Real-time traffic forecast by means of macroscopic models for Dynamic Traffic Assignment: theory and practice Guido Gentile Sapienza University of Rome The availability of big data on traffic in real-time (speeds, flows) allows in most cases to produce a reliable picture of the current conditions on the road network. The off-line analysis of these data improves the calibration of the supply model (free flow speeds, capacities). Finally, the historical trajectories of vehicles and people trips permit to observe demand and route choice by sampling the O-Ds and measuring the splitting rates to specific destinations. Thus the role of dynamic assignment models in traffic forecast is that of integrating the different sources of information providing the sensitivity to parameters changes (demand and supply) in a sort of pivoting approach, rather than that of simulating an abstract reality. The main model components for real-time applications become then rerouting and dynamic network loading, with less role for equilibrium, which is instead important to asses the base case of each day-type. Critical features of the assignment model are robustness and efficiency. Both are provided by macroscopic models, which are more precise and faster then microscopic simulation. The General Link Transmission Model, which is an implementation of the kinematic wave theory on cumulative flows, can support short term traffic prediction on large metropolitan networks in less then a minute, allowing to represent most relevant congestion phenomena in urban contexts (horizontal queues, spillback, node priorities, and turn restrictions). Its implementation in a web-based GIS solution for DSS, called Optima, has been installed in several traffic control rooms in different continents. We will present both the theoretical backgrounds of this solution and show its applications on real cases.