Exploring the potential of GEDI fusion with multi-sensor imagery for mapping the canopy height of diverse tropical forests of Colombia

Abstract:

The integration of GEDI measurements with multispectral and SAR datasets opens new possibilities for mapping forest structural variables across large areas. However, the accuracy of these maps could vary among tropical forest types due to differences in forest structure complexity, atmospheric properties, topography, and other factors. Ongoing efforts, in conjunction with additional GEDI data releases and version updates, are focused on understanding the impact of different ground-finding algorithms and model formulations on canopy height model error. We created the first GEDI-derived canopy height map for Colombia at high spatial resolution (30m) by integrating the relative height at the 98^th percentile of returned energy (RH98) of GEDI footprints as ground truth measures for canopy height and annual metrics of Landsat, PALSAR, and Sentinel-1 as predictors to model continuous maps (Figure 1a). Likewise, we created maps of canopy height for each of the five natural regions of Colombia using the same methodology to evaluate the differences in the error estimation among the national map and the regional maps. The five regions of Colombia present different types of forest and environmental conditions (Figure 1b). Five-fold cross validations from 5,000 to 50,000 training data selected randomly presented significant differences for the RMSE and MAE among the regional maps. The regions dominated by moist forest, Chocó and Amazonas, showed the highest errors (p < 0, 001). The regions dominated by dry forest and savannas, Caribe and Orinoquía presented the lowest errors (p < 0, 01). Our results show that the integration of regional models of canopy height could be an appropriate alternative for creating national maps or maps for other large areas that encompass heterogeneous regions.