Estimates of Annual Carbon Dioxide Fluxes from the Soil of Spruce Forests of the Ural-Carbon Carbon Supersite based on Incomplete Time Series using Classical Regression Approaches and Machine Learning

The annual flux of carbon dioxide from soils across different biomes plays a key role in global climate models and terrestrial carbon cycle analysis. However, there are significant gaps in such research at a regional scale. Due to the high labor intensity of obtaining daily soil respiration indicators, various modeling methods are used. In this work, based on 2760 measurements of soil respiration in spruce forests of the Ural-Carbon carbon supersite (Middle Urals), carried out in the fall of 2021 and from April to October 2022, using classical regression approaches and machine learning, annual soil respiration indicators were estimated. We also investigated the dependence of the results on the complexity of the model (number of predictors) and the methods used (random forest model extrapolation and combined approaches for estimating winter CO2 fluxes). The “simplified” model with seven predictors showed only a slight decrease in accuracy compared to the full model with 21 predictors (R2 = 0.89, MSE = 0.22 vs. R2 = 0.92, MSE = 0.31). Remote sensing-based predictors contributed more to model accuracy than field data. While initial results varied across methods, incorporating literature-based winter respiration values into the random forest model and averaging combined-approach estimates yielded consistent annual soil respiration values: 830.3 ± 6.4 and 851.6 ± 8.0 g C/m2 year, respectively.