A direct evaluation of long-term global Leaf Area Index (LAI) products using massive high-quality LAI validation samples derived from Landsat archive

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.

Add a Comment

You must log in to post a comment.


There are no comments or no comments have been made public for this article.


Download Preprint


Junjun Zha, Muyi Li, Zaichun Zhu, Sen Cao, Yanan Zhang, Weiqing Zhao, Yue Chen


The long-term global Leaf Area Index (LAI) products are critical supports for characterizing the changes in land surface and its interactions with other components of the Earth system under the dramatic global change. However, intercomparisons between current available long-term global LAI products present significant spatiotemporal inconsistencies which have been a persistent source of uncertainties in global change ecology. Yet, a direct and systematic evaluation of current long-term LAI products is still lacking due to the absence of appropriate LAI references, especially before 2000. Here, we proposed a novel evaluation framework to directly evaluate the mainstream long-term global LAI products (GIMMS LAI3g, GLASS LAI, and GLOBMAP LAI) using massive high-quality LAI validation samples. The LAI validation samples, derived from the Landsat archive using machine learning and MODIS LAI, have a global distribution, a long temporal coverage (1982−2020), and a large amount of 4.9 million. They substantially address the issue of insufficient LAI reference data and can enable quantitative LAI assessments. The long-term global LAI products showed reasonable quality in terms of absolute value, with GIMMS LAI3g having better performance (R:0.96; MAE: 0.29 m^2 m^(-2); RMSE: 0.49 m^2 m^(-2)), followed by GLASS LAI (R:0.96; MAE: 0.31 m^2 m^(-2); RMSE: 0.51 m^2 m^(-2)) and GLOBMAP LAI (R:0.90; MAE: 0.52 m^2 m^(-2); RMSE: 0.91 m^2 m^(-2)). For all LAI products, the data quality after 2000 was better than before 2000. Their annual maximum LAI trends presented mediocre consistencies with the LAI validation samples (R: 0.20−0.29) which showed a significantly larger area of greening. The evaluation of ten state-of-the-art ecosystem models demonstrated varied capabilities in simulating global LAI trends, with the standard deviations ranging from ~0.01 to 0.04 m^2 m^(-2) a^(-1). Although the Multi-Model Ensemble Mean LAI agreed with satellite-based LAI products, they differed with vegetation biomes especially for the tropics. The Landsat LAI validation dataset produced in this study can facilitate the development of long-term global LAI products and provide a quantitative reference for vegetation dynamic studies.




Biochemistry, Biophysics, and Structural Biology


Vegetation trend, Long-term global LAI products, LAI validation samples, Landsat archive, TRENDY, Random Forests regressor


Published: 2023-09-13 11:14


CC BY Attribution 4.0 International

Additional Metadata

Data Availability (Reason not available):
The data associated with the manuscript is under organization. It will soon be available. We desire to properly label and better present the tens of thousands of LAI samples of varying locations, dates, and Landsat scenes.