Potassium estimation of cotton leaves based on hyperspectral reflectance

Potassium estimation on plant leaves can help monitor metabolic processes and plant health. The detailed study of hyperspectral data on leaves can therefore be a strong ally in the nutritional diagnosis of plants and can be applied to an on the go systems for precision farming application. In this study, reflectance spectra of cotton leaves were analysed for an assessment of potassium deficiency in cotton plants (Gossypium hirsutum L. ‘BRS 293’). The crop was planted in a greenhouse in the experimental area of the Federal University of Ceara (UFC), Fortaleza, Brazil.  Irrigated cotton plants were submitted to four different doses of potassium with twenty replications (n= 80) over 119 days. The following treatments were applied: 50%, 75%, 100% and 125% of the recommended potassium dose. Hyperspectral reflectance spectra data were collected using a Fieldspec ProFR 3 during full flowering, the phenological stage most demanding of potassium. Multivariate statistical techniques were applied to the raw data, the transformed data by derivative analysis, and by the technique of continuum removal. Band selection was carried out by the stepwise method in order to fit a PLSR model focused on identifying bands that are most sensitive to variations in potassium leaf concentrations. Model performance was evaluated by adjusted correlation coefficients – R²adj, root mean square error - RMSE, and residual prediction deviation - RPD. Validation results indicated that the PLSR model accounted for 82.0% of the variation in leaf potassium concentration, with a RMSE of 3.74 and RPD of 1.61. Therefore, the discrimination of potassium deficiencies in cotton using hyperspectral data was satisfactorily performed by a PLSR model composed of 13 wavelengths, of which most are commonly associated with moisture, lignin, cellulose, sugar and protein concentrations in cotton leaves.