Most studies of the effects of forest fragmentation have been conducted in areas where fragmentation is recent and related to human land use. As forests will become increasingly fragmented in the coming decades, information about the long-term effects of fragmentation on ecosystem processes is crucial to predict future scenarios. We investigated the long-term consequences of forest fragmentation caused by aridization during the Pleistocene in coastal hills of semiarid Chile (30^o S). We assessed the patterns of litterfall in a mosaic of relict rain forest patches dominated by evergreen trees and maintained by oceanic fogs, at 600 m above sea level in Fray Jorge National Park. We also analyzed the resilience of this ecosystem process in relation to patch size and temporal variation in rain and fog inputs. We set up a total of 36 litter traps (0.17 m² each) in six forest patches, two small (0.2-0.3 ha), two medium (1.3-7.5 ha) and two large ones (24- 28 ha). Fallen litter and fine woody debris dry mass were estimated monthly during three years. Additionally, water inputs from fog condensation and rain to each patch were recorded monthly using throughfall collectors.

Results/Conclusions

Litterfall input from the evergreen canopy was significantly influenced by month (F_{(35, 174)}= 30.4 p << 0,001) and patch size (F_{(1, 1045)}= 4.9 p< 0.05). A strong interaction between month x patch size (F_{(35, 1045}= 4.0 p << 0.001) explained 89% of the variance in total litterfall. Regression analysis showed a strong negative relationship between patch size and the temporal coefficient of variance of litter inputs over the three years of study (r² = 0.85, p<0.01). Litterfall mass correlated positively with throughfall water inputs only in the smaller patches.

We show strong effects of forest patch size on total litter input and seasonal dynamics in these evergreen rain forests. Small patches were more variable and dependent on water inputs to produce litter than larger patches. Litterfall in larger patches showed less temporal variance and less dependency on climatic variability.