Upscaling of soil methane fluxes from topographic attributes derived from a digital elevation model in a cold temperate mountain forest

Abstract. Forest soils are generally considered a sink for atmospheric methane (CH4), but their uptake rate can vary considerably in space and time. This study aimed to investigate the temporal patterns of spatially distributed soil CH4 fluxes in a topographically complex cold-temperate mountain forest in central Japan. Soil CH4 fluxes were measured nine times during the snow-free season at multiple locations within a 40-ha area in a forested watershed. A machine-learning approach was developed to upscale measured upland fluxes to the landscape scale, using topographic attributes derived from a digital elevation model and vegetation types. Upland soils were a sink of CH4, while small wetland patches emitted CH4 consistently throughout the study period. The accuracy of predicted upland fluxes varied seasonally, with the highest model performance observed in early autumn (R² = 0.67) and the lowest in mid-summer (R² = 0.28). Within the study landscape, predicted upland CH4 fluxes varied significantly across topographic positions, with greater uptake on ridges and slopes than on the plain and foot slopes. Predicted upland CH4 fluxes ranged from −0.35 to −0.60 g CH4 ha−1 h−1 in spring, −0.41 to −1.25 g CH4 ha−1 h−1 in summer, and −0.50 to −0.89 g CH4 ha−1 h−1 in autumn. Seasonal upland fluxes were highly correlated with the 20-day antecedent precipitation index (R² = 0.71), revealing the importance of seasonal moisture conditions in regulating CH4 flux dynamics. This study highlighted the importance of topography in controlling the soil CH4 fluxes and the efficiency of remote sensing and machine learning approaches in scaling field measurements to the landscape level, enabling visualization of spatial patterns of fluxes across the landscape over time.