Untitled Item

Gross Primary Productivity (GPP) and Ecosystem Respiration (ER) are pivotal elements in the ecosystem carbon cycle. Over the past few decades, significant advancements have been made in estimating GPP and ER, with numerous released datasets. These datasets are derived from optical remote sensing observations, machine learning models, or process-based models, whose production processes were of high uncertainties due to the lack of an effective evaluation framework.In this dataset, we constructed a more broadly distributed global meteorological station daily carbon flux dataset with Quasi observational properties using machine learning (ML) assessment framework for ecosystem flux station migration to meteorological stations Utilizing data from global ecosystem flux stations and corresponding remote sensing biophysical parameter data, we employ a cross-validation strategy known as the 'leave-one-out location' method to construct hundreds of carbon flux ML models. These models reflect surface spatial heterogeneity and utilize the coefficient of determination (R²) to depict the similarity between model training and testing data. The heterogeneity between training and testing data is captured using an Aggregated Euclidean distance index (AED). Subsequently, we establish a negative correlation between the R² values and the Aggregated Euclidean distance index to identify an optimal AED threshold. This threshold aids in assessing the transferability of carbon flux models to meteorological sites and in constructing a fuzzy membership curve to evaluate the models' applicability at meteorological stations, leading to the generation of carbon flux data.This study analysed the transfer of carbon flux models established on global ecosystem flux stations and corresponding remote sensing information to global meteorological stations validates the efficacy of our research approach and methodology: There is a significant negative correlation between R² and AED, indicating that AED is a reliable indicator of the carbon flux model's generalization ability. The constructed fuzzy membership curve enhances the applicability evaluation system of carbon flux models at meteorological sites. This research provides a more extensively distributed site-scale carbon flux dataset and offers a promising evaluation criterion for the transfer application of ecological models. It merits further investigation to elucidate its wide applicability.