International Journal of Applied Earth Observations and Geoinformation 103 (2021) 102473 Available online 18 August 2021 0303-2434/© 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Development of a consistent global long-term burned area product (1982–2018) based on AVHRR-LTDR data Gonzalo Ot´ on * , Joshua Lizundia-Loiola , M. Lucrecia Pettinari , Emilio Chuvieco * Universidad de Alcal´ a, Departamento de Geología, Geografía y Medio Ambiente, Grupo de Investigaci´ on en Teledetecci´ on Ambiental, Colegios 2, 28801 Alcal´ a de Henares, Madrid, Spain A R T I C L E INFO Keywords: Burned area AVHRR-LTDR Multi-temporal Random Forest FireCCILT11 CCI Otsu thresholding ABSTRACT This paper presents the generation of a global long-term Burned Area (BA) product based on Advanced Very High Resolution Radiometer (AVHRR) images. The BA product was derived from the Land Long Term Data Record (LTDR), which provides a continuous dataset of geometrically and radiometrically corrected AVHRR images at 0.05 ◦ resolution (≈5 km). The BA algorithm improves a Beta version of this dataset (named FireCCILT10) previously released. The new version incorporates an enhanced Random Forest (RF) classifcation process based on two models, one using a global sample and another one using only Boreal regions. Solar zenith angle (SZA) corrections were introduced to mitigate the impact of satellite orbital drift. Binary classifcations were obtained applying probability thresholds, and BA proportions were assigned to each burned pixel. The fnal product in- cludes the date of detection at 0.05 ◦ resolution and the total burned area at 0.05 ◦ and 0.25 ◦ resolution, both covering the period from 1982 to 2018 (excluding 1994). The resulting product, called FireCCILT11, estimated that 165.26 Mkm 2 were globally burned between 1982 and 2018, with an annual average of 4.59 Mkm 2 . The largest BA was found in 2011 with 5.18 Mkm 2 and the lowest was observed in 1991 with 4.09 Mkm 2 . The month with the highest mean BA was August, with 0.63 Mkm 2 , and the one with the lowest was March with 0.15 Mkm 2 . Africa included 66% of total BA. Inter-comparison showed high correlation values with MODIS BA products for annual BA of the common years (r > 0.6, %MAE < 14%). Comparison with national fre statistics of Australia, Canada and Alaska showed also high correlation values (r > 0.8, %MAE < 42%). 1. Introduction Fire disturbance was identifed by the Global Climate Observing System (GCOS) programme as an Essential Climate Variable (ECV) because of the impacts of burnings in atmospheric composition and the carbon cycle (Yue et al. 2015). To answer GCOS’ needs, the European Space Agency (ESA) created in 2010 the Climate Change Initiative Programme (CCI), which included then 10 ECVs and now 21 (https://climate.esa.int/en/, last accessed July 2021). The Fire Distur- bance project was part of the initial programme and therefore has been active for more than 10 years. Within this project, several global BA products from different sensors have been generated and released, including inputs from Envisat-MERIS (FireCCI41: Alonso-Canas and Chuvieco 2015), MODIS (FireCCI51: Lizundia-Loiola et al. 2020) and AVHRR (FireCCILT10: Ot´ on et al. 2019) data. The National Aeronautics and Space Administration (NASA) have also developed BA products based on MODIS (MCD64A1: Giglio et al. 2018), in addition to those oriented towards detecting active fres. Currently, the FireCCI51 and MCD64A1, based on MODIS 250 m and 500 m bands, respectively, are the most accurate global products, widely used in different modelling studies (Chuvieco et al. 2019). One of the recommendations of the GCOS programme was to extend the BA datasets backwards to the 80s (GCOS 2016), as all existing global products were based on sensors that were launched at the end of the 1990s or early 2000s. This requirement, also indicated by many climate modellers (Mouillot et al. 2014;Chuvieco et al. 2019), implies using the historical catalogue of the AVHRR sensor, on board the National Oceanic and Atmospheric Administration (NOAA) satellites since 1979. How- ever, this sensor presents many limitations for generating BA datasets, as it has a coarse resolution and it is affected by changes in sensor versions, orbital drift of the satellite, acquisition problems, random noises, and location errors (McGregor and Gorman 1994; Weber and Wunderle 2019). In spite of these problems, several BA products have been generated from AVHRR images, mostly for selected regional areas * Corresponding authors. E-mail addresses: gonzalo.oton@uah.es (G. Ot´ on), emilio.chuvieco@uah.es (E. Chuvieco). Contents lists available at ScienceDirect International Journal of Applied Earth Observations and Geoinformation journal homepage: www.elsevier.com/locate/jag https://doi.org/10.1016/j.jag.2021.102473 Received 15 February 2021; Received in revised form 14 July 2021; Accepted 31 July 2021