经网络理论与MATLAB R2007实现. 北京: 电子工业出版社, 2005.

[本文引用: 1]

GE Z

, SUN Z

. Neural Network Theory and MATLAB R2007 Realization. Beijing: Publishing House of Electronics Industry, 2005. (in Chinese)

[本文引用: 1]

[38]

丁世飞, 齐丙娟, 谭红艳. 支持向量机理论与算法研究综述
电子科技大学学报, 2011,40(1):2-10.

[本文引用: 1]

DING S

, QI B

, TAN H

. An overview on theory and algorithm of support vector machines
Journal of University of Electronic Science and Technology of China, 2011,40(1):2-10. (in Chinese)

[本文引用: 1]