Geotechnology in the analysis of forest fragments in northern Mato Grosso, Brazil

Pasture implantation fragments and reduces the Amazonian forest area. The objective was to quantify landscape changes in 1985, 2000 and 2015 in northern Mato Grosso, Brazil. The study was carried out in three scenes obtained by the LANDSAT satellite of a microbasin (2742.33 ha) in the municipality of Alta Floresta. Forest, water bodies, pasture and exposed soil were the thematic classes determined to e mapping the land use evolution. The edge, density and shape indexes of the fragments were measured. Normalized vegetation difference (NDVI) values were high in 1985. Land use and occupation over 15 years (1985–2000) reduced forest cover by 69.8%, but it increased by 1.7% over the next 15 years (2000–2015). The number of exposed soil patches increased between the periods, but the total area and number of the patches of the forest fragments decreased. The high values of NDVI in 1985 showed vegetated areas with high density. Reducing forest cover decreases the size of the fragments, increases the isolation and the number of soil patches exposed. The mapping of land use showed a reduction of the Amazon forest in the microbasin in the north of Mato Grosso, in the years 2000 and 2015 compared to 1985.

In the early 1980s, the Brazilian government encouraged deforestation of native forest by colonizers to occupy and own Amazonian lands 1 . Currently, the Brazilian legislation is focused on the conservation and restoration of this native vegetation 2 .
Forest fragments are areas of natural vegetation interrupted by anthropogenic or natural barriers, reducing the animal wild, pollen and seed flow 3 . The implantation of pastures for livestock replaces the natural landscape and fragments the southern environment of the Amazon forest in the northern region of Mato Grosso, Brazil 4 . The ecological characterization of fragments contributes to proper managing and conserving these forest remnants, including at the microbasin level 5 .
Landscape ecology can be analyzed with computational tools, especially the Geographic Information System (GIS) with image processing 6 . A set of procedures and measures, known as landscape metrics, allows quantitative understanding and estimating the landscape structure patterns 7 . GIS quantifies the particularities of the landscape 8 and, when incorporated into Remote Sensing, analyzes the physical environment through a geo-referenced database at different dates and scales 9 .
Remote densing data are used to monitor vegetation and distinguish anthropic events 10 Vegetation indexes are based on linear combinations of spectral data, enhancing vegetation presence 11 . The normalized difference vegetation index (NDVI) emphasizes variations of band density for environmental analysis with conclusions based on the vegetal cover dynamics 12 . NDVI has been used to classify the distribution 13 and to study the variability of vegetation biophysical parameters such as phytomass production 14 , leaf area index 15 , land use and occupation 16 , vegetation fragmentation 17 and estimating agricultural productivity 18 .
Using maps and satellite imagery information allows evaluating the digital classification accuracy of topics from data classified and expressed as an error matrix 19 . Kappa index is important to supervising classification confidence analysis representing all the elements of the matrix 20 . Land use and land cover maps and the fragmentation dynamics analysis are environmental monitoring and preservation mechanisms for decision-making, mainly in priority areas 21 , as well as microbasins in regions with great deforestation.
The objective was to verify and to quantify structural changes of the landscape in 1985, 2000 and 2015 in a microbasin of the Teles Pires river at the Alta Floresta municipality, Mato Grosso, Brazil (Fig. 1), colonized in early 1980s, using GIS and Remote Sensing. Kappa indexes were 1.00, 0.94 and 0.98 in 1985, 2000 and 2015, respectively. The classification of 1985 did not present confusion among classes because only forest was detected. In 2000, the classification showed confusion between the pasture class and soil exposed with 99.2 and 96.5% of precision, respectively, but the accuracy for the other classes was 100%. The precision in 2015 was 99.7 e 99.6% for exposed soil and pasture and 100% for the other classes.

Results
In 1985, 100% of the vegetation was native, reduced to 30% 15 years later, followed by an increase of ≈2% in 2015 (Table 1 (Table 2).

Discussion
The NDVI in 1985 showed that the microbasin was covered by forests due to darker shades, and the NDVI values of 0.60 are common in regions of tropical rainforest 22,23 . Negative NDVI values in 2000 and 2015 show reduced native vegetation with soil exposure and watercourses. NDVI variability highlights changes in vegetation cover that may be related to anthropic action (pastures, roads, occupations) 24 , as reduction observed in forest areas for agricultural use in Chile from 1975 to 2005 25 and in Ecuador from 1982 to 2015 26 .
Kappa indexes higher than 0.75 validated the supervised classifications, with accuracy above 96% for all classes, and excellent concordance for the forest class 27 . The low confusion between pasture and exposed soil is related to the similarities of the spectral responses that these classes have with each other 28 . These indexes confirm that the data collected correctly represent the measured variables, and consider that the classifications are statistically correct 29 . The reduction by 30% of the native vegetation between 1985 and 2000 was due to conversion of forest areas to pasture and exposed soil by the anthropogenic activities, being the pasture implementations the highest responsible for the reduction (58.7%). Deforestation in Chile by the implementation of forest plantations increased from 5.5% in 1975 to 42.4% in 2007, corroborating as a direct deforestation cause and biodiversity loss 30 . The analysis of the causes of these changes in the landscape allows to predict which areas are most vulnerable to changes and to prevent socioenvironmental adversities 25 . The reduction of native vegetation in Mato Grosso impacted negatively the microbasin with increased runoff, erosion and silting of rivers 31 . However, the financial and technical support, through the "Olhos D' Água da Amazônia" project for the recovery of natural areas along the watercourse, may explain the native vegetation increase between 2000 to 2015 in the Alta Floresta municipality 32 . Inadequate pasture management, such as overgrazing, compaction and soil exposure in Mato Grosso, increased over the years with a peak in 2015, similar to that reported in the vicinity of the private reserve of the national patrimony in Cafundó, Espirito Santo from 1970 to 2007 33 .
The decrease in bodies of water was due to the increase in exposed soil class 34 . Soil use and occupation results in deforestation with potential to impact processes of the hydrological cycle (precipitation, increasing surface runoff, temperature and relative humidity) which has a close relationship with evapotranspiration 35 . Water bodies decreased by 45% between 1984 and 2015 in Egypt, mainly due to increased use of land exposed by anthropogenic activities 36 . The reduction of evapotranspiration affects the climate-vegetation equilibrium by leading to a warmer and drier condition in the Amazonian ecosystems 37 . Forests sustain biodiversity, reduce soil erosion, regulate the water cycle and sequester carbon, helping to mitigate the impacts of global warming 38 .  The Nsp increase from 1985 to 2015 in Mato Grosso resulted from the typical fragmentation process, reducing the average area of fragments during the temporal scales, presenting an increment of subdivision and less connection between them 39 . The smaller fragments represent a crucial role in reducing the isolation of larger fragments 40 . However, from 2000 to 2015, we had a decrease in Nsp's eading to increase in forest cover 41 . In 2014, there was a greater recovery of these areas, which led to a consolidation of some fragments and even its expansion, grouping other fragments.
Larger fragments as estimated by AS indicate more irregular shapes and smaller ones indicate more regular shape. Their size and shape are intrinsically bounded to the edge, that is, the smaller the fragment or more elongated, the more intense the edge effect will be, that is, reducing the inner-margin ratio 42 . It is worth noting that the more it moves away from the standard shape (the perfect circle) the more cut the shape of the spot becomes, and the more it is susceptible to the edge effect 43 . The shape of the spot is more trimmed as its pattern differs from the perfect circle being more susceptible to the edge effect 42,43 .  (Fig. 1). The microbasin was classified as a low permanent preservation rate (0-25% of APP) and forest conservation (0-25% of native vegetation) 44 . The climate is Am (Köppen classification) with average temperature of 27.6 ± 2 °C and annual precipitation of 3,000 mm 45  LANDSAT Image Processing. Scenes of LANDSAT 5 were georeferenced and the scenes of LANDSAT 8 were redesigned. The control points for the adequacy between the information plans were selected by georeferencing, aiming at the correct overlapping of the vector limits of the microbasin in the images. The typologies and vegetation patterns were differentiated by the normalized difference vegetation index (NDVI), identifying the water corps, pasture and soil exposed to generate a fragment typology by color difference 50 . The classes were differentiated by color differences. The NDVI, obtained with equations 1 and 2 for the LANDSAT 8 and LANDSAT 5 satellite images, respectively, was associated with vegetation density 51   Time evolution of land use and occupation. Satellite images were classified as "supervised" in the forest cover presence or absence with the Maximum Likelihood classifier algorithm. This classification groups the patterns of similar images into land use and occupation classes and establishes classes from training samples with those of interest to the scene 52 . The thematic classes were forest, water bodies, pasture and exposed soil, defined by the supervised classification based on the visual interpretation of the false color composition, infrared and natural color. The bands RGB 543, 432, 321, corresponding to false color, infrared and natural color, respectively, were composed with the satellite LANDSAT 5 images for 1985 and 2000 and those of th e RGB 654, 543, 432, corresponding to false color, infrared and natural color, respectively, for LANDSAT 8, relative to 2015. In addition, these results were compared with those of the NDVI e Normalized difference water index (NDWI) to increase the classification reliability. Three hundred and fifty samples, polygonal of 3 × 3 pixels per period, were collected according to the four classes. The raster file generated was converted to shapefile, after classification, to obtain the class area, comprising the respective polygons sum per year. The classification quality was evaluated from the confusion matrix of the training samples collected with the Kappa index 53 .

Conclusions
Quantitative analysis of forest fragments by landscape ecology indexes. The fragments were analyzed in shapefile files classified per year, corresponding to the forest vegetation class. Calculations of landscape metrics for the microbasin obtained through the Patch Analyst extension of ArcGis 10.4 software were applied to vector files generated per year without fragment size distinction. The metrics, fragment density, size, fragment shape and edge indexes were determined 54,55 .