GEDI is providing billions of three-dimensional forest structure measurements (points) across the tropics and mid-latitudes. However, some ecologists prefer to work with continuous (i.e. “gap-free”) maps of forest structure for their analyses (e.g. species distribution models). We have developed several methods to create continuous raster maps of GEDI forest structure metrics, like canopy height, plant area index, and canopy cover. The first method uses statistical aggregation (mean, median, standard deviation, interquartile range, and 95th percentile) of all GEDI shots within multi-resolution rasters ranging from 1 km to 25 km spatial resolution. Furthermore, we create multi-temporal maps associated with these time periods: 2019, 2020, 2021, and 2019 to 2022. The statistically-aggregated maps are publicly available for the entire GEDI observation domain. The second method provides a means to generate finer scale maps of GEDI forest structure metrics. We use the continuous change detection and classification algorithm (CCDC) to fit harmonic models to Landsat spectral bands and indices. The harmonic model coefficients, as well as synthetic spectral/index values, are associated with GEDI shots from a region and then used to train a Random Forest model. The Random Forest model is predicted across the entire region and evaluated using held-out GEDI data. The model predictions can either be made during the period of GEDI observation or hindcasted to previous years that overlap the Landsat record. This method is advantageous when an application requires finer spatial resolution maps, especially in regions where GEDI coverage is limited due to cloud cover or ISS orbital patterns. We will showcase multi-temporal maps from three regions: Southeast Asia, the Tropical Andes, and the U.S. Southwest. Furthermore, we will summarize the accuracy of the maps and discuss limitations associated with using two-dimensional multi-spectral imagery to predict three-dimensional forest structure metrics.
