TRAX GPP: Global 0.05° Monthly GPP Dataset (2001–2024) Derived from a Hybrid LSTM Framework

1. Overview Gross primary production (GPP) represents the total amount of carbon fixed through photosynthesis and is a fundamental variable for understanding terrestrial carbon and energy cycles. Persistent discrepancies among existing global GPP products—particularly in their seasonal dynamics, interannual variability, and long-term trends—highlight the need for datasets with improved temporal robustness and spatial detail. To address this gap, we developed TRAX GPP, a global GPP dataset with 0.05° spatial resolution and monthly coverage from 2001 to 2024. The dataset is derived using a hybrid LSTM-based modelling framework that incorporates Robust Seasonal–Trend decomposition (RSTL), an interannual-difference–enhanced loss function, and continuous XCO2 fields to better represent CO₂ fertilization effects. These components jointly improve the model’s ability to reproduce seasonal patterns, capture long-term trends, and ensure temporal consistency. Evaluation against eddy-covariance flux tower measurements demonstrates that TRAX GPP achieves strong agreement with observed GPP and provides more reliable seasonal and interannual variations than standard LSTM baselines and several existing GPP products. As a result, TRAX GPP reduces known biases in global totals and trend behaviors. 2. Authors Shaoyang Chen, Xinjie Liu, Qizhi Han, Yanhong Wu, Liangyun Liu 3. NetCDF Format The dataset is provided at a 0.05° spatial resolution, corresponding to a global grid of 3600 rows × 7200 columns in latitude–longitude coordinates. The dataset covers the period from January 2001 to December 2024, with monthly temporal resolution. Each NetCDF file contains a single GPP variable representing monthly gross primary production. Missing or invalid pixels are assigned a fill value of –9999. A scale factor of 0.01 is applied to the stored integer values, and users should multiply by this factor to obtain GPP in units of g C m⁻² d⁻¹. The dataset follows a standard NetCDF structure, including spatial coordinates (lat, lon), a time dimension, and variable attributes documenting units, fill values, and scaling conventions.