Resumen
Quantitative models for visible near-infrared ray spectroscopy have rarely been exploited for the measurement of soil-available potassium. These results show that the predictors of soil-available potassium exhibit different influences with 29 pretreatment methods and eight regression algorithms. We found that a combination of three methods, Savitzky?Golay, standard normal variate, and dislodge tendency, had better stability than other pretreatment methods. The boosting algorithms that form an ensemble of multiple weak predictors have better accuracy and stability than other regression algorithms. Therefore, a more robust and trustworthy visible near-infrared ray (VIS-NIR) model is proposed, which can be used across industries to quantify the soil-available potassium concentration.