In this paper, an end-to-end rate-distortion (R-D) estimation recursive model is proposed, on the basis of the statistical analysis of the spatial and temporal prediction as well as a general error concealment method by the video decoder. It takes the source quantization error, channel error, and the subsequent error spatial-temporal propagation into account. Then a corresponding integrated error resilience scheme, which adopts a global optimal intra-macroblock (MB) refreshment ratio and visual acuity selection procedure, is introduced. Moreover, it employs a combined source-channel bit allocation strategy, which could estimate an instantaneous available source transmission rate under a given time-varying network condition. The network experiments show that this scheme is effective and applicable for a better and more consistent picture quality in the receiver end.