Evaluation of Pretreatment Methods for Prediction of Soil Micronutrients from Hyperspectral Data

Abstract

An assessment of soil quality is vital for monitoring of crop growth and agricultural practices along with its management. Moreover, soil quality is essential to fulfill the requirements of agricultural as well natural resource planning. However, the conventional methods do not suffice to identify the Soil Macro Nutrients (SMN) which is useful for soil quality evaluation. Recently, visible near infrared (VNIR) reflectance spectroscopy is widely acceptable technology for detecting and estimating the soil attributes in effective and rapid manner. Nevertheless, the acquired reflectance spectra by spectroscopy are affected by sensor error or illumination errors. Though, the affected errors can be diminished by the VNIR pretreatment methods. In this study, efforts made to identify the SMN from VNIR spectroscopy. The important data has been extracted by using data mining techniques and algorithm such as the various pretreatment methods: Standard Normal Variate (SNV), First Derivative (FD) and Maximum Normalization Continuum Removal (MNCR) were computed for obtaining pure spectra. The Partial Least Squares Regression (PLSR) algorithm was used for estimating the SMN from thirty soil samples collected from agricultural sectors. The experimental results depict that, the SMN was identified and estimated better after implementing the said pre treatment methods on VNIR spectra. The R2 value was 0.87 for raw spectra and it was 0.93, 0.95 and 0.94 for SNV, FD and MNCR respectively. Whereas, Root mean square error (RMSE) was 0.037, 0.006, 0.049 and 0.028 for raw spectra, SNV, FD and MNCR spectra respectively. In conclusion, the FD method provided betters results than other tested methods. The present research is beneficial for farmers and decision makers to detect and determine SMN from soil samples in better way.