Forest Structure Analysis in the Paraguayan Chaco Based on Sentinel-1, Sentinel-2 and GEDI Data

Tropical forest ecosystems have been identified as one of the most diverse regions in the world. Offering extensive ecosystem services such as climate regulation and rich biodiversity has raised increasing concerns about their future and protection. Latest studies conducted at a global scale have defined Argentina, Brazil, and Paraguay as the countries with the highest rates of deforestation in South America. With an area of about 240 000 km², the Paraguayan Chaco covers about one fourth of the Great American Chaco which spreads out over Argentina, Bolivia and Paraguay. The Paraguayan Chaco comprises not only a great variety of ecosystems such as savannahs, shrublands, grasslands and wetlands, but also holds the largest dry forest area on earth. Furthermore, the ecoregion has been acknowledged as an important carbon sink on a global scale. Nevertheless, uninterrupted deforestation activities between 1987 and 2012 resulted in the loss of 27 % of its original cover. The constant expansion of agricultural crops, cattle ranching, and illegal logging have severely fragmented the Paraguayan Chaco transforming the last forest remnants into isolated patches, jeopardizing not only their continuity but also the biodiversity comprised within. In this context, this study focuses on the one hand on environmental statistics of forest structure and on the other, forest structure characteristics are modelled for the complete Paraguayan Chaco. Forest structure characteristics are derived from the Global Ecosystem Dynamics Investigation (GEDI), which is the first full-waveform space-borne LiDAR, that is especially designed to accurately capture footprint samples of canopy height, total canopy cover, Plant-Area-Index and Foliage-Height-Diversity-Index. More than 8.9 million valid GEDI samples have been used to determine forest characteristics of the five recognized Ecoregions of the Paraguayan Chaco, which significantly differ in climatic conditions and therefore vegetation composition. Main results are, that the more humid Ecoregions Pantanal (east) and Humid Chaco (south-east) present higher canopy heights (mean heights of 10 to 12 m) and denser canopy covers (mean canopy cover density of 33 to 39 %), which are more complex in vertical-layering, compared to the rather arid Ecoregions Médanos (north-west), Cerrado (north) and Dry Chaco (west, center) with lower canopy heights (mean heights of 7 to 8 m) and rather sparse (mean canopy cover density of 15 to 23 %), single-layered canopies. To model the GEDI-derived forest structure samples for the complete Paraguayan Chaco, Sentinel-1 and -2 data is fused in a tile-based (0.5 degree tiles) random forest regression model. Highest modelled canopy heights are located in the north-east with maximum values of greater than 15 m, whereas lowest canopy heights outside agricultural fields are present in the north-west with heights of about 5 to 6 m. In the south-eastern region (Humid Chaco) of the Paraguayan Chaco, riparian forests are described by elevated canopy heights (up to 15 m) compared to the rather low surrounding vegetation of savannahs and grasslands. Spatial-correlations are identified between canopy height, canopy cover density and foliage-height-diversity, since areas of high canopy heights, are further characterized by dense vegetation and multiple vertical foliage, whereas lower vegetation is rather sparse and holds only single-layered canopies. Correlation statistics reveal that precipitation has moderate to strong positive correlations with modelled forest structure, and therefore mainly determines the presence of different vegetation structure compositions. Models for canopy height (median R²: 60.7 % to 64.0 %, median RMSE: 1.6 m to 2.0 m) and total canopy cover (median R²: 61.4 %, median RMSE: 9.1 %) are more accurate than models of Plant-Area-Index (median R²: 50.6 %, median RMSE: 0.3) and Foliage-Height-Diversity (median R²: 48.0 %, median RMSE: 0.3). By applying the presented multi-sensor statistics and modelling approach, the first highresolution maps (10 m) of forest structure for the Paraguayan Chaco have been generated. This novel knowledge should support environmental-sound land use and conservation in the Paraguayan Chaco, to meet the challenges of continuous expansion of the agricultural frontier, illegal logging activities, and the constant demand for natural goods, which threaten the continuity of the tropical forests. There is an urgent need to develop methods and approaches in the field of remote sensing observations to support the implementation of environmental laws and conservation programs orientated toward protecting the last remnants of natural forest on the continent. Countries with large, forested areas such as Paraguay should be taking advantage of the latest natural resources available, to halt and monitor deforestation activities in the country. With upcoming data from GEDI and fusion products of optical and LiDAR sensors, global forest structure data will improve estimates of above-ground biomass models to better quantify global carbon fluxes, further highlight drastic losses of forests and promote sustainable land use