Upcoming large scale photometric surveys will require accurate photometric redshifts (photo-zs) to optimally extract astrophysical information. This thesis investigates methods of making and improving cosmological measurements using photo-zs. Gaussian Processes for Photometric Redshift Estimation (GPz) has been proven to provide accurate estimates and reliable uncertainty estimates. In Chapter 2 I evaluate the effects of adding near-IR magnitudes and angular size as features for training and find improvements in accuracy of ~15-20 per cent. A method of shifting the photo-zs based on Quantile-Quantile plots is also investigated and improves the bias by ~ 40 per cent. In Chapter 3 I use GPz based photo-zs to measure the baryon acoustic oscillations with SDSS galaxies. Two galaxy samples are used: one with all galaxy types and another with luminous galaxies. Using multi-wavelength photometry and morphological data, the errors on the photo-z estimates for the luminous galaxies showed an improvement on previous studies. Angular correlation functions are measured and BAO peaks are detected for the luminous galaxies at positions consistent with expected results. The whole galaxy sample, with a magnitude cut of i < 21 results in BAO peaks for the bins 0.5 < z < 0.6 and 0.6 < z < 0.7 at 4.09±0.16° and 3.44±0.13°. This result implies that the photo-zs are sufficiently accurate to make BAO measurements without being restricted to luminous galaxies. In Chapter 4 I forecast the constraints on the local primordial non-Gaussianity parameter fNL that can be obtained with future radio surveys and cross-matched photo-zs. I utilize the multi-tracer method, and improvements on previous work include the use of observational bias and halo mass estimates, updated simulations and realistic photo-z expectations. In the most realistic case, with photo-zs up to z = 2 the 1-s error falls between 4.07 and 6.58, rivalling the tightest constraints currently available. If photo-zs are available to z = 5 this improves to between 1.5 and 2.