Total: 1
Kilonovae, possible electromagnetic counterparts to neutron star mergers, provide important information about high-energy transient phenomena and, in principle, also allow us to obtain information about the source properties responsible for powering the kilonova. Unfortunately, numerous uncertainties exist in kilonova modeling that, at the current stage, hinder accurate predictions. Hence, one has to account for possible systematic modeling uncertainties when interpreting the observed transients. In this work, we provide a data-driven approach to account for time-dependent and frequency-dependent uncertainties in kilonova models. Through a suite of tests, we find that the most reliable recovery of the source parameters and description of the observational data can be obtained through a combination of kilonova models with time- and frequency-dependent systematic uncertainties. We apply our new method to analyze AT2017gfo. While recovering a total ejecta mass consistent with previous studies, our approach gives insights into the temporal and spectral evolution of the systematic uncertainties of this kilonova. We consistently find a systematic error below $1$ mag between $1$ to $5$ days after the merger. Our work addresses the need for early follow-up of kilonovae at earlier times, and improved modeling of the kilonova at later times, to reduce the uncertainties outside of this time window.