Multi-Temporal and Multi-Platform Satellite-Based Mapping of White Sand Ecosystems

In the past few decades, attempts have been made to map and monitor disturbances affecting white-sand ecosystems in the Amazonian lowlands. Status and dynamics of the plant communities of these ecosystems are related to a variety of factors (soil type, flooding, nutrient availability, and wildfire), and full characterization of these unique ecosystems is lacking to this date. Hence, we have selected a study area that encompasses the lowlands lying between the Orinoco and the Rio Negro rivers to the west and the borders of the Guyana highlands to the east, between 5°N and 2°N, in the Amazonas state of Venezuela. The objective of the study was to obtain an accurate white-sand land-cover and land-use change (LC-LUC) map through a synergistic integration of Global PALSAR-2 and Sentinel-1 satellite products, with the Normalized Difference Vegetation Index (NDVI) derived from Landsat data. Special attention was given to the mapping of herbaceous formations (meadows) and flooded vegetation. An interpretation key was developed combining an improved Expectation-Maximization (EM) algorithm classification with post-classification refinement (Bayesian Information Criterion), as well as the integration of contextual spatial information and high-resolution imagery (Google Earth and Bing images). The supervised classification differentiates forest (closed to open), inundated forest, dry or flooded open white-sand scrublands, shrubby meadows, shrubby or herbaceous cover regularly flooded, mosaics of plant communities, sparse vegetation, bare rocky areas, and sandbanks/sandridges. The approach of combining two different microwave sensors is a sound one as an orbital sensor can acquire multi-frequency and multi-polarization data. The implementation of integrated image classification (L- and C-band with NDVI data) allows grouping spectral combinations of classes, offering an operational way for efficient mapping of complex units of the white-sand ecosystems. The results provide novel information not only on the canopy structure but also on the vegetation greenness.