Retrieval of vertical LAI profiles over tropical rain forests using waveform lidar at La Selva, Costa Rica

This study explores the potential of waveform lidar in mapping the vertical and spatial distributions of leaf area index (LAI) over the tropical rain forest of La Selva Biological Station in Costa Rica. Vertical profiles of LAI were derived at 0.3 m height intervals from the Laser Vegetation Imaging Sensor (LVIS) data using the Geometric Optical and Radiative Transfer (GORT) model. Cumulative LAI profiles obtained from LVIS were validated with data from 55 ground to canopy vertical transects using a modular field tower to destructively sample all vegetation. Our results showed moderate agreement between lidar and field derived LAI (r2 = 0.42, RMSE = 1.91, bias = − 0.32), which further improved when differences between lidar and tower footprint scales (r2 = 0.50, RMSE = 1.79, bias = 0.27) and distance of field tower from lidar footprint center (r2 = 0.63, RMSE = 1.36, bias = 0.0) were accounted for. Next, we mapped the spatial distribution of total LAI across the landscape and analyzed LAI variations over different land cover types. Mean values of total LAI were 1.74, 5.20, 5.41 and 5.62 over open pasture, secondary forests, regeneration forests after selective-logging and old-growth forests respectively. Lastly, we evaluated the sensitivities of our LAI retrieval model to variations in canopy/ground reflectance ratio and to waveform noise such as induced by topographic slopes. We found for both, that the effects were not significant for moderate LAI values (about 4). However model derivations of LAI might be inaccurate in areas of high-slope and high LAI (about 8) if ground return energies are low. This research suggests that large footprint waveform lidar can provide accurate vertical LAI profile estimates that do not saturate even at the high LAI levels in tropical rain forests and may be a useful tool for understanding the light transmittance within these canopies.

► Deriving LAI profiles of tropical rain forest from lidar using a physical model.
► Good agreement with destructive sampling data (r2 = 0.63, RMSE = 1.36, bias = 0.0).
► Analysis suggests major errors come from inaccurate parameters, noise and slope.