Reanalysis of cross-regional data on soil chemical properties affecting soybean yield across multiple prefectures in Japan

Associations between soybean yield and soil chemical properties in Japan have largely been examined using regional-scale surveys. However, none of the previous studies in Japan modeled the strong regional heterogeneity inherent in these datasets. This study aims to reanalyze the cross-regional dataset of Takamoto, Takahashi, and Nira (2020) using a linear mixed-effects model (LMM) to identify soil chemical properties associated with soybean yield after accounting for hierarchical structure (random intercepts for prefecture and site nested within prefecture). The dataset comprised 214 sites across 16 prefectures in Japan and included soybean yield and six soil variables: total N, available P, soil pH, and exchangeable K, Ca, and Mg. Fixed effects initially included all variables and all their first-order interactions, and an exhaustive all-subsets search then selected the final fixed effect set. Random effects explained approximately 60% of the variance in yield, whereas the selected fixed effects explained less than 5%. The selected LMM retained only two fixed effects: exchangeable Ca (β = +0.24, p < 0.001) and exchangeable Mg (β = −0.22, p < 0.001). These coefficients were significantly larger in the high-yield group than in the low-yield group (groups defined by prefectural mean yields). Although soil chemical properties contribute little to soybean yield at a national scale, exchangeable Ca and Mg represent broadly applicable factors for yield improvement and may help to overcome the long-term stagnation in soybean yields in Japan.