Optimizing hybrid models for forest leaf and canopy trait mapping from EnMAP hyperspectral data with limited field samples
Loading...
Date
Publisher
Elsevier
Abstract
Accurate estimation of functional vegetation traits is essential for understanding ecosystem dynamics and supporting
environmental monitoring. This study investigates the potential of spaceborne hyperspectral data from
the Environmental Mapping and Analysis Program (EnMAP) for mapping key forest vegetation traits such as leaf
area index (LAI), leaf chlorophyll content (Cab), leaf mass per area (LMA), specific leaf area (SLA), and leaf water
content (Cw) as well as the rarely addressed leaf anthocyanin content (Canth) using a hybrid approach combining
radiative transfer calculations and machine learning. While machine learning-based studies on vegetation traits
retrieval often have relied on intensive field data collection, the present studies explicitly focus on situations with
limited field samples which is a often the case in ecosystem research. Here, we use a combination of the
PROSAIL-D leaf and canopy radiative transfer models (RTMs) and Gaussian Process Regression (GPR) optimized
with Active Learning (AL) sampling to improve the retrieval accuracy while minimizing reliance on extensive in
situ data. Validation results demonstrated that the hybrid approach successfully retrieved LAI (R2 = 0.69) and
Cab (R2 = 0.68) with relatively low errors (RMSE of 0.172 m2/m2 for LAI and 1.502 μg/cm2 for Cab), while more
complex traits, such as Canth, LMA, and Cw exhibited moderate retrieval performance only (R2 of 0.45, 0.51, and
0.31, respectively). The application of optimized hybrid GPR models to the study area enabled area-wide
landscape forest trait mapping with minimal computational effort.
Metadata
License
Except where otherwise noted, this item's license is described as Attribution 4.0 International