基于CARS-BPNN的江西省土壤有机碳含量高光谱预测

doi:10.3864/j.issn.0578-1752.2022.19.005

摘要/Abstract

摘要：

【目的】探讨光谱变量选择及依据土壤类型进行分层校准两种方法对高光谱预测土壤有机碳（SOC）精度的影响。【方法】以江西省为研究区,490个土壤样本为研究对象,对研究区内的所有样本以及不同土壤类型样本分别通过竞争性自适应重加权采样（CARS）算法筛选特征波段,并采用偏最小二乘回归（PLSR）、支持向量机（SVM）、随机森林（RF）、反向传播神经网络（BPNN）4种模型,对比不同土壤类型下SOC在全波段以及CARS算法筛选后特征波段的预测精度。进而,还对比了全局校准和分层校准下SOC在全波段以及CARS算法筛选后特征波段的预测精度。【结果】（1）红壤筛选的特征波段为484、683—714和2 219—2 227 nm,水稻土筛选的特征波段为484、689—702和2 146—2 156 nm。红壤采用CARS-BPNN模型预测效果最佳（R²=0.82）,较全波段建模验证集R²提升0.07。水稻土采用CARS-RF模型预测效果最佳（R²=0.83）,较全波段建模验证集R²提升0.13。（2）在总体样本上,分层校准相比全局校准精度有所提升。采用CARS-BPNN进行分层校准预测效果最佳（R²=0.82）,较全局校准验证集R²提升0.06。【结论】采用CARS-BPNN进行分层校准能够较好地预测江西省土壤有机碳含量,本研究可为其他类似地区预测土壤属性提供科学依据。

关键词: 土壤有机碳, 竞争适应重加权采样, 分层校准, 随机森林, 反向传播神经网络

Abstract:

【Objective】 This study explored the roles of spectral variable selection and stratified calibration based on soil type in visible-near-infrared (Vis-NIR) spectroscopy for predicting soil organic carbon (SOC) content on a large spatial scale. 【Method】 A total of 490 samples were collected in Jiangxi province (Southeast China) and used for modeling with partial least squares regression (PLSR), support vector machine (SVM), random forests (RF), and back-propagation neural network (BPNN). The competitive adaptive reweighted sampling (CARS) procedure was used to select the feature bands of different soil types and total samples (i.e., sum of red soils and paddy soils). The prediction accuracy of models incorporating full bands or feature bands was evaluated for the different soil types. Further, the prediction accuracy of these models based on their global and stratification calibration was compared for the total samples. 【Result】 (1) The feature bands of red soils were 484, 683-714, and 2 219-2 227 nm, while those of paddy soils were 484, 689-702, and 2 146-2 156 nm. The CARS-BPNN model showed the best prediction performance for red soils (validation set R² = 0.82), being 0.07 higher than that of BPNN with full bands. The CARS-RF model also had the best prediction performance for paddy soils (validation set R² = 0.83), being 0.13 higher than that of RF with full bands. (2) Based on the stratified calibration, the best prediction performance was obtained using the CARS-BPNN model (validation set R² = 0.82), which was 0.06 higher than that of the model based on global calibration. 【Conclusion】 The CARS-BPNN model combined with stratified calibration based on soil type could accurately predict SOC content in the study area.

Key words: soil organic carbon, competitive adaptive reweighted sampling, stratified calibration, random forest, back propagation neural network

吴俊,郭大千,李果,郭熙,钟亮,朱青,国佳欣,叶英聪. 基于CARS-BPNN的江西省土壤有机碳含量高光谱预测[J]. 中国农业科学, 2022, 55(19): 3738-3750.

WU Jun,GUO DaQian,LI Guo,GUO Xi,ZHONG Liang,ZHU Qing,GUO JiaXin,YE YingCong. Prediction of Soil Organic Carbon Content in Jiangxi Province by Vis-NIR Spectroscopy Based on the CARS-BPNN Model[J]. Scientia Agricultura Sinica, 2022, 55(19): 3738-3750.

图/表 9

图1

图2

表1

图3

图4

表2

图5

表3

图6

参考文献 45

[1]	ROSSEL R V, WALVOORT D J J, MCBRATNEY A B, JANIK L J, SKJEMSTAD J O. Visible, near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties. Geoderma, 2006, 131(1/2): 59-75. doi: 10.1016/j.geoderma.2005.03.007
[2]	KUANG B, MOUAZEN A M. Calibration of visible and near infrared spectroscopy for soil analysis at the field scale on three European farms. European Journal of Soil Science, 2011, 62(4) : 629-636. doi: 10.1111/j.1365-2389.2011.01358.x
[3]	ROSSEL R V, BEHRENS T. Using data mining to model and interpret soil diffuse reflectance spectra. Geoderma, 2010, 158(1/2): 46-54. doi: 10.1016/j.geoderma.2009.12.025
[4]	史舟, 王乾龙, 彭杰, 纪文君, 刘焕军, 李曦. 中国主要土壤高光谱反射特性分类与有机质光谱预测模型. 中国科学, 2014, 44(5): 978-988.
	SHI Z, WANG Q L, PENG J, JI W J, LIU H J, LI X. Development of a national VNIR soil-spectral library for soil classification and prediction of organic matter concentrations. Science China, 2014, 44(5): 978-988. (in Chinese)
[5]	MOUAZEN A M, KUANG B, DE BAERDEMAEKER J, RAMON H. Comparison among principal component, partial least squares and back propagation neural network analyses for accuracy of measurement of selected soil properties with visible and near infrared spectroscopy. Geoderma, 2010, 158(1/2): 23-31. doi: 10.1016/j.geoderma.2010.03.001
[6]	DING J, YANG A, WANG J, SAGAN V, YU D. Machine-learning- based quantitative estimation of soil organic carbon content by VIS/NIR spectroscopy. PeerJ, 2018, 6: e5714. doi: 10.7717/peerj.5714
[7]	CHENG H, WANG J, DU Y. Combining multivariate method and spectral variable selection for soil total nitrogen estimation by Vis-NIR spectroscopy. Archives of Agronomy and Soil Science, 2021, 67(12): 1665-1678. doi: 10.1080/03650340.2020.1802013
[8]	XU S, ZHAO Y, WANG M, SHI X. Comparison of multivariate methods for estimating selected soil properties from intact soil cores of paddy fields by Vis-NIR spectroscopy. Geoderma, 2018, 310: 29-43. doi: 10.1016/j.geoderma.2017.09.013
[9]	纪文君, 史舟, 周清, 周炼清. 几种不同类型土壤的VIS-NIR 光谱特性及有机质响应波段. 红外与毫米波学报, 2012, 31(3): 277-282. doi: 10.3724/SP.J.1010.2012.00277
	JI W J, SHI Z, ZHOU Q, ZHOU L Q. VIS-NIR reflectance spectroscopy of the organic matter in several types of soils. Journal of Infrared and Millimeter Waves, 2012, 31(3): 277-282. (in Chinese) doi: 10.3724/SP.J.1010.2012.00277
[10]	VOHLAND M, LUDWIG M, HARBICH M, EMMERLING C, THIELE-BRUHN S. Using variable selection and wavelets to exploit the full potential of visible-near infrared spectra for predicting soil properties. Journal of Near Infrared Spectroscopy, 2016, 24(3): 255-269. doi: 10.1255/jnirs.1233
[11]	朱亚星, 于雷, 洪永胜, 章涛, 朱强, 李思缔, 郭力, 刘家胜. 土壤有机质高光谱特征与波长变量优选方法. 中国农业科学, 2017, 50(22): 4325-4337.
	ZHU Y X, YU L, HONG Y S, ZHANG T, ZHU Q, LI S D, GUO L, LIU J S. Hyperspectral features and wavelength variables selection methods of soil organic matter. Scientia Agricultura Sinica, 2017, 50(22): 4325-4337. (in Chinese)
[12]	YANG M, XU D, CHEN S, LI H, SHI Z. Evaluation of machine learning approaches to predict soil organic matter and pH using Vis-NIR spectra. Sensors, 2019, 19(2): 263. doi: 10.3390/s19020263
[13]	KAWAMURA K, TSUJIMOTO Y, NISHIGAKI T, ANDRIAMANANJARA A, RABENARIVO M, ASAI H, RAZAFIMBELO T. Laboratory visible and near-infrared spectroscopy with genetic algorithm-based partial least squares regression for assessing the soil phosphorus content of upland and lowland rice fields in Madagascar. Remote Sensing, 2019, 11(5): 506. doi: 10.3390/rs11050506
[14]	LIU J, DONG Z, XIA J, WANG H, MENG T, ZHANG R, XIE J. Estimation of soil organic matter content based on CARS algorithm coupled with random forest. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2021, 258: 119823. doi: 10.1016/j.saa.2021.119823
[15]	于雷, 洪永胜, 周勇, 朱强, 徐良, 李冀云, 聂艳. 高光谱估算土壤有机质含量的波长变量筛选方法. 农业工程学报, 2016, 32(13): 95-102.
	YU L, HONG Y S, ZHOU Y, ZHU Q, XU L, LI J Y, NIE Y. Wavelength variable selection methods for estimation of soil organic matter content using hyperspectral technique. Transactions of the Chinese Society of Agricultural Engineering, 2016, 32(13): 95-102. (in Chinese)
[16]	国佳欣, 朱青, 赵小敏, 郭熙, 韩逸, 徐喆. 不同土地利用类型下土壤有机碳含量的高光谱反演. 应用生态学报, 2020, 31(3): 863-871. doi: 10.13287/j.1001-9332.202003.014
	GUO J X, ZHU Q, ZHAO X M, GUO X, HAN Y, XU Z. Hyper- spectral inversion of soil organic carbon content under different land use types. Chinese Journal of Applied Ecology, 2020, 31(3): 863-871. doi: 10.13287/j.1001-9332.202003.014
[17]	钟亮, 郭熙, 国佳欣, 徐喆, 朱青, 丁萌. 基于不同卷积神经网络模型的红壤有机质高光谱估算. 农业工程学报, 2021, 37(1): 203-212.
	ZHONG L, GUO X, GUO J X, XU Z, ZHU Q, DING M. Hyperspectral estimation of organic matter in red soil using different convolutional neural network models. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(1): 203-212.
[18]	YANG M, MOUAZEN A, ZHAO X, GUO X. Assessment of a soil fertility index using visible and near-infrared spectroscopy in the rice paddy region of southern China. European Journal of Soil Science, 2020, 71(4): 615-626. doi: 10.1111/ejss.12907
[19]	SHI Z, JI W, VISCARRA ROSSEL R A, CHEN S, ZHOU Y. Prediction of soil organic matter using a spatially constrained local partial least squares regression and the Chinese vis-NIR spectral library. European Journal of Soil Science, 2015, 66(4): 679-687. doi: 10.1111/ejss.12272
[20]	赵小敏, 杨梅花. 江西省红壤地区主要土壤类型的高光谱特性研究. 土壤学报, 2018, 55(1): 31-42.
	ZHAO X M, YANG M H. Hyper-spectral characteristics of major types of soils in red soil region of Jiangxi province, China. Acta Pedologica Sinica, 2018, 55(1): 31-42. (in Chinese)
[21]	LIU S, SHEN H, CHEN S, ZHAO X, BISWAS A, JIA X, FANG J. Estimating forest soil organic carbon content using vis-NIR spectroscopy: Implications for large-scale soil carbon spectroscopic assessment. Geoderma, 2019, 348: 37-44. doi: 10.1016/j.geoderma.2019.04.003
[22]	LIU Y, SHI Z, ZHANG G, CHEN Y, LI S, HONG Y, LIU Y. Application of spectrally derived soil type as ancillary data to improve the estimation of soil organic carbon by using the Chinese soil vis-NIR spectral library. Remote Sensing, 2018, 10(11): 1747. doi: 10.3390/rs10111747
[23]	唐海涛, 孟祥添, 苏循新, 马涛, 刘焕军, 鲍依临, 张美薇, 张新乐, 霍海志. 基于CARS 算法的不同类型土壤有机质高光谱预测. 农业工程学报, 2021, 37(2): 105-113.
	TANG H T, MENG X T, SU X X, MA T, LIU H J, BAO Y L, ZHANG M W, ZHANG X Y, HUO H Z. Hyperspectral prediction on soil organic matter of different types using CARS algorithm. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(2): 105-113. (in Chinese)
[24]	李冠稳, 高小红, 肖能文, 肖云飞. 基于sCARS-RF算法的高光谱估算土壤有机质含量. 发光学报, 2019, 40(8): 1030-1039. doi: 10.3788/fgxb20194008.1030
	LI G W, GAO X H, XIAO N W, XIAO Y F. Estimation soil organic matter contents with hyperspectra based on sCARS and RF algorithms. Chinese Journal of Luminescence, 2019, 40(8): 1030-1039. (in Chinese) doi: 10.3788/fgxb20194008.1030
[25]	VOHLAND M, LUDWIN M, THIELE-BRUHN S, LUDWIG B. Determination of soil properties with visible to near-and mid-infrared spectroscopy: Effects of spectral variable selection. Geoderma, 2014, 223: 88-96.
[26]	HONG Y, CHEN S, LIU Y, ZHANG Y, YU L, CHEN Y, LIU Y. Combination of fractional order derivative and memory-based learning algorithm to improve the estimation accuracy of soil organic matter by visible and near-infrared spectroscopy. Catena, 2019, 174: 104-116. doi: 10.1016/j.catena.2018.10.051
[27]	VISCARRA ROSSEL R A, HICKS W S. Soil organic carbon and its fractions estimated by visible-near infrared transfer functions. European Journal of Soil Science, 2015, 66(3): 438-450. doi: 10.1111/ejss.12237
[28]	KAWAMURA K, TSUJIMOTO Y, RABENARIVO M, ASAI H, ANDRIAMANANJARA A, RAKOTOSON T. Vis-NIR spectroscopy and PLS regression with waveband selection for estimating the total C and N of paddy soils in Madagascar. Remote Sensing, 2017, 9(10): 1081. doi: 10.3390/rs9101081
[29]	DOTTO A C, DALMOLIN R S D, TEN CATEN A, GRUNWALD S. A systematic study on the application of scatter-corrective and spectral-derivative preprocessing for multivariate prediction of soil organic carbon by Vis-NIR spectra. Geoderma, 2018, 314: 262-274. doi: 10.1016/j.geoderma.2017.11.006
[30]	KUANG B, TEKIN Y, MOUAZEN A M. Comparison between artificial neural network and partial least squares for on-line visible and near infrared spectroscopy measurement of soil organic carbon, pH and clay content. Soil and Tillage Research, 2015, 146: 243-252. doi: 10.1016/j.still.2014.11.002
[31]	HONG Y, LIU Y, CHEN Y, LIU Y, YU L, LIU Y, CHENG H. Application of fractional-order derivative in the quantitative estimation of soil organic matter content through visible and near-infrared spectroscopy. Geoderma, 2019, 337: 758-769. doi: 10.1016/j.geoderma.2018.10.025
[32]	纪文君, 李曦, 李成学, 周银, 史舟. 基于全谱数据挖掘技术的土壤有机质高光谱预测建模研究. 光谱学与光谱分析, 2012, 32(9): 2393-2398.
	JI W J, LI X, LI C X, ZHOU Y, SHI Z. Using different data mining algorithms to predict soil organic matter based on visible-near infrared spectroscopy. Spectroscopy and Spectral Analysis, 2012, 32(9): 2393-2398. (in Chinese)
[33]	郭熙, 谢碧裕, 叶英聪, 谢文. 高光谱特征辨别潴育型麻沙泥田和潮沙泥田水稻土. 农业工程学报, 2014, 30(21): 184-191.
	GUO X, XIE B Y, YE Y C, XIE W. Discrimination between hydromorphic alluvial sandy mud paddy and tide sandy mud paddy based on hyperspectral characteristics. Transactions of the Chinese Society of Agricultural Engineering, 2014, 30(21): 184-191. (in Chinese)
[34]	KAWAMURA K, NISHIGAKI T, TSUJIMOTO Y, ANDRIAMANANJARA A, RABENARIBO M, ASAI H, RAZAFIMBELO T. Exploring relevant wavelength regions for estimating soil total carbon contents of rice fields in Madagascar from Vis-NIR spectra with sequential application of backward interval PLS. Plant Production Science, 2021, 24(1): 1-14. doi: 10.1080/1343943X.2020.1785898
[35]	JI W, SHI Z, HUANG J, LI S. Correction: In situ measurement of some soil properties in paddy soil using visible and near-infrared spectroscopy. PLoS ONE, 2016, 11(7): e0159785. doi: 10.1371/journal.pone.0159785
[36]	SHI T, CHEN Y, LIU H, WANG J, WU G. Soil organic carbon content estimation with laboratory-based visible-near-infrared reflectance spectroscopy: Feature selection. Applied Spectroscopy, 2014, 68(8): 831-837. doi: 10.1366/13-07294 pmid: 25061784
[37]	VOHLAND M, BESOLD J, HILL J, FRÜND H C. Comparing different multivariate calibration methods for the determination of soil organic carbon pools with visible to near infrared spectroscopy. Geoderma, 2011, 166(1): 198-205. doi: 10.1016/j.geoderma.2011.08.001
[38]	XU L, HONG Y, WEI Y, GUO L, SHI T, LIU Y, CHEN Y. Estimation of organic carbon in anthropogenic soil by VIS-NIR spectroscopy: Effect of variable selection. Remote Sensing, 2020, 12(20): 3394. doi: 10.3390/rs12203394
[39]	HONG Y, CHEN Y, YU L, LIU Y, LIU Y, ZHANG Y, CHENG H. Combining fractional order derivative and spectral variable selection for organic matter estimation of homogeneous soil samples by VIS-NIR spectroscopy. Remote Sensing, 2018, 10(3): 479. doi: 10.3390/rs10030479
[40]	HONG Y, CHEN S, CHEN Y, LINDERMAN M, MOUAZEN A M, LIU Y, LIU Y. Comparing laboratory and airborne hyperspectral data for the estimation and mapping of topsoil organic carbon: Feature selection coupled with random forest. Soil and Tillage Research, 2020, 199: 104589. doi: 10.1016/j.still.2020.104589
[41]	ENGLAND J R, VISCARRA ROSSEL R A. Proximal sensing for soil carbon accounting. Soil, 2018, 4(2): 101-122. doi: 10.5194/soil-4-101-2018
[42]	MOURA-BUENO J M, DALMOLIN R S D, TEN CATEN A, DOTTO A C, DEMATTÊ J A. Stratification of a local VIS-NIR-SWIR spectral library by homogeneity criteria yields more accurate soil organic carbon predictions. Geoderma, 2019, 337: 565-581. doi: 10.1016/j.geoderma.2018.10.015
[43]	ARAÚJO S R, WETTERLIND J, DEMATTÊ J A M, STENBERG B. Improving the prediction performance of a large tropical vis-NIR spectroscopic soil library from Brazil by clustering into smaller subsets or use of data mining calibration techniques. European Journal of Soil Science, 2014, 65(5): 718-729. doi: 10.1111/ejss.12165
[44]	BAO Y, MENG X, USTIN S, WANG X, ZHANG X, LIU H, TANG H. Vis-SWIR spectral prediction model for soil organic matter with different grouping strategies. Catena, 2020, 195: 104703. doi: 10.1016/j.catena.2020.104703
[45]	GHOLIZADEH A, ROSSEL R A V, SABERIOON M, BORŮVKA L, KRATINA J, PAVLŮ L. National-scale spectroscopic assessment of soil organic carbon in forests of the Czech Republic. Geoderma, 2021, 385: 114832. doi: 10.1016/j.geoderma.2020.114832

土壤类型 Type of soil	样本类型 Type of sample	样本数量 Number of samples	最小值 Minimum (g·kg^-1)	最大值 Maximum (g·kg^-1)	平均值 Mean (g·kg^-1)	标准差 Standard deviation (g·kg^-1)	变异系数 Coefficient of variation
全部 Total	Ⅰ	490	4.12	34.11	16.75	6.21	0.37
	Ⅱ	367	4.63	34.11	16.89	6.10	0.36
	Ⅲ	123	4.12	29.58	16.33	6.52	0.40
红壤 Red soil	Ⅰ	242	4.44	28.89	16.17	5.91	0.37
	Ⅱ	182	4.44	28.89	16.29	5.89	0.36
	Ⅲ	60	4.44	27.20	15.83	5.99	0.38
水稻土 Paddy soil	Ⅰ	248	4.12	34.11	17.32	6.45	0.37
	Ⅱ	186	4.63	34.11	17.53	6.28	0.36
	Ⅲ	62	4.12	30.92	16.71	6.95	0.42

方法 Method	土壤类型 Type of soil	训练集 Training set			验证集 Validation set
方法 Method	土壤类型 Type of soil	R²	RMSE (g·kg^-1)	RPD	R²	RMSE (g·kg^-1)	RPD
PLSR	R	0.80	2.61	2.25	0.74	3.03	1.96
PLSR	P	0.76	3.07	2.04	0.73	3.59	1.92
SVM	R	0.84	2.38	2.47	0.76	2.91	2.05
SVM	P	0.77	2.99	2.10	0.76	3.41	2.02
RF	R	0.88	2.05	2.86	0.74	3.01	1.97
RF	P	0.76	3.06	2.05	0.70	3.78	1.82
BPNN	R	0.86	2.19	2.69	0.75	2.95	2.01
BPNN	P	0.80	2.81	2.23	0.77	3.32	2.08
CARS-PLSR	R	0.83	2.41	2.44	0.81	2.61	2.28
CARS-PLSR	P	0.81	2.74	2.29	0.79	3.12	2.21
CARS-SVM	R	0.83	2.45	2.40	0.78	2.79	2.13
CARS-SVM	P	0.79	2.87	2.18	0.77	3.31	2.08
CARS-RF	R	0.89	1.97	2.98	0.80	2.68	2.21
CARS-RF	P	0.85	2.46	2.55	0.83	2.85	2.42
CARS-BPNN	R	0.85	2.28	2.58	0.82	2.50	2.38
CARS-BPNN	P	0.82	2.69	2.33	0.81	2.99	2.31

模型 Model	全局/分类 Global/Classification	训练集 Training set			验证集 Validation set
模型 Model	全局/分类 Global/Classification	R²	RMSE (g·kg^-1)	RPD	R²	RMSE (g·kg^-1)	RPD
PLSR	G	0.71	3.26	1.87	0.67	3.73	1.75
PLSR	C	0.78	2.85	2.14	0.73	3.33	1.96
SVM	G	0.79	2.78	2.19	0.73	3.40	1.91
SVM	C	0.80	2.70	2.26	0.76	3.17	2.05
RF	G	0.84	2.45	2.49	0.61	4.08	1.59
RF	C	0.82	2.61	2.34	0.72	3.42	1.90
BPNN	G	0.79	2.78	2.19	0.61	4.05	1.60
BPNN	C	0.83	2.52	2.42	0.76	3.14	2.07
CARS-PLSR	G	0.80	2.74	2.22	0.76	3.20	2.03
CARS-PLSR	C	0.82	2.58	2.36	0.80	2.88	2.26
CARS-SVR	G	0.79	2.78	2.19	0.74	3.32	1.95
CARS-SVR	C	0.81	2.67	2.28	0.77	3.07	2.12
CARS-RF	G	0.85	2.33	2.61	0.72	3.46	1.88
CARS-RF	C	0.87	2.23	2.73	0.82	2.77	2.35
CARS-BPNN	G	0.80	2.70	2.26	0.76	3.18	2.04
CARS-BPNN	C	0.83	2.49	2.44	0.82	2.75	2.36