Optical Digital Twins (DT) require pre-trainedmodels for estimating the Quality of Transmission (QoT) ofcandidate lightpaths in multiband (MB) optical networks beforethey are provisioned. Once lightpaths are established, thosemodels need to be tuned to be used for failure management, e.g.,degradation detection. In this paper, we concentrate on the MBOCATA DT, which models optical signal propagation in the timedomain. We first evaluate strategies for creating end-to-endlightpath models from span DNNs and then, we implement modeltuning, as a building block of the MB OCATA DT. Simulationresults show the accuracy of pre-trained models for QoTestimation and how they are tuned with telemetry measurementsand improved compared to the initial pre-trained models.