Degraded Amazonian forests can take decades to recover and the ecological results of natural regeneration are still uncertain. Here we use field data collected across 15 lowland Amazon smallholder properties to examine the relationships between forest structure, mammal diversity, regrowth type, regrowth age, topography and hydrology. Forest structure was quantified together with mammal diversity in 30 paired regrowth-control plots. Forest regrowth stage was classified into three groups: late second-regrowth, early second-regrowth and abandoned pasture. Basal area in regrowth plots remained less than half that recorded in control plots even after 20–25 years. Although basal area did increase in sequence from pasture, early to late-regrowth plots, there was a significant decline in basal area of late-regrowth control plots associated with a decline in the proportion of large trees. Variation in different forest structure responses was explained by contrasting variables, with the proportion of small trees (DBH < 20 cm) most strongly explained by topography (altitude and slope) whereas the proportion of large trees (DBH > 60 cm) was explained by plot type (control vs. regrowth) and regrowth class. These findings support calls for increased efforts to actively conserve large trees to avoid retrogressive succession around edges of degraded Amazon forests.

1Degraded Amazonian forests can take decades to recover and the ecological results of natural 2 regeneration are still uncertain. Here we use field data collected across 15 lowland Amazon 3 smallholder properties to examine the relationships between forest structure, mammal diversity, 4 regrowth type, regrowth age, topography and hydrography. Forest structure was quantified together 5 with mammal diversity in 30 paired regrowth-control plots. Forest regrowth stage was classified into 6 three groups: late second-regrowth, early second-regrowth and abandoned pasture. Basal area in 7 regrowth plots remained less than half that recorded in control plots even after 20-25 years. Although 8 basal area did increase in sequence from pasture, early to late-regrowth plots, there was a significant 9 decline in basal area of late-regrowth control plots associated with a decline in the proportion of 10 large trees. There was also contrasting support for different non-mutually exclusive hypotheses, with 11 proportion of small trees (DBH <20cm) most strongly supported by topography (altitude and slope) 12 whereas the proportion of large trees (DBH >60cm) supported by plot type and regrowth class. These 13 findings support calls for increased efforts to actively conserve large trees to avoid retrogressive 14 succession around edges of degraded Amazon forests. 15 16 17 Healthy tropical forests provide goods and services to human populations. Yet tropical forests show 18 worrying rates of forest loss with an elevated loss / gain ratio and a statistically significant trend in 19 annual forest loss of 2101 km 2 /year 1 . One option to revert tropical forest loss is the restoration of 20 degraded forests and deforested landscapes 2,3 . Although the post-disturbance restoration of forest 21 ecosystems often involves passive restoration strategies (i.e. natural regeneration), the ecological 22 results of this type of restoration are still uncertain 2-4 . 23 Continuing widespread forest losses across Amazonia compromises vital ecosystem services 24 such as carbon storage, regulation of hydrological cycles and climate patterns 5-7 . Riverside forests 25 are particularly threatened and suffer losses due to the conversion of forest cover to pastures, 26 compromising the maintenance of water flows 8 . The recovery of degraded areas is necessary to 27 recuperate the standing forest value and the Amazon offers an excellent recovery opportunity due to 28 its natural potential for regeneration 9,10 . Yet, the regrowth rate of degraded Amazon forests can be 29 slow, as abandoned areas are typically on compacted poor quality soils 11,12 and due to the high 30 structural and biological diversity of the original forests 13 . 31Separating the complex interactions driving recruitment and recovery patterns of highly diverse 32 Amazon forests is challenging 2,3,14,15 , yet we know that different faunal groups can modulate and 33 generate key impacts [16][17][18][19] . Indeed, the successional trajectory of natural regeneration in degraded 34 forests ...