Leveraging unmanned aerial vehicle (UAV) is convenient to collect data from ground sensor. However, in the presence of unknown urban environment, the data collection is subject to the blockage of urban buildings. In this paper, considering the urban environment during flight, we propose dynamic adaptive modulation and height control for UAV-sensor data harvesting in urban areas. In each time slot, the modulation format and flight height are selected based on current system states, with the aim of minimizing the expected transmission energy of sensor under data volume and flight height constraints. The dynamic adaptive modulation and height control problem is formulated as constrained finite-horizon Markov decision processes (CMDP), which can be solved by backward induction algorithm. The advantage of proposed joint design over modulation selection only is illustrated via the computer simulations, where 48.23% expected transmission energy can be saved for ground sensor.