Monitoring soil quality provides indispensable inputs for effective policy advice, but very few poorer countries can implement high-quality surveys on soil. We offer an alternative, low-cost imputation-based approach to generating various soil quality indicators. The estimation results validate well against objective measures based on benchmark surveys for Ethiopia and Uganda both for the mean values and the entire distributions of these indicators based on multiple imputation (MI) methods. Machine learning methods also perform well but mostly for the mean values. Furthermore, our imputation models can be combined with other publicly available, large-scale datasets on soil quality generated by model-based analysis with earth observations to provide improved estimates. Our results offer relevant inputs for future data collection efforts.