To read the full-text of this research, you can request a copy directly from the authors.
... Lastly, the processed images of each orbit were temporally filtered (Quegan et al., 2000). All images were processed to the Sentinel-1 nominal resolution (20 m) and subsequently aggregated to 40 m to reduce speckle Belenguer-Plomer et al., 2020). ...
In this paper, we present an in-depth analysis of the use of convolutional neural networks (CNN), a deep learning method widely applied in remote sensing-based studies in recent years, for burned area (BA) mapping combining radar and optical datasets acquired by Sentinel-1 and Sentinel-2 on-board sensors, respectively. Combining active and passive datasets into a seamless wall-to-wall cloud cover independent mapping algorithm significantly improves existing methods based on either sensor type. Five areas were used to determine the optimum model settings and sensors integration, whereas five additional ones were utilised to validate the results. The optimum CNN dimension and data normalisation were conditioned by the observed land cover class and data type (i.e., optical or radar). Increasing network complexity (i.e., number of hidden layers) only resulted in rising computing time without any accuracy enhancement when mapping BA. The use of an optimally defined CNN within a joint active/passive data combination allowed for (i) BA mapping with similar or slightly higher accuracy to those achieved in previous approaches based on Sentinel-1 (Dice coefficient, DC of 0.57) or Sentinel-2 (DC 0.7) only and (ii) wall-to-wall mapping by eliminating information gaps due to cloud cover, typically observed for optical-based algorithms.
This paper presents the generation of a global burned area mapping algorithm using MODIS hotspots and near-infrared reflectance within ESA's Fire_cci project. The algorithm is based on a hybrid approach that combines MODIS highest resolution (250 m) near-infrared band and active fire information from thermal channels. The burned area is detected in two phases. In the first step, pixels with a high probability of being burned are selected in order to reduce commission errors. To do that, spatio-temporal active-fire clusters are created to determine adaptive thresholds. Finally, a contextual growing approach is applied from those pixels to the neighbouring area to fully detect the burned patch and reduce omission errors. The algorithm was used to obtain a time series of global burned area dataset (named FireCCI51), covering the 2001–2018 period. Validation based on 1200 sampled sites covering the period from 2003 to 2014 showed an average omission and commission errors of 67.1% and 54.4%. When using longer validation periods, the errors were found smaller (54.5% omission and 25.7% commission for the additional 1000 African sampled sites), which indicates that the product is negatively influenced by temporal reporting accuracy. The inter-comparison carried out with previous Fire_cci versions (FireCCI41 and FireCCI50), and NASA's standard burned area product (MCD64A1 c6) showed consistent spatial and temporal patterns. However, the new algorithm estimated an average BA of 4.63 Mkm², with a maximum of 5.19 Mkm² (2004) and a minimum of 3.94 Mkm² (in 2001), increasing current burned area estimations. Besides, the new product was found more sensitive to detect smaller burned patches. This new product, called FireCCI51, is publicly available at: http://cci.esa.int/data, last accessed on September 2019.
This paper presents a burned area mapping algorithm based on change detection of Sentinel-1 backscatter data guided by thermal anomalies. The algorithm self-adapts to the local scattering conditions and it is robust to variations of input data availability. The algorithm applies the Reed-Xiaoli detector (RXD) to distinguish anomalous changes of the backscatter coefficient. Such changes are linked to fire events, which are derived from thermal anomalies (hotspots) acquired during the detection period by the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Visible Infrared Imaging Radiometer Suite (VIIRS) sensors. Land cover maps were used to account for changing backscatter behaviour as the RXD is class dependent. A machine learning classifier (random forests) was used to detect burned areas where hotspots were not available. Burned area perimeters derived from optical images (Landsat-8 and Sentinel-2) were used to validate the algorithm results. The validation dataset covers 21 million hectares in 18 locations that represent the main biomes affected by fires, from boreal forests to tropical and sub-tropical forests and savannas. A mean Dice coefficient (DC) over all studied locations of 0.59 ± 0.06 (± confidence interval, 95%) was obtained. Mean omission (OE) and commission errors (CE) were 0.43 ± 0.08 and 0.37 ± 0.06, respectively. Comparing results with the MODIS based MCD64A1 Version 6, our detections are quite promising, improving on average DC by 0.13 and reducing OE and CE by 0.12 and 0.06, respectively.
Fire has a diverse range of impacts on Earth's physical and social systems. Accurate and up to date information on areas affected by fire is critical to better understand drivers of fire activity, as well as its relevance for biogeochemical cycles, climate, air quality, and to aid fire management. Mapping burned areas was traditionally done from field sketches. With the launch of the first Earth observation satellites, remote sensing quickly became a more practical alternative to detect burned areas, as they provide timely regional and global coverage of fire occurrence. This review paper explores the physical basis to detect burned area from satellite observations, describes the historical trends of using satellite sensors to monitor burned areas, summarizes the most recent approaches to map burned areas and evaluates the existing burned area products (both at global and regional scales). Finally, it identifies potential future opportunities to further improve burned area detection from Earth observation satellites.
A locally-adapted multitemporal two-phase burned area (BA) algorithm has been developed using as inputs Sentinel-2 MSI reflectance measurements in the short and near infrared wavebands plus the active fires detected by Terra and Aqua MODIS sensors. An initial burned area map is created in the first step, from which tile dependent statistics are extracted for the second step. The whole Sub-Saharan Africa (around 25 M km²) was processed with this algorithm at a spatial resolution of 20 m, from January to December 2016. This period covers two half fire seasons on the Northern Hemisphere and an entire fire season in the South. The area was selected as existing BA products account it to include around 70% of global BA. Validation of this product was based on a two-stage stratified random sampling of Landsat multitemporal images. Higher accuracy values than existing global BA products were observed, with Dice coefficient of 77% and omission and commission errors of 26.5% and 19.3% respectively. The standard NASA BA product (MCD64A1 c6) showed a similar commission error (20.4%), but much higher omission errors (59.6%), with a lower Dice coefficient (53.6%). The BA algorithm was processed over >11,000 Sentinel-2 images to create a database that would also include small fires (<100 ha). This is the first time a continental BA product is generated from medium resolution sensors (spatial resolution = 20 m), showing their operational potential for improving our current understanding of global fire impacts. Total BA estimated from our product was 4.9 M km², around 80% larger area than what the NASA BA product (MCD64A1 c6) detected in the same period (2.7 M km²). The main differences between the two products were found in regions where small fires (<100 ha) account for a significant proportion of total BA, as global products based on coarse pixel sizes (500 m for MCD64A1) unlikely detect them. On the negative side, Sentinel-2 based products have lower temporal resolution and consequently are more affected by cloud/cloud shadows and have less temporal reporting accuracy than global BA products. The product derived from S2 imagery would greatly contribute to better understanding the impacts of small fires in global fire regimes, particularly in tropical regions, where such fires are frequent. This product is named FireCCISFD11 and it is publicly available at: https://www.esa-fire-cci.org/node/262, last accessed on November 2018.
The two Moderate Resolution Imaging Spectroradiometer (MODIS) instruments on-board NASA's Terra and Aqua satellites have provided nearly two decades of global fire data. Here, we describe refinements made to the 500-m global burned area mapping algorithm that were implemented in late 2016 as part of the MODIS Collection 6 (C6) land-product reprocessing. The updated algorithm improves upon the heritage Collection 5.1 (C5.1) MCD64A1 and MCD45A1 algorithms by oﬀering significantly better detection of small burns, a modest reduction in burn-date temporal uncertainty, and a large reduction in the extent of unmapped areas. Comparison of the C6 and C5.1 MCD64A1 products for fifteen years (2002-2016) on a regional basis shows that the C6 product detects considerably more burned area globally (26%) and in almost every region considered. The sole exception was in Boreal North America, where the mean annual area burned was 6% lower for C6, primarily as a result of a large increase in the number of small lakes mapped (and subsequently masked) at high latitudes in the upstream C6 input data. With respect to temporal reporting accuracy, 44% of the C6 MCD64A1 burned grid cells were de-tected on the same day as an active fire, and 68% within 2 days, which represents a substantial reduction in temporal uncertainty compared to the C5.1 MCD64A1 and MCD45A1 products. In addition, an areal accuracy assessment of the C6 burned area product undertaken using high resolution burned area reference maps derived from 108 Landsat image pairs is reported.
Fires raged once again across Indonesia in the latter half of 2015, creating a state of emergency due to poisonous smoke and haze across Southeast Asia as well as incurring great financial costs to the government. A strong El Niño-Southern Oscillation (ENSO) led to drought in many parts of Indonesia, resulting in elevated fire occurrence comparable with the previous catastrophic event in 1997/98. Synthetic Aperture Radar (SAR) data promise to provide improved detection of land use and land cover changes in the tropics as compared to methodologies dependent upon cloud and haze free images. This study presents the first spatially explicit estimates of burned area across Sumatra, Kalimantan and West Papua based on high resolution Sentinel-1A SAR imagery. Here we show that 4,604,569 hectares (ha) were burned during the 2015 fire season (overall accuracy 84 %), and compare this with other existing operational burned area products (MCD64, GFED4.0, GFED4.1s). Intersection of burned area with fine-scale land cover and peat layer maps indicates that 0.89 gigatons carbon dioxide equivalents (Gt CO2e) were released through the fire event. This result is compared to other estimates based on non-spatially explicit thermal anomaly measurements or atmospheric monitoring. Using freely available SAR C-band data from the Sentinel mission, we argue that the presented methodology is able to quickly and precisely detect burned areas, supporting improvement in fire control management as well as enhancing accuracy of emissions estimation. This article is protected by copyright. All rights reserved.
Climate, land use, and other anthropogenic and natural drivers have the potential to influence fire dynamics in many regions. To develop a mechanistic understanding of the changing role of these drivers and their impact on atmospheric composition, long-term fire records are needed that fuse information from different satellite and in situ data streams. Here we describe the fourth version of the Global Fire Emissions Database (GFED) and quantify global fire emissions patterns during 1997–2016. The modeling system, based on the Carnegie–Ames–Stanford Approach (CASA) biogeochemical model, has several modifications from the previous version and uses higher quality input datasets. Significant upgrades include (1) new burned area estimates with contributions from small fires, (2) a revised fuel consumption parameterization optimized using field observations, (3) modifications that improve the representation of fuel consumption in frequently burning landscapes, and (4) fire severity estimates that better represent continental differences in burning processes across boreal regions of North America and Eurasia. The new version has a higher spatial resolution (0.25◦) and uses a different set of emission factors that separately resolves trace gas and aerosol emissions from temperate and boreal forest ecosystems. Global mean carbon emissions using the burned area dataset with small fires (GFED4s) were 2.2 × 1015 grams of carbon per year (Pg C yr−1) during 1997–2016, with a maximum in 1997 (3.0 Pg C yr−1) and minimum in 2013 (1.8 Pg C yr−1). These estimates were 11 % higher than our previous estimates (GFED3) during 1997–2011, when the two datasets overlapped. This net increase was the result of a substantial increase in burned area (37 %), mostly due to the inclusion of small fires, and a modest decrease in mean fuel consumption (−19 %) to better match estimates from field studies, primarily in savannas and grasslands. For trace gas and aerosol emissions, differences between GFED4s and GFED3 were often larger due to the use of revised emission factors. If small fire burned area was excluded (GFED4 without the “s” for small fires), average emissions were 1.5 Pg C yr−1. The addition of small fires had the largest impact on emissions in temperate North America, Central America, Europe, and temperate Asia. This small fire layer carries substantial uncertainties; improving these estimates will require use of new burned area products derived from high-resolution satellite imagery. Our revised dataset provides an internally consistent set of burned area and emissions that may contribute to a better understanding of multi-decadal changes in fire dynamics and their impact on the Earth system. GFED data are available from http://www.globalfiredata.org.
Orfeo Toolbox, OTB, is a remote sensing image processing library developed by CNES, the French Space Agency. OTB is distributed as open source software and is therefore available for any remote sensing scientist or processing chain developer. This paper describes the main features of OTB, how it can be used and the expected evolutions in the coming months.
Examination of the physical background underlying the ERS response
of forest and analysis of time series of ERS data indicates that the
greater temporal stability of forest compared with many other types of
land cover presents a means of mapping forest area. The processing chain
necessary to make such area estimations involves reconstruction of an
optimal estimate of the backscattering coefficient at each pixel using
temporal and spatial filtering so that classification rules derived from
large scale averaging are applicable. The rationale behind the filtering
strategy and the level of averaging needed is explained in terms of the
observed multitemporal behavior of forest and nonforest areas, much of
this analysis is generic and applicable to a wide range of situation in
which significant information is carried by multitemporal features of
the data. The choice of decision rules is based on the forest
observations, with the added requirement for robustness. The performance
of a classifier based only on change is assessed on a range of test
sites in the UK, Finland, and Poland. Error sources in this classifier
are identified, and the possibility of improving performance by
including radiometric information in the mapping strategy is discussed.
Brief discussions of how the classification is affected by the addition
of coherence and how the processing chain would need to be modified for
other forms of satellite data are included
A constant false alarm rate (CFAR) detection algorithm (see J.Y.
Chen and I.S. Reed, IEEE Trans. Aerosp. Electron. Syst., vol.AES-23,
no.1, Jan. 1987) is generalized to a test which is able to detect the
presence of known optical signal pattern which has nonnegligible unknown
relative intensities in several signal-plus-noise bands or channels.
This test and its statistics are analytically evaluated, and the
signal-to-noise ratio (SNR) performance improvement is analyzed. Both
theoretical and computer simulation results show that the SNR
improvement factor of this algorithm using multiple band scenes over the
single scene of maximum SNR can be substantial. The SNR gain of this
detection algorithm is compared to the previously published one. It
illustrates that the generalized SNR of the test using the full data
array is always greater than that of using partial data array. The
database used to simulate this adaptive CFAR test is obtained from
actual image scenes
Historical background and current developments for mapping burned area from satellite Earth observation
Emilio Chuvieco Florent Mouillot
Jesús Guido R Van Der Werf
Mihai San Miguel
Global fire emissions estimates during
Guido R Van Der
James T Werf
Yang Thijs T Van Leeuwen
J E Margreet
Guido R Van Der Werf, James T Randerson, Louis
Giglio, Thijs T Van Leeuwen, Yang Chen, Brendan M
Rogers, Mingquan Mu, Margreet JE Van Marle, Douglas C Morton, G James Collatz, et al., "Global fire
emissions estimates during 1997-2016," Earth System
Science Data, vol. 9, pp. 697-720, 2017.