New approach for rapid estimation of leaf nitrogen, phosphorus, and potassium contents in apple-trees using Vis/NIR spectroscopy based on wavelength selection coupled with machine learning.

Timely and rapid monitoring of apple trees nutrition status is vital for accurate management of nutrient fertilizers in order to improve the yield and quality, as well as to reduce the risk of environmental degradation. As a most frequently used method, tissue analysis used to assess the nutritional status of apple tree leaves is laborious, costly, time-consuming, environmentally unfriendly, and destructive. Ground-based sensors are able to effi-ciently provide information on nutritional status using leaf spectra reflectance. This research aims to establish a novel cost-effective and non-destructive approach for rapidly estimating the status of nitrogen (N), phosphorus (P), and potassium (K) in apple tree leaves based on Visible/Near-infrared (Vis/NIR) spectroscopy (500-1000 nm) coupled with machine learning. The Vis/NIR spectra of apple trees' leaves were acquired. Then, leaf chemical contents of NPK elements were considered as reference points. Different pre-processing techniques were used to pre-treat the spectra. Four different chemometrics analysis consist of support vector machine (SVM), Artificial neural network (ANN), Random Forest (RF) and partial least square (PLS) were applied to predict NPK contents in comparison to actual values. In order to simplify the models, the sensitive wavelengths were extracted using three wavelength selection approaches, variable importance in projection (VIP), partial least squares (PLS), and random frog (Rfrog). The extracted feature wavelengths from PLS, VIP, and Rfrog methods were widely distributed in the visible and near-infrared regions of the spectrum. The performance of the developed models was tested using Residual Prediction Deviation (RPD) and the Ratio of Performance to Interquartile Distance (RPIQ). The results demonstrated that among all models, the non-linear modeling methods were superior to the linear model. The best results for estimation of Nitrogen (N), Phosphorus (P), and Potassium (K) elements were achieved by the models of MSC + D2-Rfrog-RF (rp = 0.985, RMSEP = 0.029%, RPD = 8.77, RPIQ = 7.72), SNV + D2-Rfrog-RF (rp = 0.977, RMSEP = 0.0053%, RPD = 6.42, RPIQ = 5.09) and SNV + D2- Rfrog-RF (rp = 0.978, RMSEP = 0.018%, RPD = 8.16, RPIQ = 7.01), respectively. The findings of the current approach may provide an efficient approach to predict in-situ NPK contents of apple trees based on leaf spectral reflectance.