Estimating Carbon Sink Strength of Norway Spruce Forests Using Machine Learning

Abstract

Forests sequester atmospheric carbon dioxide (CO2) which is important for climate mitigation. Net ecosystem production (NEP) varies significantly across forests in different regions depending on the dominant tree species, stand age, and environmental factors. Therefore, it is important to evaluate forest NEP and its potential changes under climate change in different regions to inform forestry policy making. Norway spruce (Picea abies) is the most prevalent species in conifer forests throughout Europe. Here, we focused on Norway spruce forests and used eddy covariance-based observations of CO2 fluxes and other variables from eight sites to build a XGBoost machine learning model for NEP estimation. The NEP values from the study sites varied between −296 (source) and 1253 (sink) g C m−2 yr−1. Overall, among the tested variables, air temperature was the most important factor driving NEP variations, followed by global radiation and stand age, while precipitation had a very limited contribution to the model. The model was used to predict the NEP of mature Norway spruce forests in different regions within Europe. The NEP median value was 494 g C m−2 yr−1 across the study areas, with higher NEP values, up to >800 g C m−2 yr−1, in lower latitude regions. Under the “middle-of-the-road” SSP2-4.5 scenario, the NEP values tended to be greater in almost all the studied regions by 2060 with the estimated median of NEP changes in 2041–2060 to be +45 g C m−2 yr−1. Our results indicate that Norway spruce forests show high productivity in a wide area of Europe with potentially future NEP enhancement. However, due to the limitations of the data, the potential decrease in NEP induced by temperature increases beyond the photosynthesis optima and frequent ecosystem disturbances (e.g., drought, bark beetle infestation, etc.) still needs to be evaluated.