Stefany Pinheiro Costa

Ms. Stefany Pinheiro Costa

Imazon, Belém, Pará, Brazil
orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 July 2022 in Remote Sensing
Verified
Journal Article
Open access
PDF file
Jonas Paiva Botelho JunniorJonas Paiva Botelho Junnior
Jonas Paiva Botelho Junnior
SCStefany Pinheiro Costa
Stefany Pinheiro Costa
JRJúlia G. Ribeiro
Júlia G. Ribeiro
Carlos SouzaCarlos Souza
Carlos Souza
This study presents our efforts to automate the detection of unofficial roads (herein, roads) in the Brazilian Amazon using artificial intelligence (AI). In this region, roads are built by loggers, goldminers, and unauthorized land settlements from existing official roads, expanding over pristine forests and leading to new deforestation and fire hotspots. Previous research used visual interpretation, hand digitization, and vector editing techniques to create a thorough Amazon Road Dataset (ARD) from Landsat imagery. The ARD allowed assessment of the road dynamics and impacts on deforestation, landscape fragmentation, and fires and supported several scientific and societal applications. This research used the existing ARD to train and model a modified U-Net algorithm to detect rural roads in the Brazilian Amazon using Sentinel-2 imagery from 2020 in the Azure Planetary Computer platform. Moreover, we implemented a post-AI detection protocol to connect and vectorize the U-Net road detected to create a new ARD. We estimated the recall and precision accuracy using an independent ARD dataset, obtaining 65% and 71%, respectively. Visual interpretation of the road detected with the AI algorithm suggests that the accuracy is underestimated. The reference dataset does not include all roads that the AI algorithm can detect in the Sentinel-2 imagery. We found an astonishing footprint of roads in the Brazilian Legal Amazon, with 3.46 million km of roads mapped in 2020. Most roads are in private lands (~55%) and 25% are in open public lands under land grabbing pressure. The roads are also expanding over forested areas with 41% cut or within 10 km from the roads, leaving 59% of the 3.1 million km2 of the remaining original forest roadless. Our AI and post-AI models fully automated road detection in rural areas of the Brazilian Amazon, making it possible to operationalize road monitoring. We are using the AI road map to understand better rural roads’ impact on new deforestation, fires, and landscape fragmentation and to support societal and policy applications for forest conservation and regional planning.

7 November 2024 in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Verified
Journal Article
Open access
PDF file
ZWZhihao Wang
Zhihao Wang
ZLZhili Li
Zhili Li
YXYiqun Xie
Yiqun Xie
CJCarlos Souza Jr.
Carlos Souza Jr.
JFJailson Soares Souza Filho
Jailson Soares Souza Filho
SCStefany Pinheiro Costa
Stefany Pinheiro Costa
Forests play an important role in the Earth systems for carbon sequestration and climate change mitigation, yet they have been increasingly disturbed by deforestation and forest degradation at an unprecedented pace. The Brazilian Amazon, for instance, experienced a 140% rise in deforestation from 2012 to 2020, with a record loss of 13,200 km2 between August 2020 and July 2021. Alarmingly, 87% of 2019 deforestation alerts occurred on private properties, with 61% in legally restricted areas. Existing deforestation monitoring systems, such as PRODES and the Global Forest Change dataset, use about 30m resolution satellite imagery, which is insufficient for operational validation at fine scales. The Deforestation Alert System by Imazon leverages high-resolution PlanetScope data (3-4m) but faces challenges due to fewer spectral bands and variations in reflectance values across different satellite sensors and dates. As a result, current validation is based mainly on manual inspection which is highly time-consuming. To address the challenges and automate the validation process, this work develops a system based on deep learning – known for its ability to capture complex texture patterns in high-resolution images – to inspect and confirm new deforestation sites. Specifically, the system takes inputs from potential deforestation sites suggested by coarse-resolution products and uses a pair of PlanetScope images before and after the change at each site to determine new deforestation (excluding existing deforestation). Our results demonstrate that the new system achieves robust and high-quality accuracy under various test conditions.

22 December 2025 in Proceedings of the National Academy of Sciences
Verified
Journal Article
Open access
JEJayden E. Engert
Jayden E. Engert
Carlos SouzaCarlos Souza
Carlos Souza
FKFritz Kleinschroth
Fritz Kleinschroth
FIF. Yoko Ishida
F. Yoko Ishida
SCStefany Pinheiro Costa
Stefany Pinheiro Costa
WLWilliam Laurance
William Laurance
Roads act as conduits for human incursions and hence underlie many of humanity’s impacts on nature, including deforestation, wildfires, and natural-resource overexploitation. Unfortunately, existing roadmaps often drastically underestimate the true extent of road networks and future predictions of road-related impacts rely on incomplete and outdated data, undermining development planning and conservation decision-making. Here, we develop a multivariate “road expansion risk” index to identify areas prone to road building and therefore vulnerable to road-related environmental impacts. Using a massive road dataset—137 million 1-ha raster cells drawn from three different sources arrayed across the Amazon and Congo basins and insular Asia-Pacific region—we predict road-prone locations via a statistical model that integrates a range of biophysical, socioeconomic, and administrative data. This highly integrative, large-scale approach allowed us to identify areas likely to experience future road building and regions that may contain unmapped roads. Importantly, our road expansion risk index is a strong predictor of forest loss and degradation and can hence identify future road building and deforestation hotspots, even for the many tropical forest locales with grossly deficient road data.

9 February 2023 in Journal of the Brazilian Computer Society
Verified
Journal Article
Open access
PDF file
FAFábio Araújo
Fábio Araújo
FAFelipe Araújo
Felipe Araújo
DADerian Alencar
Derian Alencar
SCStefany Pinheiro Costa
Stefany Pinheiro Costa
HOHelder Oliveira
Helder Oliveira
Denis RosarioDenis Rosario
Denis Rosario
Recently, Mobile Edge Computing (MEC) and service migration policies have shown promising results to improve the user experience and optimize infrastructure resources. In addition, Unmanned Aerial Vehicles (UAVs) appear as a promising solution to provide cloud service in collaboration with traditional MEC scenarios. However, in conjunction with the resources available on users' devices, contextual information has been ignored by most policies. In this article, we propose a service migration strategy based on contextual information and evaluate the influence of user mobility on migration strategies, called DVSM. Simulation results highlight the superior performance of the DVSM compared to state-of-the-art algorithms and a performance equivalent to the optimal solution when the collection and analysis of context information are carried out correctly.