We have developed a mathematical model to evaluate the flow and erosional potential of submarine, channelized komatiite lavas at Kambalda, Western Australia. Field data from Kambalda were used to constrain the choice of important input parameters, and model results were compared with data from field studies and geochemical analyses. Our results suggest that thermal erosion is strongly dependent upon the nature and behavior of the substrate. If the substrate is treated as an unconsolidated, hydrous sediment that can be fluidized by vaporized seawater, then our model predicts that an initially 10-m-thick basal Kambalda komatiite lava could have produced very high thermo-mechanical erosion rates (~23--10 m/day), crustal thicknesses of ~5--20 cm at distances of ~5--35 km from the source, and a high degree of lava contamination (~3--12%). In contrast, if the substrate is treated as a more consolidated, anhydrous sediment that could not be fluidized, then our model predicts that a Kambalda komatiite flow would have had much lower thermal erosion rates (~1.2--0.4 m/day) and degrees of contamination (~1--3%), and would have had crustal thicknesses of ~5--20 cm at longer flow distances of ~30--165 km. Field constraints are generally consistent with our predictions for a non-fluidized substrate. The reentrant embayments at Kambalda are thought to form from either erosion of deep channels in a flat basaltic seafloor <Huppert et al., 1984>, or erosion of a thin (<5 m) sediment with minor undercutting of basalt in pre-existing topography <Lesher et al., 1984>. Our modeling indicates that the former was possible for long eruption durations (months), whereas the latter was possible for short eruption durations (<2 weeks). As the latter hypothesis is more consistent with the existing field evidence for thermal erosion at Kambalda, we believe it is the preferred interpretation. ¿ 1998 American Geophysical Union