Integrating Hyperspectral, LiDAR, and Radiative Transfer Modeling to Predict Forest GPP

Understanding forest carbon fluxes is crucial for monitoring and predicting the global carbon cycle and climate-carbon interactions. Plant physiological and structural traits (PSTs) strongly influence canopy-light interactions and, in turn, forest productivity. Hyperspectral and LiDAR observations from the US National Ecological Observatory Network’s Airborne Observation Platform (NEON AOP) enable the mapping of three-dimensional (3D) spatial variability of PSTs across observed sites. PSTs (e.g. leaf nitrogen, leaf mass area, leaf area density, woody area density) determine the optical and geometric characteristics of the canopy and have a nontrivial relationship with the distribution of photosynthetically active radiation (PAR) within the canopy, the magnitude and distribution of absorbed PAR (APAR), and the influence of the diffuse PAR fraction. Therefore, linking mapped traits to total APAR and gross primary productivity (GPP) requires the modelling of within-canopy radiative transfer as well as ecosystem function. With 3D PST estimates derived from NEON AOP data alongside measurements of PAR from the AmeriFlux network, we used the Forest Light Environmental Simulator (FLiES) to model APAR in two North American deciduous broadleaf forest sites at the University of Michigan Biological Station in the northern Lower Peninsula of Michigan (US-UMB, US-UMd) with a 30-minute temporal resolution over two months centered on the NEON AOP flight period. We then trained a random forest (RF) model using AmeriFlux daytime-partitioned (DT) eddy covariance estimates of GPP, modelled APAR, and meteorological and solar data. Our results show that our RF model reliably reproduces DT GPP (R2 > 0.8) and that FLiES-derived APAR is a key predictor of GPP. Lastly, we explore how the vertical and diurnal distributions of APAR vary with canopy structural differences, and how these structural differences relate to disturbance history.