Delta Area at Near Infrared Region (${\text{DA}}_{{\text{NIR}}}$)—A Novel Approach for Green Vegetation Fraction Estimation using Field Hyperspectral Data

Abstract

A new metric called “Delta Area at Near Infrared Region” ( ${\text{DA}}_{{\text{NIR}}}$ ) has been conceptualized and implemented for estimation of green vegetation fraction (GVF) using field spectroradiometer at different growth stages of potato over two consecutive potato growing seasons (2012–2013 and 2013–2014). Vertical photograph, collocated in time and space with spectroradiometer observation, was acquired and digitally classified for GVF. While comparing with other conventional indices, ${\text{DA}}_{{\text{NIR}}}$ showed linearity at higher GVF values and capable of capturing the movement of the curve caused by soil-vegetation mixture between inflection point and near infrared peak. Among all the univariate models, ${\text{DA}}_{{\text{NIR}}}$ showed the highest accuracy with ${\text{R}}^{2}=0.94$ and ${\text{RMSE}}=7.0$ . The new index also exhibited the highest sensitivity for the entire range of GVF while comparing with other indices; however, the sensitivity decreases at higher values especially above 70%. Stepwise multiple linear regression (SMLR) and partial least squares regression (PLSR) were performed using all the spectral variables. The prediction accuracy was further improved over univariate analysis wherein PLSR was able to predict the vegetation fraction with the highest accuracy ( ${\text{R}}^{2}=0.94$ and ${\text{RMSE}}=5.32$ ). In both SMLR and PLSR, ${\text{DA}}_{{\text{NIR}}}$ contributed significantly to improve the estimation accuracy. These findings suggest that ${\text{DA}}_{{\text{NIR}}}$ can be used as a surrogate indicator of GVF independently or in combination with other vegetation indices to further improve the estimation accuracy.