Comparative analysis of multisensor burned area products for the Brazilian Amazon - Region of the APA Triunfo do Xingu

• Author(s): Tamasauskas, Carlos ; Pacheco, Abilio P. ; Tedim, Fantina
• Date: 2022
• Persistent ID: https://hdl.handle.net/10216/144867
• Origin: Repositório Aberto da Universidade do Porto
• Subject(s): Geografia; Geography
• Document Type: Book
• Language: English

Description

Wildfires are not natural phenomena of the Amazon rainforest; therefore, they occur due to human activities, and their occurrences have increased in recent years. This situation requires continuous monitoring of this vast region, especially in areas where agricultural, livestock, mining, and infrastructure activities are located near protected areas (indigenous lands and nature conservation units). One of the conservation units that has recorded the highest increases in deforestation and fire rates is the Triunfo do Xingu Environmental Protection Area (APA Triunfo do Xingu), which since 2018 has registered the highest rates among other conservation units in the Amazon. The present study aims to develop two databases of burned areas from optical and microwave images for the years 2018, 2019, and 2020 for the APA Triunfo do Xingu using Google Earth Engine; then the results are compared with the DETER, MAPBIOMAS and MCD64A1 burned area bases to estimate existing similarities and divergences. This research uses the images of the Sentinel-2/S-2 and Landsat-8/L-8 optical satellites for the month of August of the years 2018, 2019, and 2020, a period of increased occurrences of active fires in the APA Triunfo do Xingu, to generate the burned area database that will serve as a reference for comparison with other burned area databases. Thus, images with spatial resolution, S-2 with 10 meters and L-8 with 30 meters and spectral (red, near, and medium infrared bands), are suitable for generating information with geometric and thematic quality, as the GEE allows the production of pixel mosaics excluding pixels with cloud and cloud shadows. The Sentinel-1/S-1 images used correspond to the VH cross-polarization, which is the most suitable polarization to map burned area than the VV polarization (Prasasti et al., 2020), having 10 meters of spatial resolution, being with speckle noise filter and with backscatter (DB) values. It is noteworthy that the S-1 images correspond to Band C, the wavelength of 5 cm, which reduces the interference of clouds when imaging the surface. In general, the BA-S1, BA-DETER, BA-MAPBIOMAS, and BA-MCD64A1 datasets obtained discrepant values concerning the reference dataset, which reveals that a consolidated base with values consistent with the reality of the fire regime of the study area is still to be achieved, as the discrepancies found reveal that the data and methodologies followed are not generating consistent quantitative information. Regarding the overlaps between the datasets, the spatial similarity between the datasets have a higher rate with polygons above 20 hectares, which allowed an average overlap of 63%, 69%, and 78% for the BA-S1, BA-MAPBIOMAS, and BA-MCD64A1. The BA-DETER dataset was below 20%, as it presents the lowest values of the number of polygons for the three years.