1: \begin{abstract}

2:   We study the sensitivity of weak lensing surveys to the effects of {\it

3:     catastrophic} redshift errors --- cases where the true redshift is

4:   misestimated by a significant amount. To compute the biases in cosmological

5:   parameters, we adopt an efficient linearized analysis where the redshift

6:   errors are directly related to shifts in the weak lensing convergence power

7:   spectra. We estimate the number $\nspec$ of unbiased spectroscopic redshifts

8:   needed to determine the catastrophic error rate well enough that biases in

9:   cosmological parameters are below statistical errors of weak lensing

10:   tomography.  While the straightforward estimate of $\nspec$ is $\sim10^6$,

11:   we find that using only the photometric redshifts with $z\lesssim 2.5$ leads

12:   to a drastic reduction in $\nspec$ to $\sim 30,000$ while negligibly

13:   increasing statistical errors in dark energy parameters. Therefore, the size

14:   of spectroscopic survey needed to control catastrophic errors is similar to

15:   that previously deemed necessary to constrain the core of the $z_s-z_p$

16:   distribution. We also study the efficacy of the recent proposal to measure

17:   redshift errors by cross-correlation between the photo-z and spectroscopic

18:   samples.  We find that this method requires $\sim10\%$ {\it a priori}

19:   knowledge of the bias and stochasticity of the outlier population, and is

20:   also easily confounded by lensing magnification bias.  The cross-correlation

21:   method is therefore unlikely to supplant the need for a complete

22:   spectroscopic redshift survey of the source population.

23: \end{abstract}

24: