Spectral characteristics and quantitative retrieval of free iron content in soil
-
摘要:目的
建立基于可见−近红外光谱的土壤游离铁精确预测模型,简单、快速、经济地预测土壤游离铁,有助于研究土壤发生和分类。
方法采集广西壮族自治区的铁铝土、富铁土、淋溶土和雏形土等82个旱地土壤剖面的B层土壤,进行室内土壤化学分析、光谱测定,分析不同光谱变换后的光谱反射率与土壤游离铁含量的相关性。基于特征波段利用偏最小二乘回归(PLSR)和逐步多元线性回归(SMLR)法建立土壤游离铁含量光谱预测模型,通过决定系数(R2)、均方根误差(RMSE)和相对预测偏差(PRD)确定最优模型。
结果土壤光谱曲线分别在457、800和900 nm波段附近有明显的游离铁吸收和反射峰特征;土壤游离铁含量与原始光谱反射率呈负相关;原始光谱经过微分变换后,游离铁含量与光谱反射率相关性显著提高;基于400~580和760~1 300 nm特征波段和一阶微分光谱变换的SMLR模型预测精度最高,其验证集的R2和RPD分别为0.85和2.62,RMSE为8.41 g·kg−1。
结论将可见近红外光谱技术应用于土壤游离铁含量高效快速地预测具有良好的可行性。广西旱地土壤光谱反射率与土壤游离铁含量具有高度的相关性,应用逐步多元线性回归方法可以很好地建立土壤游离铁含量反演模型。
Abstract:ObjectiveTo establish an accurate predicted model for free iron in soil based on visible and near infrared (vis-NIR) reflectance spectroscopy, provide a simple, rapid and economical method for soil free iron determination, and facilitate the pedogenesis and classification of soil.
MethodSoil samples in B horizon were collected from eighty-two upland soil profiles in Guangxi including ferralosols, ferrosols, argosols and cambosols. Chemical and spectral properties of soil samples were analyzed under laboratory condition. The correlation between spectral reflectance after transformation and free iron content in soils was analyzed. The predicted models of soil free iron were established by the method of partial least squares regression (PLSR) and stepwise multiple linear regression (SMLR) based on characteristic bands. The optimal model was determined by evaluating the coefficient of determination (R2), root mean square error (RMSE), and residual prediction deviation(RPD).
ResultSoil spectral curves had obvious characteristics of free iron absorption and reflection peaks near 457, 800 and 900 nm bands respectively. Free iron content in soils negatively correlated with the raw spectral reflectance. The correlation coefficient between spectral reflectance and free iron content in soils increased significantly after differential transformation of the raw spectrum. The predicted model of free iron content in soils established by the first-order differential spectral transformation and SMLR based on characteristic bands of 400−580 and 760−1 300 nm had the highest accuracy, R2 and RPD of the verification set were 0.85, and 2.62 respectively, and RMSE was 8.41 g·kg−1.
ConclusionIt is feasible to rapidly and cost-effectively predict free iron content in soils using vis-NIR spectral technology. Soil spectral reflectance of upland in Guangxi has a high correlation with soil free iron content. SMLR is a good method to establish the predicted model of soil free iron content.
-
Keywords:
- soil free iron /
- spectral feature /
- spectral transformation /
- predicted model /
- Guangxi
-
-
图 5 不同建模波段最佳模型的土壤游离铁含量
a:全波段倒数一阶微分逐步多元线性回归模型; b:特征波段一阶微分逐步多元线性回归模型; c:敏感波段倒数一阶微分偏最小二乘回归模型
Figure 5. Soil free iron content of the optimal model in different modeling band
a: Reciprocal first differential method using the SMLR model in the full bands; b: First differential method using the SMLR model in the feature bands; c: Reciprocal first differential method using the PLSR model in the sensitive bands
表 1 敏感波段共线性检验1)
Table 1 Collinearity test of sensitive band
光谱变换
Pectral transformationλ(敏感)/nm
Sensitive band皮尔逊相关系数
Pearson correlation coefficient容差
Tolerance方差膨胀因子
Variance inflation factor一阶微分
First order differential735 0.43** 1.00 1.00 847 −0.49** 1.00 1.00 去包络线
Continuum removal456 0.55** 0.17 5.96 1 102 −0.43** 0.99 1.01 倒数 Reciprocal 400 0.64** 0.19 5.18 510 0.41** 0.19 5.18 倒数一阶微分
Reciprocal first order differential498 −0.78** 0.54 1.84 2 191 0.63** 0.54 1.84 平方根一阶微分
Square root first order differential457 −0.64** 0.59 1.69 848 −0.48** 0.65 1.54 1 028 0.49** 0.92 1.08 1 759 −0.50** 0.90 1.11 1)**表示0.01水平显著相关
1) ** represents 0.01 level of significant correlation表 2 土壤游离铁含量预测模型1)
Table 2 Calibration and validation of the model for predicting soil free iron content
光谱变换
Spectral transformationSMLR PLSR 建模样本
Calibration set验证样本
Validation set建模样本
Calibration set验证样本
Validation setR2 RMSE/
(g·kg−1)R2 RMSE/
(g·kg−1)RPD 主成分
个数R2 RMSE/
(g·kg−1)R2 RMSE/
(g·kg−1)RPD 全波段 Full band FD 0.85 8.92 0.28 19.53 1.13 10 0.99 2.25 0.64 14.27 1.54 CR 0.76 12.16 0.62 17.86 1.23 3 0.51 16.10 0.69 15.56 1.42 RT 0.87 8.28 0.80 10.34 2.20 4 0.61 14.52 0.78 12.26 1.68 RTFD 0.89 7.61 0.82 9.97 2.21 9 0.96 2.85 0.69 13.80 1.60 SRFD 0.89 7.65 0.45 19.77 1.11 12 1.00 0.91 0.67 13.22 1.67 特征波段 Characteristic band FD 0.89 7.68 0.85 8.41 2.62 7 0.96 4.42 0.64 13.15 1.68 CR 0.83 9.77 0.78 10.00 2.20 3 0.84 9.44 0.68 11.81 1.87 RT 0.63 14.11 0.75 10.97 2.01 8 0.90 9.77 0.77 11.14 1.98 RTFD 0.89 7.73 0.40 17.08 1.29 3 0.71 12.61 0.71 13.13 1.68 SRFD 0.89 7.77 0.80 10.74 2.05 5 0.91 7.11 0.71 13.26 1.66 敏感波段 Sensitive band FD 0.56 15.34 0.59 14.22 1.55 2 0.64 13.89 0.64 14.09 1.56 CR 0.74 11.90 0.71 13.75 1.60 4 0.74 11.67 0.69 11.75 1.88 RT 0.70 12.70 0.48 15.36 1.43 4 0.63 13.96 0.61 13.13 1.68 RTFD 0.81 10.04 0.60 13.62 1.62 3 0.83 10.10 0.79 10.31 2.20 SRFD 0.65 13.71 0.65 12.52 1.76 2 0.68 13.13 0.67 12.75 1.73 1) FD为一阶微分;CR为去包络线;RT为倒数;RTFD为倒数一阶微分;SRFD为平方根一阶微分;R2为决定系数;RMSE为均方根误差;RPD为相对分析误差;全波段是400~2 400 nm;特征波段是400~580和760~1 300 nm;敏感波段是400、457、510、735、848 nm等12个波段
1) FD is first order differential; CR is continuum removal; RT is reciprocal; RTFD is reciprocal first order differential; SRFD is square root first order differential; R2 is determination coefficient; RMSE is root mean squared error; RPD is relative percent deviation; Full bands are 400−2 400 nm; Characteristic bands are 400−580 and 760−1 300 nm; Sensitive bands are twelve bands including 400, 457, 510, 735, 848 nm, et al -
[1] 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[J]. 土壤学报, 2018, 55(1): 31-42. doi: 10.11766/trxb201705220571 [2] SORIANO-DISLA J M, JANIK L J, ROSSEL R A V, et al. The performance of visible, near-, and mid-infrared reflectance spectroscopy for prediction of soil physical, chemical, and biological properties[J]. Appl Spectrosc Rev, 2014, 49(2): 139-186. doi: 10.1080/05704928.2013.811081
[3] BO S, ROSSEL R A V, MOUAZEN A M, et al. Visible and near infrared spectroscopy in soil science[J]. Adv Agron, 2010, 107: 163-215. doi: 10.1016/S0065-2113(10)07005-7
[4] 贾重建, 卢瑛, 熊凡, 等. 珠江三角洲平原不同种植年限土壤铁氧化物特征研究[J]. 土壤学报, 2017, 54(4): 894-904. [5] 龚子同, 张甘霖, 陈志诚, 等. 土壤发生与系统分类[M]. 北京: 科学出版社, 2007. [6] COLEMAN T L, MONTGOMERY O L. Soil moisture, organic matter, and iron content effect on the spectral characteristics of selected vertisols and alfisols in Alabama[J]. Photogramm Eng Rem S, 1987, 53(12): 1659-1663.
[7] 彭杰, 向红英, 周清, 等. 土壤氧化铁的高光谱响应研究[J]. 光谱学与光谱分析, 2013, 33(2): 502-506. doi: 10.3964/j.issn.1000-0593(2013)02-0502-05 [8] 夏学齐, 季峻峰, 陈骏, 等. 土壤理化参数的反射光谱分析[J]. 地学前缘, 2009, 16(4): 354-362. doi: 10.3321/j.issn:1005-2321.2009.04.035 [9] SUMMERS D, LEWIS M, OSTENDORF B, et al. Visible near-infrared reflectance spectroscopy as a predictive indicator of soil properties[J]. Ecol Indic, 2011, 11(1): 123-131. doi: 10.1016/j.ecolind.2009.05.001
[10] 谢碧裕. 南方丘陵水稻土氧化铁高光谱特性与估测模型研究[D]. 南昌: 江西农业大学, 2015. [11] 熊俊峰, 郑光辉, 林晨. 基于反射光谱的土壤铁元素含量估算[J]. 光谱学与光谱分析, 2016, 36(11): 3615-3619. [12] 郭颖, 郭治兴, 刘佳, 等. 亚热带典型区域水稻土氧化铁高光谱反演: 以珠江三角洲为例[J]. 应用生态学报, 2017, 28(11): 3675-3683. [13] XU D Y, MA W X, CHEN S C, et al. Assessment of important soil properties related to Chinese soil taxonomy based on vis-NIR reflectance spectroscopy[J]. Comput Electron Agr, 2018, 144: 1-8. doi: 10.1016/j.compag.2017.11.029
[14] CAMARGO L A, JÙNIOR J M, BARROŃ V, et al. Mapping of clay, iron oxide and adsorbed phosphate in oxisols using diffuse reflectance spectroscopy[J]. Geoderma, 2015(251/252): 124-132.
[15] SELLITTO V M, FERNANDES R, BARRÓN V, et al. Comparing two different spectroscopic techniques for the characterization of soil iron oxides: Diffuse versus bi-directional reflectance[J]. Geoderma, 2009, 149(1/2): 1-9.
[16] ROSSEL R A V, BUI E N, CARITAT P D, et al. Mapping iron oxides and the color of Australian soil using visible-near-infrared reflectance spectra[J]. J Geophys Res, 2010, 115(F4): 1-13.
[17] BAHIA A S R S, MARQUES J R J, SIQUEIRA D S. Procedures using diffuse reflectance spectroscopy for estimating hematite and goethite in oxisols of São Paulo, Brazil[J]. Geoderma Regional, 2015, 5: 150-156. doi: 10.1016/j.geodrs.2015.04.006
[18] de SOUZA BAHIA A S R, MARQUES J R J, la SCALA N, et al. Prediction and mapping of soil attributes using diffuse reflectance spectroscopy and magnetic susceptibility[J]. Soil Sci Soc Am J, 2018, 81(6): 1450-1462.
[19] 魏昌龙, 赵玉国, 邬登巍, 等. 基于光谱分析的土壤游离铁预测研究[J]. 土壤, 2014, 46(4): 678-683. [20] 张甘霖, 龚子同. 土壤调查实验室分析方法[M]. 北京: 科学出版社, 2012. [21] 南锋, 朱洪, 毕如田. 黄土高原煤矿区复垦农田土壤有机质含量的高光谱预测[J]. 中国农业科学, 2016, 49(11): 2126-2135. doi: 10.3864/j.issn.0578-1752.2016.11.009 [22] VOLKAN BILGILI A, van ES H M, AKBAS F, et al. Visible-near infrared reflectance spectroscopy for assessment of soil properties in a semi-arid area of Turkey[J]. J Arid Environ, 2010, 74(2): 229-238. doi: 10.1016/j.jaridenv.2009.08.011
[23] 石朴杰, 王世东, 张合兵, 等. 基于高光谱的复垦农田土壤有机质含量估测[J]. 土壤, 2018, 50(3): 558-565. [24] 史舟, 王乾龙, 彭杰, 等. 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[J]. 中国科学: 地球科学, 2014, 44(5): 978-988. [25] VOHLAND M, BESOLD J, HILL J, et al. Comparing different multivariate calibration methods for the determination of soil organic carbon pools with visible to near infrared spectroscopy[J]. Geoderma, 2011, 166(1): 198-205. doi: 10.1016/j.geoderma.2011.08.001
[26] 黄应丰, 刘腾辉. 华南主要土壤类型的光谱特性与土壤分类[J]. 土壤学报, 1995, 32(1): 58-68. doi: 10.3321/j.issn:0564-3929.1995.01.001 [27] 何挺, 王静, 程烨, 等. 土壤氧化铁光谱特征研究[J]. 地理与地理信息科学, 2006, 22(2): 30-34. doi: 10.3969/j.issn.1672-0504.2006.02.008 [28] GALVAO L S, VITORELLO I. Role of organic matter in obliterating the effects of iron on spectral reflectance and color of Brazilian tropical soils[J]. Int J Remote Sens, 1998, 19(10): 1969-1979. doi: 10.1080/014311698215090