Large galaxy surveys have put critical constraints on the evolution of star formation rates and the assembly of stellar mass in the Universe. However, a more complete understanding of the physical processes driving galaxy evolution requires a high angular resolution, multi-wavelength approach, alongside theoretical modelling.

In the first part of this talk, I will outline (S)HiZELS, an observational attempt to resolve star formation in galaxies at z=1.5-2.2, around the peak of cosmic star formation. I will focus on one particularly interesting, spatially extended source, which shows spatial offsets between the short and long-wavelength data. The star formation rates derived from different single-wavelength observations are very different, due to the extreme dust attenuation within this galaxy. This work illustrates the importance of combining multiple star-formation tracers, and the biases that can arise without long-wavelength data.

'observer space' can help us to understand observational data better, improve observational inference methods, and test models against data in a robust manner. Using radiative transfer, we have generated synthetic maps of multi-wavelength emission for highly-resolved zoom-in simulations of galaxies modeled with FIRE physics. I will illustrate the many exciting uses of this dataset, highlighting an investigation into the nature of highly obscured galaxies that are invisible to HST.