Uncertainty Model for NDVI Estimation from Multispectral Camera Measurements

To accurately measure vegetation, soil, and environmental feature parameters in Precision Agriculture (PA), it is important to consider the quantity of uncertainty associated with these measurements. Assessing the uncertainty of vegetation indices provides information regarding the level of confidence of the classification usually done for determining the health status of vegetation. This paper proposes an uncertainty model for the Normalized Difference Vegetation Index (NDVI) estimation, which represents a key indicator used in PA for assessing plant health. The proposed model considers the limited bandwidth of the sensors embedded in a multi-spectral camera and the variability in optical density within their nominal wavelengths, specifically in the Red and Near Infrared bands, during the process of image acquisition. This variability is an uncertainty source in the NDVI measurements. The uncertainty model is applied to a dry and fresh Douglas fir leaves dataset [1]. The obtained uncertainty values fall within the range of 0.03 to 0.09 for both dry and fresh.