*5.2. Supervised Classification*

The supervised classification technique was used to elaborate on the LULC change detection map of the RCSP and its surrounding area. In this work, we applied the supervised classification based on the Bhattacharyya distance measurement. This classification divided the homogeneous regions of the images according to the area and similarity limits indicated. As Figure 4a shows, we used an area and similarity limit equal to 30 for both images originated from PCA.

**Figure 4.** (**a**) Segmentation of the homogeneous areas of the images (limit of the area and similarity equal to 30) (**b**) Exemplification of crops in less than two months. Difference between the exposed soil and crop exchange.

The training samples were indicated after a previous study of the area and its LULC using the high-resolution image (WorldView-2) provided by ENERCAN, images from different dates to obtain a better measure, linear contrast and segmentation techniques, and fieldwork. Moreover, to aid in the classification of the areas that presented changes, but that, for their texture and well-defined form, were similar to temporary crops, we abandoned the use of the agricultural calendar from the Socioeconomic and Agricultural Planning Center (CEPA) from EPAGRI.

The data from the agricultural calendar are presented by micro-regions, encompassing the municipalities in the areas surrounding the Rio Canoas State Park. The study area presented higher representativeness of corn, soybean, tobacco, bean, and wheat crops. According to CEPA, in 2016, 75% of the corn crops and 95% of tobacco were planted in October, while the 85% of the soybean was planted between November and December, and 60% of the bean crop, in December. The winter crops, such as wheat, were planted in June, July, and August. This information associated with the orbital images from di fferent dates was essential concerning the supervised classification because, depending on the date of the image, the cultivation areas could present exposed soil, indicating the times of production rest, areas improper for cultivation or the moment between crops.

Therefore, with the objective of more accurately classifying the agricultural areas and exposed soil, we used the images from di fferent dates to befit the times of production for most crops in the region. We discriminated five categories established on the theme caption, change and no-change, monocrops, water bodies, and changes in water bodies. The monocrop class encompasses the areas with agriculture crops, reforestation, and pasture, since they often present a similar spectral behavior and distinction, not as an object of this work, but presenting or not changes in the land cover. The training and test samples were acquired in an average of 100 for the classification of the false-color composites of the principal components associated with the original band of the intermediate infrared. The PCA allowed the identification of changes and no-changes in the same image, also reducing the classification time of the area, and highlighting the distinct theme classes. After classifying the images, we conducted the post-classification procedures and matrix edition of the classes presented as changed areas with exposed soil, but that, according to many studies and analyses, we concluded that these areas were recently harvested or planted agriculture areas with no plant growth, as shown in Figure 4b.

After the classification of the image, validation has been performed visually and mathematically using the control points collected during the two field trips. More than 100 random points distributed in the image were analyzed using ground truth data from the study area with the aid of GPS. When considering all land cover classes defined, the overall accuracy index was 0.93 and the Kappa index was 0.88. This value of Kappa is associated with the quality of the classification is considered as good according to Landis and Koch [25].

#### 5.2.1. Change Detection within the Rio Canoas State Park

The Rio Canoas State Park is a conservation unit of a mixed ombrophilous forest or araucaria forest with approximately 1200 hectares. Because of this, previous to 2004, the RCSP had not ye<sup>t</sup> been created and, consequently, preservation was not required. The term Conservation Unit (CU) is defined by the MMA [26] as the territorial space and its environmental resources, including the jurisdictional waters, with relevant natural characteristics, legally instituted by the Public Power enterprise, with objectives of conservation and defined limits, under special administration regime, to which adequate protection guarantees are applied.

According to the Law n<sup>º</sup> 9985 of 18th July 2000, the conservation units integrating the National Nature Conservation Units System are divided into two groups with specific characteristics: The integral protection units and the sustainable use units. The objective of the integral protection units is to preserve nature and admits only the indirect use of its natural resources, except in certain legal cases. The general objective of the sustainable use units is to harmonize environmental conservation with the sustainable use of a portion of its natural resources.

Among the integral protection groups, the units considered are the Ecological Station, Biological Reserve, National Park, Natural Monument, and Wild Life Refuge. The National Park, the object of this work, has a specific objective of the preservation of natural ecosystems of high ecological relevance and scenic beauty, allowing the performance of scientific researches and development of environmental education activities, recreation in contact with nature, and ecological tourism. Based on this, we classified the images resultant from the PCA, comparing the periods between 1990 and 2004 and between 2004 and 2016, as presented in Figure 5a,b.

**Figure 5.** (**a**) Change detection map for Rio Canoas State Park (1990–2004); (**b**) Change detection map for Rio Canoas State Park (2004–2016).

Before the creation of the RCSP up until a few months after the Decree was approved, we verified that there were crops and plantations of exotic species within the park in areas significant to the conservation unit. However, with the consolidation of the park and over time, its interior was modified.

In Figure 4b, we observed the presence of monocrop class which encompasses areas of agricultural and silvicultural activities, such as soybean, corn, tobacco, and exotic species of the Pinus and Eucalyptus genre. In the change detection map referent to the comparison from 1990 to 2004, we verified that the areas with cultivations within the RCSP remained from 1990 until the date in which the image from 2004 (10 August 2004) was taken. These areas presented changes when comparing the images using the PCA. However, by studying other images, we verified that the area was undergoing crop exchange. Associated to this, the class measures from the SPRING software showed that 89.56% (1014.94 ha) were unchanged areas and 10.44% (118.32 ha) were areas with monocrops. Thus, we verified that, during the period from 1990 to 2004, there were no increments to the area of exposed soil, but changes of agricultural and forestry crops. Figure 5a shows an obvious distinction between the years of 2004 and 2016. We observed that 92.51% (1048.40 ha) are unchanged areas, 6.6% (75.45 ha) demonstrated changes, and 0.7% (7.92 ha) is the riverbed that entered the Park with the creation of the Campos Novos Hydroelectric Power Plant.

The multitemporal dynamic of the land cover of 2004 and 2016, aided with the fieldwork, images from other dates, and the experience of the managers of the conservation unit indicated that, over the years, the Park aimed for changing most monocrops in significant areas to regenerate other forestry species, which ratifies the objective of the conservation unit. As demonstrated in Figure 5b, we verified that two of the areas which the image-product classification of the PCA indicated as changed areas were real. These were areas previously occupied by Pinus species and removed to regenerate species native of the region.

#### 5.2.2. Change Detection in the Surrounding Areas of RCSP

We analyzed the 10 km bu ffer area surrounding the RCSP to verify the vegetation coverage in the area after the formation of the Campos Novos Hydroelectric Power Plant. This verification allowed us to raise conclusion, first, on the impact the construction of the power plant and, second, if the environmental compensation regarding the PAs of the Park has been implemented correctly. Another fact concerns the inclusion of a Bu ffer Zone to analyze whether many changes occurred in the area surrounding the Park for this study to provide a subsidy and aid in the posterior proposal of a Bu ffer Zone, since it briefly mentioned in the managemen<sup>t</sup> plan. According to the data presented in Table 1 and the same steps are taken to detect changes within the Park, the areas presenting the most changes in land cover were those of the images taken in 2004 and 2016.


**Table 1.** Classification table for change detection for the area surrounding the RCSP.

During the period from 2004 to 2016, the land cover changes in 5663. 78 ha (12.20% of the area), mostly by reforestation of Pinus species (Figure 6b). The surrounding area of the park is characterized by cellulose and paper industries, such as *Iguaçu Celulose S.A*, and wood industries. The reforestation areas of exotic species surrounding the RCSP is old and a part of the region´s economy (Figure 6c,d). However, this is a concerning subject for avoiding contamination with invasive species, due to the proximity to the park.

**Figure 6.** *Cont.*

**Figure 6.** (**a**) Change detection map for the period from 1990 to 2004; (**b**) Change detection map for the period from 2004 to 2016; (**c**) Worldview-2 image acquired in 2010 showing the areas of regenerated vegetation (Courtesy: ENERCAN); (**d**) Photograph of native species *Araucaria angustifolia* with the plantations of *Pinus eliotti*; (**e**) Photograph of various native species and *Pinus eliotti.*

In 1990 and 2004, the areas that remained unchanged were of approximately 95% of the study area and presented no changes in land cover during the 14 years (Figure 6a). This occurs mainly due to the continuous presence of agricultural monocrops in both summer and winter, with no grea<sup>t</sup> increments. Regarding the 12 years from 2004 to 2016, the changes were more recurrent. This occurs mostly due to the shallow cutting of the Pinus plantation areas, which di ffer from the crops, such as soybean, tobacco, and corn, and changes the entire landscape. The cutting of the exotic reforestation species take years and is performed in rotations that ranges from 5 to 8 years using shallow or partial cuts. Therefore, in these cases, there is change and, contrary to the soybean crops, for example, which can be replanted as soon as the wheat is harvested and germinates days later, the plantation of new seedlings takes time.

#### 5.2.3. Change Detection in the Bed of the Rio Canoas

The Campos Novos Hydroelectric Power Plant was implanted on the Rio Canoas, approximately 20 km upstream of its confluence with Rio Pelotas at the border of the municipalities of Campos Novos and Celso Ramos, according to the data provided by ENERCAN [27]. According to ENERCAN, before finishing the construction at the margins of the Rio Canoas, the dam would be formed by the flooding of the peripheral areas to the Rio Canoas, characterizing a substantial extension of flooded land.

The PCA has demonstrated the transformations as pursuant to the study of the RCSP associated with the change detection. Because the data on the flooding of the marginal areas provided by the ENERCAN Company were provided before concluding the construction in 2004, we proceeded with the analyses of the area. Therefore, the data were classified in two images with specific dates and related to the rainfall data, derived from Landsat TM (acquired on 14 July 1990) and Landsat OLI (acquired on 16 April 2016).

To obtain a better accuracy of the image classification, we acquired the rainfall data from the Hydrological Information System—Hidroweb of the National Waters Agency (Figure 7). We used the average rainfall from 1961 to 1990, historical series of 30 years, with monthly information according to the National Meteorology Institute (INMET). To classify the images, they were selected according to the dates in which there was monthly accumulated rainfall nearly the same as the average for the 30 years analyzed to exclude the influences of rain in the demarcation of the riverbed before and after flooding the area.

As presented in Figure 7, the areas in light pink identify the dates with rainfall accumulations that could influence the increase of the riverbed. We used the images according to the availability and the value nearest to the average of rain accumulated in 30 years. The image acquired on 6 January 1997 demonstrates that the amount of rain accumulated was within the average and, therefore, there were not many influences of rain. Regarding the image acquired on 16 April 2016, when considering rainfall data in October, we verified a significant increase of accumulated rainfall, but we disregarded the daily averages of accumulated rainfall for October of 2016. With this detailed information from INMET, we verified that, until the 16th of October, the rainfall accumulation was of 78 mm, also having little e ffect in the area. According to the generated map (Figure 8), an area of 6.48 km<sup>2</sup> has been indicated as an approximate measure of the riverbed on 6 January 1997 and of 30.15 km<sup>2</sup> on 16 April 2016 with the creation of the Campos Novos Hydroelectric Power Plant. The objective of the study was to analyze the size of the area surrounding the RCSP that was flooded, due to the high di fference made explicit in the change detection using the PCA technique.

**Figure 7.** Rainfall accumulated at the time of satellite data acquisition compared to the average of 30 years.

**Figure 8.** Map of the Rio Canoas in 1997 before the implementation of the Hydroelectric Plant of Rio Canoas, and in 2016.
