Forest regeneration (i.e. natural forest regrowth) is the central promise to achieve forest and landscape restoration commitments and mitigate the global biodiversity and climate crisis (Chazdon et al., 2020). This restoration approach has shown to be more scalable and cost-effective than tree planting (Crouzeilles et al., 2017), yet deciding which areas could be restored through regeneration is still a critical research challenge. It is necessary understand where forest is regenerating in agricultural landscape and what are the drivers of their development and persistence (Brancalion et al., 2016b; Reid et al., 2018). Many recent studies have tried to understand the main drivers of forest regeneration based on remote sensing analyses of land-use change and to associate the regeneration probability to spatial drivers (e.g., Crouzeilles et al., 2020a; Molin et al., 2018; Nanni et al., 2019). In general, these studies have demonstrated that the chances of forest regeneration increase in steeper slopes (>12%), close to forest remnants, inside protected areas, and far from population centers (Borda-Niño et al., 2020).

Overall, forest regeneration in a particular area would be determined by the chances of being abandoned and recolonized by native forest species (Arroyo-Rodríguez et al., 2017). For instance, terrain slope is a key determinant of agriculture mechanization, which is a relevant for land management decisions when rural labor becomes scarce or land-use changes. In contrast, proximity to forest remnants is associated with the likelihood of seed dispersal to abandoned areas. Although some spatial drivers assessed through remote sensing can be useful surrogates of forest regeneration chances, some critical drivers rely on social surveys that have been rarely employed in restoration programs (Wortley et al., 2013). The integration of social survey findings with remote sensing approaches can potentially enhance the forest regeneration model’s accuracy at the landscape scale and approximate them to land management decision-making.

Here, we explored the influence of biophysical (i.e. altitude, slope, topography and distance to rivers and remnant forest patches) and socioeconomic (related to the landowner and to the farm) drivers, including drivers obtained by a survey with local farmers, on forest regeneration in a rural landscape of Southeastern Brazil. We hypothesized that biophysical and socioeconomic drivers are significantly correlated to forest regeneration. We expect that the inclusion of socioeconomic drivers obtained by social surveys positively complement the use of spatial drivers traditionally employed in forest regeneration studies, thus highlighting the importance of social science tools to guide decisions and improve spatial prioritization in restoration programs.

The forest regeneration occurrence significantly increased at lower Euclidean distance from forest remnants and rivers, in steeper slopes (>10%), and at a lower altitude (between 470 to 600 m). Forest regeneration was most likely to occur in farms with commercial agricultural production other than permanent crops of sugarcane and beef and dairy cattle, in farms with higher number of land-use types, with few residents, and belonging to landowners with low economic reliance from agricultural activities (Table 3). GLMM results are detailed in Supplemental Material 6.

Forest regeneration is determined by biophysical and socioeconomic drivers occurring and interacting at multiple temporal and spatial scales (Arroyo-Rodríguez et al., 2017; Chazdon et al., 2020; Crouzeilles et al., 2020b). But land-use changes are consequence of decisions made by people in the landscape, and these decisions are determined not only by socioeconomic or biophysical factors, but also by policies. At the Corumbataí River basin forest regeneration between 1995 and 2018 is correlated with a combination of biophysical drivers and farmers decisions on land abandonment (Molin et al., 2018). The regeneration of native forests was more frequent on steeper slopes (>10%) previously used for extensive cattle ranching, and dependent on the economic context of the study period (Ferraz et al., 2014; Molin et al., 2017). In 1991, after deregulation of the dairy sector by the Brazilian government, high levels of price instability affected mainly small ranchers and resulted in a progressive abandonment of less productive pasturelands (Siqueira et al., 2010). The influence of slope in forest regeneration is even more pronounced in sugarcane production areas, which rely on mechanized agricultural activities that are compromised in slopes >12% where the machines are not manageable (Ferraz et al., 2014; Molin et al., 2017).

A second group of biophysical drivers related to forest regeneration was associated with regeneration processes and environmental laws, as forest regeneration tended to occur closer to older forest patches and permanent rivers. Seed dispersal is favored at higher proximity (lower distances) from forest remnants, which act as a source of both seeds and animal dispersers (Chazdon et al., 2020). The direct correlation with proximity to permanent rivers on forest regeneration was likely related to the compliance with Brazil’s Forest Code, which prohibits agricultural activities in riparian buffers and mandates their restoration (Brancalion et al., 2016a). Complementary, forest regeneration can be more likely along streams due to lower dispersal limitation, as small forest patches and isolated trees are common features of riparian areas and animals heavily use them as ecological corridors (Şekercioğlu et al., 2015).

Regarding socioeconomic drivers, forest regeneration was higher in farms where landowners have lower economic dependence from the farm’s income. However, this result can change over time since fluctuating market demands affect the revenues obtained by agricultural commodities, which in turn can encourage deforestation or natural regeneration by land abandonment (Busch and Ferretti-Gallon, 2017). The rate of land-use change is commonly associated with demands for primary commodities and is often modeled using an economic framework (Jadin et al., 2016). Nevertheless, our results suggest that the economic dependence from farming is strongly related to the economic dependence from the land area, and not only the per capita income or socioeconomic status (e.g., Newman et al., 2014).

The high economic reliance of some landowners from farming highlights the need to implement new production models that promote the more efficient use of natural resources and reduce pressure on regenerating forests. Nowadays there are alternative farming systems that can reconcile production and conservation, such as agroforestry and silvopastoral systems (Calle, 2020; Ceccon, 2013), which have been recently framed under the “regenerative agriculture” umbrella. Complementary, the regeneration of forests on lands with low agricultural potential may sequester enough carbon quantities to be subsidized by Payments for Ecosystem Services and then to ensure their longer-term permanence (Chazdon et al., 2016b; Rudel, 2015), although current carbon pricing is often insufficient to overcome land opportunity costs in the study region (Brancalion et al., 2021). Forest regeneration appears to be more probable in properties with fewer residents, where labor availability for managing less productive lands is scarcer and favors the adoption of mechanized cropping systems, which in turn may result in the abandonment of steeper areas (Rudel, 2015). The higher forest regeneration in farms with diverse land-use types was probably related to the establishment of a more favorable landscape matrix for seed dispersers movement, which may have their flows favored in more diverse mosaics of eucalypt plantations and agro-pastoral land uses (Borda-Niño et al., 2017). In fact, we report secondary forests regenerating following the abandonment of eucalypt plantations after their harvesting, which can be the object of assisted natural regeneration strategies and enrichment plantings with native species, contributing to both carbon stocking and biodiversity conservation (César et al., 2018). Other explanation is the more selective use of the land, which is also related with the permanence of the landowners in rural areas. Over time, the landowner gets better acquainted with their lands and their forests, and promote a more selective use of the natural resources of the farm areas to comply with environmental rules and of agro-pastoral systems to maximize production (Mofya-Mukuka and Hichaambwa, 2018).

Part of the dynamism of forest cover that characterized the studied landscape and resulted in an increase of native forest cover may be a temporal situation. Marginality is relative to the opportunity cost of different land uses and legal constraints, all of which may change independently of biophysical conditions (Sloan et al., 2016). For instance, if new machinery to cultivate sugarcane in sloppy areas is developed, or if the Native Vegetation Protection Law (i.e. the so called Brazilian Forest Code) is reformed again (as it was in 2012) to reduce the need for protection and restoration of native forests, a large portion of second-growth forests can be reconverted to agro-pastoral land uses. The importance of these forests for agriculture needs to reconsidered based on their origins, dynamic properties and landscape settings to better support their protection and restoration, and leverage their contributions to achieve global restoration targets (Chazdon et al., 2016a).

Recognize that the increase in native forest cover observed in the study area may not be necessarily accompanied by substantial environmental benefits results crucial. It is particularly important as the regeneration of young secondary forests, which take decades to accumulate carbon stocks and native species in marginal agricultural lands, can co-occur with the clearance and degradation of mature forests in flatter areas prone to mechanized agriculture (Brancalion and Holl, 2020). Detailed monitoring of forest structure and diversity is needed (see César et al., 2018 for forest biodiversity and carbon stocking information in the study region). It can benefit from novel remote sensing approaches and drone-lidar systems (Csillik et al., 2019) to be performed at the landscape scale.

Our local study highlights the importance of integrating social science tools to gather socioeconomic, non-spatial information on relevant drivers of forest regeneration, thus emphasizing the importance of considering farmers’ decision on restoration planning. As the Corumbataí River basin hosts a diverse gradient of socioeconomic and biophysical conditions, our findings can be considered in other geographical contexts.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.