余星兴(1996-), 女, 四川南充人, 硕士研究生, 主要从事土壤资源可持续利用研究。E-mail:
土壤颜色和游离铁均是土壤系统分类中的指标,两者之间具有一定的关系,但迄今关于两者之间定量关系的报道甚少。以川中丘陵区典型土系为研究对象,通过定量分析其Munsell颜色与游离铁含量间的关系,尝试利用Munsell颜色建立的BP神经网络模型反演土壤游离铁含量,并与利用反射光谱模型反演的土壤游离铁含量进行比较。结果表明,Munsell色调、明度、彩度值与游离铁含量呈显著正相关,据此建立的游离铁含量Munsell颜色预测模型在单隐含层神经元个数为4时,
Soil iron oxides are found in almost all the types of soils and are good indicators reflecting variation of the environment thanks to their high variability in concentration. Soil iron oxides, mostly in the form of free iron, function as important mineral binders in the soil and have a significant effect on color of the soil. Soil color is an important soil property, described by Munsell color space, in the soil taxonomy. Both soil color and free iron are indicators in the soil taxonomy, and fairly related to each other, but so far few papers have been reported on quantitative relationships between them.
Since soil color is an indicator reflecting genesis and evolution of a soil, it is often used to invert and predict soil properties via modeling. Therefore, in this paper, the typical soil series in the hilly region of Central Sichuan were taken as the research object for analysis of relationships between soil Munsell color and free iron content in the soils. On this basis, a BP neural network model is established to explore differences between the Munsell color based model and the traditional spectral model in predicting soil free iron content.
Results show significantly positive relationships of free iron content with hue, value and chroma of soil Munsell color. When the Munsell color prediction model had 4 neurons in the single hidden layer, the determination coefficient R2 of its test set was 0.94, its standard deviation RMSE 4.20, and its relative analysis error RPD 4.37; When the spectral prediction model had 6 neurons in the single hidden layer, its R2 was 0.98, RMSE 3.35, and RPD 5.99. Both models have demonstrated a high level of goodness of fit and accuracy, though the spectral model is slightly higher than the color model.
Munsell color can be used to predict soil free iron content effectively, but by comparison, the spectral model is a bit higher in goodness of fit and prediction accuracy, which may be attributed to the fewer neurons in the input layer of the color model and the dispersion degree of free iron oxides dispersion degree. Color information is easy to obtain, for some historical soil literature, which do have color data, but lack the data of free iron contents, the Munsell color-based prediction model can be used to figure out an approximate content of free iron in the soil.
土壤颜色是判断成土环境、土壤发育程度及肥力特征的重要依据之一,在土壤发生、分类研究中具有重要意义,氧化铁是其主要影响因素之一[
在众多色空间/色度指标中,土壤科学中主要采用Munsell色空间,土壤系统分类以此为依据绘制色卡判别土壤颜色[
川中丘陵区遍布紫色土,其土色包括紫红、红、红黄等一系列颜色,氧化铁平均含量介于34.7~170.9 g·kg–1,低于我国土壤平均值,但略高于世界土壤平均值,土样之间变异系数较大[
供试的27个典型土系位于川中丘陵区(
供试典型土系的剖面样点位置
Sites of studied profiles of typical soil series
于2015—2016年进行土样采集,每个剖面依据发生层次自下而上采集分析样品,共97个土样。由于新鲜土样可能因土块大小不一或水分含量不等/含水量过饱和现象[
土壤全铁(Fet,过100目筛土样)和游离铁(Fed,过60目筛土样)分别采用碳酸锂-硼酸熔融、DCB浸提,电感耦合等离子发射光谱仪(ICP-AES)法测定[
Munsell颜色及光谱数据利用日本Konica Minolta公司CM600d分光测色仪测定。将过10目的土样置于配套粉末测试装置,使样品略多于装置,拧紧装置盖;设置测定参数为观测角度2°、内置C光源,选用8 mm测色稳定片;进行零校正与白板校正后将测定端置于粉末测试装置中测定,同一土样重复测定3次,获取Munsell色空间的色调、明度、彩度参数及光谱数据。
(1)土壤颜色、光谱数据处理。Munsell色空间中的色调由数字与英文颜色缩写组合而成,如2.5R,2.5YR,2.5Y等,本研究中将2.5R计为2.5、2.5YR计为12.5、2.5Y计为22.5[
CM600d测色仪以10 nm为间隔获取400~690 nm部分可见光波段土壤反射光谱数据,并计算各波段间反射率一阶导数值,以避免原始反射光谱曲线较平滑的现象,突出光谱特征[
(2)数据处理与模型建立及检验。数据统计分析及图形绘制使用Microsoft Excel 2016软件,土壤Fed含量和铁游离度(Fed/Fet,%)与色调、明度、彩度值之间Pearson相关分析使用IBM Statistics SPSS 22.0软件,使用MATLAB R2016a将数据随机划分70%为固定训练集、30%为固定测试集并建立反演模型。
研究采用BP神经网络模型[
式中,
模型拟合度及预测精度检验采用
按铁游离度的分级列出了土样数量以及测定获取的土样游离铁和颜色指标的平均值,具体见
研究区土壤游离铁及Munsell颜色描述性统计
Descriptive statistics of Fed and Munsell color of the soils in the research area
铁游离度分级 |
样品数 |
游离铁 |
色调 |
明度 |
彩度 |
0~20 | 7 | 7.06 | 19.30 | 5.29 | 2.47 |
20~40 | 61 | 15.10 | 16.17 | 4.58 | 3.20 |
40~60 | 12 | 18.19 | 20.97 | 5.43 | 3.96 |
60~80 | 17 | 45.21 | 19.52 | 6.18 | 5.06 |
合计Total | 97 | 20.18 | 17.58 | 5.02 | 3.57 |
从
土壤游离铁与Munsell颜色的Pearson相关关系
Pearson correlation coefficients between soil Fed and Munsell color parameters
彩度Chroma | 明度Value | 色调Hue | 铁游离度Fed/Fet | |
注:**在0.01水平(双侧)上显著相关;*在0.05水平(双侧)上显著相关。Note:**:denotes significant correlation at 0.01 level(bilateral);and * significant correlation at 0.05 level(bilateral). | ||||
游离铁Fed | 0.872** | 0.659** | 0.235* | 0.867** |
铁游离度Fed/Fet | 0.888** | 0.787** | 0.553** | |
色调Hue | 0.362** | 0.758** | ||
明度Value | 0.738** |
将Munsell色调、明度、彩度作为输入层,游离铁含量作为输出层,随机划分70%为固定训练集、30%为固定测试集,采用最大最小法进行数据归一化、归一化函数采用mapminmax函数,数据反归一化利用reverse语句完成。
由
Munsell颜色模型与反射率一阶导数光谱模型比较
Comparison between Munsell color model and first derivative reflectance model
隐含层神经元个数 |
Munsell颜色模型 |
反射率一阶导数光谱模型 |
|||||||
训练集 |
测试集Test set | 训练集 |
测试集Test set | ||||||
预期值Goal | RMSE | RPD | 预期值 |
RMSE | RPD | ||||
2 | × | 0.94 | 4.01 | 4.45 | |||||
3 | × | 0.95 | 3.97 | 4.49 | |||||
4 | √ | 0.94 | 4.20 | 4.37 | |||||
5 | √ | 0.94 | 4.22 | 4.32 | √ | 0.96 | 4.79 | 4.19 | |
6 | √ | 0.92 | 4.90 | 3.42 | √ | 0.98 | 3.35 | 5.99 | |
7 | √ | 0.92 | 6.00 | 3.74 | √ | 0.97 | 3.76 | 5.49 | |
8 | √ | 0.94 | 5.85 | 4.00 | √ | 0.95 | 3.91 | 4.77 | |
9 | √ | 0.92 | 5.04 | 3.26 | √ | 0.97 | 3.46 | 5.57 | |
10 | √ | 0.93 | 4.45 | 4.08 | √ | 0.94 | 4.38 | 3.59 | |
11 | √ | 0.93 | 4.77 | 3.98 | √ | 0.96 | 3.78 | 5.29 | |
12 | √ | 0.91 | 5.21 | 3.70 | √ | 0.95 | 4.23 | 4.68 | |
13 | √ | 0.94 | 4.39 | 4.32 | |||||
14 | √ | 0.96 | 4.16 | 4.27 | |||||
15 | √ | 0.95 | 4.72 | 3.71 |
当隐含层神经元个数为4时,
游离铁实测值与Munsell颜色(a)、反射率一阶导数光谱(b)模型预测值对比
Comparison between measured and predicted Fed with the Munsell color model (a) and first derivative reflectance model (b)
作为对比模型,将400~690 nm土壤可见光波段以10 nm为间隔获取的共计30个反射率一阶导数作为输入层,土壤游离铁含量作为输出层,训练集和测试集与颜色模型相同,数据归一化、反归一化利用mapminmax函数及reverse语句完成。
从
结合测试集中预测值与实测值相关分析(
由于土壤各元素对不同波段反射率不一致,输入的光谱信息成为影响模型拟合度、精度的重要因素之一。如在研究土壤铜含量高光谱反演模型时[
因此本研究选取400~690 nm波段反射率一阶导数与土壤游离铁含量进行Pearson相关分析,如
土壤游离铁含量与反射率一阶导数Pearson相关关系
Correlation coefficients between Fed and first derivative reflectance
400 nm | 410 nm | 420 nm | 430 nm | 440 nm | 450 nm | 460 nm | 470 nm | 480 nm | 490 nm | |
注:**在0.01水平(双侧)上显著相关;*在0.05水平(双侧)上显著相关。Note:**:denotes significant correlation at 0.01 level(bilateral);and * significant correlation at 0.05 level(bilateral). | ||||||||||
Fed | 0.619** | 0.654** | 0.691** | 0.699** | 0.667** | 0.144 | –0.027 | 0.613** | 0.711** | 0.710** |
500 nm | 510 nm | 520 nm | 530 nm | 540 nm | 550 nm | 560 nm | 570 nm | 580 nm | 590 nm | |
Fed | 0.695** | 0.696** | 0.716** | 0.765** | 0.832** | 0.886** | 0.903** | 0.857** | 0.740** | 0.588** |
600 nm | 610 nm | 620 nm | 630 nm | 640 nm | 650 nm | 660 nm | 670 nm | 680 nm | 690 nm | |
Fed | 0.335** | –0.095 | –0.353** | –0.546** | –0.548** | –0.589** | –0.530** | –0.368** | –0.136 | –0.145 |
由
土壤颜色和游离铁含量均是影响土壤分类的重要性质,在《中国土壤系统分类检索(第三版)》[
“铁质特性”的本质是“土壤中游离氧化铁非晶质部分的浸润和赤铁矿、针铁矿、矿微晶的形成,并充分分散于土壤基质内使土壤红化”[
川中丘陵区土壤Munsell色调、明度、彩度值均与土壤游离铁含量具有良好的正相关关系,因此可利用土壤Munsell颜色建立BP神经网络模型预测土壤游离铁含量,同时利用反射光谱建立预测模型进行比较。结果表明,颜色模型的隐含层神经元个数为4时,
Xiong Y. Soil color research[J]. Geological Review, 1937, 2(1): 5-22.
熊毅. 土壤颜色之研究[J]. 地质论评, 1937, 2(1): 5-22.
Barouchas P E, Moustakas N K. Precipitation effect on colour characteristics of argillic horizons in well-drained soils developed under Mediterranean climate[J]. International Agrophysics, 2006, 20(2): 85-90.
Barron V, Torrent J. Use of the Kubelka—Munk theory to study the influence of iron oxides on soil colour[J]. Journal of Soil Science, 1986, 37(4): 499-510.
Wu Y Z, Tian Q J, Ji J F, et al. Soil remote sensing research theory method and application[J]. Remote Sensing Information, 2003, 18(1): 40-47.
吴昀昭, 田庆久, 季峻峰, 等. 土壤光学遥感的理论、方法及应用[J]. 遥感信息, 2003, 18(1): 40-47.
Chen J F. Oxides in soil colloids[J]. Chinese Journal of Soil Science, 1981, 12(2): 44-49.
陈家坊. 土壤胶体中的氧化物[J]. 土壤通报, 1981, 12(2): 44-49.
Wei C L, Zhao Y G, Wu D W, et al. Prediction of soil free iron oxide content based on spectral analysis[J]. Soils, 2014, 46(4): 678-683.
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.
Xiong J F, Zheng G H, Lin C. Estimating soil iron content based on reflectance spectra[J]. Spectroscopy and Spectral Analysis, 2016, 36(11): 3615-3619.
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.
Gopal B. Evaluation of topsoil iron oxide from visible spectroscopy[J]. International Journal of Research in Engineering and Technology, 2013, 2(12): 670-673.
Qin Q R, Qi Y B, Wu J, et al. Estimation of random forest model of soil free iron based on hyperspectral data[J]. Chinese Journal of Soil Science, 2018, 49(6): 1286-1293.
秦倩如, 齐雁冰, 吴娟, 等. 基于高光谱的土壤游离铁随机森林模型估算研究[J]. 土壤通报, 2018, 49(6): 1286-1293.
Gong Z T, Chen Z C, Luo G B, et al. Reference of Chinese Soil Taxonomy[J]. Soils, 1999, 31(2): 57-63.
龚子同, 陈志诚, 骆国保, 等. 中国土壤系统分类参比[J]. 土壤, 1999, 31(2): 57-63.
Mauk L. Billmeyer and Saltzman's principles of color technology[J]. Color Research & Application, 2001, 26(4): 322-324.
Kirillova N P, Vodyanitskii Y N, Sileva T M. Conversion of soil color parameters from the Munsell system to the CIE-L*a*b* system[J]. Eurasian Soil Science, 2015, 48(5): 468-475.
Wu C W, Yang Y, Xia J X. Rapid method for estimating organic matter of black soil based on RGB[J]. Chinese Journal of Soil Science, 2016, 47(4): 853-859.
吴才武, 杨越, 夏建新. 基于RGB的黑土有机质快速测定方法研究[J]. 土壤通报, 2016, 47(4): 853-859.
Islam K, McBratney A B, Singh B. Estimation of soil colour from visible reflectance spectra[J]. Super Soil 2004: 3rd Australian New Zealand Soils Conference, Sydney, 2004.
Levin N, BenâDor E, Singer A. A digital camera as a tool to measure colour indices and related properties of sandy soils in semi-arid environments[J]. International Journal of Remote Sensing, 2005, 26(24): 5475-5492.
Chen J K, Yuan D G, Yan Z M, et al. Comparison between colorimeter and new standard soil colour chart of china in determining Munsell color of soils-A case study of central Sichuan hilly region[J]. Acta Pedologica Sinica, 2019, 56(1): 78-89.
陈剑科, 袁大刚, 晏昭敏, 等. 测色仪与中国标准土壤色卡测定土壤颜色比较——以川中丘陵区为例[J]. 土壤学报, 2019, 56(1): 78-89.
Tian G L, Tang S J. Contents and distribution of mineral-forming elements in purple soils from the Sichuan Basin[J]. Geochimica, 1991, 20(1): 70-78.
田光龙, 唐时嘉. 四川盆地紫色土矿质元素的含量与分布[J]. 地球化学, 1991, 20(1): 70-78.
Liu Y, Ding X, Liu H J, et al. Quantitative analysis of reflectance spectrum of black soil as affected by soil moisture for prediction of soil moisture in black soil[J]. Acta Pedologica Sinica, 2014, 51(5): 1021-1026.
刘洋, 丁潇, 刘焕军, 等. 黑土土壤水分反射光谱特征定量分析与预测[J]. 土壤学报, 2014, 51(5): 1021-1026.
Zhang G L, Gong Z T. Soil survey laboratory methods[M]. Beijing: Science Press, 2012.
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.
Viscarra Rossel R A, Minasny B, Roudier P, et al. Colour space models for soil science[J]. Geoderma, 2006, 133(3/4): 320-337.
Ling C X, Liu H, Ju H B, et al. Identifying typical wetland vegetation types based on imaging spectrometer data: A case study in Dongting lake wetland area[J]. Journal of Northwest Forestry University, 2018, 33(3): 208-213.
凌成星, 刘华, 鞠洪波, 等. 基于地面成像光谱数据特征的湿地典型植被类型识别研究——以东洞庭湖核心区湿地为例[J]. 西北林学院学报, 2018, 33(3): 208-213.
Wang X C. Matlab neural network analysis of 43 cases[M]. Beijing: Beihang University Press, 2013.
王小川. MATLAB神经网络43个案例分析[M]. 北京: 北京航空航天大学出版社, 2013.
Lin J, Chen X M, Zhang Y, et al. Simulation of soil moisture dynamics based on the BP neural network in the typical farmland of Tai Lake region[J]. Journal of Nanjing Agricultural University, 2012, 35(4): 144-144.
林洁, 陈效民, 张勇, 等. 基于BP神经网络的太湖典型农田土壤水分动态模拟[J]. 南京农业大学学报, 2012, 35(4): 144-144.
Summers D, Lewis M, Ostendorf B, et al. Visible near-infrared reflectance spectroscopy as a predictive indicator of soil properties[J]. Ecological Indicators, 2011, 11(1): 123-131.
Guo Y K, Liu N, Liu L, et al. Hyper-spectral inversion of soil Cu content based on BP neural network model[J]. Science of Surveying and Mapping, 2018, 43(1): 135-139, 152.
郭云开, 刘宁, 刘磊, 等. 土壤Cu含量高光谱反演的BP神经网络模型[J]. 测绘科学, 2018, 43(1): 135-139, 152.
Xie W, Zhao X M, Guo X, et al. Spectrum based estimation of the content of soil organic matters in mountain red soil using RBF combination model[J]. Scientia Silvae Sinicae, 2018, 54(6): 16-23.
谢文, 赵小敏, 郭熙, 等. 基于RBF组合模型的山地红壤有机质含量光谱估测[J]. 林业科学, 2018, 54(6): 16-23.
He T, Wang J, Cheng Y, et al. Study on spectral features of soil Fe2O3 [J]. Geography and Geo-Information Science, 2006, 22(2): 30-34.
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.
Yuan X K, Li X S, Han Z Y, et al. Study on the application of quantitative analysis of iron oxides by diffuse reflectance spectroscopy in the loess of South China[J]. Geological Journal of China Universities, 2017, 23(2): 350-358.
苑晓康, 李徐生, 韩志勇, 等. 基于漫反射光谱的铁氧化物定量分析在南方黄土中的适用性研究[J]. 高校地质学报, 2017, 23(2): 350-358.
Deaton B C, Balsam W L. Visible spectroscopy; a rapid method for determining hematite and goethite concentration in geological materials[J]. Journal of Sedimentary Research, 1991, 61(4): 628-632.
Chen J, Ji J F, Balsam W, et al. Characterization of the Chinese loess-paleosol stratigraphy by whiteness measurement[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2002, 183(3/4): 287-297.
Ji J F, Balsam W, Chen J, et al. Rapid and quantitative measurement of hematite and goethite in the Chinese loess-paleosol sequence by diffuse reflectance spectroscopy[J]. Clays and Clay Minerals, 2002, 50(2): 208-216.
Hu X F, Du Y, Guan C L, et al. Color variations of the Quaternary Red Clay in Southern China and its paleoclimatic implications[J]. Sedimentary Geology, 2014, 303: 15-25.
Chinese Soil Taxonomy Research Group, Institute of Soil Science, Chinese Academy of Sciences, Cooperative Research Group on Chinese Soil Taxonomy. Keys to Chinese Soil Taxonomy[M]. 3rd ed. Hefei: University of Science and Technology of China Press, 200l.
中国科学院南京土壤研究所土壤系统分类课题组, 中国土壤系统分类课题研究协作组. 中国土壤系统分类检索[M]. 第3版. 合肥: 中国科学技术大学出版社, 2001.
Schwertmann U. Relations between iron oxides, soil color, and soil formation[M]//Soil Color. Madison, WI, USA: Soil Science Society of America, 2015: 51-69.