Testbed facilities play a major role in the study and evolution of emerging technologies, such as those related to the Internet of Things. In this work we introduce the concept of modelled testbeds, which are 3D interactive representations of physical testbeds where the addition of virtual resources mimicking the physical ones is made possible thanks to back-end infrastructure. We present the architecture of the Syndesi testbed, deployed at the premises of University of Geneva, which was used for the prototype modelled testbed. We investigate several extrap- olation techniques towards realistic value assignment for virtual sensor measurements. K-fold cross validation is performed in a dataset compris- ing of nearly 300'000 measurements of temperature, illuminance and hu- midity sensors collected from the physical sensors of the Syndesi testbed, in order to evaluate the accuracy of the methods. We obtain strong re- sults including Mean Absolute Percentage Error (MAPE) levels below 7%.