Superpressure balloons, which drift approximately on isopycnal surfaces, get displaced by gravity waves and are thus capable of detecting gravity wave signatures. The project Loon provides superpressure balloon data in the upper troposphere and lower stratosphere from 2011 to 2021. We compare Loon data from the 6 years of best data coverage with output of global storm‐resolving models from the DYnamics of the Atmospheric general circulation Modeled On Non‐hydrostatic Domains winter initiative in the tropics. We study the variance of the vertical velocity and, for the models, the gravity wave momentum flux as function of distance to closest convection. The models show large differences in the variance of the vertical wind velocity, which is crucial for calculating vertical gravity wave momentum fluxes. We find large differences between the models with respect to simulated convection, lateral propagation, and the wave background away from sources. We then sample balloons as models by optimizing the match of vertical wind distributions using a temporal low pass filter. The average distance the balloons travel during the optimum low pass filtering time turns out to correspond approximately to four times the model grid spacing. The functional dependence of the vertical velocity variance on distance to closest convection is similar between the models and the observations sampled as models. The robustness of this result across all models suggests that storm‐resolving models provide a useful resource for machine learning some characteristics of convectively generated gravity waves.