Quantitative Estimation Soil Salinization in Arid Areas Based on Machine Learning
-
摘要:
目的 探讨对土壤盐渍化进行快速、准确监测技术与方法。 方法 利用353个地面表观电导率数据,以及从Worldview-2影像获取对应采样点的波段反射率值,结合两波段组合植被指数和三波段组合植被指数,筛选最佳二维、三维波段组合方式,引入人工神经网络、K近邻和支持向量回归来构建区域土壤盐渍化定量反演模型。 结果 ① WV-2影像的红边和近红外波段与ECa呈现显著相关(P < 0.01)。② 二维植被指数(RVI(B5-B2)、NDVI(B6-B2)、DVI(B2-B6))和三维植被指数(3DVI(B2-B6-B6)、3DVI(B3-B5-B6)、3DVI(B5-B2-B1)、3DVI(B2-B1-B6)、3DVI(B2-B1-B6)、3DVI(B6-B1-B2)、3DVI(B5-B3-B7))的波段组合计算提高了其对土壤盐渍化的敏感性。③ 基于不同维度数据的机器学习估算模型中,3DVI和KNN算法结合对土壤盐渍化估算效果最为突出,且模型精度为R2 = 0.773,RMSE = 1.659 dS m−1,RPD = 2.216。 结论 所构建的多维植被指数可应用于类似环境条件下盐渍土地监测和评价研究。 Abstract:Objective Soil salinization is one of the main factors causing land degradation and desertification, especially affecting the agricultural activities and land use management in arid and semi-arid regions. There was an urgent need for rapid, accurate and economical monitoring techniquces for soil salt. Method The 353 data of surface apparent conductivity (Electronic Conductivity, ECa) were used, as well as the band reflectivity values of corresponding sampling points obtained from Worldview-2 (WV-2) images, combined with the combination of two-band combined vegetation index (Two- Dimensional vegetation index, 2DVI) and Three-Dimensional vegetation index (Three-Dimensional vegetation index, 3DVI). The best combination of two-dimensional and three-dimensional bands, artificial neural networks (Artificial Neural Network, ANN), K-nearest neighbors (K- Nearest Neighbors (KNN) and Support Vector Regression (SVR) are used to construct a quantitative inversion models of regional soil salinization. Result ① The red edge and near-infrared band of WV-2 images were significantly correlated with ECa (ρ < 0.01). ② Two-dimensional vegetation index (RVI(B5-B2), NDVI(B6-B2), DVI (B2-B6)) and three-dimensional vegetation index (3DVI(B2-B6-B6), 3DVI(B3-B5-B6), 3DVI(B5-B2-B1), 3DVI(B2-B1-B6), 3DVI(B2 -B1-B6), 3DVI(B6-B1-B2), 3DVI(B5-B3-B7)) combined calculation of the bands to improve its sensitivity to soil salinization. ③ Machine learning estimation based on data of different dimensions in the model, the combination of 3DVI and KNN algorithm had the most prominent effect on soil salinization estimation, and the model accuracy was R2 = 0.773, RMSE = 1.659 ds m−1, RPD = 2.216. Conclusion Vegetation index stablished in this study would be helpful to monitor and evaluate the saline land under similar environmental conditions. -
Key words:
- Soil salinity /
- Optimum hyperspectral mathod /
- Machine learning /
- The Kerya river
-
表 1 高分辨率WV-2影像的波段光谱信息
Table 1. Spectral band details of high-resolution WV-2 image
波段
Band波长 (nm)
Wavelength传感器分率
Resolution海岸波段 400 ~ 450 多光谱: 1.85 m GSD (星下点), 2.07 m GSD (偏离星下点20°) 蓝色波段 450 ~ 510 绿色波段 510 ~ 580 黄色波段 585 ~ 625 红色波段 630 ~ 690 红色边缘波段 705 ~ 745 全色: 0.46 m GSD (星下点), 0.52 m GSD (偏离星下点20°) 近红外-1波段 770 ~ 895 近红外-2波段 860 ~ 1040 表 2 波段组合植被指数
Table 2. The band combination vegetation index
波段组合类型
Band combination植被指数
Vegetation index缩写
Abbreviation公式
Equation文献
Reference二维植被指数 (2DVI) 比值植被指数 RVI Rλ1/Rλ2 [20] 归一化植被指数 NDVI (Rλ1 − Rλ2)/(Rλ1 + Rλ2) 差值植被指数 DVI Rλ1 − Rλ2 三维植被指数 (3DVI) 三波段植被指数 3DVI-1 Rλ1/(Rλ2 × Rλ3) [21-22] 3DVI-2 Rλ1/(Rλ2 + Rλ3) 3DVI-3 (Rλ1 − Rλ2)/(Rλ2 + Rλ3) 3DVI-4 (Rλ1 − Rλ2)/(Rλ2 − Rλ3) 3DVI-5 (Rλ2 + Rλ3)/Rλ1 3DVI-6 (Rλ1 − Rλ2)/[(Rλ1 − Rλ2) − (Rλ2 − Rλ3)] 3DVI-7 (Rλ1 − Rλ2) − (Rλ2 − Rλ3) 表 3 模型验证指标
Table 3. Model evaluation index
验证指标
Evaluation index表达式
Equation决定系数 $ R^2 = {\left[\dfrac{\displaystyle \sum\nolimits_{i=1}^{N}({x}_{i}-\bar{x})({y}_{i}-\bar{y})}{\sqrt{\displaystyle \sum\nolimits_{i=1}^{N}{({x}_{i}-\bar{x})}^{2} + \displaystyle \sum\nolimits_{i=1}^{N}{({y}_{i}-\bar{y})}^{2}}}\right]}^{2} $ 均方根误差 $ {\rm{RMSE}} = \sqrt{\dfrac{\displaystyle \sum\nolimits_{\mathrm{i}=1}^{\mathrm{N}}{({\mathrm{\gamma }}_{\mathrm{i}}-{\mathrm{\beta }}_{\mathrm{i}})}^{2}}{\mathrm{n}}} $ 相对分析误差 RPD = SD/SEP 表 4 土壤盐渍化遥感建模与验证
Table 4. Remote sensing modeling and verification of soil salinization
数据类型
Type of data模型
Model变量
Parameter建模数据集
Calibration set验证数据集
Validation setRPD R2sim RMSEsim R2pre RMSEpre 单波段反射率 KNN RB1、RB6、RB7、RB8 0.438 2.662 0.374 2.910 1.315 SVR 0.220 2.862 0.213 3.522 1.207 ANN 0.235 2.934 0.225 3.716 1.216 二维植被指数 KNN RVI(B5-B2) 0.662 1.832 0.537 1.943 1.755 SVR NDVI(B6-B2) 0.516 1.929 0.414 2.171 1.694 ANN DVI(B2-B6) 0.426 1.918 0.412 2.326 1.552 三维植被指数 KNN 3DVI(B2-B6-B6) 0.793 1.562 0.773 1.659 2.216 SVR 3DVI(B3-B5-B6)
3DVI(B5-B2-B1)
3DVI(B2-B1-B6)
3DVI(B2-B1-B6)
3DVI(B6-B1-B2)0.758 1.547 0.715 1.523 2.015 ANN 3DVI(B5-B3-B7) 0.732 1.521 0.698 1.586 1.968 -
[1] 阿斯古丽·木萨, 阿不都拉·阿不力孜, 瓦哈甫·哈力克, 等. 新疆克里雅绿洲土壤盐分、pH和盐基离子空间异质性分析[J]. 土壤, 2017, 49(5): 152 − 159. [2] Sidike A, Zhao S, Wen Y. Estimating soil salinity in Pingluo County of China using QuickBird data and soil reflectance spectra[J]. International Journal of Applied Earth Observation & Geoinformation, 2014, 26(2): 156 − 175. [3] 翁永玲, 宫 鹏. 土壤盐渍化遥感应用研究进展[J]. 地理科学, 2006, 26(3): 369 − 375. doi: 10.3969/j.issn.1000-0690.2006.03.019 [4] 关元秀, 刘高焕, 刘庆生, 等. 黄河三角洲盐碱地遥感调查研究[J]. 遥感学报, 2001, 5(1): 46 − 46. doi: 10.11834/jrs.20010109 [5] 王 爽, 丁建丽, 王 璐, 等. 基于地表光谱建模的区域土壤盐渍化遥感监测研究[J]. 干旱区地理, 2016, 39(1): 190 − 198. [6] 马 驰. 基于GF-1遥感图像土壤含盐量反演研究[J]. 中国农村水利水电, 2017, (5): 5 − 8 + 13. doi: 10.3969/j.issn.1007-2284.2017.05.002 [7] Clark R N, Roush T L. Reflectance spectroscopy’ quantitative analysis techniques for remote sensing applications[J]. Journal of Geophysical Research Atmospheres, 1984, 89(NB7): 6329 − 6340. doi: 10.1029/JB089iB07p06329 [8] Lhissoui R, Harti A E, Chokmani K. Mapping soil salinity in irrigated land using optical remote sensing data[J]. Eurasian Journal of Soil Science, 2014, 3(2): 82 − 88. [9] Asfaw E, Suryabhagavan K V, Argaw M. Soil salinity modeling and mapping using remote sensing and GIS: The case of Wonji sugar cane irrigation farm, Ethiopia[J]. Journal of the Saudi Society of Agricultural Sciences, 2016, (9): 250 − 258. [10] Gorji T, Tanik A, Sertel E. Monitoring soil salinity via remote sensing technology under data scarce conditions: A case study from Turkey[J]. Ecological Indicators, 2017, 74: 384 − 391. doi: 10.1016/j.ecolind.2016.11.043 [11] 梁 静, 丁建丽, 王敬哲, 等. 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[J]. 土壤学报, 2019, 56(2): 320 − 330. doi: 10.11766/trxb201805070182 [12] 尹小君, 宁 川, 韩 峰, 等. 天山北坡土壤盐渍化光谱响应特征与动态监测的研究[J]. 江苏农业科学, 2019, 47(16): 277 − 281. [13] 马国林、丁建丽、韩礼敬, 等. 基于变量优选与机器学习的干旱区湿地土壤盐渍化数字制图[J]. 农业工程学报, 2020, 36(19): 124 − 131. doi: 10.11975/j.issn.1002-6819.2020.19.014 [14] 王丹阳, 陈红艳, 王桂峰, 等. 无人机多光谱反演黄河口重度盐渍土盐分的研究[J]. 中国农业科学, 2019, 52(10): 37 − 48. [15] 尼加提·卡斯木, 师庆东, 郭玉川, 等. 基于EM38和WorldView-2影像的土壤盐渍化建模研究[J]. 土壤, 2019, 51(3): 594 − 601. [16] 谭灵芝, 刘 玲, 张玉进. 新疆于田绿洲荒漠交错带土地利用及覆盖遥感研究[J]. 国土资源科技管理, 2004, 21(1): 17 − 21. doi: 10.3969/j.issn.1009-4210.2004.01.005 [17] 姚荣江, 杨劲松, 邹 平, 等. 基于电磁感应仪的田间土壤盐渍度及其空间分布定量评估[J]. 中国农业科学, 2008, 41(2): 460 − 469. doi: 10.3864/j.issn.0578-1752.2008.02.020 [18] 姚荣江, 杨劲松, 姜 龙. 电磁感应仪用于土壤盐分空间变异及其剖面分布特征研究[J]. 浙江大学学报(农业与生命科学版), 2007, 33(2): 207 − 216. [19] 刘广明, 杨劲松, 鞠茂森, 等. 电磁感应土地测量技术及其在农业领域的应用[J]. 土壤, 2003, 35(1): 27 − 29. doi: 10.3321/j.issn:0253-9829.2003.01.007 [20] Inoue, Sakaiya, Zhu, et al. Diagnostic mapping of canopy nitrogen content in rice based on hyperspectral measurements[J]. REMOTE SENS ENVIRON, 2012, 126: 210 − 221. doi: 10.1016/j.rse.2012.08.026 [21] Jackson R D. Spectral indices in N-Space[J]. Remote Sensing of Environment, 1983, 13(5): 409 − 421. doi: 10.1016/0034-4257(83)90010-X [22] Tian Y C, Yao X, Yang J, et al. Assessing newly developed and published vegetation indices for estimating rice leaf nitrogen concentration with ground- and space-based hyperspectral reflectance[J]. Field Crops Research, 2011, 120(2): 299 − 310. doi: 10.1016/j.fcr.2010.11.002 [23] 姚付启, 张振华, 杨润亚, 等. 基于红边参数的植被叶绿素含量高光谱估算模型[J]. 农业工程学报, 2009, 25(13): 123 − 129. [24] 王新颖, 隽志才, 吴庆妍, 等. KNN算法的数据优化策略[J]. 吉林大学学报(信息科学版), 2010, 28(3): 94 − 98. [25] 郑立华, 李民赞, 安晓飞, 等. 基于近红外光谱和支持向量机的土壤参数预测[J]. 农业工程学报, 2010, 26(14): 81 − 87. [26] Amal Allbed, Lalit Kumar, Yousef Y. Assessing soil salinity using soil salinity and vegetation indices derived from IKONOS high-spatial resolution imageries: Applications in a date palm dominated region[J]. Geoderma, 2014, 230-231(7): 1 − 8. [27] Peng J, Biswas A, Jiang Q, et al. Estimating soil salinity from remote sensing and terrain data in southern Xinjiang Province, China[J]. Geoderma:An International Journal of Soil Science, 2019, 337: 1309 − 1319. [28] Bannari Abderrazak, El-Battay Ali, Bannari Rachid, et al. Sentinel-MSI VNIR and SWIR bands sensitivity analysis for soil salinity discrimination in an arid landscape[J]. Remote Sensing, 2018, 10(6): 1 − 20. [29] Hu J, Peng J, Zhou Y, et al. Quantitative estimation of soil salinity using UAV-Borne hyperspectral and satellite multispectral images[J]. Remote Sensing, 2019, 11(7): 736. doi: 10.3390/rs11070736