How well can we really estimate desired speeds? - A simulation study on the modified Kaplan-Meier approach

Desired speed or free speed distributions are important input parameters for microscopic traffic flow simulations. Whereas driven speeds can be measured, desired speeds are not detectable for all vehicles due to vehicles constraining each other. An established approach to estimating desired speed distributions is the modified Kaplan-Meier approach, which estimates desired speed distributions based on single vehicle data from stationary detectors. Due to a lack of a ground truth, it is not possible to validate the estimation results based on empirical data. Therefore, we employ a microscopic traffic simulation model as a virtual testbed to generate a synthetic ground truth. Our results highlight that when applying the modified Kaplan-Meier approach, it is crucial to ensure that the underlying empirical data are adequately filtered to exclude periods where the traffic density on the left lane exceeds 10 veh/km. Otherwise, the estimation may not yield valid results.