This paper proposes an integrated image nowcasting model, DeepRainX, based on Deep Learning (DL), physical models, and multiple Optical Flow (OF) models using radar image sequences. The spatio-temporal DL model is employed to predict images within short future timeframes. DL is robust against radar clutter noise and no OF estimation at the initial stage. However, DL predicts deteriorated image sequences 1 hr ahead of time. Therefore, multiple OFs are used to estimate motions from two DL output images 1 hr ahead of time. The estimated motions and the final output from DL serve as inputs to the advection equation and the Navier-Stokes equation to predict longer future image sequences after that time. Using heavy rainfall events, i.e., typhoons, the proposed DeepRainX outperforms the two worlds leading nowcast methods, i.e., Rainymotion and DL-based DGMR, in terms of accuracy in estimating precipitation amounts.