Spatiotemporal variability of leaf functional traits in central European floodplain forests: integrating in-situ, hyperspectral, and Sentinel-2 data with RTM, PLSR, and neural networks

Vegetation traits are essential indicators of plant health and ecosystem functioning, providing critical information for biodiversity assessment, forest management, and ecological modelling. This study investigates the variability of vegetation traits across spatial and seasonal scales in the floodplain forest of Lanžhot, Czech Republic, using in-situ leaf trait measurements, airborne CASI hyperspectral imagery, and Sentinel-2 satellite data. Our objectives were to: (i) compare the performance of various trait retrieval methods, including radiative transfer modelling (RTM), partial least squares regression (PLSR), and neural networks (NN); (ii) assess the temporal variability of functional traits from leaf- and canopy-level spectral retrievals; and (iii) evaluate the spatial variability of key vegetation traits derived from high-resolution CASI hyperspectral and Sentinel-2 data. In addition, we examined differences in trait dynamics between managed and unmanaged regions of floodplain forests. Our findings reveal substantial seasonal fluctuations in chlorophyll content (Cab), equivalent water thickness (Cw), dry matter content (Cm), and leaf area index (LAI), with distinct patterns across tree species. The results demonstrate the effectiveness of hyperspectral remote sensing in capturing forest trait dynamics and emphasize the relevance of both environmental drivers and management regimes in ecosystem monitoring. PLSR achieved the highest accuracy in retrieving leaf functional traits from leaf-level spectra, with R2 values of 0.84, 0.86, and 0.68 and NRMSE values of 0.20, 0.14, and 0.18 for Cab, Cw, and Cm, respectively. For canopy-level spectra, the NN model yielded the highest accuracy in Cab and LAI retrievals, with R2 > 0.5 across seasons for both CASI and Sentinel-2 data. By clearly showing how different methods influence trait estimates and by identifying key factors driving their spatial and temporal variability, this study provides valuable contributions to forest management strategies and ecological modelling, particularly in endangered floodplain forest ecosystems.