Net Carbon Fluxes in Peninsular Spain Forests Combining the Biome-BGC Model and Machine Learning

In the current context of global warming, quantifying carbon fluxes between biosphere and atmosphere and identifying ecosystems as carbon sources or sinks is essential. The goal of this study is to quantify net carbon fluxes for the main forest types in peninsular Spain and characterize them as carbon sources or sinks. A hybrid methodology is proposed. First, net primary production (NPP) is obtained through machine learning using site properties, time metrics of meteorological series, and forest inventory data as inputs. The most accurate NPP estimates (R2 ≥ 0.8 and relative RMSE ≤ 30%) were obtained by kernel ridge regression and gaussian process regression using latitude, elevation, time metrics of air temperature, precipitation and incoming solar radiation, and growing stock volume as inputs. Secondly, net ecosystem production (NEP) is obtained by subtracting heterotrophic respiration simulated by Biome-BGC from the previous NPP. All considered forest types presented small and mostly positive NPP and NEP values (greater for deciduous than for evergreen forests), thus generally acting as carbon sinks during the 2004–2018 period.