Accurate estimation of regional-scale terrestrial carbon budgets is of great importance but remains challenging. With particular advantages, the Long Short-Term Memory (LSTM) networks method show potential in improving regional carbon budget upscaling estimations. Here, based on LSTM, we upscale regional net ecosystem carbon exchange (NEE) with available flux tower measurements and satellite land surface observations in North America. With well-established ecosystem-specific LSTMs, we produced monthly NEE at a spatial resolution of 0.1°×0.1° over 2001–2021 (labeled as CROSS2023). Our estimate pointed the largest carbon sink to the Midwest Corn-Belt area during peak growing seasons and to the Southeast on an annual basis, agreeing with empirical knowledges. Moreover, the estimated seasonal variations of NEE by CROSS2023 coincided well with those by atmospheric inversions, i.e., the ensemble mean of Orbiting Carbon Observatory-2 Model Intercomparison Project (OCO-2 v10 MIP; r = 0.95, p < 0.001) and CarbonTracker2022 (CT2022) (r = 0.97, p < 0.001). The mean annual NEE was estimated at -1.27 ± 0.12 Pg C yr-1, aligning more closely with the inversions (-0.70 to -0.63 Pg C yr-1) than with existing upscaling estimates (-3.30 to -1.81 Pg C yr-1). In addition, our estimate plausibly captured the NEE spatial anomalies caused by all the recent extreme drought and flood events. We further confirmed that considering memory effects was critical for better indicating interannual variability and spatial anomalies of NEE induced by climate extremes. This study provides an improved bottom-up estimation of North American NEE, largely narrowing the gap with top-down inversions.

Upscaling flux tower measurements based on machine learning (ML) algorithms is an essential approach for large-scale net ecosystem CO2 exchange (NEE) estimation, but existing ML upscaling methods face some challenges, particularly in capturing NEE interannual variations (IAVs) that may relate to lagged effects. With the capacity of characterizing temporal memory effects, the Long Short-Term Memory (LSTM) networks are expected to help solve this problem. Here we explored the potential of LSTM for predicting NEE across various ecosystems using flux tower data over 82 sites in North America. The LSTM model with differentiated plant function types (PFTs) demonstrates the capability to explain 79.19% (R2 = 0.79) of the monthly variations in NEE within the testing set, with RMSE and MAE values of 0.89 and 0.57 g C m-2 d-1 respectively (r = 0.89, p < 0.001). Moreover, the LSTM model performed robustly in predicting cross-site variability, with 67.19% of the sites that can be predicted by both LSTM models with and without distinguished PFTs showing improved predictive ability. Most importantly, the IAV of predicted NEE highly correlated with that in flux observations (r = 0.81, p < 0.001), clearly outperforming that by the random forest model (r = -0.21, p = 0.011). Among all nine PFTs, solar-induced chlorophyll fluorescence, downward shortwave radiation, and leaf area index are the most important variables for explaining NEE variations, collectively accounting for approximately 54.01% in total. This study highlights the great potential of LSTM for improving carbon flux upscaling with multi-source remote sensing data.