Elucidation of soil CO/CH budget by integrated analysis of soil observation data

Yamanuki, Hina*; Ichii, Kazuhito*; Yamamoto, Yuhei*; Teramoto, Munemasa*; Sun, L.*; Koarashi, Jun   ; Atarashi-Andoh, Mariko  ; Nagano, Hirohiko*; Hirano, Takashi*; Takagi, Kentaro*; Ishida, Sachinobu*; Takagi, Masahiro*; Kondo, Toshiaki*; Takahashi, Yoshiyuki*; Liang, N.*

Soils exchange greenhouse gases (GHGs) such as carbon dioxide (CO) and methane (CH) with the atmosphere. In general, forest soils act as a source of CO emissions through soil respiration and as a sink for CH through methane oxidation. Recently, spatiotemporal variations in soil respiration and methane absorption have been estimated using field observation data, but the causes of these variations remain uncertain. In this study, we use data from the largest observation network of soil respiration and methane absorption in Asia and satellite observation data, and apply machine learning methods to estimate observed soil respiration and methane absorption. Using machine learning methods, we analyzed the influence of soil property data in explaining the observed spatiotemporal variation in soil respiration and methane absorption. Even when the feature importance of the soil properties was relatively small, it was found to have a significant impact on the accuracy of the estimation.