Understanding the Spatio-Temporal Behavior of Sentinel-1 SAR Vegetation Indices over the Brazilian Savanna

• Published in: IEEE transactions on geoscience and remote sensing 2024-01, Vol.62, p.1-1
• Main Authors: Bussinguer, Jales de Freitas, Baptista, Gustavo Macedo de Mello, Sano, Edson Eyji, Leal, Fabricio Assis
• Format: Article
• Language: English
• Subjects: C-band; Cerrado; Grasslands; Indexes; Meters; National parks; Precipitation; Radar; Rainfall; Rainfall regime; Rejection rate; SAR (radar); Savannahs; Scattering; Seasonal variation; Seasonal variations; Seasonality; Seasons; Spatial distribution; stochastic distances; Synthetic aperture radar; time series; Vegetation; Vegetation index; Vegetation mapping
• ISSN: 0196-2892; 1558-0644
• DOI: 10.1109/TGRS.2024.3381468

The behavior of five SAR vegetation indices, namely Dual-pol Radar Vegetation Index (DpRVI), Radar Vegetation Index dual-pol (RVI), Polarimetric Radar Vegetation Index (PRVI), Dual-pol SAR Vegetation Index (DPSVI), and Modified Dual-pol SAR Vegetation Index (DPSVIm) was investigated. The indices were derived from Sentinel-1 data between 2017 and 2021 for three vegetation classes (Forest, Savanna, and Grassland) in the Brasilia National Park, Brazil. Temporal profile analysis showed that all indices followed a seasonal pattern directly linked with rainfall regime in the study site. DpRVI and RVI were related to the rate of radar signal variation over the vegetation as a function of seasonality. PRVI, DPSVI, and DPSVIm were directly related to the classes' biomass level and its seasonal variations over time. Precipitation and seasonal vegetation structure changes were relevant drivers of the shift and spread of the indices' spatial distributions over time. A phenomenon called index equalization effect was observed mainly on rainy dates, shifting and spreading the classes' distributions to same value ranges and degrading its separability. From a populational perspective, all indices presented associative intersections between all vegetation classes, despite the influence of seasonality. From a local perspective, DpRVI and RVI had disjoint intersections; PRVI maintained the associative relationships; and DPSVI and DPSVIm provided almost a complete separation between Forest and Grassland in all seasons. In both perspectives, DPSVIm outperformed the other indices in separating the classes by presenting the less than 72.15% average overlaps between all populational distributions and the highest H0 rejection rates, above 97% over time.