Determining the controlling factors for carbon sequestration in two contrasting forests in the Boreal region and the semi-arid Mediterranean

Evergreen-needle forests are among the most adaptive ecosystems, spanning from the cold-wet Boreal to the hot-dry Mediterranean, and can provide insights into differential responses in productivity and carbon storage potential across a geographic range. Using 20 years of flux-tower data from contrasting Boreal (Hyytiälä, Finland; HYY) and semi-arid (Yatir, Israel; YAT) conifer forests, NEE sensitivity to key environmental and climate drivers was examined. We analyzed both the seasonal and the variability-driven changes in NEE with Machine Learning modeling (Random Forest; RF) and SHAP analysis and compared the results against baseline GLM and GAM outputs.  All models explained the seasonality in NEE well (RMSE<0.17, R²>0.95). However, the RF model had the advantage of capturing complex feature interactions on variability-driven NEE, with the simplicity in interpretability of the GLM (R2 values of 0.59-0.67 for RF, 0.63-0.67 for GAM, and 0.34-0.55 for GLM; with similar results in RMSE). Both forests share the sensitivity of the variability-driven changes in NEE to short- and long-wave radiation and precipitation (57%-82% of mean SHAP), but are predominantly limited by radiation duration (HYY) or intensity (YAT) in the productive season. Seasonal variations in NEE were uniquely dominated by soil water content (SWC) at the 45 cm layer in YAT (55% of meanSHAP) and by VPD in HYY (69% of meanSHAP). Based on these controlling factors, we demonstrate that observed trends in rain events that recharge deep soil layers in YAT lead to a reduction in carbon sequestration potential of 5.5 g-C/m2/year (3% of the annual mean). In contrast, no discernible trends in VPD, rainfall events, nor radiation in the productive season in HYY indicated any such changes in sequestration potential during this period. Yet, the compounding effects of a hot-dry month in tandem with a wet and warm month could reduce mean sequestration by ~70% (194 g-C/m2) in HYY, as demonstrated in summer 2020. The results indicate that across large climatic gradients, conifer forests show a shift in the predominant factor influencing NEE in the productive season between soil moisture and atmospheric moisture on the seasonal time scale, yet the variability response is consistently controlled by radiation-limiting factors.