We report the 6.5$\sigma$ detection of water from the hot Jupiter HD187123b with a Keplerian orbital velocity $K_p$ of 53 $\pm$ 13 km/s. This high confidence detection is made using a multi-epoch, high resolution, cross correlation technique, and corresponds to a planetary mass of 1.4$^{+0.5}_{-0.3}$ $M_J$ and an orbital inclination of 21 $\pm$ 5$^{\circ}$. The technique works by treating the planet/star system as a spectroscopic binary and obtaining high signal-to-noise, high resolution observations at multiple points across the planet's orbit to constrain the system's binary dynamical motion. All together, seven epochs of Keck/NIRSPEC $L$-band observations were obtained, with five before the instrument upgrade and two after. Using high resolution SCARLET planetary and PHOENIX stellar spectral models, along with a line-by-line telluric absorption model, we were able to drastically increase the confidence of the detection by running simulations that could reproduce, and thus remove, the non-random structured noise in the final likelihood space well. The ability to predict multi-epoch results will be extremely useful for furthering the technique. Here, we use these simulations to compare three different approaches to combining the cross correlations of high resolution spectra and find that the Zucker 2003 log(L) approach is least affected by unwanted planet/star correlation for our HD187123 data set. Furthermore, we find that the same total S/N spread across an orbit in many, lower S/N epochs rather than fewer, higher S/N epochs could provide a more efficient detection. This work provides a necessary validation of multi-epoch simulations which can be used to guide future observations and will be key to studying the atmospheres of further separated, non-transiting exoplanets.