GPP estimation by transfer learning with combined solar-induced chlorophyll fluorescence and eddy covariance data

This dataset provides 8-day global gross primary production (GPP) at 0.05° latitude by 0.05° longitude for 2001-2023.

(1) Model description and performance

n this study, a transfer learning (SIFEC-TL) method is proposed to obtain global long-time-series high-accuracy GPP products by combining constrained SIF and EC data. SIF data is used as the source domain to provide spatial information for pre-training, and EC GPP is used as the target domain to provide accurate GPP for fine-tuning of the machine learning model. To validate the performance of SIFEC-TL, the results are compared with those of machine learning models that use only SIF or EC GPP alone (SIFML and ECML). The results show that the SIFEC-TL model has higher temporal consistency and stronger spatial scalability than the SIFML and ECML models, which effectively addresses the underestimation/overestimation of high/low GPP values in the SIFML and ECML models.

(2) Dataset information

Non-vegetated areas in the dataset were filled using nan.

Data format: TIFF

Temporal Resolution: 8-Day

Spatial Resolution: 0.05°

Temporal Extent: 2001 to 2023

Spatial Extent: Global

Unit: g C m-2 8-day-1