Regeneration potential in South African forest fragments: extinction debt paid off or hampered by contemporary matrix modification?

Deforestation generates forest fragments within modified matrices with frequently detrimental outcomes for species persistence. In historically fragmented forests, however, species have evolved under fragmented conditions and might be more resistant to matrix modification. We examined the effect of contemporary matrix modification on forest regeneration potential in historically fragmented scarp-forest fragments in KwaZulu-Natal, South Africa. We compared tree, seedling and sapling richness, abundance and composition in four forest-fragment types varying in size and matrix modification: large and small forest fragments surrounded by indigenous grassland and small forest fragments enclosed by eucalypt and sugarcane plantations. Additionally, seedlings and saplings were classified according to their origin as local/immigrant to assess matrix permeability for seed dispersers. Trees and saplings were slightly, and seedlings were strongly affected by matrix modification: seedling richness and abundance were significantly lower in fragments with sugarcane compared to fragments with grassland matrix. Furthermore, all stage classes revealed compositional differences with more light-tolerant species occurring in fragments with modified matrices. We suggest that human activities associated with matrix modification have altered abiotic factors favouring light-tolerant species. Changes in biotic interactions seemed to be less important as immigrant recruitment occurred in all fragments implying matrix permeability. Our findings demonstrate that contemporary matrix modification might threaten forest regeneration potential in historical scarp-forest fragments. Reduced seedling recruitment might indicate a time lag in local species extinction, as trees regenerate slowly. We recommend using both recruitment stages as early warning indicators. Further, including small scarp-forest fragments into protected area networks will minimise human impact associated with matrix modification.