Current data center (DC) cooling architectures are inefficient due to (1) inherent airflow efficiencies and (2) their inability to spatiotemporally control cooling airflow and DC temperatures on demand. Rack-based cooling is a promising recent alternative since it provides more effective airflow distribution and is more amenable to rapid real-time control. A control scheme should be able to predict spatiotemporal temperature changes as the system configuration and parameters change, but a suitable method is not yet available. Existing approaches, such as proper orthogonal decomposition or machine learning are unsuitable because they require an inordinately large number of a priori simulations or experiments to generate a training dataset. We provide an alternative real-time temperature prediction tool which requires no a priori training for DC server enclosures into which a rack mountable cooling unit (RMCU) has been integrated. This new model is validated with experimental measurements and its applicability is demonstrated by separately evaluating the influence of varying IT server configuration, RMCU flowrate, step changes in system conditions, and interactions between multiple RMCUs. The resulting tool will facilitate advanced control techniques and optimize design for any DC rack-based cooling architecture. (C) 2018 Elsevier Ltd. All rights reserved.