SPACE-TIME VARIABILITY OF THE RONCADOR RIVER BASIN IN THE CHANGE OF LAND USE AND COVER AND ITS CORRELATION WITH CLIMATIC VARIABLES VARIABILIDADE ESPAÇO-TEMPORAL DA BACIA HIDROGRÁFICA DO RIO RONCADOR NA MUDANÇA DO USO E COBERTURA DO SOLO E SUA CORRELAÇÃO COM VARIÁVEIS CLIMÁTICAS

The objective of this study was to evaluate the space-time dynamics of the soil use and occupation of the Rio Roncador river basin between 1985 and 2010. The scenes were classified by two methods (partially unsupervised K-Means and supervised Maximum likelihood), the Thematic Mapper sensor products on the LANDSAT 5 orbital platform were used for both images of a 25-year time series (1985 to 2000). In order to measure the accuracy of the field the computer application Google Earth was used, in which nine classes (urban area, agricultural area, pasture, exposed soil, native forest, secondary vegetation, mangrove, altitude field and water) were collected. A multiple linear regression was performed, correlating the Normalized Difference Vegetation Index mean NDVI (dependent variable) with the independent climatic variables (global solar radiation MJmday, average air temperature °C, relative humidity -%, evapotranspiration mm d, and rain mm). According to the general classification by Kappa parameter of the images for 2005 and 2010, they were identified as very good (68% and 74%). These results confirm that the Roncador River Basin is undergoing transformation in its landscape, with an average reduction of -49% in native vegetation areas due to the increase in urban areas (25%) and agriculture (31%). The statistical analysis showed that rainfall and air temperature were the only variables that presented significant sigma (0.04) and (0.02). The obtained coefficient of determination indicated that 47% of the variations of the "vegetation index" are explained by the environmental variables.


INTRODUCTION
The human perception has been changing the conservation of natural resources, which may jeopardize the future generations. An example of the anthropic modifications and the reduction of the Atlantic Forest Biome that now owns only 7% of the original (MMA, 2016). The Atlantic Forest has been subjected to numerous changes in its ecosystem, being the land use and occupation the main causes of these transformations. Even today, impacts promoted by anthropic actions that can compromise their original preservation are verified (GASPARINI et al., 2013;SANTANA et al., 2016).
The change in the vegetation cover of the Atlantic Forest is a result of great impact activities on it, such as removal of permanent preservation areas in the vicinity of the basins (COUTINHO et al., 2013) and increase of urban areas without planning. The generation of negative impacts due to inappropriate management of land use and vegetation cover in basins can lead to climate change, which is directly correlated with environmental disasters (COUTINHO et al., 2013). Many authors have used several techniques for mapping the detection of changes in land use and occupation in basins (SILVA et al., 2014;RIBEIRO et al., 2016;BRASILEIRO et al., 2016;ALVES RÊGO et al., 2012). The study of the vegetation allows to generate information on the nature of the territory, also serving as a tool for the territorial planning of basins to mitigate the environmental impacts caused by the climatic changes caused by the intensification of anthropic actions (FREITAS et al., 2012e IPCC, 2015. In addition to providing information on vegetation dynamics, geotechnology tools have been used mainly to understand the spectral pattern of vegetation indexes associated with climatic factors, such as rainfall, air temperature, among others GOULART et al., 2015;NUNES et al., 2015). The reduction of biodiversity due to anthropic factors, followed by large episodes of forest fires, fires and extreme droughts, has a devastating impact on forest systems on a regional and global scale (CAÚLA et al., 2015). In this sense, the objective of this work was to evaluate the spatial-temporal evolution of the different classes of the land use and its relationship with the climate, in the Roncador River Basin in Magé, RJ.

Study area
Twenty five images of the TM (Thematic Mapper) sensor of Landsat-5 satellite ( Figure 1) were selected, orbit 217 and point 76 of the historical series available in the Catalog of Images on the site of the National Institute for Space Research (INPE, 2019) in order to obtain concomitantly images with lower cloud cover. In addition to the careful choice of very low cloud cover, TM images of Landsat 5 are mostly dry (April to September) ( Figure 1).

Processing of Thematic Mapper Image
The images were processed in ERDAS IMAGINE 2015 software, using the Model Maker platform. The SEBAL (Surface Energy Balance Algorithm for Land) proposed by Bastiaanssen et al. (1998) and Allen et al. (2002) was used as standard. In the software, the union of the satellite bands, the radiometric calibration, the reflectivity and the NDVI (Normalized Difference Vegetation Index) were calculated. The ArcGIS software -version 10.2 was used for the creation of vector data, database and thematic maps.
For the calculation of the spectral radiance of each band (Lλi), where the Digital Number (ND) of each image pixel is converted into monochromatic spectral radiance, representing the solar energy reflected by each pixel, per area, time, solid angle and wavelength unit measured at the Landsat 5 satellite level (approximately 705 km high) for bands 1, 2, 3, 4, 5, and 7; For band 6, this radiance represents the energy emitted by each pixel and can be obtained by the equation proposed by Markham and Baker (1987): In which: Lλi = spectral radiance of each band (Wm -2 sr -1 µm -1 ); a and b = are the minimum and maximum spectral radiances obtained in the paper by Chander et al. (2009) (Wm -2 sr -1 µm -1 ); ND = pixel intensity (digital number -integer 0 to 255); i = are the bands (1, 2, 3, 4, 5, 6 and 7) of the Landsat-5 satellite.
The monochromatic reflectance of each band (ρλi), defined as the ratio between the solar radiation flux reflected by the surface and the incident solar radiation flux was obtained according to the equation (BASTIAANSSEN et al., 1998;ALLEN et al., 2002): In which: ρλi = monochromatic reflectance; Lλi = spectral radiance of each band (Wm -2 sr -1 µm -1 ); Eλi = spectral solar irradiance of each band at the top of the atmosphere (Wm -2 µm -1 ) information in the paper by Chander et al. (2009); Z = solar zenith angle (obtained in the header of the images acquired, and it depends on the orbit, point and time of the year); dr = astronomical day of the year (U.A), according to Inverse of the square of the relative distance Earth-Sun given by Iqbal (1983); i = are the bands (1, 2, 3, 4, 5 and 7) of the TM -Landsat-5 satellite. For each image, the union of the reflectivity was performed with the Layer Stack tool of ERDAS IMAGINE 2015 software, with the exception of band 6, because it is a non-reflective thermal band.
Normalized Difference Vegetation Index (NDVI) was obtained by means of the ratio between the near infrared (ρIV) and red (ρV) reflectivities and their sum, was obtained according to the equation (ROUSE et al.,1973): In which: Near infrared (ρIV ) and; Red band (ρV ).

Elaboration of Mapping
Because the river basin region lies in an area with great altimetric amplitude (the hilltops cause the shading of the image, mistaking the partially unsupervised K-means algorithm). The K-Means algorithm was used for the six images of the studied scenarios, the shadows of the mountains caused a confusion in the algorithm, classifying the shaded regions as urban area, not being able to choose this classification, since the basin region of greatest altitude is surrounded by Montana Forest, and it is inside the park of Serra dos Órgãos (MMA, 2017).
When this situation was verified, the supervised classification was carried out, which ensured a better accuracy for the studied images, however, the difficulty of access in some points of the basin region and the good resolution of images from Google Earth only for 9/29/2005 the supervised classification for all scenes cannot be employed. This pattern justifies the choice to perform the supervised classification for the scenario (2005 and 2010), as it shows a greater wealth of detail and less confusion between the different classes of land use and cover, when using the maximum likelihood algorithm, which allowed the best representative for the location, to the other dates (1985 to 2000) the use of the non-supervised method (FREITAS et al., 2012) was adequate.
The first stage of the work consisted in the collection of samples from high resolution 2.5 meters images of SPOT (Google Earth), referring to the classes of study in the Roncador River basin located in Magé, RJ such as: (secondary vegetation, mangrove, field altitude, primary forest, urban area, agricultural area, exposed soil, degraded pasture and water) (Table 1).

Meteorological data and vegetation
Evaluating meteorological data is essential for the estimation of environmental impacts on ecosystems (XAVIER et al., 2016), therefore this study aimed to analyze the possible influences the meteorological variables (average air temperature, global solar radiation, rain, relative humidity and The files are in the (NetCDF) format, the database was made available by Xavier et al. (2016), which interpolated the climatic data for the whole Brazil. The information in their studies was collected from the National Institute of Meteorology (INMET), National Water Agency (ANA) and the Department of Water and Electric Power of São Paulo. The time series of NDVI images encompasses the same 25-year period. The procedure for generating the time series was performed in ArcGis 10.2 software, which allows the visualization through histograms of the mean values of the NDVI images.

Statistical analysis
A general multiple regression model was calculated from the "Enter" method, where all variables are entered at the same time and an R 2 is reported for the quality of the model generated. The regression analysis was estimated through R² and the weights of the coefficients, determining the linear relationships between the set of explanatory variables and the response variable. Thus, the tests performed in the multiple regression to determine which variables influence the dynamics of the basin landscape were the following: Test for significance of the multiple linear regression, Test for significance of regression parameter, Shapiro-Wilk homogeneity test, Durbin-Watson test, multicollinearity test. The multicolinarity test analyzes the linear relationship between the independent variables. This relationship was evaluated through the variance inflation factor (VIF).

RESULTS AND DISCUSSION
From the unsupervised classification through TM sensor images and subsequent quantification of soil cover classes in the study area (Table 2 and Figure 2), it was found that for 1985 there were 12147.03 ha of native vegetation cover and in 2010 the same class presented the value 7247.61 ha. Therefore, in the period of 25 years, there was an average reduction of 49% of the native vegetation cover in the area of the river basin, indicating an average annual reduction of around 2.36% over the 25 years. This reduction of plant cover represented significant growth in urban and agricultural areas of 31% and 25%. The permanent preservation areas of altitude and mangrove fields also showed reductions of 9.5% and 10%, respectively. The pasture class also decreased by 1.6%. The areas classified as exposed soil presented increase, but with values lower than 1%. In a study carried out in São Paulo in the implementation of the trench in Guarulhos, Moledo et al. (2015) also highlighted the importance of vegetation in maintaining the balance of the Tanque Grande basin, where the increasing urbanization in the metropolitan region of São Paulo contributes to the suppression of native vegetation, soil degradation and even the increase in air pollution. In the evaluation of environmental impacts in the Estiva stream in Betim, Minas Gerais, Soares et al. (2016) observed that the most significant environmental impacts that contributed to the deterioration of the environmental quality of the area were: degradation and/or removal of riparian forest, dumping of domestic sewage, irregular  Tables 3 and 4 show the results of the error matrix of the image classified by the maximum likelihood method (supervised classification). The pasture and soil class for 2005 presented the lowest error matrix of 44% and 33%, respectively (Table  3). In 2010, the agricultural area class was the lowest error 50% (Table 4). These results show that the year 2010 presented the best overall accuracy of 77%, while in 2005 the value was only 68%. The Kappa index (Table 5) estimated for the classified image by using the maximum likelihood algorithm presented a 63% accuracy for 2005 and 74% for 2010, being classified as very good, according to the criteria proposed by Landis and Kock et al. (1977). The Kappa index, when observed by class for 2005, presented low values for areas of exposed soil and altitude field (20% and 21.88%). The found may be associated to the predominance of granitic and migmatitic rocks. Additionally, the rugged terrain originates shallow soils and forms a mosaic of microhabitats (MOCOCHINSK and SCHEER, 2008). The classes of urban and mangrove areas (53% and 48%). The forest class presented representative values   widely from one year to another for the classes of exposed soil, altitude field, pasture and agricultural area (Table 5 and Figure 3). However, a good standard for forest classification remains. Similar results were found by Brasileiro et al. (2016), when studying the efficiency of the maximum likelihood classifier for mapping of land use and land cover in Quixeramobim, Ceará, the algorithm was more efficient in the vegetation classification of caatinga in relation to other features like pasture and exposed soil that presented low spectral responses. In the statistical analysis between the environmental variables and the NDVI vegetation index, the ANOVA test presented a p-value of 0.029, which indicates that the variable "NDVI" is influenced by the variables included in the model. The value obtained from R 2 was 0.47, that is, 47% of the variations of the "vegetation index" are explained by the environmental variables. Among the studied variables, it was found that the coefficients that presented the greatest significance in the equations were rainfall with growth of 0.023 and air temperature with decrease of 0.008, the other variables did not significantly influence NDVI.
In a temporal study with images from Thematic Mapper in the city of Rio de Janeiro, NDVI in 60% of the years was the index that presented the highest correlation with rainfall compared to other vegetation indexes (GOULART et al., 2015). Other authors also detected growth of covers such as: agricultural crops, natural forest and caatinga, associating the increase of tree cover with rainfall and vegetation indexes FRANCISCO et al., 2015).

CONCLUSIONS
The Roncador River Basin is undergoing transformation in its landscape, with an average reduction of (-49%) in native vegetation areas due to an increase in urban areas (25%) and agriculture (31%).
The vegetation classes found in this ecosystem, mainly in protected areas, are being replaced by Pasture formations, agriculture, urban area, as a result of anthropic activities.