Chen et al. (2025) A comprehensive assessment of the FireCCILT11 global burned area product

Identification

• Journal: International Journal of Applied Earth Observation and Geoinformation
• Year: 2025
• Date: 2025-10-08
• Authors: Xiao Dong Chen, Louis Giglio
• DOI: 10.1016/j.jag.2025.104893

Research Groups

• Department of Geographical Sciences, University of Maryland, College Park, MD, USA

Short Summary

This study comprehensively assessed the FireCCILT11 global burned area product (1982–2018) using national fire history datasets and independent satellite products, revealing significant limitations in its accuracy and consistency, especially before 2001, which renders it generally unsuitable for robust long-term fire regime analyses.

Objective

• To directly and comprehensively assess the accuracy and consistency of the FireCCILT11 global burned area product (1982–2018) using a suite of national fire history datasets and independently derived satellite-based burned area products across eight countries.

Study Configuration

• Spatial Scale: Global, with detailed assessments in the United States, Canada, Australia, Brazil, Portugal, Italy, France, Spain, and Greece.
• Temporal Scale: 1982–2018 (excluding 1994) for FireCCILT11; assessment periods focused on 1986–2018, with specific analyses for pre-2001 (1986–2000) and post-2001 (2001–2018) eras.

Methodology and Data

• Models used:
  • FireCCILT11: Multiple random forest classifiers trained using reference data from the 250 m MODIS FireCCI51 BA product.
  • Assessment methods: Consistency Index (CI) for annual total burned area (BA), Pearson correlation coefficient (R) for per-pixel comparisons, and confusion matrix-based accuracy assessment (commission error, omission error, overall accuracy).
• Data sources:
  • Target Product: FireCCILT11 (European Space Agency Climate Change Initiative, derived from 0.05° AVHRR Land Long Term Data Record (LTDR) from AVHRR Global Area Coverage (GAC) data).
  • National Fire History Data:
    • National Burned Area Composite (NBAC) (Canada, vector-based, 1986–2022, based on field survey, aerial photography, Landsat imagery).
    • National Interagency Fire Center (NIFC) Fire Perimeters (US, vector-based, 1986–2021, based on field survey, aerial photography, satellite imagery).
    • European Forest Fire Information System (EFFIS) (Portugal, Italy, France, Spain, Greece, BA summary, 1983–2022, based on satellite imagery).
    • Australia Interagency Fire History Data (AIFH) (Australia, BA summary, 1988–2019, based on field survey, aerial photography, satellite imagery).
    • MapBiomas Fire Dataset (Brazil, 30 m, 1985–2024, based on Landsat imagery).
  • Remotely Sensed Burned Area Products (for comparison/training):
    • FireCCI51 (MODIS, 231 m, Global, 2001–2020).
    • MCD64A1 (MODIS, 463 m, Global, 2001–2023).

Main Results

• FireCCILT11 exhibited generally low consistency with national fire history datasets in terms of annual total burned area (BA) across most assessed countries (US, Australia, Brazil, Portugal, Italy, France, Spain, Greece), with annual BA totals often differing by more than 50%. Canada was an exception for annual total BA consistency.
• FireCCILT11's annual BA was inconsistent with FireCCI51 in Portugal, France, Spain, and Greece, despite FireCCI51 being used as training data for FireCCILT11.
• Per-pixel assessments showed extremely low correlations between FireCCILT11 and national datasets (NIFC for US, NBAC for Canada, MapBiomas Fire for Brazil) before 2001, with US correlations never exceeding 20%. Post-2001 correlations improved but remained below 50% on average.
• Pre-2001 per-pixel correlations were often negative in Canada and Brazil, indicating conflicting trends in BA representation.
• Confusion matrix analysis revealed exceptionally high omission errors for burned areas in FireCCILT11, particularly during the pre-2001 era (e.g., 79.5% in Canada for 1986–2000, 70.7% in the US for 1986–2000). High commission errors were also observed, notably 48.7% in the US pre-2001.
• FireCCI51 and MCD64A1 consistently outperformed FireCCILT11 in all accuracy metrics for the 2001–2018 period.
• A "pixelation effect" was identified in FireCCILT11 at its native 0.05° resolution, suggesting an effective spatial resolution closer to 0.25° and compromising spatial detail and consistency, with this effect not being spatially uniform.
• The original FireCCILT11 accuracy assessment was deemed insufficient due to its heavy reliance on aggregate BA statistics and the dual use of FireCCI51 for both training/calibration and assessment.

Contributions

• Provided the first comprehensive and direct assessment of the FireCCILT11 global burned area product, offering robust evidence for its limitations.
• Quantitatively demonstrated substantial limitations in FireCCILT11's accuracy and consistency, particularly before 2001, confirming and expanding upon previous circumstantial evidence of instrumental artifacts.
• Highlighted the unsuitability of FireCCILT11 for robust long-term fire regime analyses, especially in regions and periods not well-represented in its training data.
• Identified and characterized a "pixelation effect" in FireCCILT11, indicating a lower effective spatial resolution than stated and inconsistent spatial coherence.
• Proposed best practices for the development and rigorous evaluation of future long-term, multi-sensor burned area datasets, emphasizing independent validation data and per-pixel accuracy metrics.
• Reaffirmed the critical and unsatisfied need for a reliable, multi-decadal global burned area product derived from AVHRR data.

Funding

• National Aeronautics and Space Administration grant 80NSSC24M0019.

Citation

@article{Chen2025comprehensive,
  author = {Chen, Xiao Dong and Giglio, Louis},
  title = {A comprehensive assessment of the FireCCILT11 global burned area product},
  journal = {International Journal of Applied Earth Observation and Geoinformation},
  year = {2025},
  doi = {10.1016/j.jag.2025.104893},
  url = {https://doi.org/10.1016/j.jag.2025.104893}
}