A large literature exists using optical emission lines to study warm ionized gas at T ~ 104 K in superwinds (see Heckman et al. 1993 and references therein). Given this, I will only briefly mention some of the important wind diagnostics provided by these studies, before discussing what I believe to be an important observation affecting numerical estimates of mass loss from superwinds.
Balmer lines, primarily H emission, provide kinematic information along with morphological information regarding the structure of the outflow. Spatially resolved kinematic studies (McKeith et al. 1995; Shopbell & Bland-Hawthorn 1998; Cecil et al. 2001) provide information on the entrainment and acceleration of cool gas. The [SII] doublet ( 6717, 6731) can be used as a density diagnostic. This has been used to derive densities and pressures in the warm clouds in superwinds (McCarthy et al. 1987; Armus et al. 1990). Line ratios also provide ionization source diagnostics. The gas near the starburst region is primarily photoionized by the intense UV radiation from the massive stars, but at larger distances the H emitting gas shows line ratios indicative of shock heating (Martin 1997). In the future we may hope to apply the detailed shock diagnostics used in studies of local SNRs to superwinds.
One particularly noteworthy recent development is high resolution studies with HST of the optical emission line filaments in NGC 3079 's superwind (Cecil et al. 2001) and M82's superwind (Shopbell et al, in preparation). At high resolution the filaments and arcs of warm gas break up into very small clumps or clouds. The largest clumps or clouds in NGC 3079 are ~ 30 pc in diameter, although many are unresolved even by HST. Most of the mass in a superwind is in these cool, very compact clouds. Accurately treating the entrainment and acceleration of such clouds by the wind requires that the model resolves the wind/cloud interaction. Klein, McKee, & Colella (1994) argue that this requires at least 100 cells across across a cloud diameter, implying cell sizes of << 1 pc in simulations that must cover 2-or-3-dimensional volumes 10's of kpc on a side (preferably ~ 200 kpc on a side, see Aguirre and Pettini in this volume). No current simulations achieve this level of resolution, and therefore these models may significantly underestimate mass loss in superwinds.