Gap-filling and uncertainty analysis of eight-year Eddy covariance fluxes in Dajiuhu subalpine peatland

Accurate gap-filling of ecosystem fluxes is essential for characterizing annual ecosystem-scale carbon, water and energy budgets. In this study, we focused on the Dajiuhu subalpine peatland ecosystem and use eight years of data with a half-hourly sampling frequency. By integrating the maximal information coefficient (MIC) technique with random forest (RF), support vector regression (SVR), artificial neural network (ANN) and multiple linear regression (MLR) methods, we identified the driving factors for carbon, energy and water fluxes and established optimal gap-filling models. Furthermore, we systematically evaluated the uncertainties of these flux gap-filling models from multiple dimensions, including probability density functions (PDFs), errors and confidence intervals. Through this uncertainty analysis, we identified the driving factors and optimal models for each flux. Specifically, FCO₂ is driven by photosynthetic photon flux density (PPFD), air temperature (Ta) and soil temperature (Ts3), with SVR3 co 2 demonstrating the best performance. For H and FH 2 O, the key drivers are the PPFD, relative humidity (Rh), Ta, friction velocity (u*) and soil heat flux (SHF3), with RF5 H and RF5 H 2 O performing optimally. Similarly, LE is influenced by PPFD, Rh, Ta, and u*, with RF4 LE yielding the best results. These findings provide a solid foundation for the precise gap-filling of ecosystem flux data while providing important scientific support for identifying key driving factors. • Eight years of half-hourly data enable the development of gap-filling models for four fluxes. • Maximal Information Coefficient identifies key drivers for four machine learning flux models. • Photosynthetic photon flux density, air temperature and soil temperature are the primary drivers of FCO₂. • Humidity, friction velocity and soil heat flux predominantly regulate H, LE and FH₂O. • Uncertainty analysis using 25% of data provides a robust assessment of model stability.