Guardians of the Forest: The Impact of Indigenous Peoples on Forest Loss in Chile

Abstract

The objective of this paper is to contribute to the understanding of forest cover loss patterns and the protection role of Indigenous peoples in the forests of Araucanía, Chile. Previous research indicated lower rates of forest cover loss in land managed by Indigenous peoples; however, this was primarily focused on tropical forests. This paper focuses on the temperate forests in the region of Araucanía and hypothesizes that there will be a similar trend, with lower rates of deforestation in areas owned by Indigenous peoples. A logistic regression model was used which included multiple underlying drivers that have shown to impact deforestation rates. The results of this study corroborated the hypothesis that lands owned by Indigenous peoples have lower rates of deforestation, and that protection status, agricultural function, and railway proximity have a strong influence on forest clearing, while slope, elevation, and proximity to urban areas demonstrated a minimal impact.

1. Introduction

Nobel-winning Chilean poet, Pablo Neruda once famously declared, “Anyone who hasn’t been in the Chilean forest doesn’t know this planet” [1]. Indeed, Chile’s vast territory, particularly the “Chilean Winter Rainfall–Valdivian Forests,” stands out as a biodiversity hotspot [2]. The area is enclosed by the Pacific Ocean, the Andes Mountains, and the Atacama Desert. This has resulted in the area displaying island-like characteristics, fostering high levels of endemism [3,4]. Neruda’s sentiment extends beyond nature to encompass the Indigenous peoples inhabiting the Chilean forest. Indigenous peoples are communities that share ancestral ties to the lands they live on and make up a fifth of the population of Latin America [5]. In the case of Chile, Indigenous people make up about 10% of the national population, the main groups being the Mapuche, Aymara, and Diaguita [6]. The Mapuche are the most numerous of these groups and constitute the majority of the Indigenous communities residing in the Valdivian forests [7].

Unfortunately, this particular biodiversity hotspot is one of 36 such places globally that is severely threatened by human activities like logging and plantation forests [8,9], which have led to a significant decline in both biodiversity and ecosystem services [2]. With over 70% of primary native vegetation already lost and on average another 77,000 hectares of forest disappearing annually since 2000 [10], both Chile’s forests and its Indigenous peoples are under threat, as Indigenous territories are rapidly disappearing along with the forest [11]. The political marginalization of Indigenous groups further exacerbates their vulnerability [12]. These communities have frequently engaged in conflicts with the Chilean state over land rights, which have often turned violent [8,13].

Despite their marginalized position, the Indigenous communities can play a key role in conserving the Chilean forests. In 2021, the ‘Forest Governance by Indigenous and Tribal Peoples’ report was released by the Food and Agriculture Organization of the United Nations (FAO) and the Fund for the Development of Indigenous Peoples of Latin America and the Caribbean (FILAC), recognizing Indigenous peoples as vital stewards of forests. The report states that Indigenous territories are key for conserving biodiversity and mitigating climate change, which is especially relevant for biodiversity hotspots [14]. Their resource management strategies, in combination with traditional practices, have been getting more recognition for their ability to improve sustainability [15]. Further research on this topic can expand the understanding of forest cover loss patterns, which can contribute to the design of policies and give valuable insight into where conservation efforts can be most effective [16]. The UN report (2021) mostly focused on the Amazon basin, while research has suggested that forest cover loss patterns can differ greatly between countries and even regions [17]. This becomes especially relevant, considering that the Chilean Winter Rainfall–Valdivian forests, unlike the Amazon basin, consist of temperate forests, which is the preferred biome for land conversion and agriculture [18].

Indigenous peoples are not a homogenous group; they are diverse, and their practices vary. Within the Mapuche community, for example, there is significant diversity in how individuals and subgroups interact with their environment [19]. However, many Indigenous communities share common principles that promote environmental stewardship. A prime example is the Mapuche people’s belief in the interconnectedness of all elements, which underpins their conservation practices [20]. This holistic approach ensures that their activities do not deplete natural resources, allowing ecosystems to be able to regenerate. These activities include sustainable land management practices, such as agroforestry [19]. However, the Mapuche have faced significant challenges in keeping their ancestral lands. Since the 1940s and 1950s, structural factors have driven a large portion of the Mapuche population to migrate to urban areas, with 70–80% now residing in cities such as Santiago [21]. This migration was partly due to the degradation and reduction of community lands caused by the land tenure policies imposed by the Chilean state. These policies led to the expropriation, usurpation, and irregular sale of Mapuche lands, forcing these people, especially the younger community members, to seek livelihoods in urban environments. Sadly, even in urban areas, the Mapuche have faced poverty, discrimination, and social marginalization [21,22]. The sustainability of Indigenous forest management depends significantly on whether traditional knowledge and practices are passed on to future generations. With a substantial portion of the Mapuche population living in urban areas, there is a real risk that these skills and knowledge could be lost. While some urban Mapuche communities actively work to preserve their cultural heritage, the physical disconnection from their ancestral lands presents a challenge [21,23]. The Chilean state has often failed to value or incorporate Indigenous knowledge in its environmental policies. This oversight is evident in the state’s forestry policies, which have prioritized economic interests over ecological and cultural sustainability [19]. Although the Chilean government has the intention of preserving their forests and their biodiversity, the economy’s heavy reliance on natural capital undermines these efforts, as evidenced by 70% of Chilean agricultural and forestry plantations being situated within the biodiversity hotspot [24,25]. This has prompted these communities to engage in activism to reclaim their ancestral lands through land occupation and legal processes [26]. Therefore, the recognition and support of their land rights will be vital in sustaining their role as true guardians of the forest.

In such a case, Indigenous communities could play a significant role by drawing upon their wealth of natural and cultural knowledge in order to more effectively protect biodiversity while simultaneously reducing the disadvantaged position of these communities. The purpose of this study is to contribute to the understanding of forest cover loss patterns in relation to Indigenous peoples in the forests of Chile. To understand the role of Indigenous communities in forest conservation, this study investigates the following research question: What is the impact of Indigenous land status on forest cover loss in the Araucanía region of Chile during the period of 2000–2020?

This study focuses on the Araucanía region, a part of the Chilean Winter Rainfall–Valdivian Forests and home to numerous Indigenous communities (Figure 1). It is hypothesized that there are lower rates of forest cover loss in areas managed by Indigenous peoples. Through a logistic regression analysis spanning 2000–2020, Indigenous peoples’ role in forest cover loss is assessed alongside other drivers to provide evidence and support conservation strategies. Unlike many previous studies, this model explicitly includes the location of lands managed by Indigenous peoples as a variable. This is particularly relevant, as previous research indicates that Indigenous-managed lands often have lower rates of deforestation, but this has not been extensively studied in temperate forests like those in Araucanía. By testing different distances to drivers of deforestation, the model has been specifically tailored to the unique characteristics of the Araucanía region. This customization ensures that the model accounts for local factors that may influence deforestation patterns differently than in other regions or forest types.

2. Materials and Methods

2.1. Study Area

The study area, the Araucanía region, is located in central Chile (Figure 2a) and covers an area of 3,181,791 hectares extending from 39°61′ S 73°53′ W to 37°59′ S 70°76′ W. The area is part of the Valdivian biodiversity hotspot and is surrounded by mountains in the east and the Pacific Ocean in the west. The area has impressive temperate forests, with unique plant species expediated by the higher precipitation in the south of Chile. Araucanía has a combination of land uses and includes extensive rotations of livestock and grain production [28]. It is also the home of large communities of Indigenous peoples who are spread out over the area (Figure 2b). Most of Chile’s Indigenous communities live in villages scattered across Araucanía and are mostly found in the Western parts of this region [29]. This region has experienced high rates of forest cover loss and habitat fragmentation, which are significant threats to ecosystems and their biodiversity [30,31,32,33,34]. The area is known for its conflicts between these communities and the growing number of tree plantations that can be found in this area [35,36]. A large area in the east of this region is considered a protected area known as Araucarias (Figure 2c). It is managed by Corporación Nacional Forestal (CONAF) and occupies a territory of 1,142,850 hectares, representing over 30% of the whole region. It includes nine mountain areas and thus has significant variations in elevation and slope throughout the region [37].

2.2. Determining Deforestation Variables

Spatial modeling of deforestation offers valuable insights into the drivers behind local trends, aiding decision-makers and managers [42,43]. Forest loss across Latin America results from a mixture of political, economic, social biophysical, and ecological factors [17,44]. Therefore, selecting the appropriate variables for modeling forest loss in Chile is critical, considering the variability across landscapes. While variables may vary from one landscape to another, common factors were identified in previous studies conducted by Arekhi (2013) [45], Bax et al. (2016) [46], Felicisimo et al. (2002) [42], Heilmayr et al. (2016) [28], Kaimowitz et al. (2012) [47], Kucsicsa and Dumitrică (2019) [48], Ludeke et al. (1990) [49], Pir Bavaghar (2015) [43], and Sharma et al. (2020) [50]. Factors such as elevation, slope, distance to roads, waterways, urban areas, agricultural lands, and the presence of protected areas and Indigenous lands were examined (Table 1). The expansion of agriculture stands as the most commonly cited cause of global forest loss [16,51,52]. There has also been an increase in forest plantations, which have become important direct drivers of native forest loss [16,32,53]. In 2010, plantations in Chile were more than 7% of the global forest area [54]. Tree plantations were, in most cases, established on already cleared agricultural land, but it has also become common practice to expand on these areas by removing native forests [55]. Hence, we have included the “distance to agricultural lands” in our model.

Infrastructure, crucial for transporting logging materials, is another significant driver, underscoring the inclusion of road, railway, and river distances as variables. While roads and waterways are included in most studies as important variables, railways as a variable have been uncommon in most studies. Interestingly, studies by Mamingi et al. (1996) [56] and Kaimowitz et al. (2012) [47] have produced significant results, suggesting that forests are more likely to be cleared in proximity to railways.

Urban expansion also contributes to deforestation, often resulting in the clearance of native forests to accommodate development, and several studies have integrated urbanization into their models [47,48,57,58].

Certain physical characteristics make an area prone to deforestation. Armenteras et al. (2003) [51] reported that in ‘lowland transitional areas’, deforestation is most active. Land-use decisions, such as agricultural expansion or plantation establishment, consider factors like elevation, slope, and rivers [50]. Higher elevations and steeper slopes can hinder forest clearance due to rugged terrain and challenges with the accessibility of the area, making these important variables.

This model also incorporates variables that represent information on potentially mitigating effects on deforestation, such as protected areas and Indigenous lands. Despite only being included in the deforestation model of Kaimowitz et al. (2012) [47], Indigenous territories have the potential to be an important consideration in areas with a large number of Indigenous communities. While the same can be said for protected areas, which are counteracting deforestation by turning large areas of forests into national parks and reserves, much is yet unknown about the effectiveness of these protected areas on forests in Chile [59,60].

Table 1. Variable information included in the logistic regression model.

Name	Source	Units	Descriptive
Forest cover loss *	Hansen et al., 2022 [38]	Categorical	Deforested = 1; Non-deforested = 0
Indigenous lands	Ministerio de Desarrollo Social de Chile, 2015 [39]	Categorical	Indigenous land = 1; Non-Indigenous land = 0
Protected area	Protected Planet, 2020 [40]	Categorical	Protected area = 1; Non-protected area = 0
Roads	Biblioteca del Congreso Nacional de Chile, 2020 [61]	Categorical	Within 5 km of a road = 1 More than 5 km away from a road = 0
Railways	Biblioteca del Congreso Nacional de Chile, 2020 [61]	Categorical	Within 10 km of a railway = 1; More than 10 km away = 0
Waterways	Humanitarian OpenStreetMap Team, 2021 [62]	Categorical	Within 1 km of a waterway = 1; More than 1 km away = 0
Urban areas	Biblioteca del Congreso Nacional de Chile, 2020 [61]	Categorical	Within 10 km of an urban area = 1; More than 10 km away = 0
Agricultural land	Xiong et al., 2017 [41]	Categorical	Within 500 m of agriculture = 1; More than 500 m away = 0
Slope	Jarvis et al., 2008 [63]	Degrees	Min = 0, max = 82.6, mean = 5.6
Elevation	Jarvis et al., 2008 [63]	Meter	Min = −22, max = 3732, mean = 743.9

* Dependent variable.

Table 1. Variable information included in the logistic regression model.

Name	Source	Units	Descriptive
Forest cover loss *	Hansen et al., 2022 [38]	Categorical	Deforested = 1; Non-deforested = 0
Indigenous lands	Ministerio de Desarrollo Social de Chile, 2015 [39]	Categorical	Indigenous land = 1; Non-Indigenous land = 0
Protected area	Protected Planet, 2020 [40]	Categorical	Protected area = 1; Non-protected area = 0
Roads	Biblioteca del Congreso Nacional de Chile, 2020 [61]	Categorical	Within 5 km of a road = 1 More than 5 km away from a road = 0
Railways	Biblioteca del Congreso Nacional de Chile, 2020 [61]	Categorical	Within 10 km of a railway = 1; More than 10 km away = 0
Waterways	Humanitarian OpenStreetMap Team, 2021 [62]	Categorical	Within 1 km of a waterway = 1; More than 1 km away = 0
Urban areas	Biblioteca del Congreso Nacional de Chile, 2020 [61]	Categorical	Within 10 km of an urban area = 1; More than 10 km away = 0
Agricultural land	Xiong et al., 2017 [41]	Categorical	Within 500 m of agriculture = 1; More than 500 m away = 0
Slope	Jarvis et al., 2008 [63]	Degrees	Min = 0, max = 82.6, mean = 5.6
Elevation	Jarvis et al., 2008 [63]	Meter	Min = −22, max = 3732, mean = 743.9

* Dependent variable.

2.3. Spatial Analysis

The methodology is comprised of three main stages: (1) data collection, pre-processing, and spatial analysis of the variables using GIS, (2) statistical analysis using SPSS, and (3) validation of the results. As discussed in the previous section, Indigenous lands, protected areas, urbanization, roads, railways, waterways, agriculture, elevation, and slope were identified as important factors influencing deforestation in Chile. The data collection process included a thorough analysis of the accuracy of the datasets that were available.

The data for the dependent variable, deforestation was retrieved from Hansen et al. (2022) [38]. This included a rasterized layer of years of gross forest cover loss from 2000 to 2021. To function as the dependent variable, the dataset was reclassified, where 1 meant deforested and 0 meant forested. Similar reclassification methods were applied to the Indigenous lands and protected area variables. The data used to represent Indigenous lands were retrieved from the Ministry of Social Development of Chile [39]. This dataset included polygons of areas that were acknowledged by the government as being owned by Indigenous communities. In the case of elevation, a digital elevation model (DEM) was retrieved from CGIARCSI that represented the elevation in Araucanía [63]. The Slope variable was calculated from the elevation dataset using the planar slope calculation in ArcGIS Pro (Esri, 2024) [64].

Proximity variables were transformed into dichotomous variables through a proximity analysis using the GIS tool ‘Euclidean distance’. Euclidean distance is a commonly used spatial analysis that represents the straight-line distance between two points in a plane and represents the distance at which a variable may have influence on deforestation observations. Based on the literature summarized below, various distances were explored to assess their impact on the results, with different distances chosen for each variable. For instance, studies indicate that forests are more likely to be cleared when near roads [47,49,56,65,66,67,68]. However, there is no consensus on the specific distance at which this effect occurs, with varying conclusions among different studies. Some suggest that the forest clearing occurs within a distance of 50 km [65,69,70] or even 100 km [71] Barber et al. [72] argue that such large distances lack precision, as larger areas can encompass up to 60% of the study area, making it challenging to attribute deforestation solely to roads without considering other factors.

Roads are an important driver of deforestation because they make an area more accessible [43,45,48]. Yet, roads are not the only factors that increase accessibility. Secondary routes, such as waterways and railways, also play a significant role in making areas more accessible [48,49]. Barber et al. [72] observed that the majority of deforestation in the Amazon occurs within 1 km of rivers or 5.5 km of roads, while Kucsicsa and Dumitrică [48] found that 90% of maximum forest losses in the Romanian Carpathian Mountains were within 2 km of rivers. Therefore, a wide range of distances was tested in the case of roads, rivers, and railways ranging from 1 km to 25 km with increasing 5 km steps.

Regarding urban areas, the model tested distances up to 20 km, as Echeverria (2008) [57] identified a clear increase in deforestation within this distance to urban centers.

For agricultural land, there is no available literature specifically addressing the distance to agriculture. However, it is reasonable to assume that expansion would occur close to existing agricultural lands due to their higher accessibility and typically higher soil quality. Therefore, various distances were selected for analysis, ranging from 0.5 km to 15 km, with increments of 0.5 km initially and then 1 km.

Table 2. Range and steps for the variables transformed to dichotomous for the logistic regression model.

Distance Variables	Distance Tested
Roads	1–25 km, with 5 km steps
Railways	1–25 km, with 5 km steps
Waterways	1–25 km, with 5 km steps
Urban areas	1–20 km, with 5 km steps
Agricultural land	0.5–15 km, with 0.5 (0.5–2 km)/1 km steps (2–15 km)

lists the variables entered as dummy variables, with their range and steps tested in the model.

2.4. Logistic Regression Analysis

Variable information was extracted from random sample points within the region. A sample size of 5000 random locations was used, as it allows for balancing the aim of obtaining comprehensive coverage and statistical reliability while mitigating spatial autocorrelation. This is in line with sampling strategies employed by similar studies, such as Sharma et al. (2020) [50], who also looked at similar spatial variables of deforestation in the Pathro river basin and used a sample size of 4000 points in a smaller study area.

To ensure representative sampling, a stratified random sampling approach was used, ensuring an equal number of random points were selected from both forested and deforested areas [73]. This was achieved by utilizing the tree cover dataset for the year 2000 and dividing the data into distinct strata based on forest cover status. Additionally, a minimum distance between points was implemented to further mitigate spatial autocorrelation, because samples taken in close proximity are not considered independent [73]. A distance of at least 1000 m between points was chosen to minimize the impact of spatial autocorrelation, as suggested by previous studies by Linkie et al. (2004) [74] and Pir Bavaghar (2015) [43].

For each sample point, information for all variables (Table 1) was extracted, and a logistic regression analysis was conducted using the Statistical Package for the Social Sciences (SPSS 28.0) software. The objectives were to: (i) determine the contribution of each explanatory variable to forest cover loss between 2000 and 2020; and (ii) specifically investigate the impact of Indigenous lands on forest cover loss. Logistic regression was chosen given the dichotomous nature of the dependent variable (deforested/not deforested) and the numerous independent variables involved. It is similar to multi-linear regression; however, logistic regression relies on the concept of the “odds” of an event happening, which is exponential and not linear [75]. As described by Schneider and Pontius (2001) [76], logistic regression yields a monotonic curvilinear response ranging from 0 to 1:

$$p=E\left(Y\right)=\frac{exp exp \left(\beta 0+\beta 1X1+\beta 2X2+\cdots +\beta iXi\right) }{1+exp (\beta 0+\beta 1X1+\beta 2X2+\cdots +\beta iXi)}$$

(1)

where p is the probability of deforestation, E(Y) is the expected value of the dependent variable Y, β0 is the constant, and βi is the predicted coefficient of each independent variable Xi.

Logistic regression analysis is commonly used for deforestation prediction models due to its effectiveness in handling binary response variables, such as the presence or absence of deforestation [43,45,77,78]. This approach is particularly advantageous when the goal is to statistically relate deforestation occurrence to various predictor variables or covariates. Logistic regression’s popularity in deforestation research stems from its robustness and ease of interpretation [79]. One of its main advantages is the ability to provide clear probabilities of deforestation based on the predictors, which aids in understanding the influence of different factors. However, it is not the only model available; other methods, such as random forests or neural networks, can also be used for deforestation prediction [80,81]. While logistic regression offers simplicity and interpretability, it may not capture complex, non-linear relationships as effectively as some machine learning models. Nonetheless, its frequent use and proven reliability make it a standard choice for many deforestation studies [79].

2.5. Model Validation

The final step involved ensuring the robustness of both the model and the variables incorporated. To assess goodness-of-fit, the Nagelkerke R² was calculated, providing insight into the model’s effectiveness in explaining the dependent variable [82]. This metric is used to evaluate the goodness-of-fit of logistic regression models, and it indicates the extent to which the variables account for the variance in the dependent variable [78]. The Nagelkerke R² is easy to interpret due to its scale from 0 to 1, making it a comparable metric for assessing model performance, as it indicates the proportion of the variation in the dependent variable that is explained by the independent variables in the model. A higher Nagelkerke R² value suggests a more comprehensive model. The Wald test is used to further analyze the significance of the coefficients βi in the model. This is achieved by ‘comparing the maximum likelihood estimate of every βi with its estimated standard error’ [83,84]. Thus, the higher the Wald statistic, the more significant the coefficient is.

3. Results

Table 3. Logistic regression model results with Nagelkerke R² = 0.414 (Χ² = 1859.261; p <0.0001).

Variables	β	S.E.	Sig	Exp(B)
Slope ***	−0.016	0.005	<0.001	0.966
Roads	0.452	0.317	0.154	1.571
Railway ***	0.402	0.088	<0.001	1.495
Rivers *	−0.255	0.084	0.002	0.775
Elevation ***	−0.001	0.000	<0.001	0.998
Indigenous lands ***	−1.189	0.131	<0.001	0.324
Protected areas ***	−1.511	0.125	<0.001	0.200
Urban areas ***	0.000	0.000	<0.001	1.000
Agriculture ***	1.209	0.224	<0.001	3.350

Significance levels: * p < 0.05, *** p < 0.001.

shows the results of the logistic regression model and the contribution of each independent variable to the deforestation rates in our area of interest.

Notably, protected areas emerged as the most influential predictor. Being in a protected area decreases the odds of deforestation by 80%. Similarly, “Indigenous lands” exhibited a negative coefficient (β = −1.189), which means a decrease in the odds of deforestation by 67.6% in areas owned by Indigenous communities recognized by the Chilean government.

Conversely, the model indicated a heightened deforestation risk within 500 m of agricultural fields, with an increase of 235%. It also showed a positive relation for railways within 10 km, indicating that the chance of an area being deforested increases by 49.5% when within 10 km of railways. This is a noticeably larger distance than the other variables. This threshold was discovered via iterative attempts, by keeping all other variables the same but interactively increasing the threshold of railways from 1 to 10 km (only at a 10 km distance did this factor show significance).

Waterways yielded a negative influence, which means that the probability of an area being deforested was lower within 1 km of a river (β = −0.249). = Both the variables “slope” and “elevation” produced a negative relationship to deforestation, indicating a decreased likelihood of deforestation occurring on steeper terrain and at higher altitudes (β = −0.016 and −0.001 respectively). Of interest was that the proximity of urban areas showed no impact on deforestation rates (β = 0.000). “Distance to roads” failed to achieve significance, despite varied attempts at distance thresholds and other adjustments. Model validation revealed a Nagelkerke R² of 42.2%, which is a moderate explanatory power. Although relatively low, the model correctly classified 76% of cases and exhibited significant Chi-square values (1859.261, df = 9, p = 0.000), affirming its goodness of fit. Wald tests confirmed non-zero coefficients for all variables, and only “roads” and “rivers” were low, which aligns with the coefficients and the p-values. Overall, considerable effort was invested in refining variable influence within the model.

4. Discussion

This study aimed to develop a site-specific model using logistic regression to assess the impact of Indigenous lands on deforestation rates, as well as multiple variables that are commonly employed in deforestation research. The findings of the model confirmed our hypothesis that areas under Indigenous ownership exhibit lower rates of deforestation. This aligns with the findings of the FAO’s 2021 report on “Forest Governance by Indigenous and Tribal Peoples,” which revealed that nearly half of the forests managed by Indigenous communities remain fully intact, with generally lower deforestation rates compared to other forested areas. Kaimowitz et al. (2012) [47] similarly examined land owned by Indigenous peoples, officially recognized by the government in Santa Cruz, Bolivia; yet, it yielded no significant findings. This may stem from variations in governmental treatment of Indigenous peoples and land rights, influenced by factors such as corruption and social welfare policies. Nonetheless, this study showed significant results and adds to the already existing literature. This can strengthen the case for involving Indigenous peoples and their practices in forest management and conservation efforts.

A second noteworthy finding regarding conservation efforts in Chile is the observation that protected areas exhibit even lower deforestation rates compared to Indigenous lands. This aligns with the conclusions of Barber et al. (2014) [72], who similarly reported reduced deforestation rates in protected areas. This finding could be particularly valuable for the government of Chile, as it highlights the potential role of protected areas in reducing deforestation rates and preserving biodiversity in Chile. However, further research is necessary to pinpoint the exact causes behind these outcomes, especially since forest policies in Chile have not provided any economic incentive to conserve native forests. This suggests that other factors are actively contributing to the observed conservation success [59]. The disparity in deforestation rates between protected areas and Indigenous lands in this study may be attributed to the differences in political protection. While governmental land-use restrictions placed on protected areas deter parties who desire to use the land for practices like agriculture, for decades, Indigenous communities have experienced difficulties protecting their lands from these same parties, even resulting in conflicts with the Chilean government [85]. Another potential factor contributing to the effectiveness of protected areas is their relative inaccessibility, exemplified by the limited infrastructure found within these regions [48].

Agriculture was shown to have a significant impact on forest cover loss, as forests were more likely to be deforested due to agriculture expansion than any of the other variables included. Agriculture expansion has previously been cited as one of the main drivers of forest cover loss in Chile [16,51,52]. This is not surprising, given the fact that Chile is part of an area that produces a large amount of food crops [2]. An increase in intensive agriculture has thus led to forest clearing, which has significantly reduced the natural habitats of Araucanía, Chile [2].

Slope and elevation are two variables that have commonly been included in deforestation modeling research due to their negative relationship with forest cover loss, but they have not produced significant results every time [43,45,46,48,49,50]. In this study, the results of the slope analysis were similar to the results of Arekhi (2011) [45] when researching the Ilam forests in Iran, indicating that an increase in slope leads to a lower probability of deforestation. The coefficient for elevation appears low at first, as it was measured in meters. When transforming the variable to kilometers (β = −0.001 × 1000) it results in −1, meaning the odds of deforestation are e⁻¹ = 0.368. This means that the odds of deforestation decrease by 63% for every kilometre increase in elevation. This suggests that elevation indeed plays a crucial role in shaping deforestation patterns. Despite the diverging results observed in some studies [43,48], our findings align with the prevailing belief that higher elevations are associated with lower rates of deforestation.

When considering the results of previous studies, our model produced a number of unexpected results regarding infrastructure, underscoring the site-specific nature of deforestation drivers. The variables deemed significant in one region may not hold true in others. One such surprising outcome was the variable “roads”. While infrastructure is typically recognized as a key driver of deforestation, our analysis did not reveal a significant association between roads and deforestation rates in Chile, at least within the scope of this model. Despite testing various distances suggested by previous research, ranging from 2 km to 100 km, none yielded significant results. Even when considering shorter distances, the lack of significance persisted. Multiple datasets of road networks were employed to ensure data quality, yet the significance level of roads in our model remained above 0.071. A possible explanation is that Chile’s dense road network presents challenges in analyzing and distinguishing between different road types, while not all road types are suited for the extraction of wood. Future research could explore potential associations between specific road types and deforestation rates in Chile.

A similarly unexpected finding regarding the relationship between rivers and deforestation rates emerged from the model, as the analysis revealed a negative correlation. This suggests that closer proximity to a river correlates with a decreased likelihood of deforestation, contrary to existing literature [46,48,50,72]. This study initially adopted a distance of 1 km for rivers based on the findings of Barber et al. (2014) [72], while using 5 km for roads. However, this approach yielded conflicting results. This discrepancy may stem from a lack of distinction between the width and depth of rivers, which are critical factors in determining navigability by boats. Typically, forests bordering navigable rivers are more susceptible to deforestation due to easy access via water transport. The absence of comprehensive data on river characteristics hindered accurate differentiation and led to the inclusion of all rivers in the model. This limitation likely influenced the results, particularly given the stark contrast with previous research findings. However, rivers are also a source of conflict in Chile due to a number of parties wanting to exploit the rivers for projects like damming and hydroelectric power projects, all impacting the ability to traverse these waterways [86]. The granting of water rights to large corporations has curtailed state intervention in such conflicts [86]. While these conflicts may impact deforestation rates near rivers, they fail to fully explain the unexpected negative coefficient observed in this study.

A third component of Chile’s infrastructure that was incorporated into our model was the distance to railways. Unlike the roads and waterways, our analysis revealed that the likelihood of forest clearance increased within a ten-kilometer radius to a railway. These findings are consistent with the conclusions drawn by Mamingi et al. (1996) [56] and Kaimowitz et al. (2012) [47], underscoring the significant role of railways in forest cover loss. This observation suggests that while roads and waterways did not exhibit significance in our model, infrastructure does indeed influence deforestation rates in Chile. This could be due to the fact that railways are considered a better and more efficient way of transporting trees.

Finally, our variable “Urban areas” seemingly did not have an impact on deforestation rates in the last twenty years. The lack of impact of this variable could be explained by the fact that the urbanization rate in Chile has stayed virtually constant from 2000 to 2022. Like most parts of the world, Chile experienced a population shift from rural to urban areas. This resulted in an urbanization rate of 86% in 2000 [87]. However, since this time, the urbanization rate has stayed relatively constant, with 88% in 2020 [87]. Additionally, Bax et al. (2016) [46] noted that urban areas have an indirect effect on deforestation due to higher demands for resources such as food. This effect is therefore likely most apparent in the expansion of agriculture in an area. However, spatially, it can be hard to account for, except when cross-referencing the origins and destinations of agricultural produce. It would require tracking the flow of agricultural goods from rural to urban areas to help identify patterns of land conversion driven by urban consumption, which would involve sophisticated spatial analysis techniques to accurately map the indirect pressures caused by urban centers. For this reason, future deforestation models could consider leaving this variable out.

This study employed a similar approach to research such as Arekhi (2013) [45], Kaimowitz et al. (2012) [47], and Pir Bavaghar (2015) [43]. It replicates and validates the use of logistic regression models to analyze deforestation patterns, demonstrating its effectiveness in explaining deforestation patterns. One of the novel contributions of this research is the incorporation of Indigenous land ownership into the analysis of deforestation. While previous studies have often overlooked the role of Indigenous peoples in conservation, this research highlights the potential importance of their land management practices in conservation. It therefore not only broadens the scope of deforestation research but also highlights the value of Indigenous land management in biodiversity conservation.

The limitations of this study stem primarily from the constraints associated with data availability and the scope of variables included in the model. While the model was constructed using open data from reputable sources, this caused inherent limitations in the data selection process. Despite efforts to ensure data soundness by comparing available datasets and verifying their sources, challenges persisted in obtaining comprehensive data coverage. While the model provides valuable insights into deforestation patterns in Chile, its limitations highlight the need for further research and data refinement to enhance the comprehensiveness and accuracy of future models. Completeness could be improved by adding new variables and improving the accuracy of the data sets by using remote sensing techniques to extract data that are not readily Available online.

This study underscores the critical roles that Indigenous communities play in conserving forest ecosystems, particularly in the Araucanía region of Chile. The findings of this study contribute to strengthening the case for Indigenous land rights and forest management, both in Chile and around the world, highlighting the intertwined benefits of environmental conservation and the preservation of Indigenous cultures. it is crucial to strengthen legal frameworks to protect the land rights of Indigenous peoples. This includes formally recognizing Indigenous land tenure systems and taking robust measures to prevent illegal expropriations and sales of these lands. Additionally, improving the monitoring and enforcement of environmental regulations is essential to prevent activities that degrade biodiversity on Indigenous lands. These recommendations can help governments and authorities protect the biodiversity of Indigenous lands while supporting the cultural and socio-economic well-being of Indigenous peoples.

5. Conclusions

This study conducted a spatial analysis of deforestation rates from 2000 to 2020 using GIS and logistic regression techniques. It successfully validated the hypothesis that Indigenous-owned lands in Araucanía, Chile, exhibit lower rates of deforestation. The model developed in this study offers valuable insights into the underlying drivers influencing forest loss patterns in Chile. The results highlighted significant correlations between deforestation rates and variables such as Indigenous lands, protected areas, agriculture, and railways. Interestingly, topographical variables emerged as stronger influencers compared to accessibility variables. Notably, the model did not establish a significant relationship with distance to roads and waterways.

A better understanding of the influence of Indigenous stewardship of land on deforestation patterns can make the policy and decision-making process more effective for the Chilean government and empower Indigenous communities, as they are still a disadvantaged group. Although the logistic regression model exhibited relatively low goodness of fit, suggesting potential missing explanatory variables and incompleteness in explaining deforestation rates, statistical validation affirmed its significance. Therefore, the model offers crucial information that can benefit Indigenous communities while simultaneously combating forest loss in Araucanía, Chile.