Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
1712626 | Biosystems Engineering | 2006 | 8 Pages |
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 Fourier transform infrared attenuated total reflectance (FTIR-ATR) spectroscopy could be used to determine nitrate concentration in soil pastes. The present work further investigates this approach, and proposes to combine nitrate determination with soil identification in order to improve the determination accuracy. The study focuses on soils commonly used for agriculture, which are classified according to soil taxonomy and their carbonate and clay contents. Soil identification is investigated using the 800–1200 cm−1 and 1250–1550 cm−1 intervals of the spectrum, using either cross-correlation with a reference library or principal component analysis (PCA) decomposition followed by neural network (NN) classifier. When applied to the 1250–1550 cm−1 interval, the PCA-NN method leads to correct identification of all the samples, while the other approaches lead to poorer results. Nitrate determination is achieved using several partial least-squares regression models, each model being associated with a soil type. Determination errors range from 6·2 to 13·5 mg[N]/kg[dry soil], depending on the soil type, with the lowest errors for light sandy soils. These determination errors are appreciably smaller than those obtained using a single model calibrated using all the data (19·1 mg[N]/kg[dry soil]).