The highest Aerosol Optical Depth (AOD) values occurred in Alta Floresta and Ji-Paraná (AERONET = 2.02 and CAMS = 2.24), respectively (Fig. 3). During the dry season, there is a significant increase in average AOD in several regions of the southern Amazon (ROCHA; YAMASOE, 2013), especially in sites that are located in or close to the arc of deforestation, such as Alta Floresta and Ji-Paraná. This substantial increase is mainly attributed to regional emissions resulting from biomass burning, as well as the transport of aerosols from distant areas (Morais et al., 2022; Palácios et al., 2022). In addition, the central region of the Amazon is influenced by the descending branch of the Hadley cell, which causes dry conditions, favoring the concentration of aerosols in the atmosphere (Rocha; Yamasoe, 2013).
AOD values were lowest during the rainy season in Manaus and Cuiabá (AERONET and CAMS = 0.02), which is probably only related to the cities urban emissions, such as thermal power plants near the site.
The highest AOD values were observed by AERONET and CAMS in 2007. It is important to note that according to Lizundia et al. (2020), one of the largest burnt areas detected in South America was in 2007 in Brazil, Paraguay and Colombia. From 2004 to 2008, there was a consistent upward trend in AOD (Fig. 3). In addition, during the period from 2008 to 2016, there was a downward and stable trend in the concentration of aerosols, except for a notable increase in 2011, observed mainly at sites located in the south and southeast. This particular year was drier in these regions, conditioned by the La Nina phenomenon, which favored uncontrolled deforestation of large areas and fires that occurred mainly in pastures (De Andrade et al., 2020). From 2017 onwards, the data indicates a clear upward trend in the concentration of aerosols again, occurring mainly in the Amazon and Cerrado biomes (Fig. 3). These were responsible for 86% of total particulate matter emissions in Brazil from 2003 to 2020 (Pereira et al., 2022).
Thus, the variability of AOD over the years is evident, which can be influenced by different precipitation rates, large-scale meteorological phenomena such as El Niño and La Niña, and government policies for managing deforestation (De Andrade et al., 2020). This variability reflects the complexity of the natural and anthropogenic factors that affect atmospheric aerosol concentrations at the sites analyzed and highlights the importance of considering these aspects when interpreting the results and implementing strategies to control and mitigate aerosol emissions (Marengo; Espinoza, 2016).
The Alta Floresta, São Paulo, Cuiabá, São Martinho, Petrolina, Campo Grande and Itajubá sites showed no statistically significant differences between the CAMS estimates and the AERONET measurements (Fig. 4). However, at the other sites, such as Ji-Paraná, Rio Branco, Manaus, São Paulo (EACH) and ATTO, significant differences were observed with the AERONET measurements in certain months of the year. In March, Rio Branco recorded an overestimate of 66% (Fig. 4). In Manaus, overestimates occurred between March and June, with values ranging from 75 to 166%. ATTO was also overestimated during the months of April to June, with a range from 75 to 128% (Fig. 4). CAMS AOD estimates are severely limited during the assimilation of terrestrial and satellite data (Gueymard; Yang, 2020). In addition, the high occurrence of cumulus cloud clusters in the Amazon probably affected AOD detections, influencing the overestimation of estimates at these sites (Pereira et al., 2022).
The results showed a strong correlation (indicated by *** with a significance level of p < 0.001) between the AERONET measurements and the CAMS product estimates at most of the sites assessed. Furthermore, when analyzing the agreement indices, represented by Willmott index "d", it was found that the CAMS product estimates also showed good agreement with the observed AOD values, with values approaching 1. However, the lowest values in the AOD statistical metrics were observed in Petrolina (Table 2).
The error parameters, such as RMSE and MAE, showed relatively low values at all the sites analyzed, especially in São Paulo-EACH, Petrolina and Itajubá. However, there was variability in the results between the different sites. For example, Ji-Paraná, Rio Branco, São Martinho, Manaus, ATTO and Itajubá had high PBIAS values, indicating that the CAMS product tends to overestimate the AOD values in these locations. On the other hand, in Petrolina the CAMS product showed a tendency to underestimate AOD (Table 2).
Table 2
Statistical parameters for Aerosol Optical Depth (AOD) measured by AERONET and estimated by the Copernicus Atmosphere Monitoring Service (CAMS) product. The parameters include: Pearson correlation index "r", Willmott index "d", Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Percentage Bias (PBIAS). Correlations with *** indicate a significance level of p < 0.001.
Sites
|
R
|
d
|
RMSE
|
MAE
|
PBIAS
|
Alta Floresta
|
0.86***
|
0.92
|
0.16
|
0.07
|
2.71
|
Ji-Paraná
|
0.93***
|
0.96
|
0.11
|
0.06
|
15.77
|
Rio Branco
|
0.94***
|
0.96
|
0.09
|
0.06
|
15.86
|
São Paulo-EACH
|
0.87***
|
0.92
|
0.04
|
0.03
|
9.30
|
São Paulo
|
0.78***
|
0.88
|
0.05
|
0.04
|
2.93
|
Cuiabá
|
0.95***
|
0.97
|
0.07
|
0.05
|
-5.79
|
São Martinho
|
0.65***
|
0.77
|
0.09
|
0.05
|
20.70
|
Petrolina
|
0.65***
|
0.72
|
0.03
|
0.03
|
-20.13
|
Campo Grande
|
0.94***
|
0.97
|
0.05
|
0.03
|
3.26
|
Manaus
|
0.88***
|
0.86
|
0.08
|
0.07
|
27.71
|
Itajubá
|
0.85***
|
0.90
|
0.03
|
0.03
|
12.19
|
ATTO
|
0.86***
|
0.79
|
0.07
|
0.06
|
29.19
|
All the sites showed a linear relationship between the AOD measured by AERONET and the CAMS product estimates (Fig. 5). This behavior is evidenced by the significant values of the angular coefficient, interception coefficient, coefficient of determination (R²), as well as extremely low p-values in the linear regressions carried out at each site. This linear trend suggests that the AOD measured by AERONET and that estimated by CAMS have a well-defined relationship, allowing a regression line to be established that represents the relationship between them (Fig. 5).
The sites in Ji-Paraná, Rio Branco, Cuiabá and Campo Grande showed relatively high R², around 0.87 to 0.90, which shows that the data fit the regression line well (Fig. 5). On the other hand, São Martinho and Petrolina showed lower R² coefficients, around 0.42, which indicates a less precise fit of the data to the regression line. The Campo Grande station stood out with the highest intercept coefficient value, reaching 0.97. In addition, the Manaus, Cuiabá and Ji-Paraná sites obtained considerable intercept coefficients of between 0.70 and 0.95. On the other hand, the Petrolina station had the lowest intercept coefficient value, at 0.40 (Fig. 5).
Spatially, during certain times of the year, such as the dry season, an increase in aerosol concentrations was recorded (Fig. 6). This is due to the fires that occur during this period, releasing fine particles and gaseous pollutants into the atmosphere, resulting in an increase in the concentration of AOD. This is the result of the fires that have affected almost half of Brazil territory, mainly the Amazon and Cerrado biomes, which have been responsible for 86% of particulate matter emissions in Brazil over the last two decades (Pereira et al., 2022).
On the other hand, during the rainiest months, there was a reduction in these concentrations (Fig. 6). Rain has the effect of temporarily clearing the atmosphere, leading to a decrease in aerosol concentrations. The monthly spatial distribution of AOD in Brazil is influenced by various factors, such as climatic conditions, human activities, fires, vegetation and natural processes. Brazil diverse climate and ecosystems also contribute to seasonal variations in aerosol concentrations.
It is important to note that the regions with the highest concentrations of aerosols vary according to the period analyzed. Notably, the areas located in the arc of deforestation are the most affected by high concentrations of aerosols, especially during the month of September (Fig. 6). This region, located on the southern border of the Brazilian Amazon, is known for suffering from intense deforestation and agricultural activities (Pope et al., 2020). The increase in aerosol concentrations in this region is directly linked to deforestation and fires (Pope et al., 2020). These particles can have adverse impacts on air quality, human health and the environment.
In the Midwest and Southeast, harvesting and agricultural fires in the dry season contribute to increased concentrations of particles in the atmosphere (Palácios et al., 2016; Tariq et al., 2023). The Northeast faces fires during the dry season, mainly in semi-arid areas (De Oliveira et al., 2019; Oliveira et al., 2021). In the South, industrial and urban activities generate pollutants that increase aerosols, especially in densely populated cities (Calado et al., 2021). In addition, in this region the values do not vary greatly due to the presence of a more homogeneous precipitation regime throughout the year (Reboita et al., 2015).