The study area is based on the investigation proposed by [48], which encompassed the entire state of Amazonas. However, considering the need to use the highest possible spatial resolution and the limitations of computational capacity and processing time, the focus was narrowed to the BR-319 highway and its associated access roads (existing and planned). This configuration was selected to better represent the climatic impact within the largest feasible area for high-resolution simulation. Therefore, the study domain was defined between the geographic coordinates 67˚W to 59˚W and 8˚S to 2˚S (Figure 2).

Figure 2. Study area coverage, BR-319 highway, and adjacent existing and planned road network around BR-319, including Federal and State Conservation Units, Indigenous Lands, Federal Settlement Projects, and Military Areas. Map created by the authors. Data sources: [54]-[57].

The study focuses on the impact of the reconstruction of the BR-319 highway, which is located between the Madeira and Purus river basins, connecting the cities of Manaus and Porto Velho. BR-319 is the main terrestrial access route to the municipal seats of Careiro, Manaquiri, Careiro da Várzea, and Autazes, and it also facilitates access to Humaitá, Lábrea, and Manicoré. It serves as the only land connection to the communities of Vila Realidade (a district of the municipality of Humaitá) and Igapó-Açu (a district of Borba). BR-319 also provides road access to Roraima state via the BR-174 highway, and to Pará state via BR-230 (Transamazon Highway).

The currently existing network of official roads in Amazonas state connected to the 885 km of BR-319 totals 1934 km, comprising the federal highways BR-230 (827 km from Lábrea-AM to the border with Pará), BR-174 (85 km section between BR-319 and Manicoré), and the state highways AM-254 (94 km from BR-319 to Autazes) and AM-354 (43 km from BR-319 to Manaquiri). Additionally, the Amazonas state government has proposed new roads connecting BR-319 to the municipalities of Tapauá, Tefé, Juruá (via AM-366), and Coari (via AM-343), advancing into the area west of the Purus River. This would facilitate deforestation in one of the most pristine forest areas in the state, known as the “Trans-Purus” region [53], which was also included in our modeling framework.

The dominant vegetation types in Amazonas state include dense ombrophilous forest, campinarana (in the far north), open ombrophilous forest, and savanna in the south [54]. In the northern part of the Purus-Madeira interfluve, lowland dense ombrophilous forest predominates, while in the southern part, lowland open ombrophilous forest is more common [54]. The study area features an equatorial rainforest climate (Af) and monsoon climate (Am), according to the Köppen-Geiger classification [58]. The central and western portions experience extremely humid conditions with no dry season, whereas the southern portion has a humid climate with 1 to 3 dry months in the east and south [54]. Due to the high solar radiation received throughout the year, air temperature shows little seasonal variation, except in the far south where frontal systems occur. The mean annual air temperature is approximately 27˚C, and annual rainfall is around 2000 mm [2][59].

The simulated deforestation maps used in this study were generated by Santos et al. (2023) and cropped to match the spatial extent proposed here (Figure 3). These maps were produced using the DINAMICA-EGO platform (Environment for Geoprocessing Objects) [60].

The modeling approach considered the historical trends of land occupation expansion by different local groups, reflecting the dynamics of land use and conflicts that influence landscape changes along road networks (Business as Usual, BAU) [40][42][44]. Therefore, the projections did not assume improved environmental governance or land management practices.

Two environmental forecast scenarios were developed starting from the year 2021, with projections for the years 2050 and 2100: Scenario 1 (BAU_1): BR-319 without asphalt paving (current condition, assuming the reconstruction and paving process remains unapproved); Scenario 2 (BAU_2): BR-319 with paving (assuming paving is authorized and begins in 2025). For additional methodological details, see [48].

Figure 3. Land-use and land-cover maps used in the climate simulations for the BR-319 highway area of influence: A) control year 2021; B) Scenario 1 (BAU_1-2050), without paving; C) Scenario 1 (BAU_1-2100), without paving; D) Scenario 2 (BAU_2-2050), with paving; and E) Scenario 2 (BAU_2-2100), with paving [48].

For the climate simulations, the Eta Regional Climate Model from the National Institute for Space Research (INPE) was employed [26][61]. The Eta model is a limited-area atmospheric model originally developed at the University of Belgrade in collaboration with the Hydrometeorological Institute of Yugoslavia in 1993 [62].

An enhanced version of the model, known as Eta-CPTEC, was developed by the Center for Weather Forecast and Climate Studies (CPTEC) at INPE. This version incorporated significant modifications to improve the representation of climate change interactions in South America [26][61][63][64]. The principles, parameterizations, and configuration settings of the model are described in AppendixA.

In this study, five numerical experiments were performed using the Eta regional climate model, named as follows: CNT-2021, BAU_1-2050, BAU_1-2100, BAU_2-2050, and BAU_2-2100. The control experiment (CNT) used vegetation cover data for the Legal Amazon region from the National Institute for Space Research (INPE) and deforestation data from the PRODES project for the baseline year 2021 (INPE, 2021). The other experiments employed deforestation scenarios developed by [48], projected for the years 2050 and 2100, respectively.

Each Eta model experiment consisted of one continuous 10-year integration (120 months), starting at 00:00 UTC on November 1, 2011 (wet season). Each experiment included a spin-up period of one year (12 months).

Initial and boundary atmospheric conditions were derived from ERA5 reanalysis data produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) [65]. These reanalyses have a horizontal resolution of 0.25˚ × 0.25˚ and cover the domain from 57˚S to 79.5˚N and from 180˚W to 180˚E, encompassing all continents except Antarctica. The data are available at: https://www.ecmwf.int/en/forecasts/datasets/browse-reanalysis-datasets.

The model assimilated the reanalysis fields every six hours via dynamic downscaling. Sea surface temperature (SST) values were derived from monthly ECMWF reanalysis means and were updated daily using linear interpolation. Initial soil moisture and surface albedo conditions were based on monthly climatology for the first month of integration and on seasonal climatology, respectively.

For this study, the Eta-ERA5 model was configured with a horizontal resolution of 3 km, 38 vertical layers, a timestep of 10 seconds, and a spatial domain from 67.0˚W to 59.0˚W and from 8.0˚S to 2.0˚S. During the numerical integrations, the atmospheric CO₂ concentration in the model was held constant at the mean value of 414 ppmv, since the primary focus was to analyze climate changes resulting specifically from the BR-319 highway project.

Validation of the Eta Regional Model

To assess the performance of the Eta regional model in simulating the climate of the study area, rainfall estimates from CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data) [66] and temperature data from SAMeT (South American Mapping of Temperature) [67] were used. The validation was based on mean seasonal fields for the wet season (austral summer: December-January-February, DJF) and the dry season (austral winter: June-July-August, JJA). For the quantitative evaluation of the Eta model performance, the following statistical metrics were applied: bias and root mean square error (RMSE).

The performance evaluation of the Eta regional climate model in representing climate patterns was carried out for the year 2021. The evaluated components were air temperature and precipitation. Figure 4 presents the observed mean air temperature from the SAMeT dataset and the simulated data from the Eta model for the wet (DJF) and dry (JJA) seasons in 2021. In addition, model performance was assessed using bias and Root Mean Square Error (RMSE).

The Eta model successfully reproduced the spatial variability of air temperature across the study area during both wet and dry seasons. When comparing the simulated values to observed data, the mean temperature difference for the wet season (DJF) was 0.003˚C, while for the dry season (JJA) it was −0.29˚C. These differences correspond to 0.01% and −1.08% relative to the observed temperatures for each respective season. Regarding RMSE, lower values were observed during DJF (0.19˚C), whereas greater discrepancies were found during JJA (0.39˚C). Despite these differences, the values were close to the annual mean variation, indicating that the model reasonably captured the seasonal temperature patterns. Therefore, the results suggest that the Eta model was effective in simulating air temperature in the region.

Figure 5 shows the observed precipitation data from CHIRPS and the simulated precipitation fields produced by the Eta regional climate model for the wet season (DJF) and dry season (JJA) of the year 2021. Additionally, bias and root mean square error (RMSE) metrics were calculated to assess model performance.

Figure 4. Evaluation of the Eta climate model in representing 2-m air temperature (˚C) for the wet season (DJF): (a) Observed SAMeT 2021, (b) Simulated Eta Model 2021, (c) Bias (˚C), (d) RMSE (˚C); and the dry season (JJA): (e) Observed SAMeT 2021, (f) Simulated Eta Model 2021, (g) Bias (˚C), (h) RMSE (˚C).

Figure 5. Evaluation of the Eta regional climate model performance in simulating precipitation (mm∙day⁻¹) for the wet season (DJF): (a) observed CHIRPS 2021; (b) simulated Eta Model 2021; (c) bias (mm∙day⁻¹); (d) RMSE (mm∙day⁻¹); and for the dry season (JJA): (e) observed CHIRPS 2021; (f) simulated Eta Model 2021; (g) bias (mm∙day⁻¹); (h) RMSE (mm∙day⁻¹).

Overall, the Eta model successfully reproduced the spatial variability of precipitation in both seasons across most of the study area. However, it systematically underestimated precipitation values, with an average deviation of −37.5% in DJF and −5.7% in JJA compared to the observed data. This underestimation may be attributed to limitations in the model’s physical parameterizations of precipitation or to the numerical integration period, which may have been insufficient for the model to reach dynamic equilibrium. Furthermore, as noted by [69], the Eta model generally tends to underestimate precipitation over large portions of South America.

It should be noted that the Eta model systematically underestimates precipitation, particularly during the wet season (by −37.5%). This bias affects the absolute values of rainfall, but it does not compromise the model’s ability to capture spatial patterns and seasonal contrasts. Consequently, the simulated changes in precipitation across deforestation scenarios should be interpreted as robust in terms of direction and relative magnitude, although the absolute values may be somewhat underestimated.

Regarding the residual error, similar RMSE values were observed for both seasons: 0.17 mm∙day⁻¹ for DJF and 0.15 mm∙day⁻¹ for JJA. Despite the underestimation during DJF, the simulated values captured the seasonal variation pattern of annual precipitation. Therefore, the model is considered to have performed satisfactorily in representing precipitation behavior and is suitable for investigating the climate impacts associated with deforestation.

Detailed statistical parameters used for model evaluation (bias and RMSE) are provided in the Supplementary Material (Appendix B, TablesA2-A3), along with a comparison of the seasonal cycles of observed and simulated air temperature and precipitation (Appendix B, FigureA1).

Figure 6 shows an increase in air temperature during both the wet season (DJF) and the dry season (JJA) when comparing the BAU_1 scenario (unpaved BR-319 highway) and the BAU_2 scenario (paved BR-319 highway) to the current climate baseline (reference year 2021). Relative to the current climate, the BAU_1 scenario projects an average temperature increase of 0.1˚C in both wet and dry seasons by 2050. In the BAU_2 scenario, the average increase by 2050 is 0.1˚C during the wet season and 0.5˚C during the dry season. For the year 2100, the temperature increase reaches 0.3˚C (wet season) and 0.5˚C (dry season) in BAU_1, and 0.4˚C (wet season) and 0.7˚C (dry season) in BAU_2. As illustrated in Figure 6, local temperature increases may exceed 2.5˚C in some deforested regions during the dry season.

The increase in air temperature in deforested areas is related to various factors associated with vegetation removal, particularly changes in surface energy and water balance. Alterations in soil and vegetation parameters, such as porosity, leaf area index, roughness length, and stomatal conductance, modify the surface energy and water flux components, consequently affecting the regional climate in deforested areas. In this study, reduced latent heat flux (evapotranspiration) and increased sensible heat flux over deforested regions contributed to the rise in air temperature, especially during the dry season.

During the wet season, the experiment indicated an increase in latent heat flux and a decrease in sensible heat flux, which acts in the direction opposite to that of the temperature increase. The increase in latent heat flux may have partially mitigated local air warming but did not entirely reverse it. This phenomenon might be linked to changes in surface energy availability. A similar result was reported by [70] in studies on the climatic impacts of deforestation in central Amazonia, which suggested that surface temperature increases, alongside reduced sensible heat flux, can be explained by a reduction in the modeled temperature gradient.

Figure 6. Impact on air temperature (˚C) at 2 m above the ground in 2050 and 2100 resulting from the land-use change scenarios BAU_1 (unpaved BR-319) and BAU_2 (paved BR-319), compared to the current climate (CNT-2021), for the wet season (DJF panels b, c, d, e) and dry season (JJA panels g, h, i, j).

Figure 7 illustrates the impact of paving the BR-319 highway (BAU_2) on air temperature at 2 m, relative to the unpaved scenario (BAU_1). The greatest impact is observed over deforested areas in the northern portion of BR-319 and the central region of the maps, notably around the AM-366 highway, where new deforestation patches are emerging within the forest. Temperature increases are present throughout all analyzed periods, reaching nearly 1.0˚C during the wet season and 1.5˚C during the dry season in deforested regions.

Figure 7. Impact on surface air temperature (˚C) at 2 m above the ground in 2050 and 2100, resulting from the comparison between scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), for the wet season (DJF: panels a, d) and dry season (JJA: panels c, d).

A limitation of our approach is that atmospheric CO₂ concentration was held constant in all experiments to isolate the effects of land-use change. However, it is well known that rising CO₂ interacts with vegetation physiology and structure in complex ways. Higher CO₂ concentrations tend to reduce stomatal conductance and evapotranspiration, intensifying regional warming and drying, but may also stimulate biomass growth and leaf area expansion, partly mitigating these impacts [71][72]. These aspects suggest that the combined effects of land-use change and increasing CO₂ are likely to produce more complex climate responses.

Figure 8 presents the impact on precipitation under scenarios BAU_1 and BAU_2 during the wet (DJF) and dry (JJA) seasons for the study area. An increase in precipitation is observed, especially in areas of new “fishbone” deforestation patterns, where deforestation has intensified due to the reconstruction of the BR-319 highway and the opening of new connecting roads. Comparing the simulated scenarios for 2050 with the current climate, there is an average increase of 0.1 mm∙day⁻¹ in both the wet and dry seasons for scenarios BAU_1 and BAU_2. By 2100, scenario BAU_1 shows an average increase of 0.4 mm∙day⁻¹ during the wet season and 0.3 mm∙day⁻¹ during the dry season, while scenario BAU_2 exhibits an increase of 0.5 mm∙day⁻¹ and 0.4 mm∙day⁻¹ for the wet and dry seasons, respectively. As illustrated in Figure 8, precipitation increases exceeded 2.5 mm∙day⁻¹ in some deforested regions. We note that throughout the simulated period the deforestation is in the fishbone pattern, and that, as has occurred on other Amazonian frontiers, it will later be “consolidated” into vast expanses of cleared land. The increase in precipitation in the intermediate deforestation phase represented in the simulation is as expected, but, with progressive consolidation of the deforested landscape, the effect will change to greatly decreased rainfall [73]. The high spatial heterogeneity, here in the form of the “fishbone” pattern, can have an effect at least as important as the area of deforestation in forming the anomalous circulations that led to our simulated precipitation increases during the time span of our study [74]. This effect is temporary, and various conservative assumptions in the deforestation simulation [48] may mean that decreased precipitation begins sooner than the simulated climate implies.

Although it is well established that advancing deforestation results in local warming, impacts on precipitation remain uncertain and continue to be the subject of extensive study [73][75]. Generally, the literature points to a reduction in precipitation as a consequence of deforestation. However, some studies, such as [18][28][30]-[34], report contrasting results. They suggest that in certain areas, precipitation may increase due to forest fragmentation and the dominance of local and mesoscale circulations driven by surface heterogeneity, which is determined by the extent of deforestation. This implies that in areas of smaller-scale deforestation, heterogeneity in sensible and latent heat fluxes can generate intense mesoscale circulations, affecting the planetary boundary layer structure and influencing cloud formation and precipitation patterns [29][35][76].

Figure 9 illustrates the impact of paving the BR-319 highway (scenario BAU_2) on precipitation by 2050 and 2100 compared to the unpaved scenario (BAU_1). The greatest impact occurs in the areas influenced by the BR-319 and AM-366 highways, where new deforested areas are opening. The impact intensifies through 2100 in both seasons, with an average increase of 0.1 mm∙day⁻¹ for the study area. Notably, larger discrepancies are observed during the wet season, with precipitation increases exceeding 1.5 mm∙day⁻¹ in deforested regions, alongside reductions in precipitation within surrounding forest patches and in areas where deforestation becomes more extensive, particularly in the southeastern portion of the study area.

The results corroborate previously mentioned studies indicating that, as deforestation expands beyond the early “fishbone” phase, an overall decrease in precipitation occurs. The scenario comparisons suggest that increasing deforestation leads to diminished rainfall and heightened regional aridity due to reduced moisture availability and altered atmospheric dynamics.

By 2100, the simulated landscape still represents a transitional stage where both fragmented fishbone deforestation and more consolidated clearings coexist. In areas where deforestation is already extensive, precipitation decreases are evident, whereas in regions with relatively smaller fragments rainfall increases are still observed. It is important to emphasize that 2100 does not represent the endpoint of deforestation in the Amazon, but rather a temporal snapshot under the Business as Usual scenario. As deforestation continues to expand and consolidate beyond 2100, precipitation trends are expected to shift toward significant long-term declines.

Figure 8. Impact on precipitation (mm∙day⁻¹) in 2050 and 2100 resulting from land-use change scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), compared to the current climate (CNT-2021), during the wet (DJF) panels b, c, d, e) and dry (JJA) panels g, h, i, j) seasons.

Figure 9. Impact on precipitation (mm∙day⁻¹) in 2050 and 2100 resulting from the comparison between scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), for the wet (DJF) panels a, d) and dry (JJA) panels c, d) seasons.

Figure 10 illustrates the impact of the scenarios on evapotranspiration, showing an increase in evapotranspiration effects in both scenarios during the wet season (DJF), and a reduction during the dry season (JJA). It is evident that during the wet season, evapotranspiration increases, as represented by the rise in latent heat flux.

Evapotranspiration encompasses both soil and surface water evaporation and plant transpiration, which is explained by the increase in latent heat flux. Deforestation reduces transpiration due to the removal of trees and other vegetation that previously performed this function. Consequently, the amount of water vapor released into the atmosphere decreases significantly.

Conversely, results indicate an increase in evapotranspiration during the wet season (DJF), which can be attributed to enhanced local precipitation. This modifies water vapor availability, since forest removal compromises water recycling via evapotranspiration and intensifies direct evaporation from exposed soil surfaces or water bodies. This process may increase local humidity, promote cloud formation, and potentially enhance rainfall occurrence, particularly during wet periods [77][78]. Additionally, stomatal resistance is lower in pasture areas than in forests, contributing to increased evaporation in non-forested patches [79].

Figure 10. Impact on evapotranspiration (mm∙day⁻¹) in 2050 and 2100 resulting from land-use change scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), compared to the current climate (CNT-2021), during the wet (DJF) panels b, c, d, e) and dry (JJA) panels g, h, i, j) seasons.

The increase in latent heat flux can also be explained by intensified wind speeds over fragmented areas and the effect of turbulent mixing over wet surfaces in the Amazon [70][80]. From a micrometeorological perspective, trees at forest edges tend to transpire more than those in the forest interior [81]. This suggests that, at a local scale, forest fragmentation may lead to increased evapotranspiration, partially offsetting reductions caused by deforestation [35].

During the dry season (JJA), evapotranspiration decreases while precipitation increases. This pattern results from mesoscale circulations induced by landscape heterogeneity. According to [31], these processes enhance moisture convergence, which outweighs the reduction in evapotranspiration and consequently increases precipitation.

Figure 11. Impact on evapotranspiration (mm∙day⁻¹) in 2050 and 2100 resulting from the comparison between scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), for the wet (DJF) panels a, d) and dry (JJA) panels c, d) seasons.

Figure 11 depicts the impact of paving the BR-319 highway (scenario BAU_2) on evapotranspiration compared to the unpaved scenario (BAU_1). The most pronounced impact is observed in deforested areas in the northern portion of the BR-319 and central parts of the maps, particularly near the AM-366 highway, where new deforestation patches have formed within the forest matrix. In these deforestation islands influenced by road development, increases in evapotranspiration may exceed 0.5 mm∙day⁻¹.

Figure 12 shows an increase in moisture convergence within the study area across all scenarios compared to the control experiment, particularly over newly deforested regions. This result aligns with the previously presented data, indicating that the increase in precipitation is a consequence of enhanced moisture convergence.

Figure 12. Impact on moisture convergence (mm∙day⁻¹) in 2050 and 2100 resulting from land-use change scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), compared to the current climate (CNT-2021), during the wet (DJF) panels b, c, d, e) and dry (JJA) panels g, h, i, j) seasons.

It is noteworthy that during the dry season (JJA), the impacts on moisture convergence and evapotranspiration acted in opposite ways. The increase in moisture convergence over deforested areas was sufficiently strong to compensate for the reduction in evapotranspiration. This rise in precipitation can be explained by changes in local circulation, as alterations in surface heterogeneity due to deforestation generate mesoscale circulations, favoring increased moisture convergence and, consequently, enhanced precipitation over deforested regions [18][31][82].

Figure 13 presents the impact on moisture convergence from the comparison between the two analyzed scenarios. The greatest impact is observed along the highways, with a reduction in moisture convergence (positive sign) in the eastern portion of the BR-319 highway and an increase in convergence (negative sign) in deforestation islands, especially considering the opening of new roads such as the AM-366, located more centrally in the maps.

Figure 13. Impact on moisture convergence (mm∙day⁻¹) in 2050 and 2100 resulting from the comparison between scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), for the wet (DJF) panels a, d) and dry (JJA) panels c, d) seasons.

Figure 14 presents the mean total runoff simulated by the Eta model for the wet (DJF) and dry (JJA) seasons. During both seasons, runoff values were positive (Precipitation > Evapotranspiration) across most of the study area, with higher values (>1.5 mm∙day⁻¹) observed in deforested areas during the wet season.

Figure 14. Impact on surface runoff (mm∙day⁻¹) in 2050 and 2100 resulting from land-use change scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), compared to the current climate (CNT-2021), during the wet (DJF) panels b, c, d, e) and dry (JJA) panels g, h, i, j) seasons.

The increase in surface runoff was primarily driven by the increase in precipitation within the study area. Conventionally, the conversion of forests to agriculture or pasture increases surface runoff partly due to reduced evapotranspiration [83], which was observed during the dry season (JJA) of the experiment. However, during the wet season (DJF), the increased rainfall resulted in a proportional increase in surface runoff. This water availability favored an increase in latent heat flux, which is responsible for water evaporation, and a reduction in sensible heat flux [84].

Figure 15shows the impact of paving the BR-319 highway (BAU_2) on surface runoff compared to the unpaved scenario (BAU_1). An average increase in surface runoff of 0.1 mm∙day⁻¹ was observed across the study area for both seasons until 2100. In some deforested areas, runoff values may exceed 1.0 mm∙day⁻¹ in both the wet and dry seasons by 2100, following the same pattern observed for precipitation.

Figure 15. Impact on surface runoff (mm∙day⁻¹) in 2050 and 2100 resulting from the comparison between scenarios BAU_1 (unpaved BR-319 highway) and BAU_2 (paved BR-319 highway), for the wet (DJF) panels a, d) and dry (JJA) panels c, d) seasons.

The Eta regional model was adapted to perform long-term integrations (Pesquero et al., 2010; Chou et al., 2014a, b). Due to the higher resolution of the Eta regional model, it is possible to simulate mesoscale phenomena in greater detail, such as frontal systems, orography, sea breeze, severe storms, squall lines, and organized mesoscale systems (Chou, 1996). The dynamics of the regional model are developed using a vertical coordinate (η) in the form of discrete steps (Mesinger, 1984).

where:

where, η s f c is the conversion from the sigma vertical coordinate (which follows the terrain) to the Eta coordinate (in the form of discrete steps); P t is the pressure at the top of the domain; P s f c and Z_sef are the pressure and elevation at the lower boundary of the model, respectively. The reference pressure ( P r e f ) is a function of the height above mean sea level.

The physical parameterizations included in the model are:

1) Turbulent processes in the atmosphere, represented by the Mellor-Yamada (1974) scheme, updated with the parameters of Nakanishi & Niino (2009), which calculates turbulent kinetic energy and vertical fluxes;

2) Turbulent energy diffusion in the surface planetary boundary layer (PBL), determined using the Mellor-Yamada 2.0 scheme (Mellor & Yamada, 1974);

3) The shortwave radiation scheme is based on the model by Lacis & Hansen (1974);

4) The longwave radiation scheme follows the model of Fels & Schwarzkopf (1975);

5) Precipitation in the model is produced by cumulus parameterization schemes proposed by Betts-Miller-Janjic (Janjic, 1994);

6) Cloud microphysics follows the approach described by Zhao et al. (1997);

7) The surface layer over land is resolved using the Monin-Obukhov scheme with stability functions from Paulson (1970), while the surface layer over the ocean uses the Charnock (1955) scheme;

8) Biosphere-atmosphere interaction processes are represented by the NOAH land surface scheme (Ek et al., 2003), which distinguishes 16 vegetation types, 15 soil classes, and eight soil layers with roots extending to a depth of up to 12 meters (Pilotto et al., 2023).

TableA1. Parameterizations included in the Eta-CPTEC regional model.

The land use and land cover map is derived from the ESA-CCI-LC (European Space Agency - Climate Change Initiative Land Cover) for the year 2000. The soil map is from STATSGO/FAO (State Soil Geographic/Food and Agriculture Organization) (Miller & White, 1998), with a spatial resolution of 1 km.

The model will be configured with 38 vertical levels, with the model top defined at 25 hPa. Time integration uses the split-explicit technique. The terms adjusted by inertial gravity waves are integrated separately from the advective terms. A forward-backward scheme handles the terms responsible for this adjustment (Janjic, 1979), while a modified Euler-backward scheme is applied to the horizontal and vertical advection terms (Janjic, 1984). The prognostic variables are: air temperature, specific humidity, horizontal wind, surface pressure, turbulent kinetic energy, and cloud hydrometeors. These variables are distributed on an Arakawa E-grid (Arakawa & Lamb, 1977), in which the distance between adjacent mass or wind points is used to define the model’s horizontal resolution.

Bias indicates the systematic tendency of the simulations. In other words, if the bias is greater than zero, the forecasts are, on average, overestimated; whereas if it is less than zero, the forecasts are underestimated.

The root mean square error (RMSE) method will also be used to assess the accuracy of the simulations, where high RMSE values indicate high levels of discrepancy between the simulations (ϕ) and the observed and reanalysis data (ψ). The RMSE formula is given by:

TableA2. Monthly average values observed and simulated by the Eta model for Temperature (SAMeT) and Precipitation (CHIRPS) for the year 2021, and monthly observed averages for the period 2012 to 2021.

TableA3. Values found for bias and root mean square error for average temperature (˚C) at 2 m above ground level (SAMeT) and precipitation (mm∙day⁻¹) (CHIRPS) in the simulation of the Eta regional model for the area influenced by highway BR-319.

FigureA1.Seasonal cycle of (a) air temperature (˚C) at 2 m above ground level and (b) precipitation (mm∙day⁻¹). The black lines correspond to the Eta model simulation (Model 2021); the lilac shaded area represents the standard deviation (2012-2021 observed climate range); the blue line is the mean (2012-2021 observed climate average); and the red line is the monthly observed mean for the year 2021.