Abstract <p>Traditionally, Bayesian models formulated within a probabilistic programming language (such as PyMC or Stan) incorporate an explicit likelihood function alongside prior parameter distributions. However, when dealing with complex models, specifying the likelihood function may prove unfeasible, though one may construct a simulator capable of generating data samples corresponding to each set of parameter values. Bayesian inference employing such a likelihood-free methodology is commonly referred to as simulation-based inference (SBI). This technique is realized through a sequential approximation of the posterior distribution and relies on a set of parameters governing the approximation process. The paper illustrates how SBI may be applied to the analysis of dynamic imagery from orbital multichannel detectors. This method is exemplified through the task of reconstructing thunderstorm discharge parameters from their ionospheric ‘‘fingerprints’’, termed ELVES.</p>