Global retrieval of canopy chlorophyll content from Sentinel-3 OLCI TOA data using a two-step upscaling method integrating physical and machine learning models
Dong Li , Holly Croft , Gregory Duveiller , Adam P. Schreiner-McGraw , Anirudh Belwalkar , Tao Cheng , Yan Zhu , Weixing Cao , Kang Yu
{"title":"Global retrieval of canopy chlorophyll content from Sentinel-3 OLCI TOA data using a two-step upscaling method integrating physical and machine learning models","authors":"Dong Li , Holly Croft , Gregory Duveiller , Adam P. Schreiner-McGraw , Anirudh Belwalkar , Tao Cheng , Yan Zhu , Weixing Cao , Kang Yu","doi":"10.1016/j.rse.2025.114845","DOIUrl":null,"url":null,"abstract":"<div><div>Canopy chlorophyll content per ground area (CCC, g·m<sup>−2</sup>) is tightly related to vegetation photosynthesis and is a promising indicator of photosynthetic capacity. However, a global operational CCC product is not yet available. To fill this gap, we developed a two-step upscaling method to estimate global CCC from Sentinel-3 OLCI top-of-atmosphere (TOA) reflectance. In the first step, a physically-based PROSAIL-D inversion model produced accurate CCC maps from over 20,000 high-spatial resolution (1 m) airborne hyperspectral images collected across 50 sites within the National Ecological Observatory Network (NEON) between 2019 and 2021. The validation against ground CCC measurements showed an R<sup>2</sup> of 0.89 and an RMSE of 0.30 g·m<sup>−2</sup>. In the second step, these high-resolution CCC maps were resampled or upscaled to a spatial resolution of 300 m, and combined with Sentinel-3 OLCI TOA reflectance images to train random forest (RF) models. The RF model demonstrated robust performance with leave-one-site-out cross-validation, yielding an R<sup>2</sup> of 0.92 and RMSE of 0.14 g·m<sup>−2</sup>. The two-step method also showed minimal sensitivity to angular effects and land cover variations, underscoring its robustness. In comparison, the traditional direct inversion method (the one-step method) led to underestimation of CCC by 0.16 g·m<sup>−2</sup> and a moderate estimation accuracy (R<sup>2</sup> = 0.65, RMSE = 0.30 g·m<sup>−2</sup>). We generated a long-term global OLCI CCC product using Sentinel-3 OLCI TOA reflectance data from 2016 to 2024, which can also be continuously updated using current data. This global CCC product can provide important plant physiological information, for parameterizing terrestrial biosphere models and capturing spatiotemporal photosynthetic patterns, thereby advancing research on vegetation carbon dynamics cycles at the global scale.</div></div>","PeriodicalId":417,"journal":{"name":"Remote Sensing of Environment","volume":"328 ","pages":"Article 114845"},"PeriodicalIF":11.1000,"publicationDate":"2025-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Remote Sensing of Environment","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0034425725002494","RegionNum":1,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENVIRONMENTAL SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
Canopy chlorophyll content per ground area (CCC, g·m−2) is tightly related to vegetation photosynthesis and is a promising indicator of photosynthetic capacity. However, a global operational CCC product is not yet available. To fill this gap, we developed a two-step upscaling method to estimate global CCC from Sentinel-3 OLCI top-of-atmosphere (TOA) reflectance. In the first step, a physically-based PROSAIL-D inversion model produced accurate CCC maps from over 20,000 high-spatial resolution (1 m) airborne hyperspectral images collected across 50 sites within the National Ecological Observatory Network (NEON) between 2019 and 2021. The validation against ground CCC measurements showed an R2 of 0.89 and an RMSE of 0.30 g·m−2. In the second step, these high-resolution CCC maps were resampled or upscaled to a spatial resolution of 300 m, and combined with Sentinel-3 OLCI TOA reflectance images to train random forest (RF) models. The RF model demonstrated robust performance with leave-one-site-out cross-validation, yielding an R2 of 0.92 and RMSE of 0.14 g·m−2. The two-step method also showed minimal sensitivity to angular effects and land cover variations, underscoring its robustness. In comparison, the traditional direct inversion method (the one-step method) led to underestimation of CCC by 0.16 g·m−2 and a moderate estimation accuracy (R2 = 0.65, RMSE = 0.30 g·m−2). We generated a long-term global OLCI CCC product using Sentinel-3 OLCI TOA reflectance data from 2016 to 2024, which can also be continuously updated using current data. This global CCC product can provide important plant physiological information, for parameterizing terrestrial biosphere models and capturing spatiotemporal photosynthetic patterns, thereby advancing research on vegetation carbon dynamics cycles at the global scale.
期刊介绍:
Remote Sensing of Environment (RSE) serves the Earth observation community by disseminating results on the theory, science, applications, and technology that contribute to advancing the field of remote sensing. With a thoroughly interdisciplinary approach, RSE encompasses terrestrial, oceanic, and atmospheric sensing.
The journal emphasizes biophysical and quantitative approaches to remote sensing at local to global scales, covering a diverse range of applications and techniques.
RSE serves as a vital platform for the exchange of knowledge and advancements in the dynamic field of remote sensing.