Spectral sensitivity of vegetation indices: GNDVI, NDVI and EVI in the riparian forest of the Serrinha II Reservoir – PE, Brazil

Abstract

Riparian forests are forests associated with streams, rivers and lakes that have an influence on the protection of ecosystems from man-made actions. The objective of this study was to identify changes in the Caatinga riparian forest areas, based on the spectral sensitivity of vegetation indices (IVs): GNDVI, NDVI and EVI, considering a temporal space analysis (1985 to 2019) in the surroundings of the Serrinha II Reservoir - PE - Brazil. For this purpose, satellite images from TM - Landsat 5, OLI - Landsat 8 and MSI - Sentinel 2 were used. The IVs were validated comparing spectral profiles and correlation of the products generated by Landsat 8 and Sentinel 2. Through the binary classification of the indices, changes in vegetation cover were identified between 1996 and 2019, resulting in a loss of 55.40% of vegetation. These events, associated with drought effects were also observed for the reservoir, between 2012 and 2019, with 32.36% reduction in the water mirror. Among the IVs, the GNDVI and NDVI showed greater sensitivity to identify the riparian forest with a Pearson coefficient equal to 0.93 for Landsat 8; 0.94 for Sentinel 2 and an overall accuracy of over 95%. On the other hand, EVI was more sensitive to identify the reservoir area, with overall accuracy of 99.48%. The monitoring of artificial reservoirs and riparian forests of seasonally dry forests such as the Caatinga benefit from the use of IVs where the choice of a given isolated IV should be interpreted with caution according to the characteristics of the region studied.