基于Worldview-3与Sentinel-1 SAR数据的草原矿区复垦植被生物量反演方法研究

doi:10.13745/j.esf.sf.2020.10.10

地学前缘 ›› 2021, Vol. 28 ›› Issue (4): 219-228.DOI: 10.13745/j.esf.sf.2020.10.10

• 水生态、草地生态及污染土壤修复 • 上一篇下一篇

基于Worldview-3与Sentinel-1 SAR数据的草原矿区复垦植被生物量反演方法研究

刘艳慧¹^,²(), 杨晓宇¹^,³, 包妮沙¹^,³^,^*(), 顾晓薇¹^,³

1.东北大学资源与土木工程学院, 辽宁沈阳 110819
2.中国地震局第二监测中心, 陕西西安 710000
3.东北大学智慧水利与资源环境科技创新中心, 辽宁沈阳 110819

收稿日期:2020-09-29 修回日期:2020-11-22 出版日期:2021-07-25 发布日期:2021-07-25
通讯作者: 包妮沙
作者简介:刘艳慧(1994—),女,硕士研究生,主要从事植被遥感方面的研究。E-mail: 2513178156@qq.com
基金资助:
国家自然科学基金联合基金项目(U1903216);辽宁省重点研发计划项目“工业矿区生态修复及资源综合利用研究”(2019JH2/10300051)

Estimating biomass of reclaimed vegetation in prairie mining area: Inversion method based on Worldview-3 and Sentinel-1 SAR data

LIU Yanhui¹^,²(), YANG Xiaoyu¹^,³, BAO Nisha¹^,³^,^*(), GU Xiaowei¹^,³

1. College of Resources and Civil Engineering, Northeastern University, Shenyang 110819, China
2. Second Monitoring Center, China Earthquake Administration, Xi’an 710000, China
3. Science and Technology Innovation Center of Smart Water and Resource Environment, Northeastern University, Shenyang 110819, China

Received:2020-09-29 Revised:2020-11-22 Online:2021-07-25 Published:2021-07-25
Contact: BAO Nisha

摘要/Abstract

摘要：

利用多源遥感数据定量反演矿区复垦植被生物量是高效、动态、大面积监测土地复垦和生态恢复效果的必要手段之一。本文以内蒙古草原露天煤矿为研究区,联合遥感光学与雷达数据各自的优势,探索基于Worldview-3(WV-3)与Sentinel-1 SAR数据的矿区复垦植被生物量反演方法,选择主成分-小波变换(W-PCA)算法对WV-3与Sentinel-1 SAR数据进行融合,揭示波段反射率、植被指数、后向散射系数及纹理特征等参数与生物量之间的相关关系,建立多变量的生物量反演模型,并分析不同生物量模型的空间不确定性。结果表明:(1)通过W-PCA算法得到融合后的图像,信息熵的提高反映了融合图像与光学WV-3图像相比具有更多的细节信息,平均梯度的提高反映了融合图像与Sentinel-1 SAR图像相比具有更高的清晰度和更丰富的纹理信息。融合后的第8波段相关系数最高、光谱扭曲度最低、光谱保真度最高。(2)通过相关性分析,发现增强型植被指数(EVI)、归一化植被指数(NDVI)、VH极化、VH均值纹理以及融合后第8波段与生物量显著正相关。WV-3的NDVI与Sentinel-1的VH_ME建模精度(R²=0.834 0,RMSE=16.464 6 g/m²,Ac=81.52%)最高,融合后的第8波段验证精度(R²=0.798 3,RMSE=22.828 3 g/m²,Ac=74.64%)最高。(3)基于不同模型的残差不确定性分析,Sentinel-1 SAR数据变量建立的模型估测结果更容易出现高估及饱和现象,两者联合变量建立的模型可实现优势互补,利用融合数据建立的模型明显改善生物量小于40 g/m²时的高估计现象以及生物量大于100 g/m²时的两者饱和现象,其不确定性降低2.42~9.68 g/m²。因此,利用光学和雷达遥感融合能够有效提高复垦植被生物量的估算精度,为草原矿区复垦植被精细监测提供有效的数据支持。

关键词: 草原矿区, 复垦植被, Worldview-3, Sentinel-1 SAR, 数据融合, 生物量反演

Abstract:

Quantitative inversion of reclaimed vegetation biomass in prairie mining area based on the remote sensing technology is the basis of dynamic monitoring and evaluation of mining ecological environment. In this research, focussing on the reclaimed vegetation in the grassland open-pit coal mine in Inner Mongolia, we combine the advantages of optical and radar remote sensing to explore the inversion method for biomass estimation based on Worldview-3 and Sentinel-1 SAR data. The principal component-wavelet transform algorithm was selected for data fusion. We revealed the correlation between parameters such as band reflectivity, vegetation index, backscatter coefficient or texture feature and biomass, established multivariate biomass inversion models, and analyzed the spatial uncertainty of different biomass models. The results are as follows: (1) After image fusion using W-PCA method, both data entropy and average gradient of the fusion data were significantly improved, and the fused 8th band (NIR2) had the highest correlation coefficient, the lowest spectral distortion, and the highest spectral fidelity. (2) Correlation analysis revealed a significant positive correlation between biomass and EVI, NDVI, VH polarization scattering coefficient, VH mean texture or the 8th band after fusion. Compared with a single variable, using the joint variables, NDVI of WV-3 and VH mean texture of Sentinel-1, it achieved the highest model accuracy (R²=0.8340, RMSE=16.4646 g/m², Ac=81.52%), while the 8th band after fusion gave the highest verification accuracy (R²=0.7983, RMSE=22.8283 g/m², Ac=74.64%). (3) According to the residual uncertainty analysis of different models, the Sentinel-1 variables are more prone to overestimation and saturation, whereas the joint variables can achieve complementary advantages. Using fusion data significantly improved the biomass overestimation for biomass below 40 g/m² and saturation for biomass greater than 100 g/m², reducing the model uncertainty by 2.42-9.68 g/m² on average. It can be seen that the combination of optical and microwave cooperative remote sensing can effectively improve the estimation accuracy of vegetation biomass, thereby providing effective data support for fine monitoring of reclaimed vegetation in mining areas.

Key words: prairie mining area, reclaimed vegetation, Worldview-3, Sentinel-1 SAR, data fusion, biomass inversion

中图分类号:

X171.4
X87

刘艳慧, 杨晓宇, 包妮沙, 顾晓薇. 基于Worldview-3与Sentinel-1 SAR数据的草原矿区复垦植被生物量反演方法研究[J]. 地学前缘, 2021, 28(4): 219-228.

LIU Yanhui, YANG Xiaoyu, BAO Nisha, GU Xiaowei. Estimating biomass of reclaimed vegetation in prairie mining area: Inversion method based on Worldview-3 and Sentinel-1 SAR data[J]. Earth Science Frontiers, 2021, 28(4): 219-228.

图/表 13

图1 研究区及采样点位置

Fig.1 Locations of the research area and sampling sites

图2 复垦植被种类

Fig.2 Types of reclaimed vegetation

表1 遥感数据参数

Table 1 Remote sensing data parameters

传感器	分辨率	波段名称/极化方式	波长
		海岸波段	400~450 nm
		蓝波段	450~510 nm
		绿波段	510~580 nm
Worldview-3	1.24 m	黄波段	585~625 nm
		红波段	630~690 nm
		红边波段	705~745 nm
		近红外1	770~895 nm
		近红外2	860~1 040 nm
Sentinel-1	5 m×20 m	VH和VV极化	7.5~3.75 cm

图3 图像融合流程图

Fig.3 Image fusion flow chart

表2 7种植被指数计算公式及其优点

Table 2 Calculation formulas and their advantages for 7 vegetable indexes

植被指数	计算公式	优点
NDVI	$NIR 1 - R NIR 1 + R$	NDVI指数是目前遥感影像植被分类研究应用最广泛的植被指数,是植被长势及覆盖度的最佳指标
DVI	NIR1-R	DVI指数对土壤背景变化敏感,适用植被发育早中期或植被覆盖度较低的区域
RVI	$NIR 1 R$	RVI指数对茂盛、覆盖度较高的植被敏感,绿色植被RVI值较高,非植被RVI值较低
NDGI	$NIR 1 - G NIR 1 + G$	NDGI指数在NDVI指数的基础上将红波段换成了绿波段,可用来对不同活力植被形式进行检验
ARVI	$NIR 1 - 2 R + B NIR 1 + 2 R - B$	ARVI指数根据蓝波段与红波段对大气响应的差异,用红蓝波段组合替代NDVI指数中的红光波段,以减少大气对植被指数的影响
EVI	$NIR 1 - R NIR 1 + R + B$	EVI指数不仅能克服大气对光谱反射的衰减,还能削弱土壤背景的影响
$NDV I 705$	$RE - R RE + R$	$NDV I 705$ 指数是NDVI指数的改进型,它对植被叶冠层的微小变化、林窗片段和衰老变化非常灵敏

表2 7种植被指数计算公式及其优点

Table 2 Calculation formulas and their advantages for 7 vegetable indexes

植被指数	计算公式	优点
NDVI	$NIR 1 - R NIR 1 + R$	NDVI指数是目前遥感影像植被分类研究应用最广泛的植被指数,是植被长势及覆盖度的最佳指标
DVI	NIR1-R	DVI指数对土壤背景变化敏感,适用植被发育早中期或植被覆盖度较低的区域
RVI	$NIR 1 R$	RVI指数对茂盛、覆盖度较高的植被敏感,绿色植被RVI值较高,非植被RVI值较低
NDGI	$NIR 1 - G NIR 1 + G$	NDGI指数在NDVI指数的基础上将红波段换成了绿波段,可用来对不同活力植被形式进行检验
ARVI	$NIR 1 - 2 R + B NIR 1 + 2 R - B$	ARVI指数根据蓝波段与红波段对大气响应的差异,用红蓝波段组合替代NDVI指数中的红光波段,以减少大气对植被指数的影响
EVI	$NIR 1 - R NIR 1 + R + B$	EVI指数不仅能克服大气对光谱反射的衰减,还能削弱土壤背景的影响
$NDV I 705$	$RE - R RE + R$	$NDV I 705$ 指数是NDVI指数的改进型,它对植被叶冠层的微小变化、林窗片段和衰老变化非常灵敏

图4 融合结果对比图

Fig.4 Comparison of fusion results

图5 融合前后WV-3各波段以及Sentinel-1SAR的信息熵

Fig.5 Information entropy values for all available bands of WV-3 before and after image fusion and for Sentinel-1 VV-polarized SAR image

图6 融合前后WV-3各波段以及Sentinel-1SAR的平均梯度

Fig.6 Average gradient values for all available bands of WV-3 before and after image fusion and for sentinel-1 VV-polarized SAR image

图7 WV-3与融合图像各波段的相似系数和光谱扭曲度

Fig.7 Similarity coefficient and spectral distortion between WV-3 image and fused image for all spectral bands

表3 生物量与各变量相关性分析

Table 3 Correlation analysis between biomass and different variables

Worldview-3
波段反射率		纹理特征					极化/纹理特征				融合波段
变量	相关性		变量		相关性		变量		相关性		变量		相关性
海岸波段	-0.323	波段4均值		-0.741^**		VH		0.504^**		波段1 (RHB1)		-0.759^**
蓝波段	-0.474^*	波段5均值		-0.792^**		VV		0.410^*		波段2 (RHB2)		-0.806^**
绿波段	-0.372	波段5相关性		-0.358^*		VV均值		0.412^**		波段3(RHB3)		-0.837^**
黄波段	-0.620^**	波段6均值		0.511^**		VV方差		0.174		波段4 (RHB4)		-0.834^**
红波段	-0.657^**	波段6相关性		-0.412^*		VV均一性		-0.238		波段5 (RHB5)		-0.800^**
红边波段	0.215	波段7均值		0.760^**		VV对比度		0.165		波段6 (RHB6)		0.564^**
近红外1(NIR1)	0.694^**	波段7方差		0.700^**		VV相异性		0.201		波段7 (RHB7)		0.842^**
近红外2(NIR2)	0.718^**	波段7均一性		-0.593^**		VV信息熵		0.257		波段8 (RHB8)		0.874^**
RVI	0.790^**	波段7相异性		0.535^*		VV二阶矩		-0.219
NGVI	0.871^**	波段7信息熵		0.510^**		VV相关性		0.169
NDVI₇₀₅	0.835^**	波段7二阶矩		-0.366^*		VH均值		0.505^**
NDVI	0.874^**	波段8均值		0.748^**		VH方差		-0.061
EVI	0.876^**	波段8方差		0.669^**		VH均一性		0.069
DVI	0.833^**	波段8均一性		-0.600^**		VH对比度		0.064
ARVI	0.845^**	波段8相异性		0.558^**		VH相异性		-0.016
		波段8信息熵		0.612^**		VH信息熵		-0.133
		波段8二阶矩		0.520^**		VH二阶矩		0.298
						VH相关性		-0.195

表4 各数据集最优模型及精度

Table 4 Optimal model and its accuracy for each data set

变量	模型	建模精度 (n=21)			验证精度 (n=11)
变量	模型	R²	RMSE/ (g·m^-2)	Ac/%		R²		RMSE/ (g·m^-2)		Ac/%
EVI	$y = 2.840 8 × EVI + 16.829$	0.816 3	17.318 7	80.56	0.709 8		24.201 8		73.12
VH	$y = 2.272 6 × V H 2 + 96.275 × VH + 1 094.4$	0.430 6	30.488 8	65.78	0.324		42.110 4		53.22
NDVI,VH_ME	$y = 3.256 2 × NDVI - 0.622 4 × V H ME + 8.593$	0.834 0	16.464 6	81.52	0.696 3		23.680 1		73.69
RHB8	$y = 1.055 7 × RHB 8 - 26.676$	0.784 2	18.767 9	78.94	0.798 3		22.828 3		74.64

表4 各数据集最优模型及精度

Table 4 Optimal model and its accuracy for each data set

变量	模型	建模精度 (n=21)			验证精度 (n=11)
变量	模型	R²	RMSE/ (g·m^-2)	Ac/%		R²		RMSE/ (g·m^-2)		Ac/%
EVI	$y = 2.840 8 × EVI + 16.829$	0.816 3	17.318 7	80.56	0.709 8		24.201 8		73.12
VH	$y = 2.272 6 × V H 2 + 96.275 × VH + 1 094.4$	0.430 6	30.488 8	65.78	0.324		42.110 4		53.22
NDVI,VH_ME	$y = 3.256 2 × NDVI - 0.622 4 × V H ME + 8.593$	0.834 0	16.464 6	81.52	0.696 3		23.680 1		73.69
RHB8	$y = 1.055 7 × RHB 8 - 26.676$	0.784 2	18.767 9	78.94	0.798 3		22.828 3		74.64

图8 模型残差图

Fig.8 Residual plots for models

图9 生物量不确定性分布图 a—EVI;b—Sentinel-1 VH极化;c—NDVI与VH纹理联合;d—融合数据第8波段。

Fig.9 Distribution of estimation uncertainties for different biomass surface concentrations. (a) 60-100 g/m2; (b) 40-60 g/m2; (c) <40 g/m2; (d) >100 g/m2.

参考文献 26

[1]	杨勤学, 赵冰清, 郭东罡. 中国北方露天煤矿区植被恢复研究进展[J]. 生态学杂志, 2015, 34(4):1152-1157.
[2]	张泽民, 吕昌河, 谢苗苗, 等. 基于WorldView 2影像的矿区植被重建效果评估[J]. 生态学报, 2018, 38(4):1301-1310.
[3]	张兆彤, 王金满, 张佳瑞. 矿区复垦土壤与植被交互影响的研究进展[J]. 土壤, 2018, 50(2):239-247.
[4]	PETROPOULOS G P, PARTSINEVELOS P, MITRAKA Z. Change detection of surface mining activity and reclamation based on a machine learning approach of multi-temporal Landsat TM imagery[J]. Geocarto International, 2013, 28(4):323-342.
[5]	张耀, 周伟. 安太堡露天矿区复垦地植被覆盖度反演估算研究[J]. 中南林业科技大学学报, 2016, 36(11):113-119.
[6]	曾纳, 任小丽, 何洪林, 等. 基于神经网络的三江源区草地地上生物量估算[J]. 环境科学研究, 2017, 30(1):59-66.
[7]	安海波, 李斐, 赵萌莉, 等. 基于优化光谱指数的牧草生物量估算[J]. 光谱学与光谱分析, 2015, 35(11):3155.
[8]	赖炽敏, 赖日文, 薛娴, 等. 基于植被盖度和高度的不同退化程度高寒草地地上生物量估算[J]. 中国沙漠, 2019, 39(5):127-134.
[9]	阮兰君, 杨燕琼. 东莞市针叶类森林生物量遥感模型研究[J]. 林业与环境科学, 2018, 34(1):32-36.
[10]	WANG G Q, LIU S M, LIU T X, et al. Modelling above-ground biomass based on vegetation indexes: a modified approach for biomass estimation in semi-arid grasslands[J]. International Journal of Remote Sensing, 2019, 40(10):3835-3854.
[11]	乔正年, 耿庆宏, 徐雁南. 运用卫星遥感数据对杨树人工林生物量的估算[J]. 东北林业大学学报, 2019, 47(5):66-71.
[12]	CHANG J G, SHOSHANY M, OH Y. Polarimetric radar vegetation index for biomass estimation in desert fringe ecosystems[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(12):7102-7108. DOI URL
[13]	潘磊, 孙玉军. 应用Sentinel-1影像纹理信息模型估测杉木林生物量[J]. 东北林业大学学报, 2018, 46(1):58-62.
[14]	HUANG C D, YE X Y, DENG C B, et al. Mapping above-ground biomass by integrating optical and SAR imagery: a case study of Xixi national wetland park, China[J]. Remote Sensing, 2016, 8(8):647. DOI URL
[15]	SHAO Z F, ZHANG L J, WANG L. Stacked sparse autoencoder modeling using the synergy of airborne LiDAR and satellite optical and SAR data to map forest above-ground biomass[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2017, 10(12):5569-5582. DOI URL
[16]	行敏锋, 何彬彬. 主被动遥感数据协同估算干旱区草原植被生物量[J]. 遥感技术与应用, 2015, 30(6):1122-1128.
[17]	秦立厚, 张茂震, 钟世红, 等. 森林生物量估算中模型不确定性分析[J]. 生态学报, 2017, 37(23):7912-7919.
[18]	郭山川, 汤傲, 李效顺, 等. 融合主被动遥感的乌海矿区土地损伤测度[J]. 生态与农村环境学报, 2018, 34(8):678-685.
[19]	申艳琴. 半干旱区煤炭开采对植被扰动规律的研究: 以神东矿区为例[D]. 徐州: 中国矿业大学, 2014.
[20]	李小静. 彬长煤矿地表沉陷区植被变化遥感监测研究[D]. 西安: 西安科技大学, 2013.
[21]	LIU L Y, WANG J H, HUANG W J, et al. Estimating winter wheat plant water content using red edge parameters[J]. International Journal of Remote Sensing, 2004, 25(17):3331-3342. DOI URL
[22]	DONG T F, LIU J G, SHANG J L, et al. Assessment of red-edge vegetation indices for crop leaf area index estimation[J]. Remote Sensing of Environment, 2019, 222:133-143. DOI URL
[23]	MAIMAITIJIANG M, GHULAM A, SIDIKE P, et al. Unmanned Aerial System (UAS): based phenotyping of soybean using multi-sensor data fusion and extreme learning machine[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2017, 134:43-58.
[24]	BARATI S, RAYEGANI B, SAATI M, et al. Comparison the accuracies of different spectral indices for estimation of vegetation cover fraction in sparse vegetated areas[J]. The Egyptian Journal of Remote Sensing and Space Science, 2011, 14(1):49-56. DOI URL
[25]	ELMORE A J, MUSTARD J F, MANNING S J, et al. Quantifying vegetation change in semiarid environments: precision and accuracy of spectral mixture analysis and the normalized difference vegetation index[J]. Remote Sensing of Environment, 2000, 73(1):87-102. DOI URL
[26]	LIU Y N, GONG W S, XING Y Q, et al. Estimation of the forest stand mean height and aboveground biomass in Northeast China using SAR Sentinel-1B, multispectral Sentinel-2A, and DEM imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 151:277-289.

基于Worldview-3与Sentinel-1 SAR数据的草原矿区复垦植被生物量反演方法研究

Estimating biomass of reclaimed vegetation in prairie mining area: Inversion method based on Worldview-3 and Sentinel-1 SAR data

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 26

相关文章 1

编辑推荐

Metrics

本文评价