Bhim,

I'm not sure I understand your question. Are you trying to model time-to-time differences in rainfall as the variable of interest? If so, then yes, you necessarily lose the first year.

You might look into latent difference score models or autoregressive latent trajectory models for various ways to handle such data.

Pat