fiting a model to data with two peaks (bimodal?)

I have plant growth data dependent on temperature on a yearly basis for 6 years. The data exhibit two 'peaks' in growth during the season, with a dip in between. Individually, the yearly data does not appear to be normal using a probability plot. But when the data is collapsed into one column, it conforms to a normal distribution.
I want to be able to model the general/idealized plant growth in response to temperature . I tried fitting a polynomial (3rd and 4th degree) to the collapsed data, but it is not picking up the 'dip' in the middle of the data. I think this is due to the fact that the magnitude of growth is different each year, but the shape remains the same (although slightly offset). This is resulting in a much nosier dataset than one or two of the years together.
Does anyone know of a model that could be fit that will pick up the dip in between the higher data values on either side? Could I perhaps weight the values in the dip differently to the other values or something like that?