Elevated roads are built worldwide to alleviate urban traffic congestions. An elevated road may overlap with the ground surface road. This poses a great challenge to auto navigation systems. It is particularly difficult for auto navigation systems to determine if an vehicle is running on an elevated road because altitude readings from GPS (Global Positioning Systems) receivers are of low accuracy. Without correct positioning of elevated or surface roads, an auto navigation system may recommend wrong routes which may lead to serious traffic accidents or cause drivers' anxiety. In this paper, we design a smartphone-based system called DEEL for detecting road elevation as the vehicle moves on urban roads. HMM-based models are built to characterize the patterns of vehicle dynamics with respect to the elevation of the road that the vehicle is moving on. The system leverages smartphone sensors to sense the instantaneous vehicle dynamics, including the 3D acceleration, the speed, and the number of perceived GPS satellites. Based on the sensed dynamics, it determines whether the vehicle is running on an elevated road or not in an online fashion. We have successfully implemented our system in Android-based smartphones. Extensive experiments demonstrate the system achieves a high accuracy of 92.1% with strong robustness to variations of vehicles, drivers and roads.