Natural ecosystem loss in an Atlantic Forest agricultural basin: drivers of land-use change and future landscape trajectories

The Atlantic Forest, one of the world's most threatened biodiversity hotspots, faces intense pressure from agricultural expansion and inadequate land governance. Understanding the spatial drivers and future trajectories of land-use and land cover (LULC) change is critical for safeguarding the ecosystem services and biodiversity that remnant natural areas provide. This study investigated the relative importance of multiple biophysical, socioeconomic, and demographic drivers of natural ecosystem loss in the Guaporé River Basin (GRB), a 246,300-ha subtropical Atlantic Forest agricultural watershed in southern Brazil. We combined a municipality-level temporal trend analysis to characterize historical LULC dynamics (1998–2019) with the Dinamica EGO cellular automata model, calibrated using Weight of Evidence (WoE), to simulate future landscapes under 2030 and 2050 scenarios. Our results reveal divergent ecological dynamics across the basin. The Northern region experienced extensive remnant vegetation loss (1.55% yr −1 conversion to altered ecosystems), driven by flat terrain and high agricultural productivity, while the Southern region showed partial natural regeneration (1.40% yr −1 transition from altered to regenerating vegetation), linked to steeper slopes and vegetation extraction practices. Despite apparent recovery, closer analysis reveals a regressive process of ecological simplification, in which remnant vegetation is progressively replaced by regenerating vegetation. Future projections suggest that, without strengthened enforcement of Brazil's Natural Vegetation Protection Law, the Northern basin will lose an additional 15,903 ha of remnant vegetation by 2050. These findings highlight critical spatial mismatches between legal protection frameworks and actual land-use pressures, with direct implications for conservation planning, watershed management, and biodiversity protection in the Atlantic Forest biome.

Land-use and land-cover (LULC) change is one of the primary drivers of global environmental change. Over the past six decades, approximately one-third of the Earth's terrestrial surface has undergone some form of LULC transformation (Winkler et al., 2021), generating profound consequences for biodiversity, climate regulation, hydrological processes, and the provision of ecosystem services (Ellis and Ramankutty, 2008; Foley et al., 2011; Garibaldi et al., 2021). In tropical and subtropical regions, land-use change driven by the conversion of natural vegetation into pastureland's, agricultural expansion, urbanization, and infrastructure development has resulted in extensive habitat fragmentation, reduced carbon storage capacity, declining freshwater quality, and accelerated rates of species extinction (Bongaarts, 2019; Joly et al., 2014).

The Brazilian Atlantic Forest (Mata Atlântica) represents a particularly critical example of this global trend. Once covering approximately 1.3 million km 2 along the Brazilian coastline, this internationally recognized biodiversity hotspot has already lost between 84 and 89% of its original vegetation cover, leaving highly fragmented remnants concentrated predominantly on private lands (Joly et al., 2014; Ribeiro et al., 2009; Souza et al., 2020). Because nearly 70% of the remaining Atlantic Forest vegetation occurs on private properties located outside formally protected areas (Rezende et al., 2018), land governance mechanisms regulating private land use are central to long-term conservation efforts.

In Brazil, land use on private rural properties is regulated by the Natural Vegetation Protection Law (NVPL; Law n° 12,651/2012), which requires landowners to maintain both Legal Reserves (LR) and Areas of Permanent Preservation (APP). However, compliance with these regulations remains inconsistent, and enforcement mechanisms are often insufficient. Continued deforestation within APP areas and failure to maintain legally required reserve areas continue to undermine conservation objectives and weaken the effectiveness of environmental governance frameworks (Rajao et al., 2020; Rezende et al., 2018).

Understanding the drivers and spatial patterns of LULC change is therefore essential for environmental management, conservation planning, and sustainable land governance. Comprehensive LULC analysis requires simultaneously mapping spatial landscape transformations while identifying the most relevant biophysical, socioeconomic, and demographic drivers responsible for these changes. Recent advances in remote sensing (RS) and Geographic Information Systems (GIS) have substantially improved our capacity to map land-use dynamics and analyze spatiotemporal patterns of landscape change at increasingly large spatial scales (Amani et al., 2019; Bauer et al., 2021; Chasmer et al., 2020; de Bem et al., 2020; Souza et al., 2020). In parallel, emerging technologies now enable the generation of near real-time and high-resolution global LULC datasets, considerably improving our ability to monitor landscape conditions, long-term trends, and anthropogenic pressures affecting both natural and managed ecosystems (Brown et al., 2022; Souza et al., 2020).

Furthermore, increasing availability of geospatial datasets combined with advances in spatial statistics has stimulated the development of a wide range of spatially explicit models capable of analyzing complex landscape dynamics. These approaches combine different mathematical and statistical frameworks, including artificial neural networks, cellular automata (CA), agent-based models, and regression-based approaches (Liu et al., 2008; Noszczyk, 2018). Among these approaches, CA models have become particularly important because of their ability to simulate and predict the spatiotemporal evolution of complex geographical phenomena, including urban growth, forest dynamics, and LULC transitions in both agricultural and non-agricultural landscapes (Leite-Filho et al., 2021).

Cellular automata models represent landscapes as regular multidimensional arrays of cells, where each cell corresponds to a specific land-use type or landscape state. Through synchronized discrete time steps, cells update their states according to transition rules influenced by neighboring cells and environmental drivers (Soares-Filho et al., 2002). CA models can be implemented as stand-alone frameworks or integrated into environmental modeling platforms such as Dinamica EGO. Dinamica EGO (hereafter referred to as Dinamica) is a widely used environmental modeling platform that simulates land-use change through a rule-based framework integrating environmental variables, socioeconomic interactions, and decision-making processes (Soares-Filho et al., 2002). This framework combines map algebra, cellular automata algorithms, and tabular data processing to represent complex environmental and socioeconomic systems.

Several studies have successfully applied Dinamica for LULC modeling under different environmental contexts. Applications include modeling spatial conflicts generated by competing land-use demands in China (Liu et al., 2023), assessing the effectiveness of protected area programs in reducing deforestation within the Brazilian Amazon (Soares-Filho et al., 2023), and projecting future landscape dynamics within the Atlantic Forest biome (Molin et al., 2017). Although these studies have substantially improved our understanding of broad-scale deforestation dynamics, fewer studies have investigated land-use and land-cover dynamics at watershed and landscape scales in highly fragmented Atlantic Forest regions (Lira et al., 2012; Alvarenga et al., 2016), particularly considering the combined influence of environmental conditions and local socioeconomic drivers. In particular, limited attention has been given to understanding how multiple drivers interact spatially across sub-regions and how current land-use trends may influence long-term ecosystem persistence.

The objective of the present study is to investigate the relative importance of multiple land-use drivers in explaining recent natural ecosystem changes observed within a highly fragmented Atlantic Forest watershed located in the Guaporé River Basin (GRB), southern Brazil. Specifically, we: (i) assessed recent deforestation trends within the GRB and across municipality boundaries; (ii) evaluated the influence of multiple biophysical, socioeconomic, and demographic drivers on observed LULC dynamics; and (iii) modeled and projected future land-use scenarios for 2030 and 2050 based on current transition trends. We hypothesize that: (a) historical spatiotemporal changes in natural ecosystem cover have disproportionately affected remnant vegetation, leading to its progressive replacement by regenerating vegetation; (b) legally protected native vegetation located on private lands continues to experience measurable land conversion pressure, indicating that the current Brazilian environmental legal framework, particularly the Natural Vegetation Protection Law, remains insufficient to ensure the long-term conservation of remaining natural ecosystems; and (c) biophysical factors associated with agricultural suitability, together with socioeconomic factors related to agricultural production and land management, are expected to exert an important influence on LULC changes.

Study area

The study area is the Guaporé River Basin (GRB), covering approximately 246,300 ha in the state of Rio Grande do Sul, southern Brazil. The Guaporé River is part of the Jacuí River system, which supplies freshwater resources to the Metropolitan Region of Porto Alegre, the capital of Rio Grande do Sul (Figure 1). The native vegetation covered approximately 38% of the GRB, comprising predominantly Atlantic Forest formations (approximately 37% of the basin area) and Pampa formations (approximately 1%) (Souza et al., 2020).

Currently, more than 60% of the basin area has been subjected to human influence, mainly consisting of agricultural lands, pastures, and scattered urban infrastructure (Souza et al., 2020). The extent of anthropogenic activities, combined with the limited adoption of sustainable agricultural practices (Tiecher et al., 2017) and increasing fragmentation of natural ecosystems, represents a major challenge for sustainable landscape management in the region. These pressures threaten both natural resources and the long-term provision of ecosystem services, particularly freshwater availability (Bastos et al., 2021; Becker et al., 2009; Ferreto et al., 2021; Kaiser et al., 2010, 2015; Reichert et al., 2021b; Valente et al., 2021) and the maintenance of productive agricultural soils (Ambus et al., 2018; Reichert et al., 2021a; Tiecher et al., 2017).

The GRB is located within two climatic zones according to the Köppen climate classification system: humid subtropical climate with hot summers (Cfa) and subtropical highland climate with uniform rainfall (Cfb). Temperatures typically range between −3 and 22°C, while mean annual precipitation varies between 1,600 and 2,200 mm (Alvares et al., 2013). Geologically, the basin is predominantly underlain by volcanic rocks of the Serra Geral Formation, comprising basic to acidic volcanic rocks distributed among different geological facies, including Caxias, Gramado, and Paranapanema (CPRM–Servico Geologico do Brasil, 2006). Basin elevation ranges from 49 to 836 m above sea level, resulting in substantial variation in terrain slope and topographic conditions. According to the World Reference Base (WRB) classification system, the dominant soils within the GRB belong to five major soil groups: Acrisols, Regosols, Ferralsols, Luvisols, and Nitisols.

Data sources and data preparation

Initial landscape data

To simulate historical land-use and land-cover (LULC) dynamics, land-use maps corresponding to the years 1998 and 2019 were obtained from the MapBiomas Collection database (available at http://plataforma.brasil.mapbiomas.org), which provides reliable and highly accurate LULC maps for Brazilian biomes, with reported overall classification accuracy reaching up to 96.6% at Level 1 (Souza et al., 2020). This dataset reconstructs annual LULC information at 30-m spatial resolution for all Brazilian biomes using a Random Forest classification algorithm applied to the Landsat image archive through the Google Earth Engine platform.

Each annual map consists of six major LULC classes: Forest, Non-Forest Formation, Farming, Non-Vegetated Area, Water, and Not Observed. However, for the objectives of this study, the original classification scheme was adapted into two broader ecosystem categories: Altered Ecosystems and Natural Ecosystems.

The Altered Ecosystems category included all pixels originally classified as Forest Plantation, Pasture, Agriculture, Mosaic of Agriculture and Pasture, Non-Vegetated Area, and Urban Infrastructure. In contrast, the Natural Ecosystems category included Forest Formation, Non-Forest Natural Formation, Wetland, Grassland Formation, Rocky Outcrop, and Other Non-Forest Formations. The distinction between Non-Forest Natural Formation and Other Non-Forest Formations follows the original hierarchical classification adopted by MapBiomas; both classes were aggregated into the Natural Ecosystems category in this study.

In a subsequent step, the Natural Ecosystems class was further subdivided into two vegetation categories: Remnant Vegetation and Regenerating Vegetation. Remnant vegetation was defined as areas that remained continuously classified as natural vegetation throughout the historical MapBiomas record used in this study. Regenerating vegetation was defined as areas that had previously been classified as altered ecosystems and subsequently transitioned back to natural vegetation. For the 1998 baseline map, the MapBiomas annual series from 1988 to 1997 was used to distinguish remnant from regenerating vegetation.

This additional reclassification was performed using a raster reclassification procedure in order to better capture long-term anthropogenic disturbance dynamics and quantify the relative susceptibility of persistent and regenerating vegetation to land-use transitions over time. The final reclassified land-use maps used in the analysis are presented in Figure 2.

The adoption of a simplified classification scheme dividing the landscape into Natural Ecosystems and Altered Ecosystems was a deliberate methodological decision motivated by the central objective of this study: to model the net conversion of natural habitats into anthropogenic land uses, and vice versa, as the ecologically most critical transition within the Atlantic Forest context.

Although a more detailed classification scheme, for example distinguishing forest subtypes or different categories of agricultural land use, could provide additional ecological detail, it would substantially increase model parameterization complexity and reduce statistical robustness given the spatial extent of the study area. The ecological relevance of this simplified classification is strongly supported by previous research demonstrating that, in highly fragmented landscapes such as the GRB, even relatively small losses of natural vegetation can produce disproportionate impacts on habitat connectivity, edge effects, and biodiversity conservation (Ribeiro et al., 2009).

The additional subdivision of the Natural Ecosystems class into persistent and regenerating vegetation provides the ecological resolution necessary to identify and track processes of ecological simplification, which emerged as one of the central ecological findings of this study. Potential limitations associated with this classification framework are discussed in Section 4.

Driving forces

To replicate the historical pattern of annual land-use and land-cover (LULC) change between 1998 and 2019, we quantified the empirical relationships between observed LULC transition events and a suite of spatial driver variables using the Dinamica LULC model (see model details below). These driving force variables were selected based on a literature review (Gomes et al., 2020; Hersperger et al., 2010; Richards, 2011; Stan et al., 2015), and data were collected from nationwide surveys and official databases maintained by Brazilian public institutions, including the Brazilian Institute of Geography and Statistics (IBGE), the National Institute for Colonization and Agrarian Reform (INCRA), and the Rural Environmental Registry (SICAR; Table 1), at the municipality level in multiple formats. The selected variables were grouped into three categories: biophysical, socioeconomic, and demographic.

All spatial driver variables were harmonized to a common spatial resolution of 30 × 30 m per pixel and projected in the WGS84 coordinate system to ensure that all raster layers contained the same number of rows and columns as the reference LULC map. Bilinear interpolation was used to resample continuous variables, whereas nearest-neighbor resampling was used for categorical variables. Subsequently, the Functor Cube algorithm implemented in Dinamica was used to combine all auxiliary variables into a single multilayer raster structure. In terms of temporal resolution, the MapBiomas LULC maps used as model inputs represent annual snapshots for the years 1998 and 2019. The socioeconomic and demographic variables derived from Brazilian national surveys (IBGE, INCRA, and SICAR) are updated at decadal (census) or annual (agricultural production) intervals.

Biophysical variables were incorporated into the model as either continuous or categorical predictors. Continuous variables included the Digital Elevation Model (DEM) and hydrography layers, where only perennial watercourses were retained and watercourse density was calculated using the Line Density function in ArcGIS. Categorical variables included slope classes defined as Flat (0%–3%), Gentle (3%–8%), Moderate (8%–20%), Steep (20%–45%), Extremely Steep (45%–75%), and Excessively Steep (>75%) (Santos et al., 2018). Additional categorical variables included aspect classes (N, NE, E, SE, S, SW, W, and NW) and soil order classifications.

Socioeconomic and demographic variables were classified into categories and assigned weights proportional to the relative basin area represented by each municipality within the Guaporé River Basin (GRB). Socioeconomic indicators related to vegetation extraction, silviculture, agricultural production, and livestock production were divided into productivity classes based on the relative distance of each value from the mean value (x ̄), where x ̄ represents the sum of each socioeconomic factor divided by the total number of municipalities within the GRB. Productivity classes were defined as follows: High (x > 1. 10 x ̄), Medium-High (x > 1. 05 x ̄), Medium (x > 0. 95 x ̄), Medium-Low (x > 0. 90 x ̄), and Low (x < 0. 90 x ̄).

Agricultural practices were classified according to the proportion of farmland adopting conservation practices and grouped into three categories: High (up to 95% adoption), Medium (75%–95% adoption), and Low (< 75% adoption). Farm structure was characterized according to the number of fiscal modules following Brazilian land tenure regulations and classified as smallholding (< 1 fiscal module), small farms (1–4 fiscal modules), medium farms (4–15 fiscal modules), and large farms (>15 fiscal modules).

Land tenure and legal conservation categories were derived from the Sistema Nacional de Cadastro Ambiental Rural (SICAR) vector database and classified as Alternative Land Use (AL), Legal Reserve (LR), and Areas of Permanent Preservation (APP). For the Socioeconomic Development Index (IDESE), values were divided into three classes (Low, Medium, and High), whereas demographic density (inhabitants km −2) was classified into five categories: < 1, 1–10, 10–25, 25–100, and >100 inhabitants km −2.

Model framework

The environmental modeling platform Dinamica EGO version 5.2.1 was used to develop a spatially explicit land-use and land-cover (LULC) change model (see Figure 3) to simulate annual land-use dynamics from 1998 to 2019 and to project future transitions for two time horizons: 2030 and 2050. The software environment contains a set of algorithms, referred to as functors, capable of performing map algebra operations, multi-scale stochastic simulations, and incorporating dynamic feedback mechanisms for analyzing and simulating spatiotemporal phenomena such as LULC change (Oliveira et al., 2019; Rodrigues and Soares-Filho, 2018).

Following standard cellular automata (CA) modeling approaches, the LULC model implemented in Dinamica represents the landscape as a two-dimensional array in which each pixel corresponds to a specific LULC class. Initially, the map algebra algorithm Calc Net Transition Rates was used to compute transition matrices based on initial (T i) and final (T f) landscape maps. These transition matrices represent the proportion of land that transitions from one state to another during a given time period.

The next step consists of estimating transition probabilities for each land-use pixel using the Weights of Evidence (WoE) method. First, auxiliary variables representing the driving factors of LULC change (e.g., vegetation, soil properties, altitude, and slope) were loaded as categorical raster layers. Subsequently, an internal algorithm was used to calculate WoE coefficients and assign transition probabilities to each pixel. The WoE method applies a modified Bayesian conditional probability framework (Agterberg et al., 1990; Goodacre et al., 1993) to estimate empirical relationships between spatial variables, identifying the most probable occurrence of a transition event given the presence or absence of specific spatial patterns. The effect of each auxiliary variable on a transition is calculated independently rather than through a combined multivariate solution (Soares-Filho et al., 2009).

The spatial independence of input variables was assessed using Cramér's coefficient (V) and Joint Information Uncertainty (JIU) through the algorithm Correlation Between Variables. This routine performs pairwise association tests between categorical maps. If significant correlation is detected, correlated variables must either be excluded or combined before recalculating transition probabilities (Bonham-Carter, 1994).

To initialize the simulation framework, the model requires four primary inputs: the initial landscape map, the multi-step transition matrix, the WoE coefficients, and the categorical maps representing auxiliary driver variables. A set of pre-implemented CA transition algorithms, referred to as Patcher and Expander, were used to simulate landscape structure while transition potential maps were simultaneously generated using the Calc WoE Probability Map algorithm. Each pixel within the transition probability map represents the probability of a LULC transition from one state to another.

The Expander algorithm controls the expansion or contraction of existing land-use patches, whereas the Patcher algorithm generates new patches through a seeding mechanism. In both cases, patch growth and patch size are determined according to a log-normal probability distribution controlled by the mean and variance of patch-size parameters. Another critical parameter controlling CA dynamics is patch isometry, which regulates the aggregation level and spatial compactness of simulated patches. Together, these three parameters can be calibrated to reproduce complex spatial patterns and replicate observed landscape dynamics and structure (Soares-Filho et al., 2002).

To evaluate the spatial similarity between observed and simulated landscapes, the Calc Reciprocal Similarity Map algorithm was employed. This routine implements modified versions of the Kfuzzy and Fuzzy Minimum Similarity Index methods. These metrics use distance-decay functions to compute similarity values for each pixel while accounting for neighboring spatial context. First, difference maps were generated between the initial and observed final maps, as well as between the initial and simulated final maps. Subsequently, a two-way similarity comparison was performed.

Similarity values were calculated using either an exponential distance-decay function or a constant decay function with multiple moving windows. The constant decay approach with multiple windows allows model performance to be evaluated across progressively decreasing spatial resolutions. Initially, similarity values are calculated for each pixel according to neighboring cells weighted by their distance from the focal cell. Similarity values are then averaged across multiple neighborhood windows (e.g., 1 × 1, 3 × 3, …, 15 × 15 cells), allowing model performance to be evaluated across different spatial scales (Almeida et al., 2008).

The two-way similarity procedure generates two similarity values, and the minimum value is retained in order to avoid artificially inflated similarity estimates caused by stochastic variation. Once similarity values equal to or greater than 50% are achieved, the model is considered to have acceptable performance (Nascimento et al., 2019). At this stage, the validation phase is considered complete and the model can be used to project future LULC trajectories.

Based on the conceptual framework of the model, the basin was divided into two sub-regions (Northern and Southern; see Figure 1) and simulations were conducted using six sequential sub-models: transition matrices calculation, transition probability estimation, map correlation analysis, LULC simulation, simulation validation, and future LULC trajectory projection. The observed LULC differences between 1998 and 2019 were used as the reference period for model calibration and validation.

Once model validation produced satisfactory results, future annual land-use changes were projected and analyzed for the years 2030 and 2050. Additionally, transition matrices for the full period between 1998 and 2050 were calculated to identify long-term trends in annual LULC transitions. The 2030 and 2050 projections were developed under a business-as-usual assumption, in which the historical LULC transition rates and the relationships between land-use change and the explanatory variables observed during the calibration period were assumed to remain broadly consistent throughout the projection period.

Landscape spatial patterns analysis

Changes in the annual spatial patterns of land-use and land-cover (LULC) within the Guaporé River Basin (GRB) over the 21-year study period (1998–2019) were assessed at the municipality level using a temporal trend analysis of the annual proportion of natural ecosystems (persistent and regenerating vegetation) within each municipality. Temporal trends were identified using the modified Mann–Kendall non-parametric trend test (τ) (Patakamuri and O'Brien, 2021), which corrects for serial correlation in the annual time series, considering statistically significant trends when p ≤ 0.05 and |τ|≥0.60.

Municipalities were then classified according to the magnitude and statistical significance of their Mann–Kendall trend statistic (Z) into a five-class gradient ranging from a significant decline (loss) to a significant increase (gain) in natural ecosystem cover, with intermediate classes representing weaker or non-significant trends. For descriptive purposes, we hereafter refer to municipalities in the strongest-decline and strongest-increase classes as “hot spots” and “cold spots,” respectively; this terminology reflects the magnitude of each municipality's own trend and does not, by itself, constitute a formal test of spatial autocorrelation. These analyses were conducted in the R statistical environment (R Core Team, 2024) using the modifiedmk (Patakamuri and O'Brien, 2021) and raster (Hijmans et al., 2021) packages. This classification allowed the identification of municipalities exhibiting the most pronounced long-term loss or regeneration of natural ecosystems across the basin.

Dinamica EGO performance and modeling trajectories of LULC

Comparison between the observed land-use and land-cover (LULC) maps from 1998 and 2019 and the simulated maps generated by the Dinamica EGO model under different constant decay functions with multiple windows showed that similarity values for both the Northern and Southern regions of the Guaporé River Basin (GRB) ranged from 34 to 45% using a 1 × 1 pixel window (equivalent to 30 × 30 m or 0.09 ha). Similarity values reached satisfactory levels ranging from 51 to 63% when a 3 × 3 window was applied (Table 2).

The highest similarity values were obtained by setting the mean patch size to 0.1 ha for the transition from regenerating vegetation to altered ecosystems and 1.5 ha for the transition from remnant vegetation to altered ecosystems. Patch size variance was set to 0.6 ha and 12.5 ha for these transitions, respectively, while patch isometry was fixed at 1.5.

Since each pixel corresponds to 0.09 ha, the simulated patches had an average size ranging from approximately 2 to 16 cells, with variances ranging from 6 to 39 cells. Overall, these results indicate that the model performs best when reproducing observed landscape dynamics at a spatial resolution of 0.81 ha. Although model validation reached satisfactory performance at this resolution, the relatively coarse spatial aggregation resulted in a systematic overestimation of persistent and regenerating vegetation extent, particularly within the Southern region.

The relationship between driving factors and land-use transitions assessed through the Weights of Evidence (WoE) approach revealed distinct dominant drivers in the Northern and Southern regions (Supplementary Figures A1, A2). Overall, biophysical variables associated with areas located near previously altered ecosystems, elevations above 400 m a.s.l., low stream density, gentle slopes (0%–8%), and north-facing slopes showed a higher probability of transition toward altered ecosystems and anthropogenic disturbance (Supplementary Figures A3–A6). These landscapes are not currently intensively used for agriculture, and their physical characteristics facilitate access to natural ecosystems, increasing vulnerability to future land-use conversion.

In contrast, socioeconomic variables such as conservation practices, vegetation extraction, and livestock production exerted influence primarily at local scales. For example, in the Southern region, WoE coefficients indicate that lands managed for native resource extraction, including timber, latex, seeds, fibers, fruits, and roots, are less likely to experience deforestation. In the Northern region, areas located within riparian buffer zones classified as Areas of Permanent Preservation (APP) and Legal Reserves (LR) showed a negative association with deforestation of both persistent and regenerating vegetation (Supplementary Figures A3–A6).

Changes in the spatial patterns of LULC from 1998 to 2019

In 2019, natural ecosystems within the GRB, including persistent and regenerating vegetation, covered approximately 106,000 ha, corresponding to nearly 30% of the total basin area. Approximately 70,000 ha (66% of all natural ecosystems) were located within areas designated for alternative land use on private properties. Under Brazilian legislation, conversion of native vegetation on private land may be authorized only where the area is not protected as an Area of Permanent Preservation (APP), the required Legal Reserve (LR) is maintained, and other applicable legal restrictions are satisfied. In the Atlantic Forest biome, additional restrictions are established by the Atlantic Forest Law (Law n° 11,428/2006) (Brasil, 2006).

Despite the legal protection provided by Areas of Permanent Preservation and Legal Reserves, approximately 15,000 ha of natural ecosystems were converted into altered ecosystems between 1998 and 2019, with most disturbed areas located within distances of less than 10 m from APP buffer zones associated with small streams.

Between 1998 and 2019, distinct spatial patterns of transition involving persistent and regenerating vegetation conversion into altered ecosystems were identified between the Northern and Southern regions of the basin (Figure 4). In the Northern region, persistent and regenerating vegetation were converted into altered ecosystems at annual transition rates of 1.55 and 2.78%, respectively (Table 3).

In contrast, the Southern region showed simultaneous patterns of natural vegetation conversion and agricultural land abandonment. In this region, altered ecosystems transitioned into regenerating vegetation at an annual rate of 1.40%, indicating an active process of vegetation regeneration. This regeneration process partially reduced the overall trend of natural vegetation loss in the Southern region, suggesting that distinct drivers control patterns of anthropogenic conversion and ecological recovery across different sectors of the GRB.

These findings were further supported by the analysis of WoE coefficients and transition probability maps associated with the conversion of persistent and regenerating vegetation into altered ecosystems (Supplementary Figures A3–A6).

Projected spatial patterns of LULC in 2030 and 2050

Future simulations indicate continued expansion of altered ecosystems throughout both Northern and Southern regions of the basin as anthropogenic land use progressively replaces persistent and regenerating vegetation. Historical transition patterns indicate that the dominant long-term process is the conversion of remnant vegetation into regenerating vegetation followed by subsequent transition into altered ecosystems.

By overlaying projected LULC maps for 2030 and 2050 onto the baseline map from 1998, the model reveals a consistent sequence of landscape transformation characterized by initial loss of remnant vegetation, followed by expansion of regenerating vegetation and progressive increase in altered ecosystems (Table 4).

Figure 5 highlights the projected landscape changes expected for 2030 and 2050 across the GRB. The model predicts that the Northern region of the basin will lose approximately 15,903 ha of remnant vegetation, primarily driven by the expansion of altered ecosystems. Historical transition patterns captured through WoE analysis indicate that flatter terrain and highly productive soils, which favor agricultural activities, are particularly susceptible to conversion into altered ecosystems. In contrast, the Southern region is projected to experience a progressive transition of remnant vegetation into regenerating vegetation. In this region, steeper terrain limits agricultural suitability, while vegetation extraction activities, which are generally associated with the maintenance of natural areas, are expected to promote the continued expansion of natural vegetation cover.

The 2019 land-use map of the Guaporé River Basin (GRB) indicates that a substantial proportion of the remaining natural ecosystems occurs on privately owned farms where land-use conversion may be authorized under specific legal conditions. Under the Natural Vegetation Protection Law (NVPL), conversion is subject to the protection of Areas of Permanent Preservation (APP), maintenance of the required Legal Reserve (LR), and compliance with other applicable restrictions. Within the Atlantic Forest biome, native vegetation is also subject to the Atlantic Forest Law (Law n° 11,428/2006) (Brasil, 2006). Similar patterns have been observed across Brazil. Although the NVPL protects approximately 193 ± 5 Mha of natural ecosystems within LR and APP areas, Soares-Filho et al. (2014) estimated that nearly 88 ± 6 Mha of Brazil's native vegetation could still be legally converted for anthropogenic purposes.

Despite the legal framework, a large proportion of landowners remain in noncompliance with environmental regulations (Soares-Filho et al., 2014; Sparovek et al., 2010). Even in consolidated agricultural regions, compliance with environmental legislation remains limited. Similar to our findings, Sparovek et al. (2010) reported widespread illegal occupation of APP and LR areas across Brazil, including southeastern regions. This pattern can be partially explained by weak enforcement of environmental regulations, including the NVPL and the Atlantic Forest Law (Brock et al., 2021; Soares-Filho et al., 2014; Soterroni et al., 2018), as well as Brazil's persistent difficulties in controlling illegal deforestation (Rajao et al., 2020).

The level of compliance with the NVPL directly affects nature's contribution to ecosystems and society, including food production, water-flow regulation, and water quality maintenance. Areas of Permanent Preservation were originally designed to protect environmentally sensitive regions such as steep slopes, springs, riparian zones, and mountaintops. Consequently, land-use change within these protected areas compromises ecosystem functioning and reduces the capacity of natural systems to sustain both ecological integrity and human wellbeing.

In the present study, even considering landowner self-declared information, we identified substantial environmental liabilities within APP areas, particularly those associated with small streams. Riparian zones play a critical ecological role by controlling soil erosion and protecting water bodies from sediment transport and nutrient loading (Guidotti et al., 2020; Hanna et al., 2020; Monteiro et al., 2016; Valera et al., 2019; Vieira et al., 2018). Degradation of these systems can therefore have cascading consequences for watershed functioning and long-term water resource sustainability.

Anthropogenic land use, particularly agriculture, is widely recognized as a major driver of water quality degradation and biodiversity loss (Johnson et al., 2021). Several studies have already demonstrated strong relationships between agricultural land use and declining water quality in watersheds located within the southern Atlantic Forest biome (Bastos et al., 2021; Bonuma et al., 2014; Didoné et al., 2014; Kaiser et al., 2010, 2015; Tiecher et al., 2017). Recent work conducted specifically within the GRB associated degraded water quality with inadequate rural sanitation infrastructure combined with intensive agricultural activity (Bastos et al., 2021).

The continued conversion of natural ecosystems, particularly within Legal Reserves and Areas of Permanent Preservation, combined with increasing water pollution pressures, raises important concerns regarding the long-term capacity of ecosystems within the GRB to continue providing critical water-related ecosystem services. This concern becomes particularly relevant considering that the basin plays a strategic role in supplying freshwater resources to the metropolitan region of Rio Grande do Sul.

Drivers of land-use change

According to our results, the Northern sector of the Guaporé River Basin (GRB) has historically represented a hotspot for losses of persistent and regenerating vegetation, whereas the Southern sector was identified as a cold spot due to the expansion of regenerating vegetation. Although differences in vegetation-cover trends justify dividing the basin into two distinct regions, additional local drivers must be considered to better understand the observed spatial patterns of land-use change. Our results indicate that land-use and land-cover (LULC) changes driven by biophysical factors operate at similar spatial scales in both regions despite the basin's complex geographical characteristics.

Biophysical characteristics such as topography, distance to human settlements, altitude, and aspect have been consistently associated with land-use change and ecosystem degradation (Calaboni et al., 2018; Molin et al., 2017). In general, the conservation of natural ecosystems and the expansion of secondary forests were concentrated in areas with low agricultural suitability, particularly steep slopes and poor soils (Lingner et al., 2020; Silva et al., 2016). In contrast, the expansion of altered ecosystems occurred predominantly in areas characterized by flat to gently undulating terrain.

These landscapes provide more stable physical conditions and favorable environments for human occupation and agricultural production due to improved soil physical properties and greater water retention capacity compared with more concave or topographically complex areas (Lemos et al., 2021; Reichert et al., 2016). Additionally, crop transport logistics and mechanized agricultural operations are facilitated by flatter terrain (Rezende et al., 2015), increasing pressure for the conversion of natural ecosystems into anthropogenic land uses.

Similarly, slope, distance to previously altered ecosystems, altitude, and aspect emerged as strong predictors of land-use change. Areas located near pre-existing settlements and at lower altitudes are generally preferred for agricultural expansion due to easier accessibility and lower implementation costs (Mendoza-Ponce et al., 2021). In contrast, areas with lower agricultural productivity, particularly southern-, southeastern-, and southwestern-facing slopes, often associated with reduced solar radiation, are more frequently maintained under natural vegetation cover (Lemos et al., 2021).

Although all biophysical factors considered in this study showed statistically significant relationships with land-use change, they are insufficient to fully explain land conversion processes without accounting for interactions among multiple driving forces (Geist and Lambin, 2002; van Vliet et al., 2016). Consequently, previous studies have adopted conceptual frameworks that integrate multiple environmental, socioeconomic, and institutional factors in order to better capture the complexity of land-use dynamics and associated trade-offs in coupled human-environment systems (van Vliet et al., 2016).

These findings support our hypothesis that LULC dynamics within the GRB are strongly influenced by the combined effects of biophysical conditions associated with agricultural suitability and socioeconomic factors related to land use and agricultural production.

Following the conceptual framework proposed by Geist and Lambin (2002), and based on our literature review, we observed that biophysical constraints are often amplified by proximate causes and underlying socioeconomic drivers. At the global scale, increasing international trade has triggered unprecedented transformations within the Brazilian agricultural sector (Cattelan and Dall'Agnol, 2018; Winkler et al., 2021), including already consolidated agricultural regions in southern Brazil. At the local scale, agricultural intensification, expansion of mechanized farming systems, land tenure structure, and smallholder production systems increase pressure on natural ecosystems (Maranhão et al., 2019; Song et al., 2021).

In contrast to biophysical variables, which largely explain the spatial suitability for land conversion at the regional scale, socioeconomic factors appear to determine the decision-making processes that ultimately drive landscape transformation. In the Northern region, expansion of altered ecosystems appears strongly associated with farmers' decisions regarding the adoption of conservation practices, whereas in the Southern region vegetation extraction activities are more closely associated with the maintenance of natural areas.

However, simplifying land-use change as a direct cause-and-effect relationship may lead to misleading interpretations of complex socioeconomic processes (Lambin et al., 2001). The underlying mechanisms behind socioeconomic drivers are often linked to broader economic opportunities, market incentives, and environmental awareness among rural communities (Prokopy et al., 2019). Within the GRB, two major commercial crops dominate agricultural production, with soybean cultivation concentrated in the Northern region and tobacco production prevailing in the Southern sector.

Over recent decades, global demand and profitability of soybean production have increased substantially (Cattelan and Dall'Agnol, 2018; Song et al., 2021). In response to this economic opportunity, Brazilian agriculture has progressively shifted toward export-oriented commodity production systems (Garrett et al., 2013). Even in regions where land availability for grain expansion is limited, such as southern Brazil, soybean cultivation continues to expand primarily through replacement of pastureland and other agricultural land uses (Flexor and Leite, 2017; Lapola et al., 2013).

Soybean expansion has also been associated with greater adoption of environmentally favorable agricultural practices. Compared with other crops, soybean is commonly cultivated during summer under no-tillage management systems (Wade et al., 2015). Relative to conventional tillage systems, no-tillage practices can improve soil structure, increase soil organic carbon storage, enhance water retention capacity, reduce soil compaction, improve aggregate stability, and increase soil porosity (Ambus et al., 2018; Lozano et al., 2016; Peres da Rosa et al., 2021; Reichert et al., 2018). By improving soil health and reducing land degradation, conservation practices such as no-tillage may indirectly reduce pressure for further land conversion, potentially sparing land for natural ecosystems (Villoria et al., 2014).

In contrast to soybean-producing systems, regions where economic activity is strongly dependent on tobacco production, such as the Southern sector of the GRB, are often associated with less environmentally sustainable land-use practices. Tobacco cultivation is widely recognized as one of the most environmentally damaging agricultural systems, particularly in rural areas characterized by lower socioeconomic development and limited economic diversification (Leppan et al., 2014; WHO, 2017). Previous studies have shown that tobacco cultivation contributes to native forest loss, depletion of soil fertility, increased soil erosion, and water contamination (Bastos et al., 2021; Becker et al., 2009; Bonuma et al., 2014; Kaiser et al., 2015; Reichert et al., 2021a; Thomaz and Antoneli, 2021).

By degrading environmental quality, tobacco production may also contribute to reinforcing rural poverty cycles. Tobacco companies frequently encourage smallholder families to cultivate tobacco by providing financial support, agricultural inputs, and technical assistance (Reichert et al., 2021a). However, many smallholder farmers earn only marginal profits and often fail to cover production costs. As a consequence, many producers accumulate debt under highly asymmetric contractual relationships established by tobacco companies (Leppan et al., 2014). This process of economic vulnerability may further intensify environmental degradation, as impoverished farmers are more likely to overexploit locally available natural resources (Masron and Subramaniam, 2018).

An alternative pathway for improving environmental quality while reducing rural dependency on intensive cash-crop systems involves promoting more diversified and sustainable production systems. Within the GRB, many small family farms generate income through integrated production systems involving multiple food crops such as corn, oats, wheat, and dairy production.

Forests also represent an important economic resource, particularly in areas dominated by Ilex paraguariensis St. Hil. (Erva-mate), a native species traditionally used for tea production and naturally occurring in the intermediate canopy of the Brazilian Atlantic Forest (Eibl et al., 2017). Traditional Erva-mate production systems support high forest-species diversity and promote nutrient cycling processes (Ilany et al., 2010), while often requiring little or no chemical inputs (Chaimsohn and Radomski, 2016). Altogether, the cultural importance of Erva-mate combined with sustainable use of natural resources provides important evidence that local communities possess substantial capacity to recognize the ecological and economic value of conserving remaining forest fragments (Christo et al., 2012).

Conservation and policy implications

The spatial patterns identified in this study have important implications for conservation planning and environmental governance within the Atlantic Forest biome. The contrasting dynamics observed between the Northern and Southern sectors of the Guaporé River Basin (GRB) suggest that spatially differentiated management strategies may be more effective than uniform basin-wide conservation policies. In the Northern region, where relatively flat terrain and high agricultural suitability continue to drive the conversion of remnant vegetation, stronger enforcement of Legal Reserve requirements combined with targeted Payments for Ecosystem Services (PES) schemes could provide economic incentives for landowners to maintain or restore native vegetation.

In contrast, the Southern region, where steeper terrain naturally constrains agricultural expansion and regenerating vegetation is expanding, represents a strategic priority for assisted natural regeneration initiatives and long-term ecological monitoring. Our projections indicate that, in the absence of effective intervention, the Northern basin may lose approximately 15,903 ha of remnant vegetation by 2050. Such losses would likely compromise key ecosystem functions at the watershed scale, including carbon storage capacity, freshwater regulation, and habitat connectivity.

Achieving the objectives established under Brazil's National Biodiversity Strategy (NBio), which is aligned with the Kunming–Montreal Global Biodiversity Framework target of protecting 30% of terrestrial ecosystems by 2030, will require prioritizing watershed units such as the GRB, where remaining remnant vegetation is simultaneously ecologically valuable and exposed to persistent land-use pressure.

An important finding of this study is the continued replacement of remnant vegetation by regenerating vegetation, even in landscapes where total natural vegetation cover appears relatively stable. Although the LULC classification does not directly quantify vegetation structure or biodiversity, this transition may indicate ecological simplification, as regenerating forests generally require time to recover the structural complexity and ecological attributes associated with older forest remnants.

The biodiversity implications associated with this transition are particularly concerning because late-successional forest specialists, including epiphytes, shade-tolerant understory species, and forest-interior vertebrates, are strongly associated with structural characteristics typically found in old-growth forests. Secondary forests may require decades to recover these ecological conditions, if they recover them at all (Brancalion et al., 2019; Joly et al., 2014).

Brazil's Natural Vegetation Protection Law (NVPL), together with the Atlantic Forest Law (Law n° 11,428/2006), establishes legal protections for native vegetation and restricts its conversion under specific conditions (Brasil, 2006). However, the continued conversion observed within the GRB suggests that some of these changes may occur in violation of environmental legislation, highlighting potential shortcomings in enforcement and monitoring.

This finding highlights an important limitation in current environmental governance and underscores the need for more robust monitoring systems capable not only of detecting forest cover change, but also distinguishing vegetation age classes, successional stages, and ecosystem quality. Improving this level of ecological monitoring will be critical for preventing progressive degradation processes that remain largely invisible under conventional land-cover assessments.

Future trajectory of land use

Overall, future land-use and land-cover (LULC) simulations indicate that anthropogenic activities will continue to reshape natural ecosystems throughout the Guaporé River Basin (GRB). Although the model calibration showed that a spatial resolution of 3 × 3 pixels (0.81 ha) was adequate for reproducing historical landscape dynamics, spatial resolution remains an important source of uncertainty in land-use simulations (Rodrigues and Soares-Filho, 2018). Coarser spatial resolutions generally improve statistical fit with historical land-use patterns; however, they often fail to accurately reproduce fine-scale landscape heterogeneity. Despite these limitations, coarse-resolution models remain capable of capturing broader long-term trajectories of landscape change and identifying general patterns of future ecosystem transformation (Cheng et al., 2020).

Projecting the spatial expansion of altered ecosystems through time remains inherently challenging because the relative importance of driving factors is not static and may fluctuate under changing environmental, economic, and social conditions (de Brito et al., 2021). Furthermore, assessing the influence of both direct and indirect drivers on natural ecosystems remains a complex task, given the limited capacity of LULC models to fully represent the complexity of real-world human–environment interactions (Resende et al., 2020). Consequently, future simulations should be interpreted as plausible landscape trajectories under current conditions rather than deterministic forecasts.

Considering the overall historical LULC trends, observed transition rates from natural ecosystems to altered ecosystems within the GRB fall within the range previously reported for other Atlantic Forest regions, although deforestation rates remain comparatively high (Molin et al., 2017). This pattern is particularly evident in the Northern sector of the basin, where continued conversion of remnant vegetation remains a major long-term concern.

In contrast, the Southern sector presents a more complex ecological trajectory. Historical transitions from altered ecosystems toward regenerating vegetation may initially suggest a positive process of ecological recovery, consistent with broader regional observations indicating widespread forest regrowth and relatively stable forest cover conditions across portions of the Atlantic Forest biome (Brancalion et al., 2019; Ferreira et al., 2019; Rezende et al., 2018). However, a more detailed interpretation reveals a less optimistic trajectory, in which remnant vegetation is progressively being replaced by regenerating vegetation. Although regeneration represents an important component of landscape recovery, younger regenerating forests may not immediately provide the structural complexity, biodiversity, and ecosystem functions associated with older forest remnants.

Similar patterns have been reported by Rosa et al. (2021), who identified an ongoing decline in older native forest cover accompanied by an expansion of younger secondary forests (less than 10 years old). This process suggests that apparent forest recovery may conceal progressive losses of ecological complexity, biodiversity, and ecosystem functionality, even in landscapes where total vegetation cover appears relatively stable.

Although the Atlantic Forest is protected under specific legislation that restricts deforestation of forests in intermediate and late successional stages (greater than 10 years old) (Brasil, 2006; Silva et al., 2018), conflicts among competing land uses, particularly within Areas of Permanent Preservation (APP) and Legal Reserves (LR), continue to be reported across multiple regions (Patrício et al., 2019). Similar patterns observed in the GRB suggest that the basin may be undergoing a forest transition process, in which declining deforestation rates are not accompanied by equivalent gains in ecologically mature forest cover.

At the local scale, these long-term landscape transformations may amplify future climate-related vulnerabilities while reducing the capacity of ecosystems to sustain critical environmental services and generate long-term socioeconomic benefits. Consequently, preserving remaining mature forest ecosystems should be considered a strategic priority not only for biodiversity conservation, but also for strengthening watershed resilience under future environmental change.

Limitations and uncertainties

This study is subject to several limitations that should be considered when interpreting the results. The MapBiomas land-use and land-cover (LULC) maps used as model inputs, although widely validated (overall accuracy reaching up to 96.6% at Level 1; Souza et al., 2020), are derived from Random Forest classification of Landsat imagery and may introduce classification errors, particularly when distinguishing spectrally similar vegetation types along forest–agriculture boundaries and within heterogeneous mosaic landscapes. Because such errors can propagate through the cellular automata (CA) model, they may affect the accuracy of transition probability estimates, particularly for transition types that are spatially diffuse or occur at relatively low frequencies, and should be kept in mind when interpreting both historical transition dynamics and future simulations.

Some of this uncertainty is reinforced by the simplified classification scheme adopted here, which groups landscapes into natural and altered ecosystems (see Section 2.2) and necessarily reduces ecological heterogeneity within each class. Distinctions between structurally intact and degraded forests, or among agricultural systems with different management intensities, are not explicitly represented in the model and may influence the estimated relationships between spatial drivers and observed land-use transitions.

Beyond the input data, the Dinamica EGO modeling framework itself assumes temporal stationarity in both transition rates and the spatial relationships among driving factors calibrated during the 1998–2019 period. Future changes in climate conditions, fluctuations in agricultural commodity markets, technological innovation, or major policy reforms, such as changes in the implementation of the Cadastro Ambiental Rural (CAR) system, may substantially alter these relationships in ways not captured by the current modeling framework. The future scenarios presented in this study should therefore be interpreted as extrapolations of historically observed trends rather than deterministic predictions of specific future landscape states, representing plausible outcomes under a business-as-usual scenario that assumes relative continuity in the dominant environmental, socioeconomic, and institutional drivers observed during the calibration period.

A further constraint applies specifically to the municipality-level trend classification presented in Figure 4, which was derived from a modified Mann–Kendall trend test applied independently to each municipality's own natural-ecosystem cover time series and does not itself constitute a formally tested measure of spatial autocorrelation. The terms “hot spot” and “cold spot” are used descriptively throughout this study to denote municipalities with the most pronounced loss or gain trends, based on the magnitude and significance of each municipality's own trend statistic, so the resulting spatial pattern should be interpreted qualitatively rather than as evidence of a statistically confirmed spatial cluster. A dedicated local spatial-autocorrelation analysis, such as Getis–Ord Gi* or Local Indicators of Spatial Association (Bivand and Wong, 2018), applied directly to the trend statistic would be needed to formally test whether the observed loss and gain patterns are more spatially clustered than expected under complete spatial randomness. Four municipalities also intersect less than 5% of their total administrative area with the GRB (Água Santa, 0.6%; Passo Fundo, 2.5%; Encantado, 3.2%; and Soledade, 3.8%), so their trend estimates rest on comparatively few pixels. All municipalities were nonetheless retained in the loss–gain gradient, and classifications for these smaller-area units should be interpreted with additional caution relative to municipalities with larger areas within the basin.

The same concern about spatial resolution applies to the socioeconomic and demographic variables, which were aggregated at the municipality level and subsequently downscaled proportionally according to basin area, an approach that may smooth important intra-municipal spatial variation associated with local land-use decision-making processes. Access to finer-scale socioeconomic information, particularly property-level spatial data available through the Sistema Nacional de Cadastro Ambiental Rural (SICAR) database, would likely improve the spatial precision of driver estimation and strengthen future modeling efforts.

Based on the analysis of land-use drivers and the application of a spatially explicit predictive model, this study demonstrates that land-use dynamics within the Guaporé River Basin (GRB) are not only causing net losses of natural vegetation, but are also driving an ecological transition characterized by the progressive replacement of remnant vegetation by regenerating vegetation. These trends highlight the importance of effective land-use planning and stronger environmental governance mechanisms to ensure the long-term conservation of remaining remnant vegetation.

The contrasting land-use and land-cover (LULC) trajectories observed between the Northern and Southern sectors of the GRB demonstrate that landscape transformation processes are spatially heterogeneous and strongly dependent on the interaction of multiple drivers. Biophysical characteristics, particularly terrain slope, altitude, and soil properties, operate at the basin scale by determining the physical suitability for land conversion, while socioeconomic factors, including agricultural productivity, land tenure structure, and conservation practices, largely determine local-scale land-use decisions.

Future simulations under current landscape trends indicate that the Northern sector of the GRB may lose an additional 15,903 ha of remnant vegetation by 2050, primarily in relatively flat areas with high agricultural suitability. Without strengthened environmental governance, more effective enforcement mechanisms, and targeted spatial planning strategies, this trajectory will likely accelerate fragmentation of the remaining Atlantic Forest and further compromise ecosystem integrity.

Private lands emerged as critical areas for long-term conservation, yet observed transition patterns indicate continued conversion of natural vegetation into anthropogenic land uses. These patterns suggest that current environmental legislation and existing enforcement mechanisms remain insufficient to effectively prevent ongoing deforestation and ecosystem degradation, even within legally protected areas.

Overall, the LULC model applied in this study indicates that continued land-use change within the basin may lead to further deforestation and progressive replacement of remnant vegetation by regenerating vegetation. In ecological terms, such transitions may contribute to simplification where regenerating forests have not yet recovered the structural complexity and biodiversity associated with older forest remnants. Such transformations may substantially reduce nature's capacity to sustain critical ecosystem services, including carbon sequestration, freshwater regulation, and biodiversity conservation. Consequently, long-term land-use management strategies must adopt a comprehensive and spatially differentiated approach capable of balancing biodiversity conservation, ecosystem resilience, economic development, and social wellbeing.