Long-term ecological records are rare yet essential for understanding how biological assemblages respond to environmental changes. Here we present a 33-year eDNA metabarcoding time series (1992–2024) of leaf-associated fungal communities at Ogawa Forest Reserve, Japan, spanning three co-occurring host tree species (Quercus serrata, Fagus crenata, and Fagus japonica) and based on 310 samples rarefied to a standardized depth of 25,000 reads. Although OTU richness showed a weak but significant positive trend across years (Spearman’s test, P = 0.022), this trend was driven by a reduction in OTU richness observed during the first five years following litter collection. We attribute this to an artifact of DNA degradation. However, OTU richness remained remarkably stable for the preceding period, extending back 33 years, which strongly suggests that, while initial degradation occurs, preserved litter trap specimens remain highly effective for investigating the long-term temporal dynamics of phyllosphere fungal assemblages.
On the other hand, Shannon diversity remained stable across the entire study period (Spearman’s test, P = 0.962). Beta diversity analyses reinforced this nuanced view of long-term trends at Ogawa. Temporal distance-decay analysis revealed that Bray-Curtis dissimilarity increased by only 0.00079 per year across the 33-year span (P < 0.001); however, the extremely low explanatory power (R2 = 0.0013) suggests that the vast majority of compositional variation is independent of a linear temporal trend. This characterizes the fungal assemblages as highly dynamic and unpredictable, governed more by stochastic turnover than by directional succession. Despite the high degree of stochastic turnover, local fungal diversity was significantly influenced by annual climate. Both total annual precipitation and maximum daily precipitation intensity were negatively correlated with OTU richness (ρ = −0.186 and −0.178, respectively, P < 0.01) and Shannon diversity (ρ = −0.173 and −0.156, respectively, P < 0.05). Although the modest effect sizes suggest that precipitation is one of several interacting drivers, its consistent significance highlights a deterministic filtering process that operates amidst the otherwise high compositional noise. Collectively, these findings reveal a system where community composition is dominated by stochasticity, yet local diversity remains measurably sensitive to inter-annual precipitation variation.