Hyperspectral predictors for monitoring biomass production in Mediterranean mountain grasslands: Majella National Park, Italy

Abstract

The research objective was to determine robust hyperspectral predictors for monitoring grass/herb biomass production on a yearly basis in the Majella National Park, Italy. HyMap images were acquired over the study area on 15 July 2004 and 4 July 2005. The robustness of vegetation indices and red‐edge positions (REPs) were assessed by: (i) comparing the consistency of the relationships between green grass/herb biomass and the spectral predictors for both years and (ii) assessing the predictive capabilities of linear regression models developed for 2004 in predicting the biomass of 2005 and vice versa. Frequently used normalized difference vegetation indices (NDVIs) computed from red (665–680 nm) and near‐infrared (NIR) bands, the modified soil adjusted vegetation index (MSAVI), the soil adjusted and atmospherically resistant vegetation index (SARVI) and the normalized difference water index (NDWI), were highly correlated with biomass (R 2⩾0.50) only for 2004 when the vegetation was in the early stages of senescence. Although high correlations (R 2⩾0.50) were observed for the NDVI involving far‐red bands at 725 and 786 nm for 2004 and 2005, the predictive regression model for each year produced a high prediction error for the biomass of the other year. Conversely, predictive models derived from REPs computed by the three‐point Lagrangian interpolation and linear extrapolation methods for 2004 yielded a lower prediction error for the biomass of 2005, and vice versa, indicating that these approaches are more robust than the NDVI. The results of this study are important for selecting hyperspectral predictors for monitoring annual changes in grass/herb biomass production in Mediterranean mountain ecosystems.