*4.4. Fieldwork*

The fieldwork was conducted in two stages: From 26 November 2015 to 30 November 2015 and from 3 February 2017 to 4 February 2017 within the RCSP and its surrounding area aiming at a future elaboration of a proposal for bu ffer zones. The sampling points were selected using the Random Points tool of ArcGis ® 10.2 and posteriorly guided according to the need for validating the multitemporal dynamics to identify the land cover at the locations presenting or not changes, as demonstrated on the map elaborated using the PCA. A total of 107 points were analyzed. The tools used in this work are Garmin ® Etrex30 GPS, a digital camera (Sony Cyber Shot DSC H300), a clipboard for notes, and maps elaborated for the study area. To identify the areas with the need for change detection in the RCSP, we elaborated a spreadsheet with the features acquired from Landsat OLI images and compared the information with the photographs acquired at the location.

#### **5. Results and Discussion**

#### *5.1. Principal Component Analysis*

The Principal Component Analysis reduced the redundancy of the information between the spectral bands, which presented very similar behavior. The number of principal components is equal to the number of bands in which each component is associated with a variance of the digital levels, with the first component presenting the highest variance, successively decreasing the values [17,24]. Thus, the application of this technique, manipulated with multiple band association tests, can demonstrate the areas of use dynamic and land cover from 1990 to 2004 and 2004 to 2016.

Similar to the observations made by Ding et al. [20], the first three PCs of each year corresponds to more than 99.5% of the total covariance. For example, for the Landsat TM image in 2004, PC1 presented 68.6% of the total covariance of the set and PC2 and PC3 presented 20.5% and 10.7%, respectively (total 99.9%). Based on this information, we perceived that the second and third components are not correlated. This analysis associated with RGB combination tests between the bands allowed the identification of the changes that occurred in 1990 and 2004.

After the parameter analysis, we proceeded to study the combination of the principal components and the bands on the RGB composition. The composition that demonstrated the areas in which changes occurred or not was the second component regarding the green channel (G), the third component regarding the red channel (R), and band 5 (intermediate infrared–L5–2004) regarding the blue channel (B), as demonstrated in Figure 3a. This composition of bands best met the objectives after numerous tests for detecting changes in the reforestation areas from 1990 to 2004 and from 2004 to 2016.

**Figure 3.** *Cont.*

**Figure 3.** (**a**) False-colored composites using 2nd and 3rd Principal Components and Band 5 of the Landsat TM (2004); (**b**) Change detection in the reforested areas (2004 to 2016).

From Figure 3a, we can analyze that, the areas in dark green as LULC that had already existed in 1990 and continued in 2004. The areas colored in light and dark orange demonstrated the changes that occurred in the last 14 years. The areas in dark blue are the regions that existed in 1990 and showed changes in 2004. This combination indicated very effectively the respective changes/no-changes, as shown in Figure 3b, which established the comparison between the images acquired in 2004 and 2016.

It is worth mentioning that the result of the PCA highlighted the information measured and the changes that occurred between years (of image acquisition). However, if in an image of a specific date the soil was exposed, and, in another, the areas were cultivated, the color composition image will consequently present an area of land use and cover change. At this moment, the analyst must add the ground-truth data from the study area obtained during the fieldwork, images from other dates, which corroborated with the identification of areas with temporary crops and reforestation areas distinct from those used for the classification.
