Soil microbial communities are integral to nutrient cycling and biological functioning in agroecosystems. The objective of this study was to examine how different NPK fertilization regimes, season and sampling time influence selected soil biological indicators in winter wheat and to evaluate the use of predictive and ML approaches for interpreting these responses. A two-year field experiment was conducted at a single site using 20 NPK treatments encompassing nitrogen rates of 0–150 kg ha −1 and phosphorus and potassium rates of 0–130 kg ha −1 . Soil samples were collected at three growth stages per season, and culturable microbial groups (ammonifiers, Azotobacter spp., oligonitrophils, actinomycetes, fungi, and total microorganisms) together with dehydrogenase activity were quantified. Data were analyzed using three-way ANOVA and principal component analysis, complemented by interpretable machine-learning models. Balanced NPK fertilization, particularly nitrogen applied at 130 kg ha −1 combined with moderate phosphorus and potassium inputs, was associated with higher abundances of ammonifiers and increased dehydrogenase activity across seasons. Treatments with balanced or elevated NPK inputs also showed higher values of Azotobacter spp., oligonitrophils, and total microorganisms, whereas excessive or unbalanced fertilization was less consistently associated with positive microbial responses. Machine-learning analysis was used to identify nonlinear associations and to rank the relative importance of fertilization regime, sampling time, and season variability, without implying causal relationships. Taken together, the results provide insight into soil biological responses to NPK fertilization in winter wheat and suggest how combining conventional field measurements with interpretable approaches can support their interpretation under the studied conditions. Combining predictive strength with interpretability, Random Forest models provided consistent predictive performance and interpretable insights into the relative importance and associations of fertilization regime, sampling time, and interannual variability with microbial parameters.
Pešić et al. (Mon,) studied this question.