Our study area spans the paved road verge areas within the states of Minas Gerais (MG) and Goiás (GO) (including the Federal District - FD) (MMA, 2023; Fig. 1). The Cerrado vegetation in this region ranges from open grasslands to dense woodlands and shrublands (Eiten, 1972). The mean annual temperature in the study region varies from around 22 °C–26 °C and with an average annual rainfall ranging from 1000 to 1600 millimeters, with more than 70% of the rains occurring between November and March (Ab’Sáber, 2003; Alvares et al., 2013). The dominant soils of the region are Oxisols, with poor nutrient availability and moderate to strong acidity (Lopes and Cox, 1977; Araújo and Haridasan, 1988).

Fig. 1.

Left: The area covered by the Cerrado biome in Brazil (in green). Right: The network of paved roads (black lines) in the Cerrado region of Minas Gerais, Goiás, and the Federal District.

(0.39MB).

We accessed the road shapefile database of 2022 from the Infrastructure and Transport Federal Department’s website (https://www.gov.br/dnit/pt-br/assuntos/planejamento-e-pesquisa/dnit-geo). We filtered paved roads, and we categorized them into two groups: multi-lane roads, which are approximately 40 meters wide, and two-lane roads, which are approximately 15 meters wide (Fig. 2).

Fig. 2.

(a) Schematic view of the two-lane and multi-lane roads, along with the buffer width allocated on each road verge side. (b) Frequency distribution of the widths of the areas covered with native vegetation along the roads in the states MG and GO (FD) based on remote sensing analysis of 100 randomly selected road sections.

(0.12MB).

Under Brazilian federal land allotment legislation (Federal Law 6.766/1979), road verges must maintain a setback of 15 meters on each side of the road, often covered with native plants to promote environmental conservation. However, our field observations revealed considerable variation in the width of these vegetated areas, likely influenced by differences in local land use, road construction practices, and management policies. To accurately quantify these variations and establish a suitable buffer width for our study, we measured, using remote sensing, 100 randomly selected road sections across the two states. The analysis revealed an average vegetated width of 27 meters (Fig. 2), exceeding the minimum legal requirement. Based on these findings, we adopted a 30-meter buffer as the standard for our study, ensuring it encompasses the typical extent of vegetation while providing a consistent framework for ecological assessments (Fig. 2).

We used satellite imagery (Sentinel-2 MSI (MultiSpectral Instrument) Level 2A) to categorize land use classes. We accessed Sentinel-2 (10-meter resolution) with Google Earth Engine (GEE) (Gorelick et al., 2017) from January to July of 2017, 2020, and 2023. For each year, we obtained a single mosaicked image, using the composite algorithm to select and average (using median values) cloud-free pixels.

We cropped the images to our study area and performed a land cover classification over three land use classes: native vegetation, roads, and other human uses (that included agricultural land, pastures, and bare soil). Our ground control point (GCP) dataset consisted of 3,600 points distributed across the three land use classes selected from the high-resolution data from Google Earth 2023. We additionally included 100 points for native vegetation for which we have ground information.

We used the Random Forest algorithm for our land use classification (Zhang and Yang, 2020). In the context of GEE-based classification using Random Forest, five key variables must be configured: the number of decision trees, maximum leaf nodes per tree, randomization seed, input fraction per tree, and the number of variables per split. In our study, we set the number of decision trees to 10, the input fraction per tree to 0.8, and left the remaining parameters at their default values.

To maximize the accuracy of the land classification, we used Sentinel-2 MSI spectral bands BLUE, GREEN, RED, and two vegetation indices: the normalized difference vegetation index (NDVI) and the soil-adjusted vegetation index (SAVI) to enhance our land use classification (Xue and Su, 2017). NDVI serves as a vital indicator of vegetation cover and greening, and SAVI, on the other hand, is a vegetation index designed to reduce the influence of soil brightness by incorporating a soil-brightness correction factor.

We obtained land use maps for the study area in 2017, 2020, and 2023. We allocated 80% of the GCP dataset for training a Random Forest classifier. Subsequently, we assessed classification accuracy against the remaining 20% of the GCP, achieving an overall accuracy of 75%.

We evaluated the accuracy of the Random Forest classification using the overall classification accuracy and the Kappa coefficient (He et al., 2022b). We found that the overall classification accuracy rates were 73% in 2017, 74.26% in 2020, and 75% in 2023. Importantly, all Kappa coefficients exceeded 0.60, indicating that the classification process was both effective and precise. We calculated the land use change over the periods (2017 – 2020 – 2023) by calculating the area that had changed in the period considered.

To estimate the aboveground carbon stock, we used data from 1,663 trees sampled across 100 plots (50 m × 10 m each) established in areas of native Cerrado vegetation along road verges, across four regions: 18 plots in northwest Goiás (15º58’29” S, 50º06’18” W, along the GO-164, GO-070, and GO-060 highways), 29 plots in southern Goiás (18º16’55” S, 49º14’38” W, along the BR-153), 13 plots in southeastern Goiás (17º42’13” S, 48º09’31” W, along the GO-330, GO-213, and BR-490 highways), and 40 plots in the Triângulo Mineiro and Alto Paranaíba region of Minas Gerais (18º49’04” S, 47º32’17” W, along the MG-190, MG-223, BR-365, and BR-352 highways) (Rios et al., 2023). A minimum distance of 1 km was maintained between plots to ensure spatial independence. Areas with forests, an absence of vegetation, or high levels of degradation with little to no arboreal vegetation were excluded. For more methodological details see Rios et al., 2023.

We calculated the aboveground biomass (AGB) of each sampled tree using the allometric formula developed by Ribeiro et al. (2011): ln B = −3.352 + 2.9853*lnD+1.1855*lnWD, where B is aboveground biomass in kg; D is the diameter at breast height in cm, and WD is the wood density in g cm−3. Subsequently, we extrapolated the value from our calculation to the total area of the road verges covered by native vegetation. For the carbon stock, we assumed that 50% of AGB of each tree is represented by carbon (Chave et al., 2005).

We used a linear model to evaluate the differences in the amount of native vegetation between regions, road types and to evaluate if the amount of native vegetation was increasing, decreasing, or remaining stable over time (from 2017 to 2023). The percentage of native vegetation cover was the response variable, while the predictors included region, road type and year as a covariate. The analysis was performed in R version 4.4.1 (R Core Team, 2024).

We found 24,566 kilometers of paved roads within the study region, of which 12,975 km were in Minas Gerais (MG) and 11,590 km in Goiás (GO) (including the Federal District - FD). Most paved roads (84.2%) were two-lane roads (totaling an extension of 20,689 km). Considering a buffer of 30 meters on each side of the road, road verges amount to a total area of 147,399 ha (of which 77,848.47 ha in MG and 69,550.42 ha in GO and FD).

In total, 32.08% of the area of the road verges was covered by native vegetation, while the remaining area (67.92% or ca. 150,000 ha) had been converted into anthropogenic land uses before 2023. Two-lane roads had comparatively more native vegetation than multi-lane ones (33.1% versus 25.1%; F1,7 = 914.11, p < 0.001; Fig. 3). Similarly, in total, MG had comparatively more native vegetation along road verges than GO and FD (35.4% vs 28.6%; F1,7 = 14.30, p = 0.007), notably along the two-lane roads (Table 1; Fig. 3). Overall, the amount of native vegetation cover along road verges remained relatively stable through the study period (from 2017 to 2023) and this was true in both states and in the two types of roads (F1,7 = 1.23, p = 0.30; Fig. 3).

Fig. 3.

Temporal variation in the amount of native vegetation cover along road verges for two-lane and multi-lane paved roads in the Cerrado region of Minas Gerais and Goiás (including the Federal District).

(0.1MB).

Based on the tree inventory data, we found that the average tree aboveground biomass along road verges is 25.32 tons per hectare, whereas the average aboveground carbon stock amounts 12.66 tons per hectare. When extrapolating these values to the total area covered by native vegetation, we determined that road verges collectively store more than 600,000 tons of carbon in existing native vegetation (Table 1). Given the observed differences in vegetation cover between states and road types, we found that the carbon stored along two-lane roads is 7.04 times greater than that along multi-lane roads, whereas there is 1.40 times more carbon stored along the roads of MG compared to those of GO and FD (Table 1). The carbon lost by 67.92% of already converted land before 2023 is estimated to be 3.8 million tons.

Despite the high potential of vegetation along road verges in the Cerrado, numerous challenges must be addressed concerning the conservation and restoration of these areas. From a biological perspective, careful planning is imperative for restoration to avoid unintended negative consequences, such as the spread of invasive species, and to ensure that native plant species well-adapted to the local ecosystem are selected (Phillips et al., 2020a). Furthermore, road verges are frequently subjected to human activities like illegal logging, poaching, and littering, making effective management and prevention crucial for successful conservation efforts and necessitating the involvement of multiple stakeholders (Nemec et al., 2021).

Financially, conservation and restoration efforts demand significant resources, and securing funding for long-term projects can be particularly challenging, especially in regions with limited financial means. Additionally, when financial support is available, competing priorities compel governments to make strategic investment decisions (Nemec et al., 2021). This underscores the complexity of managing road verges and highlights the need for collaborative efforts and strategic planning to overcome these challenges and ensure the long-term sustainability of Cerrado ecosystems.

Given the costs and challenges associated with conserving and restoring Cerrado road verge vegetation, we recommend a few next steps that we believe are easier to implement. Firstly, we advocate for the implementation of Integrated Road verge Vegetation Management (Bernes et al., 2017; Nemec et al., 2021), a strategy already employed in other countries. This plan can establish a framework for preserving these areas, particularly against illegal activities and improper management practices. Such initiatives may involve leveraging existing law enforcement mechanisms and incorporating adjacent landowners to recognize the benefits of maintaining native areas near their lands.

Furthermore, considering the current and future potential of these areas in storing native vegetation and carbon stock, implementing REDD + and robust payment for ecosystem services programs would also be vital in reducing deforestation rates in the biome (da Conceição Bispo et al., 2023) and avoiding further carbon emissions into the atmosphere. As restoration of the degraded area (i.e., almost 70% of the area without native cover) might be costly and unachievable, another step involves establishing an initiative to strategically identify road verge hotspots (Lewis et al., 2023) that can connect native Cerrado remnants, thereby facilitating focused restoration efforts and biodiversity maintenance.