Data gathering algorithm based on compressive sensing(CS) has enormous application potential in wireless sensor network(WSN) in which there is limited energy and a lot of redundant data. However, most existing studies assume that network is based on ideal link. This paper illustrates a situation by experiment that existing CS reconstruction quality will be seriously affected by lossy link, and proposes a CS data gathering algorithm based on retransmission and time series correlation prediction(CS-RTSC). The type of packet loss is modeled as element random loss(ERL) and block random loss(BRL). The loss type prediction algorithm based on sliding window statistics is designed to determine the type of packet loss when link packet loss occurs. Retransmission recovery is applied for ERL, and time series correlation prediction algorithm is designed to recover the loss for BRL. The simulation result indicates that the proposed algorithm can effectively reduce the impact of lossy link in CS data gathering. When the packet loss ratio is up to 30%, the relative error of CS reconstruction signal is only 0.1% higher than that of the CS reconstruction signal in the ideal link.