@InProceedings{CsillikKLBBCCDFGJOVV:2022:AmFoSt,
               author = "Csillik, Ovidiu and Keller, Michael and Longo, Marcos and Bonal, 
                         Damien and Burban, Benoit and Chave, J{\'e}rome and Coomes, David 
                         A. and Derroire, Geraldine and Feldpausch, Ted and G{\"o}rgens, 
                         Eric Bastos and Jackson, Tobias and Ometto, Jean Pierre Henry 
                         Balbaud and Valdivia, Mar{\'{\i}}a Isabel Vilalba and Vincente, 
                         Gr{\'e}goire",
          affiliation = "{Jet Propulsion Laboratory} and {NASA Jet Propulsion Laboratory} 
                         and {Lawrence Berkeley National Laboratory} and {Universit{\'e} 
                         de Lorraine} and INRAE and CNRS and {University of Cambridge} and 
                         CIRAD and {University of Exeter} and {Universidade Federal dos 
                         Vales do Jequitinhonha e Mucuri} and {University of Cambridge} and 
                         {Instituto Nacional de Pesquisas Espaciais (INPE)} and 
                         {Jard{\'{\i}}n Bot{\'a}nico de Missouri} and {University of 
                         Montpellier}",
                title = "Amazon forest structural diversity estimated using field inventory 
                         plots, airborne lidar and GEDI spaceborne lidar",
                 year = "2022",
         organization = "AGU Fall Meeting",
            publisher = "AGU",
             abstract = "Tropical forests store more than half of terrestrial aboveground 
                         biomass and are key to understanding the global carbon cycle. 
                         Knowing the structure of a forest today is critical to predicting 
                         its future structure and productivity. Unfortunately, field 
                         inventories of tropical forests are sparse and vast areas have no 
                         representative inventory information. Reliance on sparse field 
                         measurements can lead to large uncertainties when extrapolating to 
                         vast regions such as the Amazon forest. The advent of airborne and 
                         spaceborne lidar expands the coverage from local to regional and 
                         continental measurements of forest structural properties. We 
                         produced Amazon-wide estimates of forest structure (aboveground 
                         biomass, basal area, leaf area index, and stem number density) by 
                         integrating information from forest inventory plots, airborne 
                         lidar and GEDI (Global Ecosystem Dynamics Investigation) 
                         spaceborne lidar using a two-step approach. The first step relates 
                         forest properties from about one thousand forest inventory plots 
                         and relative heights from simulated GEDI metrics based on airborne 
                         lidar overlapping the plots. We compared several regression models 
                         and machine learning approaches, with best results obtained by 
                         Ordinary Least Squares and Random Forests that reached R2 values 
                         from 0.69 for stem density (relative RMSE 22%) to 0.79 for basal 
                         area (20%), 0.80 for aboveground biomass (24%) and 0.83 for leaf 
                         area index (19%). Second, we used 11,280 contemporary co-located 
                         GEDI-simulated shots from airborne lidar with spaceborne GEDI 
                         shots to build linear models between their corresponding relative 
                         heights. Ultimately, we combined the two models to upscale the 
                         measurements from field inventory to GEDI data over the Amazon, 
                         consisting of more than 250 million quality-filtered shots 
                         acquired over 2.5 years. We analyzed environmental and climatic 
                         factors that influence regional and local variations in forest 
                         structure. Preliminary results indicate that the dry season length 
                         was a strong predictor of major gradients of stem number density 
                         across the Amazon basin. These results fill a gap where systematic 
                         forest inventories are lacking and can be used to study a wide 
                         range of ecosystem processes and model predictions of forest 
                         carbon budget dynamics.",
  conference-location = "Chicago, IL",
      conference-year = "12-16 Dec. 2022",
        urlaccessdate = "18 jun. 2024"
}