High-precision inversion of vegetation parameters in the AI era: Integrating hyperspectral remote sensing and deep learning

Vegetation traits and parameters serve as key indicators of ecosystem structure, processes, and functioning while also playing crucial roles in biodiversity assessments and the global carbon and water cycles. Remote sensing technologies have emerged as indispensable ecological tools for capturing the spatial and temporal dynamics of vegetation parameters/traits across diverse landscapes and scales. Instead of relying on empirical relationships between remote sensing and vegetation parameters, more sophisticated data models can now be developed that leverage both vegetation spectral and structural signals to account for the complex interactions between radiation and vegetation canopies and provide a more comprehensive and accurate assessment of vegetation parameters. The proliferation of remote sensing data, particularly with the increasing availability of satellite-based imaging spectroscopy, has created an unprecedented dataset of information about the Earth’s terrestrial biosphere. This exponential growth in data, coupled with an increasing demand for more precise vegetation parameter retrievals, has spurred the development of new methodologies aimed at creating efficient, accurate, and adaptable data analysis techniques and applications for deriving vegetation parameters from remote sensing data.

Based on the Web of Science Core Collection, we collected over 40,000 papers published since 1990 with the topic of remote sensing inversion of vegetation parameters. CiteSpace software was subsequently applied for cluster analysis to identify key vegetation parameters, which were then cross validated against previous authoritative studies. A targeted search within the Web of Science Core Collection was further conducted to locate relevant articles specifically addressing these parameters. The retrieved articles were categorized by vegetation parameters and research topics and their numbers summarized (Figure 1A). Given the extensive volume of literature retrieved, ChatGPT was employed to assist in efficiently analyzing and summarizing the abstracts. This process enabled systematic classification and analysis of the papers based on their methodologies, conclusions, and contributions. Finally, an in-depth analysis of the advancements in remote sensing inversion methods for vegetation parameters was performed.