EGU24-513, updated on 08 Mar 2024
https://doi.org/10.5194/egusphere-egu24-513
EGU General Assembly 2024
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Post-fire forest recovery in Protected Areas of the Amazon: Disentangling natural processes from human disturbances using multi-source remote sensing data and machine learning

Qianhan Wu1, Calvin K.F. Lee1, Jonathan A. Wang2, Yingyi Zhao1, Guangqin Song1, Eduardo Eiji Maeda3,4, Yanjun Su5,6, Alfredo Huete7, and Jin Wu1
Qianhan Wu et al.
  • 1University of Hong Kong, Faculty of Science, School of Biological Sciences, Hong Kong (qhwu@connect.hku.hk)
  • 2School of Biological Sciences, The University of Utah, Salt Lake City, Utah, United States of America
  • 3Department of Geosciences and Geography, P.O. Box 68, FI-00014, University of Helsinki, Finland
  • 4Finnish Meteorological Institute, FMI, Finland
  • 5State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing, China
  • 6University of Chinese Academy of Sciences, Beijing 100049, China
  • 7Faculty of Science, University of Technology Sydney, Sydney, New South Wales, Australia

Establishing protected areas (PAs) in Amazon forests is crucial for safeguarding tropical forest ecosystem from human land use and mitigating forest degradation. However, PAs across the Amazon basin have increasingly suffered from intensified fires. Understanding post-fire recovery trajectories in these protected forests is essential for assessing the resilience and effectiveness of PAs. However, recovery trajectories under natural conditions remain unclear, as human settlements often disrupt or influence the recovery process, potentially diminishing recovery rates and forest potential. To address this challenge, we investigated 4,036 fire events that occurred from 2001 to 2020 within PAs in the eastern Amazon detected by Moderate Resolution Imaging Spectroradiometer (MODIS) satellite. Furthermore, we explored the effectiveness of multi-source earth observation data and eXtreme Gradient Boost machine learning model in distinguishing fire areas where recovery of local forests undergoes natural conditions (N-recovery) from those impacted by human activities (H-recovery). We then analyzed temporal trends in fire burn severity (based on the relationship between fire year and Landsat-derived burn severity metrics) and post-fire canopy structure recovery (based on the relationship between GEDI lidar-derived canopy structure metrics and fire age using a space-for-time substitution approach) for both recovery types. Our model accurately differentiated N-recovery (n=2019) from H-recovery (n=2017) with an overall classification accuracy of 87.61%.  Our analysis further reveals a clear increasing trend in fire burn severity for N-recovery from 2001 to 2020, while the trend for H-recovery was relatively stable with no significant change. Moreover, the recovery rates of relative heights (RH), canopy ratio (CR), and plant area index (PAI) in N-recovery areas were significantly higher than those in H-recovery areas over 20 years, highlighting the importance of separating these two recovery types. By focusing on N-recovery areas, we found that forest structural traits related to understory recovery and plant vertical space use (i.e., PAI values across the entire vertical strata) exhibited stronger recovery rates than traits related to height metrics (i.e., RHs), revealing their utility for characterizing more complex ecosystem recovery processes. These findings demonstrate the potential and necessity of using multi-source earth observation data to distinguish between the two types of post-fire forest recovery. This distinction contributes to an improved understanding of ecological recovery rates and processes of post-fire forest successional dynamics under natural conditions, offering new opportunities to further study their biogeographical distribution, recovery rate variabilities, and impacts on carbon sequestration and ecosystem resilience under climate change. 

How to cite: Wu, Q., Lee, C. K. F., Wang, J. A., Zhao, Y., Song, G., Maeda, E. E., Su, Y., Huete, A., and Wu, J.: Post-fire forest recovery in Protected Areas of the Amazon: Disentangling natural processes from human disturbances using multi-source remote sensing data and machine learning, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-513, https://doi.org/10.5194/egusphere-egu24-513, 2024.