Air-cooling system in power plant is of typical multi-scale system, of which the characteristic length covers from millimeter to hundreds of meters. In order to predict the thermo-flow performance of 2 x 300 MW air-cooled power generating units, a cross scale modeling methodology was proposed in the present study. Two neighbor characteristic lengths were considered, including that of the air cooling island region containing air-cooled condensers (ACCs) with 10(1) m length scale, and the large-scale one consisted of power plant buildings and the geomorphology with 10(2) m length scale. The small-scale region was described by reduced order model based proper orthogonal decomposition (POD), and the large-scale one by Navier-Stokes equations. The interfaces between the two regions were coupled by the measure that the lumped parameters on the interfaces obtained from information in small-scale region were transferred to the large-scale region as boundary condition. The air side flow velocity, turbulent kinetic energy, turbulent dissipation rate and temperature of air cooling island were solved by POD method under various natural winds. The results indicated that the errors of the thermo-flow characteristics were less than 7.75% relative to that of the multi-gird CFD results for the overall calculating region, implying that the parameters could be well coupled in the overlap regions between small and large scales by the present approach. The inverse flow phenomena of the air cooling island caused by environmental natural wind, as well as the heat transfer deterioration of downwind ACC cells were revealed. The calculating time and resources were significantly saved through the present cross scale modeling methodology with acceptable accuracy. (C) 2014 Elsevier Ltd. All rights reserved.