EGU General Assembly 2023
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Tropical leaf phenology characterization by using an ecologically-constrained deep learning model with PlanetScope satellites

Guangqin Song1, Jing Wang1,2, Michael Liddell3, Patricia Morellato4, Calvin K.F. Lee1, Dedi Yang5, Bruna Alberton4,6, Matteo Detto7, Xuanlong Ma8,9,10, Yingyi Zhao1, Henry C.H. Yeung1, Hongsheng Zhang11,12, Michael Ng13, Bruce W. Nelson14, Alfredo Huete15, and Jin Wu1,12
Guangqin Song et al.
  • 1The University of Hong Kong, Faculty of Science, School of Biological Sciences, Hong Kong (
  • 2School of Ecology, Shenzhen Campus of Sun Yat-sen University, Shenzhen, Guangdong, China.
  • 3Centre for Tropical Environmental and Sustainability Science, College of Science and Engineering, James Cook University, Cairns, Queensland 4878, Australia
  • 4Department of Biodiversity, Bioscience Institute, São Paulo State University UNESP, Rio Claro, São Paulo, Brazil
  • 5Department of Environmental and Climate Sciences, Brookhaven National Laboratory, NY11973, USA
  • 6Instituto Tecnológico Vale, Belém, Pará, Brazil
  • 7Department of Ecology and Evolutionary Biology, Princeton University, Princeton, NJ 08544, USA
  • 8College of Earth and Environmental Sciences, Lanzhou University, Lanzhou, China
  • 9International Research Center of Big Data for Sustainable Development Goals, Beijing, China
  • 10Key Laboratory of Geographic Information Science (Ministry of Education), East China Normal University, Shanghai, China
  • 11Department of Geography, The University of Hong Kong, Hong Kong, China
  • 12Institute for Climate and Carbon Neutrality, The University of Hong Kong, Hong Kong, China
  • 13Institute of Data Science and Department of Mathematics, The University of Hong Kong, Hong Kong, China
  • 14National Institute for Amazon Research (INPA), Manaus, Brazil
  • 15School of Life Sciences, University of Technology Sydney, Sydney, NSW 2007, Australia

Tropical leaf phenology signals leaf-on/off status and exhibits strong variability from individual tree crowns to forest ecosystems, which importantly regulates carbon and water fluxes. The availability of daily PlanetScope data with high spatial resolution offers a new chance to monitor phenology variability at both the fine scale and the ecosystem scale across pan-tropics. However, a scalable method for tropical leaf phenology monitoring from PlanetScope with clear biophysical meaning still needs to be developed. To advance tropical leaf phenology monitoring, we developed an index-guided, ecologically constrained autoencoder (IG-ECAE) method to automatically generate a deciduousness metric (percentage of upper tree canopies with leaf-off status within an image pixel) from PlanetScope. The IG-ECAE includes three steps: (1) extracting the initial reflectance spectra of leafy/leafless canopies based on their spectral indices characteristics; (2) training an autoencoder deep learning method with the guidance of derived reflectance spectra and additional ecological constraints to refine the reflectance spectra; and (3) estimating the relative abundance of leafless canopies (or deciduousness) per PlanetScope image pixel with the integration of refined spectra reflectance and linear spectral unmixing method. To test the IG-ECAE method, we compared the PlanetScope-derived deciduousness to the corresponding measures derived from WorldView-2 (n = 9 sites) and local phenocams (n = 9 sites) at 16 tropical forest sites spanning multiple continents and a large precipitation gradient (1470-2819 mm year-1). Our results show that PlanetScope-derived deciduousness agrees: 1) with WorldView-2-derived deciduousness at the patch level (90 m × 90 m) with r2 = 0.89 across all sites; and 2) with phenocam-derived deciduousness to quantify ecosystem-scale seasonality with r2 ranging from 0.62 to 0.96. These results demonstrate that IG-ECAE can accurately characterize the wide variability in deciduousness across scales from pixels to forest ecosystems, and from a single date to the entire annual cycle, indicating the feasibility of tracking the large-scale phenological patterns and responses of tropical forests to climate change with high-resolution satellites.

How to cite: Song, G., Wang, J., Liddell, M., Morellato, P., Lee, C. K. F., Yang, D., Alberton, B., Detto, M., Ma, X., Zhao, Y., Yeung, H. C. H., Zhang, H., Ng, M., Nelson, B. W., Huete, A., and Wu, J.: Tropical leaf phenology characterization by using an ecologically-constrained deep learning model with PlanetScope satellites, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-13177,, 2023.

Supplementary materials

Supplementary material file