Spatio-temporal dynamics of the evolution of atmospheric pollutants in the itasy region

Revue Internationale de la Recherche Scientifique (Revue-IRS) ISSN: 2958-8413 Vol. 3, No. 6, Décembre 2025 This is an open access article under the CC BY-NC-ND license. http://www.revue-irs.com 7304 Spatio-temporal dynamics of the evolution of atmospheric pollutants in the itasy region HERITAHINA Rambeloson1,3, RASOLOMANANA Eddy Harilala 1,2, RANDRIANJA Kanto Volahasina1,3 1Ecole Doctorale Ingénierie et Géosciences, 2Ecole Supérieure Polytechnique d’Antananarivo, 3Université de l’Itasy Abstract: This study aims to characterize the spatiotemporal dynamics of major pollutant gases—NO₂, CO, SO₂, O₃, and CH₄—in the Itasy region of Madagascar over the period 2019-2024, using Sentinel-5P/TROPOMI satellite data series. Monthly TROPOMI data were analyzed using statistical methods (nonparametric trend tests, correlations). The results show that CO and NO₂ exhibit strong seasonal variations, with peaks at the end of the dry season, such as CO increasing from approximately 0.021 to 0.037 mol/m² between February and October, and NO₂ from 9×10⁻⁶ to 2.3×10⁻⁵ mol/m². Tropospheric ozone follows a regular annual cycle, with a maximum (0.129 mol/m²) in October (dry season) and a minimum in May. Average SO₂ remains very low throughout the year (values close to zero). Spatially, CO, NO₂, and O₃ are more concentrated around urban centers and roads (Arivonimamo, Miarinarivo) than in rural areas. In addition, methane shows an upward trend from 1830 ppb in 2019 to 1875 ppb in 2024, an increase of +2.5%. These results confirm the strong seasonality of primary pollutants linked to anthropogenic activities such as vegetation fires, traffic, and weather conditions. Keywords: Sentinel 5P image, atmospheric pollutants, time series, spatial correlation, meteorological parameters Digital Object Identifier (DOI): https://doi.org/10.5281/zenodo.17848032 1 Introduction The analysis of air quality in the Malagasy Highlands, more specifically in the Itasy region, is a relevant area of study for better understanding local atmospheric dynamics. The peri-urban and rural areas of Itasy are characterized by slash-and-burn farming, limited waste management, and informal activities such as brick making, which have been identified as major sources of atmospheric emissions. These factors contribute to worrying levels of air pollutants and significant spatial variability within the territory. [1] This local assessment is part of a national context in which climate change has become a major issue: Madagascar is one of the country’s most vulnerable to climate change and least prepared to deal with it, which increases the risks associated with hazards (cyclones, droughts, floods) and environmental exposure. [2] In this context, the period 2019-2024 is of particular scientific interest: it corresponds to the continuous availability of Sentinel-5P/TROPOMI satellite data, which enable the mapping and monitoring of the spatiotemporal dynamics of the main gas tracers in a predominantly rural area such as Itasy. [3] This study, entitled “Spatiotemporal dynamics of atmospheric pollutant evolution in the Itasy region,” has the overall objective of characterizing the spatiotemporal dynamics of major atmospheric pollutants—nitrogen dioxide (NO₂), carbon monoxide (CO), sulfur dioxide (SO₂), ozone (O₃), and methane (CH₄)—in the Itasy region, based on continuous satellite data series, in order to better understand the climate issues facing this territory. Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7305 In light of this, two working hypotheses are proposed. First hypothesis: In the Itasy region, NO₂ and CO concentrations vary greatly throughout the year, with increases expected during the dry season and during vegetation fires, which should be clearly visible in the Sentinel-5P/TROPOMI series. Second hypothesis: O₃ levels, and to a lesser extent SO₂ levels, change with the seasons, influenced by sunlight. Dry periods are particularly conducive to the formation of tropospheric ozone and the accumulation of gases that contribute to its production. To better understand this, we will first review the materials and methods used, including the datasets and adopted methodology. Then, we will examine the results and interpretation of the characteristics of temporal variations in atmospheric pollutants, followed by their spatiotemporal variations and an analysis of temporal trends. In the discussion section, we will examine the correlation between meteorological factors and atmospheric pollutants, as well as the limitations and prospects of our research. 2 Materials and Methods 2.1 Study Area The Itasy region, located in the center of Madagascar's highlands, covers an area of approximately 6,993 km². It is bordered to the north by the Analamanga region, to the west by Bongolava, to the south by Vakinankaratra, and to the east by Amoron'i Mania. Itasy is characterized by volcanic terrain with altitudes ranging from 950 to 1,300 m, and by the presence of Lake Itasy, a major water reservoir that supports fishing, irrigation, and tourism activities. [4] [5] The region has a tropical highland climate, with hot, humid seasons and annual rainfall ranging from 1,330 to 1,575 mm, which, according to the Bank, directly influences the dispersion of air pollutants and the seasonal formation of tropospheric ozone (O₃). Traditional agricultural practices, such as biomass combustion and the burning of agricultural residues. The climate of the Highlands is characterized by a rainy season from November to March/April and a dry, cooler season from May to October, which imposes a strong seasonality on the atmospheric and hydrological processes observed in the study area. Antananarivo thus has a subtropical climate characterized by temperate (not too hot) summers. These climatic characteristics, confirmed by climatological summaries and monthly precipitation profiles, provide the framework for the spatio-temporal analysis of pollutants that the study proposes to conduct. [6] Our study focuses on the Itasy region, specifically in the districts of Soavinandriana, Miarinarivo, and Arivonimamo. The following figure shows a map of the Itasy region. Figure 1. Map showing the location of the Itasy regions Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7306 2.2 Data set used The data set includes all key data such as Sentinel-5P data sets and climate variability data. 2.2.1 Key data The table below summarizes the Sentinel-5P/TROPOMI OFFL/L3 datasets used, as well as the unit, nominal resolution, and date of availability for each pollutant. Table 1. Data used Polluant Earth Engine (ID & bande) Unit Resolution Start of data Nitrogen dioxide COPERNICUS/S5P/OFFL/L3_NO2-Band :tropospheric_NO2_column_number_density mol/m² ~1 113 m (0,01°) June 28, 2018 Carbon monoxide COPERNICUS/S5P/OFFL/L3_CO-Band : CO_column_number_density mol/m² ~1 113 m (0,01°) June 28, 2018 Sulfur dioxide COPERNICUS/S5P/OFFL/L3_SO2-Band : SO2_column_number_density mol/m² ~1 113 m (0,01°) December 5, 2018 Ozone COPERNICUS/S5P/OFFL/L3_O3-Band : O3_column_number_density mol/m² ~1 113 m (0,01°) September 8, 2018 Méthane COPERNICUS/S5P/OFFL/L3_CH4 - Band :CH4_column_volume_mixing_ratio_dry_air ppb/ppm ~1 113 m (0,01°) February 8, 2019 2.2.2 Additional data The table below summarizes the climate datasets used (CHIRPS v2) for precipitation (0.05° grid, 1981-present), ERA5 Monthly for temperature, and TerraClimate for monthly climate variables (1/24°, ~4 km, 1958-present), specifying for each the key variables, unit, resolution, and update period. [7] Table 2. Additional datasets used Dataset Key variables Unit Resolution Start of data Precipitations CHIRPS v2 Precipitation mm/j 0,05° (5,6 km) 1981 Temperature ERA5 Monthly mean_2m_air_temperature (T2m), m/s 0,25° (31 km) 1979 Monthly climate TerraClimate tmin, tmax, ws (wind), ppt, vapor_pressure °C 4 km (1/24°) 1958 Fire data MCD64A1 MCD64A1 Burn Date (day of burning), derived burned area (pixelArea) No unit, area in ha (m²/10⁴) 500 m (0,005°) 2000 (start of MODIS) 2.3 Data processing flowchart The study area is defined by isolating the region from the GAUL level 1 (ADM1) administrative boundaries, which provides an official and consistent perimeter for all subsequent spatial aggregations. In parallel with geographic information system operations in QGIS (import, cartographic consistency checks, and layout), the Sentinel-5P/TROPOMI collections in OFFL/L3 mode are opened in Google Earth Engine in order to load the NO₂, CO, SO₂, O₃, and CH₄ gas tracers required for spatio-temporal analyses. The Earth Engine datasets describe, in particular, the OFFL/L3 NO₂ and OFFL/L3 CH₄ products, used here as references for the period 2019-2024. Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7307 The pre-processing applied in GEE includes a 2019-2024 temporal filter, a spatial filter on Itasy via filterBounds then clip, and the use of gridded L3 products that facilitate regular aggregation at the regional level. Spatio-temporal aggregations are produced in the form of monthly and annual composites by pollutant and accompanied by zonal statistics calculated by region and district in order to quantify average levels, gradients, and intra-regional contrasts. Visualization and analysis combine maps by pollutant and period with time series covering the entire study window, making it possible to identify hotspots, seasonality, and possible interannual anomalies. Figure 2. Data processing procedures 2.4 Methodology for analyzing monthly trends in pollutant concentrations using Mann-Kendall To statistically evaluate monotonic trends in pollutant concentrations over time, we conducted the nonparametric Mann-Kendall (MK) trend test on the monthly average values of CO, NO₂, SO₂, and O₃ concentrations over the period 2019-2024. The Mann-Kendall test is widely used in the exploration of environmental time series because of its robustness to missing data, non-normal distributions, and the absence of linearity assumptions. The test was performed on the time series for each pollutant, yielding Kendall's τ coefficient, the p-value, and the trend class (increasing, decreasing, or no trend). A significance level of α = 0.05 was used to assess statistical significance. This step was decisive in quantifying long-term changes in concentrations and validating the patterns observed in the spatiotemporal maps. The Mann-Kendall results were then cross-referenced with spatial analysis to visualize statistically significant trends and interpret the dynamics. [8] 2.4.1 Monthly series by pollutant and region For a pollutant p (CO, NO₂, SO₂, O₃) and a region r, the series of monthly averages {x_t }_(t=1…n), the Mann Kendall statistic is written as: 𝑆=∑∑𝑠𝑔𝑛(𝑥𝑗−𝑥𝑖), 𝑠𝑔𝑛(𝑢)={+1 𝑢>0 0 𝑢=0 −1 𝑢<0 𝑛 𝑗=𝑖+1 𝑛−1 𝑖=1 And Variance with correction for ties (𝑡1, …,𝑡𝑔) : 𝑉𝑎𝑟 (𝑆)= 𝑛(𝑛−1)(2𝑛+5)−∑𝑡𝑝(𝑡𝑝−1)(2𝑡𝑝+5) 𝑔 𝑝=1 18 The normalized statistic with continuity correction is: Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7308 𝑍= { 𝑆−1 √𝑉𝑎𝑟(𝑆) 𝑖𝑓 𝑆>0, 0 𝑖𝑓 𝑆=0, 𝑆+1 √𝑉𝑎𝑟(𝑆)𝑖𝑓 𝑆<0. 2.4.2 Seasonal Kendall Adaptation Pollutants exhibit monthly seasonality (weather chemistry, burning cycles, heating, etc.). To avoid this, we apply the Seasonal Mann Kendall method for each month. 𝑚∈{1,…,12} construct the interannual sub-series {𝑥𝑚,𝑦}𝑦=2019−2024, calculate 𝑆𝑚 and 𝑉𝑎𝑟(𝑆𝑚) as above, comparing only the same months (January with January, etc.), then aggregate [10] 𝑆𝑆𝐾=∑𝑆𝑚, 12 𝑚=1 𝑉𝑎𝑟 (𝑆𝑆𝐾)=∑𝑉𝑎𝑟(𝑆𝑚) 12 𝑚=1 2.4.3 Air-specific magnitude and interpretation Kendall's tau: 𝜏=𝑆/(𝑛2) (or with Seasonal MK, τ is calculated from 𝑆𝑆𝐾 ummarizes the direction and strength of the trend (values close to ±1= clear trend). [6] 2.4.4 Pearson correlation In order to analyze the linear relationships between the different variables considered (air pollutants and possibly meteorological parameters), a Pearson correlation matrix was calculated. This matrix groups together the correlation coefficients r that exist between all pairs of variables observed in a square table. Pearson's linear correlation coefficient, which can be calculated and understood for two quantitative variables, makes it possible to assess the intensity and direction of the linear relationship between them. [11] For two variables 𝑋 and 𝑌, Pearson's correlation coefficient 𝑟𝑋𝑌 is defined by : 𝑟𝑋𝑌=∑(𝑥𝑖−𝑥)(𝑦𝑖−𝑦) 𝑛𝑖=1 √∑(𝑥𝑖−𝑥)2 𝑛𝑖=1 √∑(𝑦𝑖−𝑦)2 𝑛𝑖=1 where 𝑥𝑖 et 𝑦𝑖 represent the observed values of variables X et 𝑌, 𝑥 and 𝑦 their respective means, and 𝑛 the number of observations. The coefficients 𝑟 take values between -1 and +1. A value close to +1 indicates a strong positive linear correlation (the two variables tend to move in the same direction), a value close to -1 indicates a strong negative linear correlation (an increase in one is associated with a decrease in the other), while a value close to 0 suggests the absence of a marked linear relationship. For this study, Pearson's correlation matrix was established based on the series of monthly averages for each year of the period under consideration. This approach makes it possible to identify pollutants that behave similarly over time, likely due to common emission sources or similar formation mechanisms, and to track the evolution of these relationships from one year to the next. The correlation matrices obtained in this way are a diagnostic tool for understanding the signatures of sources and interactions between pollutants in the region under study. Pearson's correlation matrix 𝑅 brings together all the coefficients 𝑟𝑖𝑗 between each pair of variables (pollutants, meteorological parameters). 𝑅=(1 𝑟12 𝑟21 1 ⋮ ⋮ 𝑟𝑝1 𝑟𝑝2 … 𝑟1𝑝 … 𝑟2𝑝 ⋱ ⋮ … 1) 2.4.5 Estimating the trend using Sen's slope In order to quantify the magnitude of the trends identified by the Mann-Kendall test, we used Sen's slope estimator (Theil-Sen). This is a nonparametric method that provides a robust estimate of the linear trend in a time series. It is widely used in hydrology, climatology, and air quality studies because of its low sensitivity to outliers and non-normality of data [12]. For each air pollutant (CO, NO₂, SO₂, O₃, CH₄), the series of monthly averages x_t was considered over the study period. The Sen slope β is calculated as the median of all slopes between all pairs of points (𝑡𝑖,𝑥𝑖) et (𝑡𝑗,𝑥𝑗) with 𝑗>𝑖∶ Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7309 𝛽=𝑚𝑒𝑑𝑖𝑎𝑛 (𝑥𝑗−𝑥𝑖 𝑡𝑗−𝑡𝑖) ,∀ 𝑗>𝑖. A positive slope (β>0) indicates an upward trend in pollutant concentrations, while a negative slope (β<0) indicates a downward trend (relative improvement in air quality for that pollutant). The combination of the Mann-Kendall test (for the direction and significance of the trend) and Sen's slope (for the intensity of the trend) thus provides a robust characterization of the temporal evolution of air pollutants in the region studied. The estimated Sen slope values for each pollutant are then plotted graphically in order to visually compare the relative importance of trends between different pollutants and/or between different periods considered. [13] 3 RESULTS AND INTERPRETATION 3.1 Variation of meteorological parameters in the Itasy region between 2019 to 2024 The figure below illustrates the monthly evolution of the main meteorological parameters in the Itasy region, with the observed series shown as a solid line and a smoothed curve shown as a dotted line for each variable. The alternation between the rainy season and the dry season typical of the tropical highland climate is clearly visible. Figure 1. Variation in meteorological parameters in the Itasy region: (a) average temperature, (b) average precipitation, and (c) average relative humidity (a) Average temperature The average temperature generally ranges between 15°C and 22°C. Each year, a maximum is observed between November and March, when temperatures regularly exceed 20-21°C, followed by a marked minimum during the cool, dry season from June to September, around 15-16°C. From one year to the next, variations are moderate: the temperature regime remains relatively stable, with only slight differences in the intensity of the maximum and minimum temperatures, confirming a fairly consistent climate on an interannual scale. (b) Average precipitation Precipitation shows a very contrasting seasonal cycle. The rainy season is concentrated each year between approximately November and March, with peaks that can exceed 400-500 mm/month (particularly around the beginning of 2020 and 2024), reflecting very intense rainfall events. Conversely, the dry season (mainly from June to September) is almost completely rainless, with monthly totals close to 0 mm. There are interannual differences: some rainy seasons are more pronounced (very heavy rainfall), while other years have slightly more moderate peaks, suggesting more or less humid seasons. (c) Average relative humidity Relative humidity logically follows rainfall patterns: the highest values (above 80%) occur during the rainy season, when precipitation is abundant and the atmosphere is more humid. In the dry season, relative humidity decreases significantly, falling to around 55-60%, reflecting drier air during cooler periods. Wet years result in higher or more sustained humidity plateaus, while slightly drier years show slightly lower values and more pronounced minimums. Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7310 3.2 Characteristics of temporal variations in atmospheric pollutants in the Itasy region In this section, we present the variation curves between 2019 and 2024 for the main atmospheric pollutants studied (CH₄, C, NO₂, O₃, and SO₂), in order to characterize the spatio-temporal dynamics of air pollution in the Itasy region. In general, the respective variations in CO, NO₂, SO₂, CH₄, and O₃ above the Itasy region tend to be explained by a similar set of probable causes, which are, on the one hand, anthropogenic emissions from fuel combustion in road traffic, older vehicles, generators, and certain small industries, all of which are likely to be significant sources of CO, NO₂, and SO₂; and, on the other hand, the widespread use of solid biomass and charcoal for cooking and domestic heating in Madagascar, as well as the burning of waste, which together account for the majority of CO, NO₂, SO₂, and VOCs, while methane (CH₄) comes from agricultural activities (flooded rice cultivation, livestock farming, organic waste management) and natural wetlands, which are known to be significant sources of CH₄ in tropical areas. Finally, Tropospheric ozone (O₃), which is formed by a series of photochemical reactions involving NOₓ, CO, and VOCs under conditions of strong sunlight, according to well-established mechanisms, means that any increase in these precursors, NOx, CO, and VOCs under the synergy of weather conditions during the dry season causes a sharp increase in O₃ production in the atmospheric column. 3.2.1 CH₄ variation between 2019 to 2024 The following figure highlights an overall upward trend in methane (CH₄) over the entire 2019-2024 period. The annual average rises from around 1,830 ppb in 2019 to 1,875 ppb in 2024, an increase of around 45 units (2.5%), reflecting moderate seasonality but a gradual increase in background methane levels above the Itasy region. Figure 2. CH4 variation between 2019 to 2024 3.2.2 CO variation between 2019 to 2024 For carbon monoxide (CO), the figure highlights strong seasonal cyclicality. On average, the lowest values are observed in February, at around 0.021 mol/m², while the maximum values reach around 0.037 mol/m² in October. Over the entire 2019-2024 period, monthly concentrations range from 0.019 to 0.046 mol/m², with an annual average rising from around 0.025 mol/m² in 2019 to 0.028 mol/m² in 2024 (an increase of around 14%). These values confirm the existence of marked peaks at the end of the dry season/beginning of the hot season. Figure 3. CO variation between 2019 to 2024 Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7311 3.2.3 NO2 variation between 2019 to 2024 Nitrogen dioxide (NO₂) in mol/m² also shows a clear seasonal cycle. Average monthly values increase from 9×10⁻⁶ (minimum in February) to 2.3×10⁻⁵ (maximum in September). Over the period studied, concentrations generally ranged between 8×10⁻⁶ and 2.8×10⁻⁵, with an average of 1.3×10⁻⁵. The annual average rose from around 1.3×10⁻⁵ in 2019 to 1.5×10⁻⁵ in 2024, representing an increase of around 14% and indicating an increasingly significant NO₂ load in the atmospheric column, probably linked to the increase in anthropogenic emissions. Figure 4. NO2 variation between 2019 to 2024 3.2.4 O3 variation between 2019 to 2024 The ozone (O₃) series in mol/m² shows almost regular seasonal variability. Monthly averages range from approximately 0.113 (minimum in May) to 0.129 (maximum in October). Over the entire period from 2019 to 2024, concentrations range from 0.111 mol/m² to 0.133 mol/m², with an average rate of around 0.120 mol/m². Annual averages are relatively stable (0.118 in 2019 and 0.118 in 2024), with slightly higher years (0.122 in 20212023), suggesting interannual variability in ozone peaks rather than a marked upward trend. Figure 7. O3 variation between 2019 to 2024 3.2.5 SO2 variation between 2019 to 2024 Finally, for sulfur dioxide (SO₂) at very low levels, monthly values generally range between -5.4×10⁻⁵ and 1.35×10⁻⁴ for an overall average close to 1.4×10⁻⁵, with an average minimum observed in April (-3.1×10⁻⁵) and an average maximum observed in July (7.9×10⁻⁵). Annual averages remain close to zero (from -2×10⁻⁶ in 2019 to 1.1×10⁻⁵ in 2024), confirming the very low background concentration of SO₂ in Itasy. Figure 5. SO2 variation between 2019 to 2024 Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7312 3.3 Spatiotemporal variation of atmospheric pollutants in the Itasy region The following figure shows the average spatial distribution of the main atmospheric pollutants (CH₄, CO, NO₂, O₃, and SO₂) over the period 2019-2024 in the Itasy region, highlighting the contrasts between the districts of Miarinarivo, Soavinandriana, and Arivonimamo. Figure 9. Average variation in air pollutants between 2019 to 2024 Generally, the maps show that average pollutant concentrations are not uniform across the region, but exhibit spatial gradients related to population density, transportation corridors, and certain local environmental characteristics. CH₄ (2019-2024): Average methane values range from approximately 1835 to 1868 ppb/ppm, with generally higher levels in the northeastern and eastern parts of the region, near Arivonimamo and more anthropized areas, while slightly lower values are observed in the southwest, around Soavinandriana. This distribution may be associated with a combination of agricultural activities (flooded rice cultivation, livestock farming) and the presence of wetlands, which promote CH₄ emissions. CO (2019-2024): carbon monoxide has average values between 0.021 to 0.028 mol/m², with a gradient indicating slightly higher levels around Arivonimamo and Miarinarivo, where roads and urban centers are concentrated, and lower values in the more rural areas of Soavinandriana. This suggests that road traffic, domestic combustion, and biomass fires play an important role in the distribution of CO. NO₂ (2019-2024): The NO₂ map clearly shows a hotspot of higher concentrations in the east of the region, particularly around Arivonimamo, with average values ranging from 1.1×10⁻⁵ to 2.8×10⁻⁵ mol/m². This area corresponds to the vicinity of the axis connecting Itasy to Antananarivo and to more urbanized areas, suggesting a marked influence of road traffic, combustion activities, and population density. Levels are lower in the west and south (Soavinandriana). O₃ (2019-2024): For ozone, average values also increase towards the extreme southeast of the region, where concentrations are higher near Arivonimamo, while lower values are observed towards the center and west. This pattern is consistent with the fact that O₃ is a secondary pollutant, formed from precursors (NOₓ, CO, VOCs) emitted in more anthropized areas; the presence of higher ozone levels in these areas therefore reflects photochemical production from precursor gases. SO₂ (2019-2024): sulfur dioxide concentrations are very low, with a map marked by a mosaic of small patches of positive and negative values, with no clear hot spots. This indicates that the average SO₂ load remains low overall across the Itasy region, with any contributions likely coming from point sources of sulfur fuel Revue Internationale de la Recherche Scientifique (Revue-IRS) - ISSN : 2958-8413 http://www.revue-irs.com 7319 [11] J. C. J. H. Y. &. C. I. Benesty, Pearson correlation coefficient. In: Noise reduction in speech processing, Springer, Berlin, Heidelberg, 2009, p. p. 1-4.. [12] R. Gilbert, Statistical Methods for Environmental Pollution Monitoring., Van Nostrand Reinhold, New York, 1987. [13] H. Theil, A rank-invariant method of linear and polynomial regression analysis, Vols. %1 sur %2 53,, Nederl. Akad. Wetensch. Proc, 1950, p. 386-392. [14] Mongabay, Top environment stories from Madagascar in 2020, 2020. [15] R. V. ,. Martin, Satellite remote sensing of surface air quality. Atmospheric Environment, vol. 42(34), 2008.