Scaling soil methane fluxes across a topographically complex landscape in a cold temperate mountain forest

Forest soils play a critical role in the global methane (CH₄) budget, but the magnitude of CH₄ fluxes varies significantly across a landscape, spatially and temporally. In complex landscapes, soil hydrology is strongly influenced by variations in topography and vegetation, which affect soil CH₄ fluxes (F_CH4). Consequently, accurately scaling F_CH4 to the landscape level is a significant challenge. This study aimed to develop a methodology for scaling seasonal F_CH4 across a topographically complex landscape in a cold temperate mountain forest.

This study was conducted in the upper watershed of the Yura River in the Ashiu Experimental Forest (area 40 ha and elevation 600-850 m). The landscape was classified into upland, wetland, and river, comprising approximately 94%, 1%, and 5% of the total study area, respectively. 52 collars were installed in upland areas covering different topographic positions and vegetation types, and F_CH4 were measured nine times from April to November. Additionally, 11 collars were installed in small riparian wetlands and measured twice during the wet-to-dry summer transition. Then, we used measured F_CH4 together with topographic attributes i.e., slope, aspect, profile curvature (PRC), vertical distance to the channel network (VDCN), topographic position index (TPI), and topographic wetness index (TWI) from remotely sensed data (digital elevation model) and vegetation type (broadleaf, coniferous, and mixed) to develop a machine-learning model (quantile regression forest) for predicting upland seasonal F_CH4 at 5 m resolution with uncertainty across the landscape level. A simple average was used to estimate the wetland fluxes.

Seven predictor variables were used to model upland F_CH4 for each season; the selected predictors and model accuracy varied with seasons. The model accuracy was high in early autumn (R² = 0.67) and low in early wet summer (R² = 0.28). TPI was consistently selected in all seasons, while TWI was chosen for most seasons except two, where VDCN was selected instead. VDCN and PRC were occasionally selected with TWI and TPI. Vegetation type was not selected for any of the seasons. Across the landscape, predicted upland median seasonal F_CH4 ranged from -0.35 to -0.60 g CH₄ hr^-1 ha^-1 in spring, -0.41 to -1.25 g CH₄ hr^-1 ha^-1 in summer, and -0.50 to -0.89 g CH₄ hr^-1 ha^-1 in autumn. This seasonal variation in upland predicted median F_CH4 was well explained by the antecedent precipitation index (R² = 0.71, p < 0.01) calculated over 20 days. When scaled at the landscape level, the average CH₄ uptake by upland soils was -25.1 (uncertainty -35.8 to -16.2) g CH₄ hr^-1. In the wet summer, small wetland patches offset 8% of the upland CH₄ uptake (-15.6 g CH₄ hr^-1 upland, 1.2 g CH₄ hr^-1 wetland), and the following dry summer, they offset only 2% because both the upland CH₄ uptake increased and the wetland emission decreased (-32.6 g CH₄ hr⁻¹ upland, 0.5 g CH₄ hr⁻¹ wetland). This study highlighted the efficiency of remote sensing and machine learning approaches to extrapolate field measurements to the landscape level and allowed us to visualize spatial patterns of fluxes over time.