Supporting data for ''Decoding tropical non-green phenology: patterns, drivers and implications to regional carbon cycle ''

My PhD thesis proposes a framework that integrates deep learning with multisource remote sensing observations to quantify tropical leaf phenology across various scales, from individual to ecosystem scale. This method was then applied to the entire Amazon intact forests region to characterize phenological patterns. The research utilizes the following data:

(1) Proximate remote sensing data: phenocam observations. A deep learning method was developed to enable monitoring of tropical leaf phenology at a crown scale.

(2) Commercial satellite data (PlanetScope). An unsupervised deep learning unmixing method was proposed to extract tropical leaf phenology monitoring across various tropical forest ecosystems.

(3) Publicly available satellite data (Sentinel-2 and Landsat-8), which were downloaded from third-party sources through the Google Earth Engine platform. These multi-source satellite observations were used to evaluate the impact of spatial scale on tropical leaf phenology monitoring and to identify the critical scale for cost-effective extensive monitoring in the tropics.

(4) Reanalysis data and climate data, such as precipitation and light. These were used to investigate the drivers of tropical leaf phenology variability across the Amazon.

Overall, these data serve as powerful tools for my PhD thesis to quantify tropical leaf phenology across the tropics and to explore the drivers of such dynamic patterns. All the data would be used for non-commercial purposes and are available from the author upon request.