Modeling the long-range propagation of electromagnetic waves in the troposphere is a topic of interest for many applications in surveillance, communication, and remote-sensing. The split-step Fourier method is widely used in this context, allowing wide steps along the propagation direction. Recently, a wavelet-based method with lower complexity and efficient in terms of memory storage has been proposed. Both methods are one way and compute the forward propagation in the troposphere. Based on the Fourier two-way method, which computes the backward propagation when reaching an obstacle, we propose here a two-way split-step wavelet method. Besides, we show that the stopping criterion is intrinsic using compression in the wavelet domain. Numerical experiments have been performed (in S band) to validate the method and highlight its advantages.