The study areas are the Moribane Forest Reserve (MFR) in Sussundenga District and Serra Chôa (SCH) in Barue District, Manica Province (Figure 1). We selected the study areas, taking into account the existence of transitional forest in environmentally sensitive places with human presence, but with different land tenure systems—protected and open access. The spatial extent is limited to two mountainous places in Manica Province, MFR and SCH, recognized as critical habitats under the International Finance Corporation criteria due to their importance to biodiversity, the existence of endemic species, the occurrence of key evolutionary processes, and high scientific value (Virtanen, 2002; Sitoe et al., 2015). The MFR is considered the most extensive and best-preserved lowland rainforest in Mozambique (Müller et al., 2005). Its topography is irregular—varying from 400 to 550 m above sea level. The non-protected area, SCH, is situated in highlands above 1,200 m (Cizek, 2009). Wooded or open grasslands dominate the SCH vegetation at the central and southern plateau, miombo dominates in the northern area, and mountainous evergreen forests dominate in steeper slopes (Mussanhane et al., 2000).

Moribane Forest Reserve is a protected area that was created in 1957 for wood production and since 2010 has been a part of the Chimanimani National Park (CNP) buffer zone (ANAC, 2021). As legislation states, the national park has the function to protect, conserve, and manage the biodiversity (Governo de Moçambique, 2014). Human presence is allowed under specific conditions set by the management plan and should not threaten the biodiversity; only indirect use of natural resources is allowed. Conservation legislation does not recognize the forest reserve as a protected area but defines objectives of a buffer zone (MITADER, 2019). According to current legislation, economic activities are allowed in the buffer zone, although priority is given to activities in line with conservation purposes like protection of water courses, landscape maintenance, and low urbanization (Governo de Moçambique, 2014). The current CNP management plan (ANAC, 2021) defines the extraction of non-timber forest products and timber in the buffer zone as only allowed for subsistence. The buffer zone of CNP is larger than the core protected area, so the modifications in land use and increase of human presence will probably negatively impact the park (Timberlake et al., 2016).

In MFR, the population lives mostly along the N260 asphalted road and in 2017 was estimated at around 2,400 inhabitants. In SCH, human settlements tend to occur in the central and western parts, and in 2017, the population was estimated at approximately 1,800 inhabitants. In 2019, it was estimated that 479 households are in MFR, while in Serra there are 343 households (J. Matias, personal communication, September 13, 2019). SCH, in general, has a low level of transport infrastructure, with one feeder road connecting Catandica and Chôa-Sede and a few tracks connecting communities. In both areas, population density is low (15 inhabitants/km² at the MFR and 5 inhabitants/km² at SCH), which is considerably lower than the province average, calculated for 2017 at 34 inhabitants/km². The population density in the study areas is also lower than that in the respective districts—Sussundenga has 24 people/km² and Barue 45 people/km² (INE, 2018). Although the density is low, there are areas with relatively dense settlement because people tend to live close to each other and to roads, leaving vast areas uninhabited.

The main economic activity for both sites is agriculture for subsistence and cash. Commercial agriculture is developed mainly to produce macadamia and potato in SCH and banana and sesame in MFR. There is no tourism activity in SCH. In contrast, MFR has the community association Kubatana Moribane, in cooperation with the NGO Eco-MICAIA, which runs the Ndzou Camp lodge (Dondeyne et al., 2012). Nevertheless, a few commercial activities are developed in both areas.

The survey was conducted in four communities in MFR (Mucuawaio, M’Punga Centro, Mutoe, and Magaraba) and the same number in SCH (Muswipa, Chôa-Sede, Doro, and Nhaterere). The total number of households surveyed was 293, with 149 in MFR and 144 in SCH. The sample size calculation was done separately for each site to ensure spatial coverage and representative sampling. Following this, the number of households to interview for each community was calculated based on the proportion of households living in the settlement to the total number of household in the study area.

The data collection occurred in two phases—in September and October 2019 in MFR and in August and September 2020 in SCH. Data were collected by two researchers with local guides who were proficient in Portuguese and the local language. Although the questionnaire was in Portuguese, the interviews were in the local languages (ChiNdau and ChiManica). In addition to each questionnaire, there was a record of the geographic location and delimitation of the area belonging to the household, done by the researchers accompanied by the owner or other adult household member. The respondents were mostly household heads, but, in their absence, we interviewed other adults. Our survey covered questions related to household profile, their economic activities, income, and access to and use of forest resources.

In 2017, the Manica Province had 1,781,968 ha forested areas, 27% of the total province area. The deforestation rates for Manica Province were assessed at 0.7% (country rate was 0.6%) from 1990 to 2002 and 1.8% (country rate was 0.8%) for the period 2003–2013 (MITADER, 2018b). The forest conversion occurred mainly from natural vegetation to agriculture (Jansen et al., 2008). Since the 1980s, four forest inventories have been conducted in Mozambique (Malleux, 1980; Saket, 1994; Marzoli, 2007; MITADER, 2018b). The sampling methodology, forest definition, and scale differ, so comparing results to evaluate deforestation in specific places is tricky. To overcome this, Global Forest Change data were used, specifically the global forest cover loss for the period 2000–2020 available in the Google Earth Engine (Hansen et al., 2013). The datasets were based on the analysis of the Landsat image time series, which captured the changes of forest cover at the global level obtained up to the year 2020 on the Google Earth Engine platform. A study by Hansen et al. (2013) set the tree cover for the year 2000 and quantified the forest losses and gains, where forest loss was considered as a complete tree cover canopy removal and forest gains were assumed as the reverse. The tree cover defined for the year 2000 captures vegetation taller than 5 m and records the percentage of the canopy for each cell, varying from 0 to 100. To define areas as forested, we applied a threshold of 30% of tree cover to the Hansen et al. (2013) data. The advantage of this dataset is its consistency and easy access, as well as the possibility to compare data from a long period (Galiatsatos et al., 2020).

The dataset in grid format was clipped to the study areas and downloaded. In ArcGIS version 10.1, the grids were converted into polygon vector data. The grid was converted into a binary raster of values, 1 for forest cover loss and 0 for no forest cover loss, and the area was calculated for the forest cover loss. The next step was to clip data with farmers’ plots to find out which households cleared forest.

We first performed an exploratory data analysis computing the basic descriptive statistics (mean, median, SD) to examine the respondents’ and households’ characteristics and livelihood strategies and to find similarities and differences between the study areas. The strategies were grouped into three main classes: on-farm, off-farm, and forest-based. Then, the possible multicollinearity was analyzed using the Spearman nonparametric test.

We used a logit regression model to assess the probability to clear the forest by households based on Greene (2002) and the following form:

The logit model is frequently applied for deforestation evaluation looking at different economic, social, and spatial variables in large and local scales (Ojeda Luna et al., 2020; Ullah et al., 2020). The dependent variable is binary and depicts the occurrence or non-occurrence of forest clearing between 2000 and 2020 at the household level (0: absence of forest clearing; 1: occurrence of forest clearing).

The explanatory variables that influence household decisions to clear forests are listed in Table 1. Household is considered a basic social unit where decisions are made regarding the production and use of available assets, namely (i) natural, (ii) social, (iii) human, (iv) physical, and (v) financial (DFID, 1999). The availability and quality of assets determine how natural resources are used and their impact on the environment (Wunder et al., 2014). In addition, external factors such as institutions, economic factors, different policies, and cultural factors influence the assets (Yang et al., 2019). We divided the values by root mean square for standardization purposes to run a regression model and better compare the effects of each independent variable (Pir Bavaghar, 2015). The variance inflation factor (VIF) revealed a moderate correlation as the values for independent variables are between 1 and 2. We performed the analysis through RStudio version 1.2.5019 (RStudio Team, 2020).

We found that in 2000 the SCH had more significant forest extensions than the MFR, but comparison of the total surface versus forested area showed higher forest occurrence in the protected area, MFR, than in non-protected SCH (Table 2). The average annual forest loss from 2000 to 2020 was 276.0 ha/year for MFR and 88.8 ha/year for SCH. In MFR, most deforestation occurred between 2014 and 2017, while in SCH the deforestation was most notable in 2017 (Figure 2).

In MFR, 41% of the interviewees were women and 59% were men; in SCH, the percentage of women was 35% and men 65%. The average age of respondents was 42.3 years in MFR and 41.7 years in SCH. In general, men are better educated than women, and respondents in SCH are better educated than in MFR, where illiteracy is relatively high and directly related to gender. The highest level attained by most respondents in both areas is primary education. Analyzing the availability of workforce within the household, we found that 70.5% of households in MFR have one to four members who are economically active. The remaining 29.5% of households have five or more adult members. In SCH, the number of households with one to four members is 86.1%, and the remaining 13.9% have five or more adult members (Table 3).

In the study areas, the local leadership plays an important role in forest management. The local leader was pointed out as the person contacted to issue permission by 50.3% of households in MFR and by 61.8% of households in SCH. When asked about any limitations imposed by conservation status or cultural reasons, 89.9% of the MFR respondents and 68.8% of SCH found no restrictions or problems in accessing forest resources (Figure 3). There are slight differences between the two areas in reasons behind perception about difficult access to the forest. The ownership of the forested parcels by private citizens is mentioned by 4.2% of households in SCH, which is in accordance to the Land Act that allows citizens in open areas to hold the right to use land and consider themselves as rightful owners despite the basic premise that land and natural resources belong to the state (Bruna, 2019). Another difference is the human/elephant conflict in MFR. The area is inhabited by a small number of elephants that sometimes invade agricultural fields (Virtanen et al., 2020).

Fallow is a common agricultural practice for 90.6% of households in MFR and is a valuable source of poles, firewood, medicinal plants, honey, and grazing. Shifting cultivation is practiced by 13.9% of households in SCH, and the remaining respondents found fallowing unsuitable because of space shortage (44.4%) or good soil fertility (34.7%). The households in SCH tend to gather non-timber forest products from the forest instead of fallow (Table 4).

Crop cultivation and livestock are the dominant economic activity in both study areas (Table 5). The agricultural and livestock products are intended for their own consumption and trade in both areas, although road access is a limiting factor in SCH. Nonetheless, in SCH, 10% more households are engaged in commercializing agricultural products than in Moribane. The difference is even greater with regard to commercialization of livestock/poultry. In SCH, 83.3% of households commercialize animal products against 47.9% in Moribane. A considerable difference is also observed in trading honey, with 41.0% of households selling this product in MFR, but only 12.5% in SCH (Table 6).

More diversification of income sources is observed in SCH. There are several opportunities to generate income outside agriculture based on forest exploitation and employment in agricultural companies producing macadamia and potatoes. In SCH, four families are engaged exclusively in the exploitation of forest resources, such as charcoal production, hunting, and timber. Poles, firewood, and honey are the most traded products (Table 7). In MFR, only honey is sold as an additional source of income. In both areas, forest products play an important role in the households’ subsistence use.

If we consider the minimum wage for the agricultural sector, which was 61.6 EU/month in 2019 (WageIndicator Foundation, 2019), 91.3% of households in MFR earn less than the minimum wage compared to 85.4% of families in SCH. Whereas households in SCH are quite evenly distributed among income intervals, 50% received 28€ or less, and in Moribane, the average monthly income of 67.8% of the households ranged from 28.1€ to 70.0€ (Figure 4).

The analysis of the variables indicates that conservation measures do not influence the households’ decision to clear forest in the conservation area in MFR and play only a small role in SCH, where the probability for forest clearing is 10% (Table 8). The coefficient is positive in the protected area and negative in the open-access area.

Demographic characteristics relative to age, gender, and education level of respondents do not contribute in a significant manner to deforestation, although they behave as was expected. The increase in household members old enough to work does not seem relevant for deforestation in either area, although the direction is negative. The size of the agricultural plot is significant in MFR while in SCH it plays a minor role—although it correlates with forest cover changes in both areas. This is especially salient in MFR, where households cultivate larger areas than in SCH. Distance to roads is equally significant in both areas. In both places, the households who live distant from roads have a higher probability to deforest than those who live near roads, contradicting the expectations.

Level of income is related to deforestation in Moribane, with an 8% probability to clear forest when income increases. In SCH, an increase in income decreases the probability to clear forest by 13%. Regarding the livelihood strategies, there is a negative direction for the on-farm strategy in both sites, showing that more diversified agricultural production correlates with reduced deforestation. The effect of an on-farm strategy is significant in SCH, with 26% less probability to deforest by households more engaged in crop and livestock diversification than in households with less variety. The off-farm and forest-based strategies are relevant for MFR, although in opposite directions.

The prevalent perception among MFR households is that the legal conservation status does not limit their activities. In both areas, the traditional rules are important means to control access to forest resources. The traditions, belief, cultural rules, and local habits play an important role in natural resource management at the community level (Araia et al., 2020). According to Alfredo (2009), the land is seen by communities as a sacred and communal property. Hence, traditional authorities are vested with powers to settle norms and rules of its use rights and ownership. The turbulent history of the areas and the delicate political situation have also reduced the population’s trust in government initiatives, negatively affecting conservation efforts (Virtanen, 2019). The perception of unrestricted use of forest resources may be a sign of moderate application of protection measures and governmental agencies’ consideration of local power structures or weak enforcement of laws. Policy without proper implementation and enforcement of rules will not influence community-based forest resource conservation and management (Yami and Mekuria, 2022). Alternatively, a strict ban on natural resource use appears insufficient as a conservation tool. Studies show that it often creates revolt, increasing unsustainable use (Milgroom, 2012; Givá and Raitio, 2017; Wamir et al., 2017). It also negatively impacts livelihood choices, especially within communities in remote and poor areas with limited options for employment and income generation, while it has marginal impact on reducing deforestation (Angelsen et al., 2014). The relative proximity of forested areas also contributes to the households’ perception about easy access to forest resources. In MFR, more than 40% of households collect forest products from fallows and, to a lesser extent, from the forest, whereas in SCH, people tend to use the forest as the main source. Studies show that fallowing can play a relevant role as a source of forest products and thus reduce deforestation (Dewi et al., 2017; Heinimann et al., 2017).

Our study identified a set of livelihood strategies developed by households. They seem diverse but still heavily reliant on natural resources. The households combine on-farm, off-farm, and forest-based activities to fulfill their daily needs. Crop growing and livestock are predominant activities in both areas and typical for Mozambican rural areas (Kidane et al., 2019). The advantage of livelihood strategies based on agriculture and livestock is their short-term compensation. The results are immediate, unlike forest usage. The rapid reward for effort implemented is a strong motivation (Miller et al., 2021). Restrictions in access to forest resources can lead to increased dependency on agriculture, observed in the MFR, so the environmental agencies aid households to diversify livelihoods and ameliorate agricultural practices (Virtanen et al., 2020). Regarding the agricultural practice in open-access areas, there is a focus on improving productivity and intensification to reduce shifting agriculture and ensure food security (Cammaer, 2016). Government policies that tend to intensify agricultural productivity and increased market opportunities can induce farmers to move from shifting agriculture to more intensive agricultural practices (Dobler-Morales et al., 2020). Although the differences in the selection of livelihood strategies are not considerable, we can see that off-farm activity, livestock, and fruit gathering play a more important role in SCH than in MFR. However, it is worth mentioning that diversification has little impact on income generation, with majority of families earning less than minimum wage. Cash crop cultivation is more notable in MFR than in SCH. Livelihood diversification and land use intensification at the smallholder level are seen as viable paths to accomplish forest transition (Rudel et al., 2020). The conservation agenda should consider agriculture-based livelihood strategies (Zafra-Calvo and Moreno-Peñaranda, 2018).

Forests in the study areas are used to provide plenty of products constituting an important component of households’ subsistence strategy and income generation (Guedes, 2008). Due to conservation status, trade of tree-derived products is reduced in MFR, so households earn money mainly from the sale of honey, cash crops, and off-farm activities. Our study shows that honey production is an important livelihood strategy for households in MFR, with 40% of families engaged in it. Beekeeping is also used as a protection against elephants (Virtanen et al., 2020). In the open-access area, forest product trade is more diversified, and 3% of households rely exclusively on forest exploitation for income. There is a moderate positive correlation between the perceptions about limitations in accessing the forest and the off-farm and on-farm strategies in MFR, indicating that households tend to adjust strategies to existing conditions. The analysis revealed only a weak effect of conservation on the households’ decisions in the open-access area and none in the protected area.

In MFR, off-farm activities, although developed by a smaller number of households than that in SCH, play an important role in reducing deforestation. The time spent outside indicates less time for agricultural activity within the household, and the earnings help with expenses. We could not see this correlation in the open-access area, where increased job opportunities do not reduce deforestation, as seen by Angelsen (2010). African rural households tend to diversify, rather than specialize their livelihoods, and wages are seen as an additional activity to buy goods that the household cannot produce (Alobo Loison, 2015). Off-farm activities decrease the probability for deforestation, whereas forest-based livelihoods correlate with increase in deforestation due to the extraction of forest resources for daily use. In our study, probit results showed that engagement in forest-based strategies could increase deforestation instead of reducing it, which is in line with previous findings (MacKenzie et al., 2017).

In attempting to reduce human/elephant conflict, decrease deforestation, and increase households’ income, the CNP authorities introduced beekeeping in the conservation area (Virtanen et al., 2020). Honey is currently the only NTFP commercialized in MFR. The lower income levels in MFR, compared with SCH, indicate little impact by the honey trade on household income, which is in line with other studies (Kimengsi et al., 2019). Although beekeeping is considered a viable activity that should decrease deforestation (Chanthayod et al., 2017; Mudzengi et al., 2020), our results show that, in MFR, this is not yet observed. Attempts to use apiculture in MFR as an alternative income source and approach to reduce the forest clearing date to 1998, with the implementation of the conservation project on the Chimanimani area (De Matos, 2011). Although the honey production experienced commercialization problems, the beekeeping promotion was not abandoned by authorities and new attempts were made (De Matos, 2011; Virtanen, 2019). In 2018, the Sustainable Development National Fund again distributed hives to communities living in CNP, including MFR (FNDS, 2018). The positive impact of the apiculture on forest conservation will probably be observed in the future, but this aspect requires more and better studies. Lowore et al. (2018) found contradicting evidences about the role of beekeeping in forest conservation among different African countries and associated it to economic and social factors, mainly land use related. Evidence from a Tanzanian study on traditional beekeeping in a forest reserve shows no considerable difference on forest conservation among areas with and without beekeeping practice (Augustino et al., 2016). The difference in income distribution shows that efforts to add honey production as an income source in the conservation area do not translate into better income for households and their willingness to protect forests. The financial motivation is pointed out as an important factor influencing the willingness to conserve forest (Sutcliffe et al., 2012; Musinguzi et al., 2018). This shows that the implementation of conservation policies should better address the socioeconomic aspects and, principally, poverty reduction (Miller et al., 2021). Connecting the preservation of natural resources with the economic utility of resources as attractive assets is—in theory—a good strategy. It produces incentives for local communities and allows them to be a part of an economic network (Dondeyne et al., 2012). However, despite the importance of forest products for livelihoods, agricultural production has more economic value for households in the study areas.

Our study observed increasing deforestation in both areas, especially in the protected area, despite conservation efforts from government and civil society agencies. This corroborates other studies, which show limited effectiveness of conservation areas to halt deforestation (Shah and Baylis, 2015; Spracklen et al., 2015; Wade et al., 2020). Reduction of the forest area in both study areas can be attributed, among other reasons, to the underlying driver of population growth, with agriculture as the main direct driver, which is in line with previous studies on the drivers of deforestation (Ryan et al., 2014; Sitoe et al., 2016; MITADER, 2018a). Babigumira et al. (2014) found that an increase in agricultural area augments the probability for forest clearing if the farm size is 3 ha or less and that increases in the size of the plot heighten the probability of deforestation. Households intensify their exploration of forest resources, including conversion to agriculture, when there are limited opportunities to generate income (Angelsen et al., 2014; Belcher et al., 2015). The results are also in line with the findings of Babigumira et al. (2014), which indicate that access to markets and the existence of good road connections are important factors in forest cover change in the initial phases of forest transition, because agriculture is more rewarding than low-disturbance forest exploration. Difficulty in reaching a market for agricultural products discourages production, thus reducing the need to increase the cultivated area (Schmook and Vance, 2009). In the study areas, access to roads and markets determines the feasibility of selling products. This is notable, especially in SCH, where the population lives in hard-to-reach areas with no passable roads. The feeder road connecting Chôa-Sede and Catandica, the district headquarters, does not allow safe traffic during the rainy season. The inhabitants from communities must walk long distances to reach Chôa-Sede and use animal traction to transport goods. The lower deforestation rates in SCH point to the lower opportunities for trade due to weak road infrastructure, as well as a preference for economic activities other than agriculture, notably livestock breeding, which does not require the clearing of forest since open grasslands are available. In particular, we see the influence of road improvement in MFR and better trade opportunities as a trigger to increase agricultural areas. Our model showed that access to roads leading to better trade is a more relevant aspect for households than conservation policies. The forest cover reduction is most notable for MFR after 2010, which is especially interesting. In 2010, the CNP management plan was drawn up, preceded by community consultations and awareness campaigns (Ghiurghi et al., 2010). It could be expected that extensive awareness campaigns and revitalization of conservation area would have a positive impact and reduce deforestation after 2010. The increased deforestation between 2014 and 2017 in MFR probably occurred due to improvement in the main road, which triggered the commerce of agricultural products, especially bananas (Eriksson, 2020). Good road connection and proximity with large banana production areas influence the choice of households in MFR to engage in agricultural activities (Guedes, 2008).

The existing conservation policies only reduce the economic opportunities but do not decrease the attractiveness of conservation areas for a new settlement. Once someone is assigned a piece of land by traditional authorities or receives land from a family member, that person finds himself/herself entitled to change land use/cover according to necessities. Considering the existence of large areas covered by forests and the tendency for population to increase, there is a great probability that deforestation will continue as the need for new agricultural plots will increase.