Mapping canopy phenolics in European mixed temperate forests using air- and space-borne imaging spectroscopy

Abstract

Phenolics are a rarely quantified plant biochemical trait that plays a vital role in plant physiology and ecosystem functioning, contributing to plant's chemical defence and influencing nutrient cycling and soil microbial compositions. Spatially continuous information on foliar phenolics is essential for assessing plant health and ecosystem functional diversity. However, previous efforts to predict and map phenolics have been confined to aircraft-based hyperspectral data in limited biomes. The potential of next-generation imaging spectroscopy, whether airborne- or spaceborne-based, for mapping phenolics remains underexplored, particularly in structurally complex and heterogeneous ecosystems such as European mixed temperate forests. Furthermore, much is still unknown about the consistency and uncertainties of predicting forest canopy phenolics across different acquisition levels (airborne vs. spaceborne), limiting our ability to generalise and upscale local trait estimates to broader spatial extents. In this study, we sampled sunlit top-of-canopy leaves from three dominant tree species across mixed temperate forests in southeast Germany. Leveraging next-generation airborne (AVIRIS-NG) and spaceborne (PRISMA) imaging spectroscopy (400–2400 nm), we modelled two ecologically important phenolics (total phenol and tannin) expressed in three forms (foliar mass-based, foliar area-based, and canopy-based). The predictive accuracy of two data-driven approaches, partial least squares regression (PLSR) and Gaussian processes regression (GPR), was compared to assess performance across different spatial scales. Our results demonstrate that phenolics in sunlit canopy leaves can be accurately estimated from both airborne and spaceborne data, with foliar area-based phenolics showing the strongest relationship with spectral reflectance (total phenol: R² = 0.64–0.69, NRMSE = 13.28%–15.65%; tannin: R² = 0.49–0.65, NRMSE = 15.86%–21.29%). We observed several similar patterns in model coefficients across airborne and satellite levels, with informative wavelengths aligning with known phenolic features. While the model accuracy declined slightly when scaling from canopy to landscape scale, phenolic maps derived from AVIRIS (aggregated to 30 m) and PRISMA showed good spatial agreement and linearity (GPR: r = 0.68, slope = 0.86; PLSR: r = 0.57, slope = 0.49). These maps successfully captured inter- and intra-species phenolic variability across the test site with low prediction uncertainty. Our findings provide valuable insights into mapping canopy traits across different observational scales, demonstrating how next-generation imaging spectroscopy can characterize the spatial and temporal dynamics of plant phenolics. This research paves the way for improved global monitoring of ecosystem functioning, as well as the pattern of phenolics across forested landscapes and trees' potential ‘chemical’ defences against herbivory and other environmental stressors.