The standard way to model periodic function is with Fourier series (though in your case the function seems approximately periodic only).
A simple second order approximation would be
$$a+b\cos\omega t+c\sin \omega t+d\cos 2\omega t+e\sin 2\omega t$$ where $\omega=\dfrac{2\pi}{12}$ if $t$ is in months.
You can find the extrema by canceling the derivative,
$$-b\sin\omega t+c\cos\omega t-d\sin2\omega t+e\cos2\omega t=0.$$
The coefficient $a$ will be the annual average and the remaining coefficients can be found by giving the positions and amplitudes of the two maxima, and solving a linear system of equations.