| Article ID | Journal | Published Year | Pages | File Type |
|---|---|---|---|---|
| 1712825 | Biosystems Engineering | 2006 | 12 Pages |
Fast, precise and affordable soil analytical techniques are needed for the determination of soil fertility of each zone of a field in site-specific land management. The objective of this study was to develop a reliable calibration model with the visible (VIS) and near infrared (NIR) spectroscopy method to predict the phosphorus (P) content in fresh soil as a preliminary step towards the development of an automated variable rate fertiliser distributor. The Zeiss Corona® 45 visnir fibre-type spectrophotometer was used to measure soil spectra in reflectance mode in the wavelength range of 305–1710 nm. A total of 185 soil samples were collected from arable fields and grassland distributed over Flanders (Belgium) amended with 29 and 14 samples taken from two fields near Leuven, (Belgium) to obtain a balanced P data set. Only samples with texture classes of silt loam, sandy loam and loamy sand texture were held, to avoid possible averse effects of small amounts of soil samples with different textures. Fresh samples were subject to chemical analyses and optical reflectance measurements. Soil spectra were correlated with available P using the partial least-squares (PLS) regression and full cross-validation. After removing the noise, the VIS-NIR wavelengths in the range 401–1663 nm were used. From the same data set, two suitable calibration models were developed using different data pre-processing methods. In the validation stage both models predicted available P in the soil with a coefficient of determination R2 of 0·75 and 0·73. On two extra date sets (126 and 36 samples, collected from two fields near Leuven), both models were evaluated. The prediction of P on these samples, provided acceptable results (R2=0·68 and 0·63). The two P models, developed in this research, can be used for rapid prediction of P in fresh soil samples in the laboratory and in situ in fields in Flanders having texture classes: loamy sand, sandy loam and silt loam. This detailed information could be employed to provide input to master a site-specific P management. These promising P models could lead to provide the necessary information for an automatic adaptation of the fertiliser rates of a P applicator.
