Mapping of Leaf Nutrient Content Using Multispectral Imagery in Oil Palm Based on Unmanned Aerial Vehicles

Abstract

Adequate nutrition is one of the critical factors in determining the productivity of oil palm. Usually, to obtain sufficient nutrients for plants, fertilization is performed based on the results of soil and leaf nutrient content analyses. Measuring leaf nutrient content conventionally lacks flexibility, is impractical, is labour-intensive, and takes time and money, so remote sensing can be an alternative to addressing this problem. The study used the extraction of the reflectance values of each band as well as the transformation of the NDVI and GNDVI vegetation index from 3 bands of multispectral cameras (red, green, nears infrared) as an independent variable (predictor) to estimate leaf nutrient on oil palm plantations. A multiple polynomial regression model is built from a laboratory analysis of 35 examples of oil palm leaves used as a dependent variable. Predictive models of N, P, K, Ca, and Mg using multiple polynomial regression of order 4 resulted in R² values in the succession of 0,986; 0,975; 0,981; 0,970; and 0,968; Adjusted R² values in the sequence of 0,861; 0,761; 0,812; 0,710; and 0,690; RSE values consecutive of 0,065; 0,003; 0,076; 0,074; and 0,036; as well as MAPE values successive of 5,23; 3,22; 10,38; 13,40; and 16,59%. The predictive value of the leaf nutrients of each tree processed and classified spatially resulted in the content of N, P, and Ca being dominated by average criteria of respectively 95,51%, 100%, and 80,58% of the total tree, while K and Mg dominated the low criteria of 77.86% and 90.39%, respectively.