Mapping tropical dry forest age using airborne waveform LiDAR and hyperspectral metrics

摘要

Tropical dry forest (TDF) regeneration has been extensively characterized as three deterministic successional stages, i.e., early, intermediate, and late, for the past few decades. This deterministic definition, however, ignores many biophysical and biochemical processes in the forest regeneration. This study mapped a second TDF as a function of regeneration age at the Santa Rosa National Park (SRNP) Environmental Monitoring Super Site, Costa Rica, using an airborne full-waveform LiDAR (Laser Vegetation Imaging Sensor or LVIS), a hyperspectral dataset (Hyperspectral MAPper or HyMap) and three advanced machine learning methods. We defined five age groups (0-10 years, 10-20 years, 20-30 years, 30-50 years, 50 + years) based on historical forest cover maps, and analyzed their effective LiDAR waveforms and cumulative return energy curves (derived from LVIS) and their reflectance (derived from the HyMap). Then, nine LVIS metrics and eleven HyMap indices were calculated and their abilities to differentiate the age groups were evaluated using Multiple Comparison Analysis (MCA). We found that six of the LVIS metrics which describe the vertical structure of the forests can significantly differentiate all age groups. None of HyMap metrics can differentiate all age groups, but some of them can identify certain age groups. Selected LVIS metrics and HyMap indices were used to map TDF age, through Support Vector Machine, Artificial Neural Network and Random Forest (RF). We found LVIS plus HyMap metrics generally produced more accurate forest age maps than either LVIS metrics or HyMap indices and RF better performed than other two classifiers. We finally proposed a method to synthesize different forest age maps into one age map which had the highest accuracy for all age groups. Our study highlighted the importance to consider the forest regeneration as a continuous stochastic phenomenon, and also highlighted the advantages to incorporate multiple remote sensing techniques to describe the forest regeneration. Our method to synthesize the forest age map can also benefit other researchers who need to take advantage of multiple mapping results.