|M.Sc Student||Weiner Michal|
|Subject||Reduction of Soil Interference in the Direct Determination|
of Nitrate by ATR-FTIR Spectroscopy
|Department||Department of Agricultural Engineering||Supervisors||Professor Raphael Linker|
|Professor Emeritus Abraham Shaviv|
Direct determination of nitrate in soil is required for improving N-application management, which would help reduce soil and water pollution.
Several works have demonstrated that mid-infrared FTIR-ATR (Fourier transform infrared attenuated total reflectance) spectroscopy could be used to determine nitrate concentration in soil extracts and soil pastes at high nitrate concentrations. It was also reported that the main obstacle to accurate determination of nitrate in soil pastes is a strong interfering band around 1450cm-1, which was attributed to soil carbonate components.
The main objective of this study is to investigate the possibility of using soil identification as a tool for reducing errors of direct determination of nitrate concentration in soil pastes.
The present work included five soil types: Hamra, Grumosol, Loess, Terra Rossa and Rendzina commonly used for agriculture, which are divided into six groups according to soil taxonomy and their carbonate and clay contents.
The data processing consisted of a two stage method (1) soil classification - determination of the soil type by comparing a specific region in the spectrum to a reference spectral library, and (2) nitrate prediction - determination of the nitrate concentration using the model corresponding to the identified soil type.
Principal component analysis (PCA) decomposition followed by a classifier based on a neural network or Mahalnobis distance was used for classification.
Nitrate determination was made by integration, principal component regression (PCR) and partial least squares (PLS).
Applying PCA to the 1250-1550cm-1 interval and using Mahalnobis distance led to correct classification of 84% of the samples.
Applying PCA to two intervals that contain information about the soil type and carbonate content namely (950-1150cm-1 and 1250-1550cm-1) and a neural network classifier led to the correct classifications of 99.6% of the samples.
Nitrate concentration best results were obtained using partial least squares (PLS). All soil groups showed lower determination errors when soil-type specific models were used rather that a straightforward PLS analysis that used all the soils without group separation.
Determination errors range from 5 to 18 mg[N]/kg[dry soil], depending on the soil type, with the lowest errors for light sandy soils.
It can be concluded that ATR-FTIR spectroscopy in the mid infrared region can be used for soil classification, and that following soil identification, nitrate prediction can be achieved using model customized for each soil type. This allows for the effects of each soil specific interferences to be accounted for and leads to lower determination errors.