FLAML version 2.3.3 model-based assessment of gross primary productivity at forest, grassland, and cropland ecosystem sites

• Published in: Geoscientific Model Development Vol. 18; no. 16; pp. 5115 - 5142
• Main Authors: Lai, Jie, Zhang, Yuan, Wang, Anzhi, Fei, Wenli, Diao, Yiwei, Li, Rongping, Wu, Jiabing
• Format: Journal Article
• Language: English
• Published: Katlenburg-Lindau Copernicus GmbH 22.08.2025; Copernicus Publications
• Subjects: Agricultural land; Algorithms; Analysis; Automation; Carbon; Carbon cycle; Climate change; Climate models; Climatic conditions; Climatic extremes; Coniferous forests; Crops; Drought; Drought conditions; Ecosystems; Eddy covariance; Efficiency; Estimation; Forests; Grasslands; High temperature; Limiting factors; Machine learning; Meteorological data; Mixed forests; MODIS; Photosynthesis; Primary production; Productivity; Radiation; Remote sensing; Respiration; Temporal variations; Terrestrial ecosystems; Vapor pressure; Vapour pressure; Vegetation
• ISSN: 1991-9603; 1991-959X; 1991-962X; 1991-9603; 1991-962X
• DOI: 10.5194/gmd-18-5115-2025

Accurately estimating gross primary productivity (GPP) in terrestrial ecosystems is essential for understanding the global carbon cycle. Satellite-based light use efficiency (LUE) models are commonly employed for simulating GPP. However, the variables and algorithms related to environmental limiting factors differ significantly across various LUE models, leading to high uncertainty in GPP estimation. In this work, we developed a series of FLAML-LUE models with different variable combinations. These models utilize the Fast Lightweight Automated Machine Learning (FLAML) framework, using variables of LUE models, to investigate the potential of estimating site-scale GPP. Incorporating meteorological data, eddy covariance measurements, and remote sensing indices, we employed FLAML-LUE models to assess the impact of various variable combinations on GPP across different temporal scales, including daily, 8 d, 16 d, and monthly intervals. Cross-validation analyses indicated that the FLAML-LUE model performs excellently in GPP prediction, accurately simulating both its temporal variations and magnitude, particularly in mixed forests and coniferous forests, with average R2 values for daily-scale simulations reaching 0.92 and 0.91, respectively. However, the model performed less effectively in alpine shrubland and typical grassland ecosystems, though it still outperformed both MODIS GPP and PML GPP in terms of performance. Furthermore, the model's adaptability under extreme climate conditions was evaluated, and the results showed that high temperatures and high vapor pressure deficit (VPD) lead to a slight decrease in model accuracy, though R2 remains around 0.8. Under drought conditions, the model's performance improved slightly in croplands and evergreen broadleaf forests, although it declined at some sites. This study offers an approach to estimate GPP fluxes and evaluate the impact of variables on GPP estimation. It has the potential to be applied in predicting GPP for different vegetation types at a regional scale.