Effective spectral pre-processing methods enhance accuracy of soil property prediction by NIR spectroscopy
Abstract Near-infrared (NIR) spectroscopy is a fast, non-invasive, and effective technique that has gained widespread use in soil analysis. Pre-processing plays an essential role in enhancing the precision of calibrating NIR spectra with laboratory-measured soil properties. This research assessed th...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-08-01
|
| Series: | Discover Applied Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s42452-025-07580-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Near-infrared (NIR) spectroscopy is a fast, non-invasive, and effective technique that has gained widespread use in soil analysis. Pre-processing plays an essential role in enhancing the precision of calibrating NIR spectra with laboratory-measured soil properties. This research assessed the efficacy of two two-band index transformations—simple ratio indices (SRI) and normalized difference indices (NDI)—in addition to four distinct three-band index transformations (TBI) for predicting various soil characteristics using NIR spectroscopy in a laboratory setting. A total of 333 soil samples were sourced from different farms across Northern Germany, analyzed using two NIR spectrometers, and their properties were measured in a certified lab. Several feature selection approaches, including recursive feature elimination (RFE) and the least absolute shrinkage and selection operator (LASSO), were employed to identify the most significant wavebands. Calibration models were developed using partial least squares regression (PLSR) and LASSO regression. The results indicated that index transformations considerably enhanced the predictive performance of the models. Model performance was assessed through several metrics, including the coefficient of determination (R2), root mean square error (RMSE), and the ratio of performance to deviation (RPD). Compared to unprocessed data, R2 values improved by up to 0.13, 0.30, and 0.23 for organic matter (OM), pH, and phosphorus (P2O5), respectively. The optimal models for estimating OM (R2=0.59, RMSE = 1.61%, RPD = 1.79), pH (R2=0.63, RMSE = 0.28, RPD = 1.73), and P2O5 (R2=0.46, RMSE = 16.1 mg/100 g, RPD = 1.46) were attributed to TBI transformations on selected wavebands, calibrated using PLSR. These findings highlight that NIR spectroscopy, even with a limited spectral range (950–1650 nm), can provide reliable estimates of soil properties when combined with suitable pre-processing methods. |
|---|---|
| ISSN: | 3004-9261 |