@article{fdi:010094446,
  title = {{R}econstructed global monthly burned area maps from 1901 to 2020},
  author = {{G}uo, {Z}. {X}. and {L}i, {W}. and {C}iais, {P}. and {S}itch, {S}. and van der {W}erf, {G}. {R}. and {B}owring, {S}. {P}. {K}. and {B}astos, {A}. and {M}ouillot, {F}lorent and {H}e, {J}. {Y}. and {S}un, {M}. {X}. and {Z}hu, {L}. and {D}u, {X}. {M}. and {W}ang, {N}. and {H}uang, {X}. {M}.},
  editor = {},
  language = {{ENG}},
  abstract = {{F}ire is a key {E}arth system process, driving variability in the global carbon cycle through {CO}2 emissions into the atmosphere and subsequent {CO}2 uptake through vegetation recovery after fires. {G}lobal spatiotemporally consistent datasets on burned area have been available since the beginning of the satellite era in the 1980s, but they are sparse prior to that date. {I}n this study, we reconstructed global monthly burned area at a resolution of 0.5 degrees x 0.5 degrees from 1901 to 2020 using machine learning models trained on satellite-based observations of burned area between 2003 and 2020, with the goal of reconstructing long-term burned area information to constrain historical fire simulations. {W}e first conducted a classification model to separate grid cells with extreme (burned area >= the 90th percentile in a given region) or regular fires. {W}e then trained separate regression models for grid cells with extreme or regular fires. {B}oth the classification and regression models were trained on a satellite-based burned area product ({F}ire{CCI}51), using explanatory variables related to climate, vegetation and human activities. {T}he trained models can well reproduce the long-term spatial patterns (slopes = 0.70-1.28 and {R}-2 = 0.69-0.98 spatially), inter-annual variability and seasonality of the satellite-based burned area observations. {A}fter applying the trained model to the historical period, the predicted annual global total burned area ranges from 3.46x10(6) to 4.58x10(6) km(2) yr(-1) over 1901-2020 with regular and extreme fires accounting for 1.36x10(6)-1.74x10(6) and 2.00x10(6)-3.03x10(6) km(2) yr(-1), respectively. {O}ur models estimate a global decrease in burned area during 1901-1978 (slope ==-0.009x10(6) km(2) yr(-2)), followed by an increase during 1978-2008 (slope = 0.020x10(6) km(2) yr(-2)), and then a stronger decline in 2008-2020 (slope = -0.049x10(6 )km(2) yr(-2)). {A}frica was the continent with the largest burned area globally during 1901-2020, and its trends also dominated the global trends. {W}e validated our predictions against charcoal records, and our product exhibits a high overall accuracy in simulating fire occurrence (>80 %) in boreal {N}orth {A}merica, southern {E}urope, {S}outh {A}merica, {A}frica and southeast {A}ustralia, but the overall accuracy is relatively lower in northern {E}urope and {A}sia (<50 %). {I}n addition, we compared our burned area data with multiple independent regional burned area maps in {C}anada, the {USA}, {B}razil, {C}hile and {E}urope, and found general consistency in the spatial patterns (linear regression slopes ranging 0.84-1.38 spatially) and the inter-annual variability. {T}he global monthly 0.5 degrees x 0.5 degrees burned area fraction maps for 1901-2020 presented by this study can be downloaded for free from https://doi.org/10.5281/zenodo.14191467 ({G}uo and {L}i, 2024).},
  keywords = {{MONDE}},
  booktitle = {},
  journal = {{E}arth {S}ystem {S}cience {D}ata},
  volume = {17},
  numero = {7},
  pages = {3599--3618},
  ISSN = {1866-3508},
  year = {2025},
  DOI = {10.5194/essd-17-3599-2025},
  URL = {https://www.documentation.ird.fr/hor/fdi:010094446},
}