Selection of characteristic wavebands to minimize soil moisture effects with in-situ soil spectroscopy

摘要

Visible and near-infrared spectroscopy (VNIRS) has been used to supply fast, nondestructive and environmentally-friendly estimation ofprofile soil properties, but most of the previous research has been limited to VNIRS spectroscopy of air-dried and sieved soil. The need to air-dry and sieve the soil slows the estimation process and increases labor requirements, so real-time, in-situ measurements of moist soil would be preferable. Soil moisture is recognized as a major factor reducing the utility andaccuracy of soil property estimation by real-time VNIRS spectroscopy. Therefore, the objective of this paper is to explore pretreatment methods ofaoil spectra to improve the estimation of soil total nitrogen (TN) based on real-time, in-situ measurementswithout the need for artificial drying and sieving of soil cores. A commercial soil profile sensing instrument, the Veris P4000, was used to acquire real-time VNIRS data to a Im depth in 22 fields across Missouri and Indiana, USA. Simultaneously, soil cores were obtained and TN content was measured in the laboratory with standard methods. The derivative method was used to reduce the interference of soil moisture. The correlation coefficients between VNIRS spectroscopy and TN content demonstrated that the derivative method was effective at decreasing the effect of soil moisture at some moisture levels. Seven sensitive wavebands were selected using the particle swarm optimization algorithm. Finally, two modelling approaches were compared. Partial least squares regression models were calibrated based on the full spectrum, and back-propagation neural network models were calibrated based on sensitive wavebands. Both approaches reduced the effect of soil moisture, suggesting an improvement in prediction accuracy could be achieved. These results demonstrate that other soil moisture reduction approaches should be. evaluated to further improve the prediction accuracy of VNIRS.