Spatial Trends in Tree Phenology in the Bohemian Forest Ecosystem over the Last Two Decades

Tree phenology is both a crucial proxy for ecosystem services such as biomass production and an important indicator for the impact of climate change on forests due to its dependence on local climatic conditions. Both traditional monitoring and phenological cameras (pheno cams) offer only few observation points, limiting the study of large-scale spatial patterns. Optical satellite time-series can provide the necessary spatio-temporal extent needed. However, their agreement with in situ data is often unknown and uncertainties are high.

This study is part of the AI-Klima project and aims to fill these gaps for the Bohemian Forest, where traditional and pheno cam observations are available, but no large-scale spatio-temporal analysis of tree phenology has been conducted to date. The studied period stretches over two decades, from 2000 to 2024. The study area, the Bohemian Forest, is a cross-border region including the German Bavarian Forest National Park (BFNP) and the Czech Šumava National Park. It contains a diverse forest composition and with both managed and unmanaged areas.

Forest phenology in satellite data is analyzed via an innovative method combination and validated with pheno cam data. EVI time series data from harmonized Landsat 4-9 and Sentinel 2 data is used to calculate six phenological timings per pixel and year: (1) start of the green up in spring, (2) point of fastest growth (spring inflection point (SIP)), (3) end of spring growth, (4) start of senescence in autumn, (5) point of fastest decline (autumn inflection point), (6) end of the autumn decline. This is achieved through a novel dynamic combination of filtering and fitting methods for time series smoothing. This approach accounts for the high variability of available data. Phenological timings are then obtained trough geometric analysis of the smoothed time series curve.

To validate the results, time series of 15 pheno cams are analyzed in a similar way, obtaining reference phenological timings. Here, the Green Chromatic Coordinate (GCC) is used as a spectral index instead of the EVI. To insure accuracy, the pheno cam images are pre-processed to exclude bad-quality images using both traditional and AI-driven methods.

The resulting phenological timings for every year over the whole study area are analyzed for temporal trends and spatial patterns in forest phenology. For spring phenology, first results show spatially heterogeneous patterns of change in the inflection point timing. Lower elevation areas display comparatively little change, while many higher elevation areas show an unexpected strong trend of later spring inflection points. This goes against the expected trend of earlier springs due to climate change, but might be explained by the effect of mass bark beetle outbreaks and consequent shift in vegetation composition. However, the influence of data availability, alongside data quality, will need to be discussed critically as well, since for some years, only a small number of cloud free satellite scenes is available.

Combined with other work package outputs in AI-Klima, the phenology trends will be utilized to identify the forest stands most impacted by climate change in the Bohemian Forest Ecosystem.