Acta Optica Sinica, Volume. 39, Issue 9, 0930002(2019)

Estimation of Soil Organic Matter Content Based on Characteristic Variable Selection and Regression Methods

Guanwen Li1,2、**, Xiaohong Gao、*, Nengwen Xiao2, and Yunfei Xiao1
Author Affiliations
  • 1 Qinghai Provincial Key Laboratory of Physical Geography and Environmental Process, School of Geography Sciences, Qinghai Normal University, Xining, Qinghai 810008, China
  • 2 Chinese Research Academy of Environmental Sciences, Beijing 100012, China
  • show less
    Figures & Tables(16)
    Location of the study area and distribution of soil sampling sites
    Spectral reflectance curves of soil samples. (a) Raw spectra; (b) spectra after MSC-MF-1st Derivative pre-processing
    Variable selection process by sCARS. (a) Changing trend of variables; (b) 10-fold RMSECV values; (c) regression coefficients of variables
    Characteristic variable selection process by GA
    Characteristic variable selection process by sCARS-SPA from the pre-processing spectrum. (a) Number of variables in the model; (b) variable index
    Distribution of characteristic variables with different variable selection methods
    Scatter plot for the measured and predicted value by sCARS-PLSR model
    Scatter plot for the measured and predicted value by SPA-SVM model
    Scatter plot for the measured and predicted value by IRIV-RF model
    Results of PLSR, SVM and RF models with different variable selection methods
    Scatter plots for the measured and predicted value by sCARS-RF model before and after artificially eliminating outliers. (a) Contain outliers; (b) eliminate outliers
    • Table 1. Soil organic matter content statistics of calibration set and validation set

      View table

      Table 1. Soil organic matter content statistics of calibration set and validation set

      Sample setSamplenumberMin /(g·kg-1)Max /(g·kg-1)Mean /(g·kg-1)SD
      Calibration set2684.86148.7432.4723.52
      Validation set1338.26133.5632.1622.44
    • Table 2. Accuracies of PLSR model with different variable selection methods

      View table

      Table 2. Accuracies of PLSR model with different variable selection methods

      Selection methodVariable numberPCCalibration setValidation set
      Rcal2RMSECALR2valRMSEVALRPD
      Full-spectrum200050.8429.3260.8359.0692.5
      sCARS5150.8748.3270.8837.7972.9
      SPA550.8509.1030.8588.5252.6
      GA18640.8429.3420.8618.4152.7
      IRIV6360.8439.3000.8758.0432.8
      sCARS-SPA1740.76511.3910.8488.7912.6
    • Table 3. Accuracies of SVM model with different variable selection methods

      View table

      Table 3. Accuracies of SVM model with different variable selection methods

      Selection methodVariable numberOptimal parameterCalibration setValidation set
      g (nuclear function)c (punishment coefficient)Rcal2RMSECALRval2RMSEVALRPD
      Full-spectrum20000.0363.0310.917.2210.7411.5461.9
      sCARS510.0213.0310.8818.1160.8777.9182.8
      SPA50.0041.7410.8588.8550.8897.4773.0
      GA1860.0121.7410.8678.5770.8718.0932.8
      IRIV630.0213.0310.8698.4930.8648.3072.7
      sCARS-SPA170.0112.8580.8778.2460.8738.0522.8
    • Table 4. Accuracies of RF model with different variable selection methods

      View table

      Table 4. Accuracies of RF model with different variable selection methods

      Selection methodVariable numberCalibration setValidation set
      Rcal2RMSECALRval2RMSEVALRPD
      Full-spectrum20000.9425.8170.9574.8404.6
      sCARS510.9425.7810.9584.7804.7
      SPA50.9306.5850.9545.0824.4
      GA1860.9395.8940.9594.6994.8
      IRIV630.9415.9270.9604.6564.8
      sCARS-SPA170.9405.9100.9554.9714.5
    • Table 5. Model accuracy after manually eliminating outliers

      View table

      Table 5. Model accuracy after manually eliminating outliers

      ModelCalibration setValidation set
      Rcal2RMSECALRval2RMSEVALRPD
      sCARS-PLSR0.9435.5380.9265.9873.7
      sCARS-SVM0.9266.0920.9574.9034.6
      sCARS-RF0.9853.2040.9882.8657.8
    Tools

    Get Citation

    Copy Citation Text

    Guanwen Li, Xiaohong Gao, Nengwen Xiao, Yunfei Xiao. Estimation of Soil Organic Matter Content Based on Characteristic Variable Selection and Regression Methods[J]. Acta Optica Sinica, 2019, 39(9): 0930002

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Spectroscopy

    Received: Mar. 5, 2019

    Accepted: May. 5, 2019

    Published Online: Sep. 9, 2019

    The Author Email: Guanwen Li (lgw126522@163.com), Xiaohong Gao (xiaohonggao226@163.com)

    DOI:10.3788/AOS201939.0930002

    Topics