What properties of the refinement do the different lines on the graph describe?

White This curve gives a measure of the degree of completeness of the partial structure model as a function of resolution: correlation coefficients close to unity indicate that the model for the partial structure accounts for most of the observed amplitudes. This curve has troughs where the missing structure and/or the bulk solvent have strong Fourier components
Yellow Correlation coefficients close to unity here indicate that the fragment, prior and bulk solvent models can account for most of the observed amplitudes at that resolution. They can be used to monitor the improvement during the course of the refinement, or at successive stages of model building: you should see them approach unity as the model improves
Blue After cycle 1 of a refinement the Fc,Fxpct curve should be on the same scale as Fo,Fcalc: this means that the variance estimate are adequate. If they are lower or higher, you might want to change the parameters that affect the variances. This is done by selecting the generalised MaxLik scaling via the Refine scaling parameters buttons in the input form
Green The curve can be useful to confirm the maximum resolution up to which the signal is above the noise, or spot noisier resolution ranges, such as those where ice-ring scattering might peak