Mapping Tropical Forest Cover and Deforestation with Planet NICFI Satellite Images and Deep Learning in Mato Grosso State (Brazil) from 2015 to 2021

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(2), P. 521 - 521

Published: Jan. 16, 2023

Monitoring changes in tree cover for assessment of deforestation is a premise policies to reduce carbon emission the tropics. Here, U-net deep learning model was used map monthly tropical Brazilian state Mato Grosso between 2015 and 2021 using 5 m spatial resolution Planet NICFI satellite images. The accuracy extremely high, with an F1-score >0.98, further confirmed by independent LiDAR validation showing that 95% pixels had height >5 while 98% non-tree <5 m. biannual then built from map. showed relatively consistent agreement official Brazil (67.2%) but deviated significantly Global Forest Change (GFC)’s year forest loss, our product closest made visual interpretation. Finally, we estimated 14.8% Grosso’s total area undergone clear-cut logging 2021, increasing, December last date, being highest. High-resolution imagery conjunction techniques can improve mapping extent regions.

Language: Английский

---

Integration of deep learning algorithms with a Bayesian method for improved characterization of tropical deforestation frontiers using Sentinel-1 SAR imagery

Remote Sensing of Environment, Journal Year: 2023, Volume and Issue: 298, P. 113821 - 113821

Published: Sept. 29, 2023

Language: Английский

---

Leveraging past information and machine learning to accelerate land disturbance monitoring

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 305, P. 114071 - 114071

Published: Feb. 26, 2024

Language: Английский

---

Monitoring road development in Congo Basin forests with multi-sensor satellite imagery and deep learning

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: unknown, P. 114380 - 114380

Published: Sept. 1, 2024

Language: Английский

---

On the use of Earth Observation to support estimates of national greenhouse gas emissions and sinks for the Global stocktake process: lessons learned from ESA-CCI RECCAP2

Carbon Balance and Management, Journal Year: 2022, Volume and Issue: 17(1)

Published: Oct. 1, 2022

The Global Stocktake (GST), implemented by the Paris Agreement, requires rapid developments in capabilities to quantify annual greenhouse gas (GHG) emissions and removals consistently from global national scale improvements GHG inventories. In particular, new are needed for accurate attribution of sources sinks their trends natural anthropogenic processes. On one hand, this is still a major challenge as inventories follow globally harmonized methodologies based on guidelines established Intergovernmental Panel Climate Change, but these can be differently individual countries. Moreover, many countries capability systematically produce detailed annually updated lacking. other spatially-explicit datasets quantifying carbon dioxide, methane nitrous oxide Earth Observations (EO) limited uncertainty. While diverse depending availability activity data different countries, proposed comparison with EO-based estimates help improve our understanding comparability published Indeed, EO networks satellite platforms have seen massive expansion past decade, now covering wide range essential climate variables offering high potential quantification regional budgets advance process understanding. Yet, there no that quantifies fluxes directly, rather observations or proxies transformed into using models. Here, we report results lessons ESA-CCI RECCAP2 project, whose goal was engage National Inventory Agencies about methods used each community estimate GHGs evaluate in-situ estimates. Based dialogue recent studies, discuss approaches provide compared those We outline roadmap implementation an carbon-monitoring program contribute Agreement.

Language: Английский

---

Inter-comparison of optical and SAR-based forest disturbance warning systems in the Amazon shows the potential of combined SAR-optical monitoring

International Journal of Remote Sensing, Journal Year: 2023, Volume and Issue: 44(1), P. 59 - 77

Published: Jan. 2, 2023

ABSTRACTMore than half a decade after the launch of Sentinel-1A C-band SAR satellite, several near real-time forest disturbances detection systems based on backscattering time series analysis have been developed and made operational. Every system has its own particular approach to change detection. Here, we compared performance main SAR-based operational disturbance produced by research agencies (INPE, in Brazil, CESBIO, France, JAXA, Japan, Wageningen University, Netherlands), them state-of-the-art optical algorithm, University Maryland's GLAD-S2. We implemented an innovative validation protocol, specially conceived encompass all analysed systems, which measured every system's accuracy speed four different areas Amazon basin. The results indicated that, when parametrized equally, Sentinel-1 methods outperformed reference method terms sample-count F1-Score, having comparable among them. GLAD-S2 showed superior user's (UA), issuing no false detections, but had lower producer (PA, 84.88%) (PA∼90%). University's system, RADD, proved be relatively faster, especially heavily clouded regions, where RADD warnings were issued 41 days before ones, one that better performs small disturbed patches (<0.25 ha) with UA 70.11%. Of high-resolution methods, CESBIO's best regarding (99.0%). Finally, tested potential three hypothetical combined optical-SAR systems. show these would excellent capabilities, exceeding largely producer's at cost slightly diminished accuracy, constitute promising feasible for forthcoming monitoring systems.POLICY HIGHLIGHTSRecently automated tropical accuracies, even small, difficult-to-spot deforested patches.SAR detections can as precise fast being more faster very cloudy or subjected illegal mining.The combination recently yield optimized results, overall accuracy. AcknowledgmentsPlanetScope data used was provided through Norway's International Climate Forests Initiative (NICFI). Contains modified Copernicus Sentinel (2021).Disclosure statementJ.D.,S.M.,A.B.,J.R. M.W. participate participated development hereby declare they not, any way, influenced process, performed L.L.Data availability statementThe support findings this study are available from corresponding author, J.D., upon reasonable request.Additional informationFundingThe first author funded National Council Scientific Technological Development, project "Monitoring Brazilian Biomes Satellite – Building new capacities"/process 444418/2018-0, grant #350303/2021-5, Coordenação de Aperfeiçoamento Pessoal Nível Superior - Brasil (CAPES) Finance Code 001, part Internationalization program PrInt-INPE. received funding Forest (NICFI), US Government's SilvaCarbon program.

Language: Английский

---

How textural features can improve SAR-based tropical forest disturbance mapping

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2023, Volume and Issue: 124, P. 103492 - 103492

Published: Sept. 20, 2023

Spatially and timely accurate information about tropical forest disturbances is crucial for tracking critical changes, supporting management, enabling law enforcement activities. In recent years, disturbance monitoring alerting using cloud-penetrating Synthetic Aperture Radar (SAR) imagery has proven effective at national pan-tropical scales. Related detection approaches mostly rely on detecting post-disturbance altered backscatter values in C or L-band SAR time series. Some are characterized by tree remnants debris. For the periods where these kinds of remain present surface, can be similar to those stable forest. This cause omission errors delayed it considered a key shortcoming current backscatter-based approaches. We hypothesized that despite fairly values, different orientation arrangement leads an heterogeneity neighboring pixel this quantified textural features. assessed six uncorrelated Gray-Level Co-Occurrence Matrix (GLCM) features dense Sentinel-1C-band Forest disturbances, based each GLCM feature pixel-based probabilistic change algorithm, were compared against results from mapped only data. studied impact speckle-filtering kernel sizes. developed method combine features, we evaluated its robustness variety natural human-induced types across Amazon Biome. Out tested Sum Average (SAVG) performed best. derived non-speckle filtered speckle-filtered data did not show noticeable accuracy. A combination SAVG resulted reduced error up 36% improved timeliness detections average 30 days, with individual showing even higher improvements level. The was found robust types. largest reduction greatest improvement evident sites large unfragmented patches (e.g., large-scale clearings, fires mining). increasing sizes, observed trade-off between combined commission errors. size 5 provide best reducing improving while introducing emphasize combining SAR-based overcome caused help improve consistency timelines short (C-band) long wavelength (L-band) operational alerting. Result maps visualized via: https://johannesballing.users.earthengine.app/view/forest-disturbance-texture.

Language: Английский

---

A near-real-time tropical deforestation monitoring algorithm based on the CuSum change detection method

Frontiers in Remote Sensing, Journal Year: 2024, Volume and Issue: 5

Published: July 26, 2024

Tropical forests are currently under pressure from increasing threats. These threats mostly related to human activities. Earth observations (EO) increasingly used for monitoring forest cover, especially synthetic aperture radar (SAR), that is less affected than optical sensors by atmospheric conditions. Since the launch of Sentinel-1 satellites, numerous methods disturbance have been developed, including near real-time (NRT) operational algorithms as systems providing early warnings on deforestation. include Radar Detecting Deforestation (RADD), Global Land Analysis and Discovery (GLAD), Real Time Detection System (DETER), Jica-Jaxa Forest Early Warning (JJ-FAST). provide online maps applied at continental/global scales with a Minimum Mapping Unit (MMU) ranging 0.1 ha 6.25 ha. For local operators, these hard customize meet users’ specific needs. Recently, Cumulative sum change detection (CuSum) method has developed disturbances long time series images. Here, we present development NRT version CuSum MMU 0.03 The values different parameters this algorithm were determined optimize changes using F1-score. In best configuration, 68% precision, 72% recall, 93% accuracy 0.71 F1-score obtained.

Language: Английский

---

The Use of Artificial Intelligence to Curb Deforestation in the Brazilian Rainforest

IGI Global eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 81 - 122

Published: Feb. 7, 2025

Tropical rainforests like the Amazon are invaluable ecosystems for human society and biodiversity. However, they facing unprecedented threats, primarily from deforestation. This chapter explores use of machine learning (ML) deep (DL) to address this pressing environmental problem. By analyzing different ML/DL methods, we show how these tools can be used understand deforestation patterns in Brazilian better. Specifically, discuss help identify drivers deforestation, improve remote sensing-based monitoring, predict future trends. Our results, particularly role providing actionable insights, empower decision-makers policymakers with knowledge make informed choices. Ultimately, strategies contribute more effective forest conservation measures sustainable land use, reassuring audience about reliability our research.

Language: Английский

---

Integrating satellite-based forest disturbance alerts improves detection timeliness and confidence

Environmental Research Letters, Journal Year: 2024, Volume and Issue: 19(5), P. 054011 - 054011

Published: April 16, 2024

Abstract Satellite-based near-real-time forest disturbance alerting systems have been widely used to support law enforcement actions against illegal and unsustainable human activities in tropical forests. The availability of multiple optical radar-based alerts, each with varying detection capabilities depending mainly on the satellite sensor used, poses a challenge for users selecting most suitable system their monitoring needs workflow. Integrating alerts holds potential address limitations individual systems. We integrated RAdar Detecting Deforestation (RADD) (Sentinel-1), optical-based Global Land Analysis Discovery Sentinel-2 (GLAD-S2) GLAD-Landsat using two confidence rulesets at ten 1° sites across Amazon Basin. Alert integration resulted faster new disturbances by days months, also shortened delay increased confidence. An rate an average 97% when combining highlights complementary cloud-penetrating radar sensors detecting largely drivers environmental conditions, such as fires, selective logging, cloudy circumstances. improvement was observed integrating RADD GLAD-S2, capitalizing high temporal observation density spatially detailed 10 m Sentinel-1 2 data. introduced highest class addition low classes systems, showed that this displayed no false detection. Considering spatial neighborhood during alert enhanced overall labeled level, nearby mutually reinforced confidence, but it led detections. discuss implications study demonstrate is important data preparation step make use more user-friendly, providing stakeholders reliable consistent information timely manner. Google Earth Engine code integrate various datesets made openly available.

Language: Английский

---