In the last decades the energy-balance-closure problem has been thoroughly investigated from different angles, resulting in approaches to reduce but not completely close the surface energy balance gap. Energy transport through secondary circulations has been identified as a major cause of the remaining energy imbalance, as it is not captured by eddy covariance measurements and can only be measured additionally with great effort. Several models have already been developed to close the energy balance gap that account for factors affecting the magnitude of the energy transport by secondary circulations. However, to our knowledge, there is currently no model that accounts for thermal surface heterogeneity and that can predict the transport of both sensible and latent energy. Using a machine-learning approach, we developed a new model of energy transport by secondary circulations based on a large data set of idealized large-eddy simulations covering a wide range of unstable atmospheric conditions and surface-heterogeneity scales. In this paper, we present the development of the model and show first results of the application on more realistic LES data and field measurements from the CHEESEHEAD19 project to get an impression of the performance of the model and how the application can be implemented on field measurements. A strength of the model is that it can be applied without additional measurements and, thus, can retroactively be applied to other eddy covariance measurements to model energy transport through secondary circulations. Our work provides a promising mechanistic energy balance closure approach to 30-min flux measurements.