Vegetation Structure Modelling and Explorative Statistics based on Sentinel-1, Sentinel-2 and GEDI in the Paraguayan Chaco

Abstract

Understanding vegetation structures in forests is crucial to estimate species composition, assess habitat quality, and preserve forest resources to ensure carbon storage and climate buffering. Tropical forests are the most endangered forest areas on Earth due to massive deforestation and degradation processes, although holding the richest biodiversity due to its complex vegetation structures. The Paraguayan Chaco is a subtropical dry forest that has experienced a total loss in forest area of about 30 % since the 1980s, resulting in a heavily fragmented landscape. To better understand and characterize the forest areas being lost and to preserve remaining forested areas, the present study implemented a workflow to model vegetation structure characteristics based on complementary satellite remote sensing data sets of Sentinel-1 (synthetic aperture radar), Sentinel-2 (multispectral), and the Global Ecosystem Dynamics Investigation (GEDI, Light Detection and Ranging, LiDAR). Since the Paraguayan Chaco comprises a study area of about 240 000 km², the processing of multi-temporal metrics of Sentinel was conducted in the cloud-computing platform Google Earth Engine (GEE). GEDI attributes of vegetation structure, such as canopy height, canopy cover density, and vertical foliage complexity, served as modelling responses in a Random Forest Regression model trained with comprehensive spatio-temporal metrics derived from Sentinel of 2019. The integration of novel GEDI samples (GEDI sensor is operating since April 2019) allows for large-scale extrapolation of vegetation structure characteristics. Therefore, the first high-resolution maps (10 m) of canopy height, total canopy cover, Plant-Area-Index, and Foliage-Height-Diversity-Index for 2019 have been generated for the Paraguayan Chaco. In addition, comprehensive statistics have been carried out, to better understand spatial patterns of vegetation structure and correlations of various environmental variables with modelled vegetation structure attributes. Distinct differences in seasonality and the expansion of agricultural fields are key influences that shape the vegetation in the Paraguayan Chaco that ranges from dense and high dry forest (maximum canopy height and total canopy cover: 17.6 m, 78.1 %) to grasslands and savannahs (maximum canopy height and total canopy cover: 1.8 m, 10 %). The canopy height model reached highest accuracy (R²: 64.0 %), followed by total canopy cover (R²: 61.4 %), Plant-Area-Index (R²: 50.6 %), and Foliage-Height-Diversity-Index (R²: 48.0 %). Explorative statistics show longitudinal gradients of vegetation structure with elevated values in the eastern part that is characterized by higher precipitation rates. In addition, the modelled characteristics of vegetation structure reflect the characteristics of an Ecoregion classification: Ecoregions with more fertile soils, higher precipitation rates, and weaker seasonality (Humid Chaco, Pantanal; eastern part of the Paraguayan Chaco) present higher, denser, and more complex forest structures than the Dry Chaco and Médanos (western and central part of the Paraguayan Chaco) which are Ecoregions that are strongly influenced by changes in temperature and precipitation due to changes in seasonality. The provision of continuous information of vegetation structure and a deeper understanding of environmental drivers will allow for a more comprehensive assessment of forest resources to support strategies for environmental-sound land use and prioritization of conservation areas in order to halt continuing deforestation activities in the Paraguayan Chaco.