Refined burned-area mapping protocol using Sentinel-2 data increases estimate of 2019 Indonesian burning

Abstract. Many nations are challenged by landscape fires. A confident knowledge of the area and distribution of burning is crucial to monitor these fires and to assess how they might best be reduced. Given the differences that arise using different detection approaches, and the uncertainties surrounding burned-area estimates, their relative merits require evaluation. Here we propose, illustrate, and examine one promising approach for Indonesia where recurring forest and peatland fires have become an international crisis. Drawing on Sentinel-2 satellite time-series analysis, we present and validate new 2019 burned-area estimates for Indonesia. The corresponding burned-area map is available at https://doi.org/10.5281/zenodo.4551243 (Gaveau et al., 2021a). We show that >3.11 million hectares (Mha) burned in 2019. This burned-area extent is double the Landsat-derived official estimate of 1.64 Mha from the Indonesian Ministry of Environment and Forestry and 50 % more that the MODIS MCD64A1 burned-area estimate of 2.03 Mha. Though we observed proportionally less peatland burning (31 % vs. 39 % and 40 % for the official and MCD64A1 products, respectively), in absolute terms we still observed a greater area of peatland affected (0.96 Mha) than the official estimate (0.64 Mha). This new burned-area dataset has greater reliability than these alternatives, attaining a user accuracy of 97.9 % (CI: 97.1 %–98.8 %) compared to 95.1 % (CI: 93.5 %–96.7 %) and 76 % (CI: 73.3 %–78.7 %), respectively. It omits fewer burned areas, particularly smaller- (<100 ha) to intermediate-sized (100–1000 ha) burns, attaining a producer accuracy of 75.6 % (CI: 68.3 %–83.0 %) compared to 49.5 % (CI: 42.5 %–56.6 %) and 53.1 % (CI: 45.8 %–60.5 %), respectively. The frequency–area distribution of the Sentinel-2 burn scars follows the apparent fractal-like power law or Pareto pattern often reported in other fire studies, suggesting good detection over several magnitudes of scale. Our relatively accurate estimates have important implications for carbon-emission calculations from forest and peatland fires in Indonesia.

Download Full-text

Uni-Temporal Multispectral Imagery for Burned Area Mapping with Deep Learning

Remote Sensing ◽

10.3390/rs13081509 ◽

2021 ◽

Vol 13 (8) ◽

pp. 1509

Author(s):

Xikun Hu ◽

Yifang Ban ◽

Andrea Nascetti

Keyword(s):

Deep Learning ◽

Large Scale ◽

Multiple Scales ◽

Detection Methods ◽

Burned Area ◽

Landsat 8 ◽

Multispectral Imagery ◽

Burned Areas ◽

Local Climate Zones ◽

Sentinel 2

Accurate burned area information is needed to assess the impacts of wildfires on people, communities, and natural ecosystems. Various burned area detection methods have been developed using satellite remote sensing measurements with wide coverage and frequent revisits. Our study aims to expound on the capability of deep learning (DL) models for automatically mapping burned areas from uni-temporal multispectral imagery. Specifically, several semantic segmentation network architectures, i.e., U-Net, HRNet, Fast-SCNN, and DeepLabv3+, and machine learning (ML) algorithms were applied to Sentinel-2 imagery and Landsat-8 imagery in three wildfire sites in two different local climate zones. The validation results show that the DL algorithms outperform the ML methods in two of the three cases with the compact burned scars, while ML methods seem to be more suitable for mapping dispersed burn in boreal forests. Using Sentinel-2 images, U-Net and HRNet exhibit comparatively identical performance with higher kappa (around 0.9) in one heterogeneous Mediterranean fire site in Greece; Fast-SCNN performs better than others with kappa over 0.79 in one compact boreal forest fire with various burn severity in Sweden. Furthermore, directly transferring the trained models to corresponding Landsat-8 data, HRNet dominates in the three test sites among DL models and can preserve the high accuracy. The results demonstrated that DL models can make full use of contextual information and capture spatial details in multiple scales from fire-sensitive spectral bands to map burned areas. Using only a post-fire image, the DL methods not only provide automatic, accurate, and bias-free large-scale mapping option with cross-sensor applicability, but also have potential to be used for onboard processing in the next Earth observation satellites.

Download Full-text

A Deep Learning Approach for Burned Area Segmentation with Sentinel-2 Data

Remote Sensing ◽

10.3390/rs12152422 ◽

2020 ◽

Vol 12 (15) ◽

pp. 2422

Author(s):

Lisa Knopp ◽

Marc Wieland ◽

Michaela Rättich ◽

Sandro Martinis

Keyword(s):

Deep Learning ◽

Network Performance ◽

Near Infrared ◽

Remote Sensing Data ◽

Economic Consequences ◽

Validation Dataset ◽

Burned Area ◽

Burned Areas ◽

Spectral Bands ◽

Sentinel 2

Wildfires have major ecological, social and economic consequences. Information about the extent of burned areas is essential to assess these consequences and can be derived from remote sensing data. Over the last years, several methods have been developed to segment burned areas with satellite imagery. However, these methods mostly require extensive preprocessing, while deep learning techniques—which have successfully been applied to other segmentation tasks—have yet to be fully explored. In this work, we combine sensor-specific and methodological developments from the past few years and suggest an automatic processing chain, based on deep learning, for burned area segmentation using mono-temporal Sentinel-2 imagery. In particular, we created a new training and validation dataset, which is used to train a convolutional neural network based on a U-Net architecture. We performed several tests on the input data and reached optimal network performance using the spectral bands of the visual, near infrared and shortwave infrared domains. The final segmentation model achieved an overall accuracy of 0.98 and a kappa coefficient of 0.94.

Download Full-text

Exploitation of Sentinel-2 Time Series to Map Burned Areas at the National Level: A Case Study on the 2017 Italy Wildfires

Remote Sensing ◽

10.3390/rs11060622 ◽

2019 ◽

Vol 11 (6) ◽

pp. 622 ◽

Cited By ~ 20

Author(s):

Federico Filipponi

Keyword(s):

Time Series ◽

Spatial Resolution ◽

Satellite Data ◽

High Spatial Resolution ◽

National Level ◽

Burned Area ◽

Commission Error ◽

Burned Areas ◽

Sentinel 2

Satellite data play a major role in supporting knowledge about fire severity by delivering rapid information to map fire-damaged areas in a precise and prompt way. The high availability of free medium-high spatial resolution optical satellite data, offered by the Copernicus Programme, has enabled the development of more detailed post-fire mapping. This research study deals with the exploitation of Sentinel-2 time series to map burned areas, taking advantages from the high revisit frequency and improved spatial and spectral resolution of the MSI optical sensor. A novel procedure is here presented to produce medium-high spatial resolution burned area mapping using dense Sentinel-2 time series with no a priori knowledge about wildfire occurrence or burned areas spatial distribution. The proposed methodology is founded on a threshold-based classification based on empirical observations that discovers wildfire fingerprints on vegetation cover by means of an abrupt change detection procedure. Effectiveness of the procedure in mapping medium-high spatial resolution burned areas at the national level was demonstrated for a case study on the 2017 Italy wildfires. Thematic maps generated under the Copernicus Emergency Management Service were used as reference products to assess the accuracy of the results. Multitemporal series of three different spectral indices, describing wildfire disturbance, were used to identify burned areas and compared to identify their performances in terms of spectral separability. Result showed a total burned area for the Italian country in the year 2017 of around 1400 km2, with the proposed methodology generating a commission error of around 25% and an omission error of around 40%. Results demonstrate how the proposed procedure allows for the medium-high resolution mapping of burned areas, offering a benchmark for the development of new operational downstreaming services at the national level based on Copernicus data for the systematic monitoring of wildfires.

Download Full-text

Characterization of wildfire regimes in Canadian boreal terrestrial ecosystems

International Journal of Wildland Fire ◽

10.1071/wf08096 ◽

2009 ◽

Vol 18 (8) ◽

pp. 992 ◽

Cited By ~ 16

Author(s):

Yueyang Jiang ◽

Qianlai Zhuang ◽

Mike D. Flannigan ◽

John M. Little

Keyword(s):

Power Law ◽

Terrestrial Ecosystems ◽

Fire Regimes ◽

Carbon Dynamics ◽

Burned Area ◽

Burned Areas ◽

Fire Recurrence ◽

Natural Fires ◽

Recurrence Intervals ◽

Frequency Area

Wildfire is a major disturbance in boreal terrestrial ecosystems. Characterizing fire regimes and projecting fire recurrence intervals for different biomes are important in managing those ecosystems and quantifying carbon dynamics of those ecosystems. This study used Canadian wildfire datasets, 1980–1999, to characterize relationships between number of fires and burned area for 13 ecozones and to calculate wildfire recurrence intervals in each ecozone. For the study period, wildfires were found to follow power–law relationships between frequency densities (number of fires normalized to unit bins) and burned areas in all ecozones. Power–law frequency–area relationships also held for both anthropogenic fires and natural fires in the 1980s and 1990s. For each Canadian ecozone using the parameters of the power–law frequency–area distributions, fire recurrence intervals were then calculated for wildfires equal to or larger than a given size of burned area. Fire recurrence intervals ranged from 1 to 32 years for burned areas >2 km2, and from 1 to 100 years for burned areas >10 km2 in every 10 000-km2 spatial area for each ecozone. The information obtained through characterizing the wildfires and the fire recurrence intervals calculated in this study will provide guidance to wildfire risk managers throughout Canada. The findings of this study will also be a benefit to future efforts in quantifying carbon dynamics in Canadian boreal terrestrial ecosystems.

Download Full-text

A view of recent forest fire in Australia by satellite derived indices

10.5194/egusphere-egu21-8562 ◽

2021 ◽

Author(s):

Kim-Anh Nguyen ◽

Yuei-An Liou ◽

Le-Thu Ho

Keyword(s):

Air Quality ◽

Optical Sensors ◽

Forest Fires ◽

Remote Sensing Data ◽

Heat Index ◽

Burned Area ◽

Landsat 8 ◽

Burned Areas ◽

High Severity ◽

Sentinel 2

<p>Bushfire is one of the dangerous natural manmade hazards. It can cause great damges to the air quality, human health, environment and bio-diversity. In addition, forest fires may be a potential and signigicant source of polychlorinated dibenzo-p-dioxins and polychlorinated dibenzofurans. In early 2020, Australia experienced serious bushfires with over an area of estimated 18.6 million hectares burned, over 5,900 buidlings (including 2, 779 homes) destroyed, and at least 34 people (including three fire fighters) and billion animals and some endangered species killed. Subsequently, air quality was degraded to hazardous levels. It was estimated that about 360 million tonnes of CO<sub>2</sub> was emitted as of 2 Jan. 2020 by NASA. Remote sensing data has been instrumental for the environmental monitoring in particular the bushfire. Many methods and algorithms have been proposed to detect the burned areas in the forest. However, it is challenging or even infeasible to routinely apply them by non-experts due to a chain of sophisticated schemes during their implementation. Here, we present a simple and effective method for mapping a burned area. The performances of different optical sensors and indices are conducted. Sentinel-2 MSI and Landsat 8 data are ultilized for the comparison of burned forest by analyzing different indices (including NDVI, NDBR and newly development index Nomarlized Difference Laten Heat Index (NDLI)). The forest damages are estimated over the Katoombar, Austrialia and the burning severity map is generated and classified into eight levels (none, high regrowth, lowregrowth, unburned, low severity, moderate low severity, moderate high severity, and high severity). The comparision in results from Sentinel-2 MSI data and Landsat image is performed and presented.</p>

Download Full-text

Intercomparison of Burned Area Products and Its Implication for Carbon Emission Estimations in the Amazon

Remote Sensing ◽

10.3390/rs12233864 ◽

2020 ◽

Vol 12 (23) ◽

pp. 3864

Author(s):

Ana Carolina M. Pessôa ◽

Liana O. Anderson ◽

Nathália S. Carvalho ◽

Wesley A. Campanharo ◽

Celso H. L. Silva Junior ◽

...

Keyword(s):

Land Cover ◽

Forest Fires ◽

Carbon Emission ◽

Regional Scale ◽

National Level ◽

Burned Area ◽

Land Cover Changes ◽

Burned Areas ◽

Great Divergence ◽

Global Emissions

Carbon (C) emissions from forest fires in the Amazon during extreme droughts may correspond to more than half of the global emissions resulting from land cover changes. Despite their relevant contribution, forest fire-related C emissions are not directly accounted for within national-level inventories or carbon budgets. A fundamental condition for quantifying these emissions is to have a reliable estimation of the extent and location of land cover types affected by fires. Here, we evaluated the relative performance of four burned area products (TREES, MCD64A1 c6, GABAM, and Fire_cci v5.0), contrasting their estimates of total burned area, and their influence on the fire-related C emissions in the Amazon biome for the year 2015. In addition, we distinguished the burned areas occurring in forests from non-forest areas. The four products presented great divergence in the total burned area and, consequently, total related C emissions. Globally, the TREES product detected the largest amount of burned area (35,559 km2), and consequently it presented the largest estimate of committed carbon emission (45 Tg), followed by MCD64A1, with only 3% less burned area detected, GABAM (28,193 km2) and Fire_cci (14,924 km2). The use of Fire_cci may result in an underestimation of 29.54 ± 3.36 Tg of C emissions in relation to the TREES product. The same pattern was found for non-forest areas. Considering only forest burned areas, GABAM was the product that detected the largest area (8994 km2), followed by TREES (7985 km2), MCD64A1 (7181 km2) and Fire_cci (1745 km2). Regionally, Fire_cci detected 98% less burned area in Acre state in southwest Amazonia than TREES, and approximately 160 times less burned area in forests than GABAM. Thus, we show that global products used interchangeably on a regional scale could significantly underestimate the impacts caused by fire and, consequently, their related carbon emissions.

Download Full-text

A Practical Method for High-Resolution Burned Area Monitoring Using Sentinel-2 and VIIRS

Remote Sensing ◽

10.3390/rs13091608 ◽

2021 ◽

Vol 13 (9) ◽

pp. 1608

Author(s):

Miguel M. Pinto ◽

Ricardo M. Trigo ◽

Isabel F. Trigo ◽

Carlos C. DaCamara

Keyword(s):

High Resolution ◽

Satellite Data ◽

Computation Time ◽

Practical Method ◽

Google Earth ◽

Burned Area ◽

High Resolution Data ◽

Burned Areas ◽

Resolution Data ◽

Sentinel 2

Mapping burned areas using satellite imagery has become a subject of extensive research over the past decades. The availability of high-resolution satellite data allows burned area maps to be produced with great detail. However, their increasing spatial resolution is usually not matched by a similar increase in the temporal domain. Moreover, high-resolution data can be a computational challenge. Existing methods usually require downloading and processing massive volumes of data in order to produce the resulting maps. In this work we propose a method to make this procedure fast and yet accurate by leveraging the use of a coarse resolution burned area product, the computation capabilities of Google Earth Engine to pre-process and download Sentinel-2 10-m resolution data, and a deep learning model trained to map the multispectral satellite data into the burned area maps. For a 1500 ha fire our method can generate a 10-m resolution map in about 5 min, using a computer with an 8-core processor and 8 GB of RAM. An analysis of six important case studies located in Portugal, southern France and Greece shows the detailed computation time for each process and how the resulting maps compare to the input satellite data as well as to independent reference maps produced by Copernicus Emergency Management System. We also analyze the feature importance of each input band to the final burned area map, giving further insight about the differences among these events.

Download Full-text

Forest Burned Area Mapping Using Bi-Temporal Sentinel-2 Imagery Based on a Convolutional Neural Network: Case Study in Golestan Forest

Engineering Proceedings ◽

10.3390/ecsa-8-11291 ◽

2021 ◽

Vol 10 (1) ◽

pp. 6

Author(s):

Fattah Hatami Maskouni ◽

Seyd Teymoor Seydi

Keyword(s):

Neural Network ◽

Convolutional Neural Network ◽

Burned Area ◽

Kappa Score ◽

Burned Areas ◽

The North ◽

North Of Iran ◽

Decision Making Processes ◽

Sentinel 2

Forest areas are profoundly important to the planet, since they offer considerable advantages. The mapping and estimation of burned areas covered with trees are critical during decision making processes. In such cases, remote sensing can be of great help. This paper presents a method to estimate burned areas based on the Sentinel-2 imagery using a convolutional neural network (CNN) algorithm. The framework touches change detection using pre- and post-fire datasets. The proposed framework utilizes a multi-scale convolution block to extract deep features. We investigate the performance of the proposed method via visual and numerical analyses. The case study for this research is Golestan Forest, which is located in the north of Iran. The results of the burned area detection process show that the proposed method produces a performance accuracy rate of more than 97% in terms of overall accuracy, with a Kappa score greater than 0.933.

Download Full-text

DSMNN-Net: A Deep Siamese Morphological Neural Network Model for Burned Area Mapping Using Multispectral Sentinel-2 and Hyperspectral PRISMA Images

Remote Sensing ◽

10.3390/rs13245138 ◽

2021 ◽

Vol 13 (24) ◽

pp. 5138

Author(s):

Seyd Teymoor Seydi ◽

Mahdi Hasanlou ◽

Jocelyn Chanussot

Keyword(s):

Neural Network ◽

Remote Sensing ◽

Burned Area ◽

High Resolution Data ◽

Burned Areas ◽

Heterogeneous Datasets ◽

Burned Area Mapping ◽

Net Framework ◽

Sentinel 2

Wildfires are one of the most destructive natural disasters that can affect our environment, with significant effects also on wildlife. Recently, climate change and human activities have resulted in higher frequencies of wildfires throughout the world. Timely and accurate detection of the burned areas can help to make decisions for their management. Remote sensing satellite imagery can have a key role in mapping burned areas due to its wide coverage, high-resolution data collection, and low capture times. However, although many studies have reported on burned area mapping based on remote sensing imagery in recent decades, accurate burned area mapping remains a major challenge due to the complexity of the background and the diversity of the burned areas. This paper presents a novel framework for burned area mapping based on Deep Siamese Morphological Neural Network (DSMNN-Net) and heterogeneous datasets. The DSMNN-Net framework is based on change detection through proposing a pre/post-fire method that is compatible with heterogeneous remote sensing datasets. The proposed network combines multiscale convolution layers and morphological layers (erosion and dilation) to generate deep features. To evaluate the performance of the method proposed here, two case study areas in Australian forests were selected. The framework used can better detect burned areas compared to other state-of-the-art burned area mapping procedures, with a performance of >98% for overall accuracy index, and a kappa coefficient of >0.9, using multispectral Sentinel-2 and hyperspectral PRISMA image datasets. The analyses of the two datasets illustrate that the DSMNN-Net is sufficiently valid and robust for burned area mapping, and especially for complex areas.

Download Full-text

OPTIMIZATION OF A RANDOM FOREST CLASSIFIER FOR BURNED AREA DETECTION IN CHILE USING SENTINEL-2 DATA

ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences ◽

10.5194/isprs-archives-xlii-3-w12-2020-337-2020 ◽

2020 ◽

Vol XLII-3/W12-2020 ◽

pp. 337-342

Author(s):

E. Roteta ◽

P. Oliva

Keyword(s):

Random Forest ◽

Visual Assessment ◽

Random Forest Classifier ◽

Burned Area ◽

National Forest ◽

Central Chile ◽

Burned Areas ◽

Area Product ◽

Sentinel 2

Abstract. Due to the high variability of biomes throughout the country, the classification of burned areas is a challenge. We calibrated a random forest classifier to account for all this variability and ensure an accurate classification of burned areas. The classifier was optimized in three steps, generating a version of the burned area product in each step. According to the visual assessment, the final version of the BA product is more accurate than the perimeters created by the Chilean National Forest Corporation, which overestimate large burned areas because it does not consider the inner unburned areas and, it omits some small burned areas. The total burned surface from January to March 2017 was 5,000 km2 in Chile, 20 % of it belonging to a single burned area in the Maule Region, and with 91 % of the total burned surface distributed in 6 adjacent regions of Central Chile.

Download Full-text