Significance: Fires burn an area comparable to Europe each year, emitting greenhouse gases and aerosols. We compared burned area (BA) based on 20-m resolution images with a BA derived from 500-m data. It represents an 80% increase in BA in sub-Saharan Africa, responsible for about 70% of global BA. This difference is predominately (87%) attributed to small fires (<100 ha), which account for 41% of total BA but only for 5% in coarse-resolution products. We found that African fires were responsible for emissions of 1.44 PgC, 31–101% higher than previous estimates and representing 14% of global CO ₂ emissions from fossil fuel burning. We conclude that small fires are critically important in characterizing the most important disturbance agent on a global scale.

This paper presents a new global burned area (BA) product, generated from the Moderate Resolution Imaging Spectroradiometer (MODIS) red (R) and near-infrared (NIR) reflectances and thermal anomaly data, thus providing the highest spatial resolution (approx. 250 m) among the existing global BA datasets. The product includes the full times series (2001–2016) of the Terra-MODIS archive. The BA detection algorithm was based on monthly composites of daily images, using temporal and spatial distance to active fires. The algorithm has two steps, the first one aiming to reduce commission errors by selecting the most clearly burned pixels (seeds), and the second one targeting to reduce omission errors by applying contextual analysis around the seed pixels. This product was developed within the European Space Agency's (ESA) Climate Change Initiative (CCI) programme, under the Fire Disturbance project (Fire_cci). The final output includes two types of BA files: monthly full-resolution continental tiles and biweekly global grid files at a degraded resolution of 0.25°. Each set of products includes several auxiliary variables that were defined by the climate users to facilitate the ingestion of the product into global dynamic vegetation and atmospheric emission models. Average annual burned area from this product was 3.81 Mkm², with maximum burning in 2011 (4.1 Mkm²) and minimum in 2013 (3.24 Mkm²). The validation was based on a stratified random sample of 1200 pairs of Landsat images, covering the whole globe from 2003 to 2014. The validation indicates an overall accuracy of 0.9972, with much higher errors for the burned than the unburned category (global omission error of BA was estimated as 0.7090 and global commission as 0.5123). These error values are similar to other global BA products, but slightly higher than the NASA BA product (named MCD64A1, which is produced at 500 m resolution). However, commission and omission errors are better compensated in our product, with a tendency towards BA underestimation (relative bias −0.4033), as most existing global BA products. To understand the value of this product in detecting small fire patches ( < 100 ha), an additional validation sample of 52 Sentinel-2 scenes was generated specifically over Africa. Analysis of these results indicates a better detection accuracy of this product for small fire patches ( < 100 ha) than the equivalent 500 m MCD64A1 product, although both have high errors for these small fires. Examples of potential applications of this dataset to fire modelling based on burned patches analysis are included in this paper. The datasets are freely downloadable from the Fire_cci website (https://www.esa-fire-cci.org/, last access: 10 November 2018) and their repositories (pixel at full resolution: https://doi.org/cpk7, and grid: https://doi.org/gcx9gf).

The file associated with this record is under a 24-month embargo from publication in accordance with the publisher's self-archiving policy. The full text may be available through the publisher links provided above.The accuracies of six global burned area (BA) products for year 2008 were compared using the same validation methods and reference data to quantify accuracy of each product. The selected products include MCD64, MCD45 and Geoland2, and three products developed within the Fire Disturbance project (fire_cci), which is part of the European Space Agency's (ESA) Climate Change Initiative (CCI) program. The latter three products were derived from MERIS and VEGETATION sensors (one product from each sensor separately, and a third one from the merging of MERIS and VGT products). The reference fire perimeters were mapped from two multi-temporal Landsat TM/ETM + images at 103 non-overlapping Thiessen scene areas (TSA) selected with a stratified random sampling design. The validation results were based on cross tabulated error matrices from which six accuracy measures were computed following the requirements of end-users of burned area products. While overall accuracy (OA) exceeded 99% for all products, overall accuracy was lower for the burned class. Burned area commission error ratio was above 40% for all products and omission error ratio was above 65% for all products. The statistical significance of differences in accuracy between pairs of products was evaluated based on theory of the stratified combined ratio estimator. Statistical tests identified the MCD64 as the most accurate product, followed by MCD45 and the MERIS product.Peer-reviewedPost-prin

This paper aims to develop a global burned area (BA) algorithm for MODIS BRDF-corrected images based on the Random Forest (RF) classifier. Two RF models were generated, including: (1) all MODIS reflective bands; and (2) only the red (R) and near infrared (NIR) bands. Active fire information, vegetation indices and auxiliary variables were taken into account as well. Both RF models were trained using a statistically designed sample of 130 reference sites, which took into account the global diversity of fire conditions. For each site, fire perimeters were obtained from multitemporal pairs of Landsat TM/ETM+ images acquired in 2008. Those fire perimeters were used to extract burned and unburned areas to train the RF models. Using the standard MD43A4 resolution (500 × 500 m), the training dataset included 48,365 burned pixels and 6,293,205 unburned pixels. Different combinations of number of trees and number of parameters were tested. The final RF models included 600 trees and 5 attributes. The RF full model (considering all bands) provided a balanced accuracy of 0.94, while the RF RNIR model had 0.93. As a first assessment of these RF models, they were used to classify daily MCD43A4 images in three test sites for three consecutive years (2006–2008). The selected sites included different ecosystems: Australia (Tropical), Boreal (Canada) and Temperate (California), and extended coverage (totaling more than 2,500,000 km²). Results from both RF models for those sites were compared with national fire perimeters, as well as with two existing BA MODIS products; the MCD45 and MCD64. Considering all three years and three sites, commission error for the RF Full model was 0.16, with an omission error of 0.23. For the RF RNIR model, these errors were 0.19 and 0.21, respectively. The existing MODIS BA products had lower commission errors, but higher omission errors (0.09 and 0.33 for the MCD45 and 0.10 and 0.29 for the MCD64) than those obtained with the RF models, and therefore they showed less balanced accuracies. The RF models developed here should be applicable to other biomes and years, as they were trained with a global set of reference BA sites.

A new supervised burned area mapping software named BAMS (Burned Area Mapping Software) is presented in this paper. The tool was built from standard ArcGIS^TM libraries. It computes several of the spectral indexes most commonly used in burned area detection and implements a two-phase supervised strategy to map areas burned between two Landsat multitemporal images. The only input required from the user is the visual delimitation of a few burned areas, from which burned perimeters are extracted. After the discrimination of burned patches, the user can visually assess the results, and iteratively select additional sampling burned areas to improve the extent of the burned patches. The final result of the BAMS program is a polygon vector layer containing three categories: (a) burned perimeters, (b) unburned areas, and (c) non-observed areas. The latter refer to clouds or sensor observation errors. Outputs of the BAMS code meet the requirements of file formats and structure of standard validation protocols. This paper presents the tool’s structure and technical basis. The program has been tested in six areas located in the United States, for various ecosystems and land covers, and then compared against the National Monitoring Trends in Burn Severity (MTBS) Burned Area Boundaries Dataset.