Determining Uncertainty Prediction Map of Copper Concentration in Pasture from Hyperspectral Data Using Qunatile Regression Forest

Hyperspectral data has high potential to predict the biochemical components of grass with high accuracy, although the accuracy can be significantly improved when hyperspectral data is combined with environmental and topographical data. In this study, a fixed wing airborne survey was conducted using a AisaFENIX hyperspectral imager on a hill country sheep and beef farm. Pasture samples were collected across the farm to determine the copper concentration. After processing the hyperspectral imagery, the data was combined with environmental and topographical data to produce spatial prediction maps with associated uncertainties (95% prediction interval) using a new approach called Quantile Regression Forest (QRF). The results from this study suggest that QRF could provide more accurate and uncertain maps of pasture chemical properties.