Deep learning derived leaf traits reveal phylogenetic and environmental drivers of diversification in Rhododendron

Leaf traits mediate plant heat and water exchange, driven by evolutionary history and environmental conditions. However, broad‐scale trait data are often limited for clades with high species richness, making it difficult to evaluate the relative contributions of historical trait divergence and contemporary environmental variation.

Here, we established an automated workflow to extract leaf area and roundness from 105,897 intact leaves detected in 44,020 of 68,802 Rhododendron specimens, representing 624 species and 26 varieties. Leaf area exhibited a strong phylogenetic signal (Blomberg's K = 1.21), whereas roundness was less constrained by shared ancestry (Blomberg's K = 0.50). Multi‐optima Ornstein‐Uhlenbeck (OU) models suggested a small ancestral leaf area and indicated six shifts in adaptive optima, most of which occurred between 30 and 10 Mya. Leaf area evolutionary rates increased from the Oligocene onwards, overlapping major phases of Himalayan uplift and associated climatic reorganisation.

Contemporary climate accounted for only a limited fraction of leaf trait variation. Leaf area was negatively associated with mean diurnal range, wind speed and elevation, but positively associated with annual mean temperature, precipitation seasonality and solar radiation ( p  < 0.05). Leaf roundness increased with precipitation seasonality, wind speed and elevation, and decreased with annual mean temperature ( p  < 0.05). Even where the two subgenera co‐occurred and occupied broadly similar climatic space, they differed in how leaf traits tracked wind exposure. In subg. Rhododendron , increasing wind speed and elevation were associated with a steeper reduction in leaf area and a stronger shift towards rounder leaves. By contrast, the divergent leaf trait patterns observed in subg. Hymenanthes may be attributed to its higher occupancy of sheltered habitats, combined with varying degrees of environmental sensitivity between the two subgenera.

Synthesis : This study reveals that the diversity of leaf morphology in Rhododendron is shaped not only by broad evolutionary history and contemporary environment, but also by distinct evolutionary lineages adopting differentiated adaptive strategies within similar environmental niches.