Why do we need two Long training sequences for Channel Estimation in 802.11a and 802.11p systems?

I am working on 802.11p and was wondering about the channel estimation part. At physical layer (PPDU), The preamble contains 10 Short Training Sequence and two Long training same sequences. Why do we need two of the long training sequence for Fine channel estimation?

Is it something like we take First Training sequence and estimate the channel taps and channel taps estimated from second training sequence and finally average them?