Prediction of SOC content using NIR spectroscopy and DSANN

Jakkan, Digambar Aggayya and Ghare, Pradnya and Kumar, Nirmal and Sakode, Chandrashekhar (2025) Prediction of SOC content using NIR spectroscopy and DSANN. AGROKÉMIA ÉS TALAJTAN, 74 (1). pp. 33-53. ISSN 0002-1873

Text 0088-article-p33.pdf - Published Version Restricted to Repository staff only Download (965kB)

Abstract

Soil organic carbon (SOC) levels directly affect the production and health of crops. Making use of a database of the usefulness of using the 350–2,500 nm Near Infra Red (NIR) spectroscopy data range on 200 soil samples from the Indian state of Uttar Pradesh was evaluated in this study. The more sophisticated Artificial Neural Network, to choose the spectral components that were used to forecast SOC, Random Forest (RF) and Ensemble Lasso-Ridge Regression (ELRR) were utilized. In the preprocessing, the inversion derivative, logarithmic(log) derivative, and logarithmic base to 10(log 10 x) derivatives were employed to duplicate the spectrum wavelength. The main characteristic of spectrum wavelength for SOC were found to be within the range of 350 and 450 nm, per the results. The best accurate estimation of SOC content was obtained by combining the suggested DSANN or Dropout Sequential ANN technique with the Log 10 x pre-processed data. The R-squared (R 2 ), RMSE, and RPIQ (Ratio of Performance in Inter Quartile Distance) values for the testing dataset were 0.83, 0.08, and 4.32, respectively.

Actions (login required)

Edit Item