Hyperspectral estimation models for total potassium content of Kuerle fragrant pear leaves

The conventional analysis of nutrient elements required destructive sampling, highly complex processes, highly time consuming and difficult nutrition diagnosis process in fruit trees. However, hyperspectral remote sensing technology has been reported to resolve the problems of destructive sampling and rapidly diagnose nutrient elements of plant. To monitor the state of nutrients in Korla fragrant pear in a non-destructive, timely and quick manner, an SVC HR-768 portable spectrometer was used to measure the spectral reflectance of leaves in the field of 20-year Korla fragrant pear tree under different K fertilization rates. The total K content was analyzed in the lab, and the relationships between total K content of leaves and original spectrum, first derivative spectrum, high spectral parameters established. The results showed that a single linear model built at 425 nm between total potassium content and original spectrum significantly described the relationship, with an adjusted determination coefficient (R²) of 0.913. Another linear model built at 630 nm between total potassium content and the first order derivative spectrum was similarly significant, with an adjusted R² of 0.986. The relationships bewteen total potassium content of leaf and green peak position (Rg), red valley position (Ro) were extremely significant in selecting hyperspectral feature variables. Results also showed that the adjusted R² was above 0.96 for all the built linear models. After evaluation of all the built models, the model Y = 1 136.835X₆₃₀ + 50.709 (X₆₃₀ is the first derivative spectrum at 630 nm) was the best for predicting total potassium content (Y) of Korla fragrant pear leaf.