Many distant objects can only be detected, or become more scientifically valuable, if they have been highly magnified by strong gravitational lensing. We use EAGLE and BAHAMAS, two recent cosmological hydrodynamical simulations, to predict the probability distribution for both the lens mass and lens redshift when point sources are highly magnified by gravitational lensing. For sources at a redshift of two, we find the distribution of lens redshifts to be broad, peaking at z=0.6. The contribution of different lens masses is also fairly broad, with most high-magnification lensing due to lenses with halo masses between 10^12 and 10^14 solar masses. Lower mass haloes are inefficient lenses, while more massive haloes are rare. We find that a simple model in which all haloes have singular isothermal sphere density profiles can approximately reproduce the simulation predictions, although such a model over-predicts the importance of haloes with mass <10^12 solar masses for lensing. We also calculate the probability that point sources at different redshifts are strongly lensed. At low redshift, high magnifications are extremely unlikely. Each z=0.5 source produces, on average, 5x10^-7 images with magnification greater than ten; for z =2 this increases to about 2x10^-5. Our results imply that searches for strongly lensed optical transients, including the optical counterparts to strongly lensed gravitational waves, can be optimized by monitoring massive galaxies, groups and clusters rather than concentrating on an individual population of lenses.

Motivated by discovering strongly-lensed supernovae, gravitational waves, and kilonovae in the 2020s, we investigate whether to build a watch-list of clusters based on observed cluster properties (i.e. lens-plane selection) or on the detectability of strongly-lensed background galaxies (i.e. source-plane selection). First, we estimate the fraction of high-redshift transient progenitors that reside in galaxies that are themselves too faint to be detected as being strongly-lensed. We find $\sim15-50$ per cent of transient progenitors reside in $z = 1-2$ galaxies too faint to be detected in surveys that reach ${\rm AB}\simeq23$, such as the Dark Energy Survey. This falls to $\ls10$ per cent at depths that will be probed by early data releases of LSST (${\rm AB}\simeq25$). Second, we estimate a conservative lower limit on the fraction of strong lensing clusters that will be missed by magnitude limited searches for multiply-imaged galaxies and giant arcs due to the faintness of such images. We find that DES-like surveys will miss $\sim75$ per cent of $10^{15}$M$_\odot$ strong lensing clusters, rising to $\sim100$ per cent of $10^{14}$M$_\odot$ clusters. Deeper surveys, such as LSST, will miss $\sim40$ per cent at $10^{15}$M$_\odot$, and $\sim95$ per cent at $10^{14}$M$_\odot$. Our results motivate building a cluster watch-list for strongly-lensed transients that includes those found by lens-plane selection.