EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

The spatio-temporal characteristics of forest fires in China: observations from hybrid remote sensing data

Lei Fang1, Zeyu Qiao2, and Jian Yang3
Lei Fang et al.
  • 1Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang, 110016, P.R.China (
  • 2Shandong University of Science and Technology, Qingdao, 266590, P.R.China (
  • 3Department of Forestry and Natural Resources, TP Cooper Building, University of Kentucky, Lexington, KY, 40546, USA (

Forest fire is a natural disaster threatening global human well-beings as well as a crucial disturbance agent driving forest landscape changes. The remotely sensed burned area (BA) products can provide spatially and temporally continuous monitoring of global fires, but the accuracies remain to be improved. We firstly developed a hybrid burned area mapping approach, which integrated the advantages of a 250 m global BA product (CCI_Fire) and a 30 m global forest change (GFC) product, to generate an improved 250 m BA product (so-called CCI_GFC product). Based on 248 fire patches derived from Landsat imagery, the results showed that the CCI_GFC product improved the CCI_Fire product substantially, which are significantly better than MCD64A1 product. According to the CCI_GFC, we found the total BA in the past 17 years was about 12.1 million ha in China, which approximately covered 6.1% of the total forested areas with a significantly decreased trend through Mann-Kendall test (Tau= -0.47, P<0.05) . We conducted a grid analysis (0.05°×0.05°) to determine the hot spots of forest fire from 2001 to 2017. We also quantified fire characteristics on frequency, spatial distribution, and seasonality in terms of Burned Forest Rate (BFR), hot spot areas, and fire seasons, respectively. We found that low frequency burns with a 0<BFR≤20% in 17 years covered 64% of total grids; the medium-low frequency burns (20%<BFR≤40%), the medium frequency burns (40%<BFR≤60%), the medium-high frequency burns (60%< BFR≤80%) accounted for 15%, 7%, 4% respectively; the high frequency burns (80%<BFR≤100%) and extremely high burns (100%<BFR≤120%) together occupy 10% of total grids which mainly distributed in Xiao Hinggan mountains, south China, and southwest China. The seasonality of forest fires differed substantially among eco-regions. The fire seasons of two temperate forest eco-regions are spring and autumn. The two peak fire months are May and October, in which about 22% and 37% of the total burned area were founded respectively. As a comparison, fire seasons in tropical and subtropical eco-regions are spring and winter (i.e., November to March of the next year), which accounted 88% of the total burned area. Our study clearly illustrated the characteristics of forest fire patterns in the past 17 years, which highlighted the remarkable achievements due to a nationwide implementation of fire prevention policy. At the same time, we emphasized that it is critically important to regard the long-term forest fire dynamics to design scientific and reasonable strategies or methods for fire management and controlling, which will be of sound significance to optimize the allocation of financial resources on fire management, and to achieve sustainable management of forests.

How to cite: Fang, L., Qiao, Z., and Yang, J.: The spatio-temporal characteristics of forest fires in China: observations from hybrid remote sensing data, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-2281,, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 1 – uploaded on 06 May 2020
  • CC1: Approach to combine FireCCI51 and GFC, M. Lucrecia Pettinari, 06 May 2020

    Hello Lei, I have some questions regarding your presentation.

    1. Are you assuming in your methodology that all the GFC in areas near a burned patch are caused by fires? Also when there is no burned patch directly associated to it? Could the forest loss be due to other causes?

    2. Is the GXAM database available somewhere? How was it produced? It would be interesting to do further evaluation of the BA products.

    By the way, thank you for your appreciation in the last slide :-)

    • AC1: Reply to CC1, Lei Fang, 06 May 2020

      Hi Lucrecia, I am so happy that you notice my presentation. Thank you very much for your comments and your efforts on developing ESA CCI_Fire product.

      For Q1: You are right. We did know some other forest disturbance may cause commission error. In our study, we applied a conservative buffer distance (~5 km) and assumed pixels around seed pixels (found in both CCI_Fire and GFC) are burned areas. In the subsequent analysis, we using spatial filter to remove fragment pixels. Please note that we are focusing on forest pixels only in this study. Harvesting and post-fire harvesting can influence our results, but I think the influence is limited as the harvesting with large size is forbidden in our country.

      For Q2: Sorry such technical details were not provided in my material. The burned patches are manually delineate using Landsat imagery. I noticed your team just developed a reference dataset. That is great work. Wish we can collaborate for BA validation. Thanks.

      • CC2: Reply to AC1, M. Lucrecia Pettinari, 06 May 2020

        Hi Lei,

        Thank you for your explanation. I hope that you will publish this analysis, to have more detailed information regarding your methods, as it is impossible to do it in a presentation format.

        Regarding the validation data, you have probably noticed the article in Earth Science System Data that is currently under revision by Franquesa et al. ( Since the BARD dataset is open to new inputs, maybe you could contact Magí Franquesa to analyse the possibility to include that dataset there.