• 10-year of CC was a cut-off point in separating soil bacterial community structures. • soil pH and P were well associated with changes of diversity and community structures. • N fixation bacteria were increased with successive year, but P, K solubilizing bacteria decreased. • Monocropped alfalfa simplified the complexity of the cooccurrence networks. Soil-borne plant diseases cause major economic losses globally. This is partly because their epidemiology is difficult to predict in agricultural fields, where multiple environmental factors could determine disease outcomes. Here we used a combination of field sampling and direct experimentation to identify key abiotic and biotic soil properties that can predict the occurrence of bacterial wilt caused by pathogenic Ralstonia solanacearum. By analyzing 139 tomato rhizosphere soils samples isolated from six provinces in China, we first show a clear link between soil properties, pathogen density and plant health. Specifically, disease outcomes were positively associated with soil moisture, bacterial abundance and bacterial community composition. Based on soil properties alone, random forest machine learning algorithm could predict disease outcomes correctly in 75% of cases with soil moisture being the most significant predictor. The importance of soil moisture was validated causally in a controlled greenhouse experiment, where the highest disease incidence was observed at 60% of maximum water holding capacity. Together, our results show that local soil properties can predict disease occurrence across a wider agricultural landscape, and that management of soil moisture could potentially offer a straightforward method for reducing crop losses to R. solanacearum |