Estimating Fertility Index by Using Field-Measured Vis-NIR Spectroscopy in the Huanghui River Basin
-
摘要: 为探讨野外实测光谱数据对土壤肥力的估算能力,采集青海省湟水流域表层0 ~ 20 cm土壤样品220份,同步测量其采样位置的野外实测光谱数据,实验室对土壤养分、机械组成含量以及pH值进行分析。基于上述数据,对野外实测光谱反射率进行多元散射校正(Multiplicative scatter correction,MSC)、SG-一阶导数变换(SG - First Derivative,SG-1st)预处理,采用稳定性竞争自适应重加权采样法(stability competitive adaptive reweighted sampling,SCARS)提取不同土壤养分、机械组成含量以及pH值的特征波段,以偏最小二乘回归(partial least squares regression,PLSR)模型对土壤全碳(TC)、有机质(OM)、全氮(TN)、碱解氮(AN)、pH、黏粒(clay)、粉粒(silt)、砂粒(sand)含量进行估算并对比分析,构建土壤养分含量、pH值以及机械组成含量的最优野外实测光谱估算模型。结果表明:通过MSC校正和SG-1st变换能够有效增强野外光谱特征;经SCARS选取的特征波段主要集中于近红外波段。基于野外实测光谱数据建立的PLSR模型能够对研究区土壤TC、OM、TN、AN含量以及pH值进行粗略估算;其中,对于TC、OM、TN含量及pH值而言,最佳估算模型为经SG-1st处理后的SCARS-PLSR模型,RPD值均达到1.70以上(RPDTC = 1.76; RPDOM = 1.82;RPDTN = 2.04;RPDpH = 1.89),RPIQ值均达到1.90以上(RPIQTC = 1.91;RPIQOM = 2.53;RPIQTN = 2.98;RPIQpH = 2.03);对于土壤AN含量而言,经MSC处理后的SCARS-PLSR模型最佳,其RPDAN值高达1.91,RPIQ值高达2.39。对土壤clay、silt以及sand含量野外光谱均无法估算,RPD值均在1.00左右,RPIQ值在1.20左右。
-
关键词:
- 野外实测Vis-NIR光谱 /
- 土壤肥力属性 /
- 特征波段 /
- 偏最小二乘模型 /
- 湟水流域
Abstract: In order to explore the ability of field vis-NIR spectroscopy for estimating soil fertility index, 220 soil samples in the depth profile of 20 cm were collected in 2015, 2016 and 2017 in the Huangshui river basin, respectively. Soil reflectance spectroscopy of 350-2500 nm were synchronously measured in field by using ASD Field Spectrometer. Soil nutrients, pH and mechanical composition were analyzed in laboratory. The pretreatments of multiplicative scatter correction (MSC) and SG-1st derivative transform (SG-1st) were used to obtain the spectral reflectance curve, and then stability competitive adaptive reweighted sampling (SCARS) was used to select characteristic wave bands of soil properties. The prediction models for the contents of soil total carbon (TC), organic matter (OM), total nitrogen (TN), alkeline nitrogen (AN), pH, clay, silt and sand were constructed by partial least squares regression(PLSR), then selecting the optimal model for every soil property by comparison analysis. The results showed that SG-1st transformation and MSC correction could effectively enhance the field spectral characteristics of soil nutrient, pH and soil mechanical composition. The characteristic bands selected by SCARS were mainly focused on the near infrared wave bands. PLSR model based on field vis-NIR spectroscopy could roughly estimate the contents of soil TC, OM, TN and AN as well as pH values. For the properties of soil TC, OM, TN and pH, the best estimation model was SCARS-PLSR by SG-1st processed with the RPD values over 1.70 (RPDTC = 1.76; RPDOM = 1.82; RPDTN = 2.04; RPDpH = 1.89), the RPIQ values over 1.90 (RPIQTC = 1.91; RPIQOM = 2.53; RPIQTN = 2.98; RPIQpH = 2.03). For soil AN contents, the best estimation model was SCARS-PLSR by MSC processed with the RPD values up to 1.91 and the RPIQ values up to 2.39. The contents of soil clay, silt and sand could not be estimated with all the RPD values around 1.00 and the RPIQ values around 1.20. -
图 3 土壤野外原始反射光谱及预处理后的光谱反射率曲线(a为去除噪声前的原始光谱;b为去除噪声后的原始光谱;c为MSC预处理光谱反射率曲线;d为SG-1st预处理光谱反射率曲线)
Figure 3. Field original spectra of soil samples and different curves of soil spectral reflectance(a is the original spectrum before noise removal;b is the original spectrum after noise removal;c is the spectral reflectance of MSC;d is the spectral reflectance of SG-1st)
表 1 土壤肥力指标含量统计
Table 1. Statistical characteristics of soil fertility indices
土壤肥力指标
Soil fertility index最大值
Maximum最小值
Minimum平均值
Mean标准差
SD变异系数(%)
CVTC(g/kg) 107.43 16.16 30.37 11.47 37.78 OM(g/kg) 130.19 4.51 30.75 21.21 68.96 TN(g/kg) 8.66 0.36 2.15 1.23 57.50 AN(mg/kg) 224.46 11.88 67.61 40.55 59.99 pH 8.93 7.00 8.01 0.28 3.46 clay(%) 18.04 2.37 6.73 1.63 24.18 silt(%) 87.85 25.89 70.10 9.41 13.43 sand(%) 71.68 6.37 23.11 10.50 45.42 表 2 土壤肥力指标与光谱最大相关系数绝对值与对应波段
Table 2. Absolute value of maximum correlation coefficient between soil fertility indices and spectra and corresponding band
土壤肥力指标
Soil fertility indexRaw MSC SG-1st 最大值
Maximum波段位置(nm)
Band position最大值
Maximum波段位置(nm)
Band position最大值
Maximum波段位置(nm)
Band positionTC 0.359 602 0.603 1350 0.663 1593 OM 0.577 602 0.657 426 0.699 840 TN 0.511 598 0.636 608 0.663 1593 AN 0.490 598 0.648 604 0.787 840 pH 0.444 599 0.779 783 0.511 562 Clay 0.061 743 0.359 553 0.231 579 Silt 0.115 516 0.248 2232 0.242 958 Sand 0.111 516 0.201 1131 0.248 1709 表 3 基于野外光谱的土壤肥力指标PLSR模型估算精度
Table 3. Estimation accuracy of soil fertility indices with field spectra using PLSR
土壤肥力指标
Soil fertility index波段类型
Band type预处理方法
Pretreatment methodPC 建模集
Calibration sets验证集
Validation setsR2cv RMSEv R2val RMSEval RPD RPIQ TC 全波段 Raw 10 0.66 6.94 0.53 6.99 1.47 1.74 MSC 9 0.68 6.90 0.53 7.04 1.50 1.75 SG 1st 5 0.66 7.09 0.54 6.94 1.48 1.74 特征波段 Raw 4 0.68 6.94 0.55 6.85 1.50 1.76 MSC 4 0.70 6.53 0.57 6.70 1.54 1.78 SG 1st 3 0.76 5.88 0.67 5.85 1.76 1.91 OM 全波段 Raw 6 0.71 11.84 0.62 12.38 1.63 2.40 MSC 6 0.68 12.24 0.68 11.42 1.80 2.49 SG 1st 5 0.72 11.75 0.61 12.59 1.60 2.35 特征波段 Raw 4 0.71 11.60 0.65 11.90 1.70 2.43 MSC 3 0.72 11.49 0.68 11.34 1.78 2.50 SG 1st 3 0.78 10.19 0.69 11.09 1.82 2.53 TN 全波段 Raw 6 0.71 0.68 0.67 0.66 1.75 2.64 MSC 4 0.62 0.79 0.57 0.75 1.50 2.37 SG 1st 3 0.68 0.72 0.64 0.69 1.67 2.38 特征波段 Raw 4 0.69 0.72 0.59 0.73 1.58 2.30 MSC 4 0.72 0.67 0.67 0.66 1.76 2.45 SG 1st 3 0.75 0.63 0.76 0.56 2.04 2.98 AN 全波段 Raw 4 0.69 23.28 0.64 23.60 1.68 2.10 MSC 3 0.60 25.80 0.64 23.56 1.70 2.10 SG 1st 3 0.73 21.30 0.70 21.55 1.84 2.30 特征波段 Raw 4 0.71 22.31 0.68 22.18 1.78 2.23 MSC 4 0.65 24.12 0.72 20.68 1.91 2.39 SG 1st 1 0.75 20.64 0.72 20.94 1.89 2.36 pH 全波段 Raw 3 0.71 0.15 0.67 0.16 1.74 1.88 MSC 2 0.61 0.17 0.59 0.17 1.60 1.83 SG 1st 4 0.69 0.16 0.62 0.17 1.63 1.85 特征波段 Raw 2 0.73 0.15 0.68 0.15 1.79 1.91 MSC 1 0.63 0.17 0.60 0.17 1.60 1.84 SG 1st 3 0.73 0.14 0.71 0.14 1.89 2.03 Clay 全波段 Raw 7 0.27 1.41 0.05 1.54 1.04 1.19 MSC 5 0.32 1.35 0.33 1.32 1.23 1.32 SG 1st 4 0.27 1.41 0.32 1.31 1.22 1.33 特征波段 Raw 3 0.27 1.41 0.11 1.50 1.06 1.23 MSC 4 0.33 1.35 0.26 1.37 1.17 1.26 SG 1st 3 0.37 1.31 0.40 1.23 1.30 1.33 Slit 全波段 Raw 7 0.26 8.34 0.09 8.61 1.06 1.22 MSC 6 0.24 8.36 0.01 9.02 1.01 1.16 SG 1st 4 0.18 8.78 0.04 8.84 1.03 1.19 特征波段 Raw 4 0.26 8.27 0.19 8.14 1.12 1.20 MSC 5 0.28 8.18 0.16 8.30 1.10 1.27 SG 1st 3 0.33 7.90 0.13 8.42 1.08 1.15 Sand 全波段 Raw 6 0.32 8.87 0.22 8.77 1.14 1.25 MSC 5 0.29 9.06 0.22 8.77 1.14 1.26 SG 1st 3 0.29 9.03 0.25 8.61 1.16 1.29 特征波段 Raw 4 0.28 9.09 0.33 8.14 1.23 1.33 MSC 5 0.30 8.95 0.34 8.09 1.24 1.34 SG 1st 4 0.44 8.04 0.29 8.35 1.20 1.31 -
[1] 王 璐, 王海燕, 何丽鸿, 刘 鑫. 基于GIS的土壤肥力质量综合评价-以天然云冷杉针阔混交林为例[J]. 土壤通报, 2016, 47(5): 1223 − 1230. [2] 袁金华, 俄胜哲, 黄 涛, 车宗贤, 包兴国, 曾希柏. 水肥管理对带田土壤肥力和作物产量的影响[J]. 土壤通报, 2017, 48(2): 433 − 440. [3] 刘秀英, 王 力, 常庆瑞, 等. 基于相关分析和偏最小二乘回归的黄绵土土壤全氮和碱解氮含量的高光谱预测[J]. 应用生态学报, 2015, 26(7): 2107 − 2114. [4] 郭 燕, 纪文君, 吴宏海, 等. 基于野外Vis-NIR光谱的土壤有机质预测与制图[J]. 光谱学与光谱分析, 2013, 33(4): 1135 − 1140. [5] 齐海军, Karnieli A, 李绍稳. Y-梯度广义最小二乘加权校正的土壤碱解氮野外原位光谱预测[J]. 光谱学与光谱分析, 2018, (1): 171 − 175. [6] 齐海军. 土壤速效养分高光谱检测方法研究[D]. 安徽农业大学, 2018. [7] 纪文君. 基于野外vis-NIR高光谱的土壤属性预测[D]. 浙江大学, 2014. [8] 尤李俊, 李润杰, 王少丽, 等. 湟水流域农业面源污染现状及对策[J]. 农业工程学报, 2019, 9(5): 44 − 49. [9] Isaksson T, Naes T. The effect of multiplicative scatter correction and linearity improvement in Nir Spectroscopy[J]. Applied Spectroscopy, 1988, 42(7): 1273 − 1284. doi: 10.1366/0003702884429869 [10] 方 向, 金 秀, 朱娟娟, 等. 基于可见-近红外光谱预处理建模的土壤速效氮含量预测[J]. 浙江农业学报, 2019, 31(9): 1523 − 1530. [11] 李冠稳, 高小红, 肖能文, 等. 特征变量选择和回归方法相结合的土壤有机质含量估算[J]. 发光学报, 2019, (5): 361 − 371. [12] Zheng K Y. Studies on model optimization and model transfer methods of near infrared spectroscopy[D]. East China University of Science and Technology, 2013. [13] Chang C W, Laird D A, Mausbach M J, et al. Near-infrared reflectance spectroscopy principal components regression analyses of soil properties[J]. Soil Science Society of America Journal, 2001, 65(2): 480 − 490. doi: 10.2136/sssaj2001.652480x [14] Nawar S., Mouazen A. M Predictive performance of mobile vis-near infrared spectroscopy for key soil properties at different geographical scales by using spiking and datamining techniques[J]. Catena, 2017, 151: 118 − 129. doi: 10.1016/j.catena.2016.12.014 [15] 全国土壤普查办公室. 中国土壤普查数据[M]. 北京: 中国农业出版社, 1998.National Soil Survey Office. Soila of China[M]. Beijing: China Agriculture Press, 1998. [16] 马 创, 申广荣, 王紫君, 等. 不同粒径土壤的光谱特征差异分析[J]. 土壤通报, 2015, 46(2): 292 − 298. [17] Rossel R A V, Lark R M. Improved analysis and modeling of soil diffuse reflectance spectra using wavelets[J]. European Journal of Soil Science, 2009, 60(3): 453 − 464. doi: 10.1111/j.1365-2389.2009.01121.x [18] 王世芳, 程 旭, 宋海燕. 水分对土壤有机质检测影响的光谱特性分析及抗水分干扰模型建立[J]. 光谱学与光谱分析, 2016, 36(10): 3249 − 3253. [19] 张颖帝, 张佳宝, 李晓鹏. 基于高光谱的砂姜黑土含水量反演研究[J]. 土壤, 2017, 49(3): 630 − 634. [20] 鲁一冰, 刘文清, 张玉钧, 等. 一种自适应层进式Savitzky-Galay光谱滤波算法及其应用[J]. 光谱学与光谱分析, 2019, 39(9): 2657 − 2663. [21] 李鑫星, 梁步稳, 白雪冰, 等. 光谱技术在土壤水分含量检测中的研究进展[J]. 光谱学与光谱分析, 2020, 40(12): 3705 − 3710. [22] Clark R N, King T V, Klejwa M, et al. High spectral resolution reflectance spectroscopy of minerals[J]. Journal of Geophysical Research, 1990, 95(B8): 12653 − 12680. doi: 10.1029/JB095iB08p12653 [23] Kuang B, Mahmood H S, Quraishi M Z, et al. Chapter four-Sensing soil properties in the laboratory, in Situ, and on-line: A Review[M]. In Donald L S, editor, Advances in Agronomy, 2012(114): 155-223. [24] Waiter T H, Morgan C L S, Brown D J, et al. In situ characterization of soil clay content with visible near-infrared diffuse reflectance spectroscopy[J]. Soil Science Society of America Journal, 2007, 71(2): 389 − 396. doi: 10.2136/sssaj2006.0211 [25] Rossel V R. Robust modeling of soil diffuse reflectance spectra by “bagging-partial least squares regression”[J]. Journal of Near Infrared Spectroscopy, 2007, 15: 39 − 47. doi: 10.1255/jnirs.694 [26] Bowers S A, Hanks R J. Reflection of radiant energy from soils[J]. Soil Science, 1965, 100(2): 130 − 138. doi: 10.1097/00010694-196508000-00009 [27] Alabbas A H, Swain P H, Baumgardner M F. Relating organic matter and clay content to the multispectral radiance of soils[J]. Soil Science, 1972, 114(6): 477 − 485. doi: 10.1097/00010694-197212000-00011 [28] 张俊辉. 基于Vis-NIR光谱的不同水分状态下土壤质地预测[D]. 河南农业大学, 2016.