Species distribution shifts are one of the most widely reported consequences of climate-driven warming across marine ecosystems, creating complex ecological and social challenges. Successfully meeting these challenges and supporting forward-looking, decision-making processes requires developing tools with the capacity to accurately predict species distribution and abundance in response to changing ecosystem conditions. The objective of this dissertation was to assess the capacity of species distribution models (SDMs) to meet this information need. In Chapter 2, I developed an approach for combining quantitative SDMs with qualitative expert climate vulnerability assessments to project the distribution and relative biomass for a suite of marine species under a “worst case” climate change scenario in the Northeast U.S. Continental Shelf Large Marine Ecosystem. Projections suggested declines in commercially-important groundfish and traditional forage species in the Gulf of Maine area as coastal fish species and warmer-water forage species historically found in the southern New England/Mid-Atlantic Bight area increased. While the proposed approach tried to leverage the strengths of each method, it had no noticeable improvement in predictive skill. In Chapter 3, I designed a simulation experiment to understand SDM prediction skill variability as a function of environmental novelty. Surprisingly, prediction skill increased under novel conditions if they overlapped with optimum environmental ranges, highlighting the importance of understanding how novel conditions map to species expected responses. In Chapter 4, I worked to develop an approach that allows modelers to make predictions from a fitted spatio-temporal SDM. Spatio-temporal models offer several potential advantages over simpler, environment-only SDMs. However, their application to making predictions has been slowed by conceptual and technological challenges. This approach overcomes these challenges and readily makes predictions from a fitted spatio-temporal SDM with new covariate values and quantifies the influence of different model components on prediction uncertainty. Collectively, this dissertation adds to our understanding of prediction skill variability and contributes targeted model developments in hopes of generating SDMs with better prediction skill. Both components are critical to continuing to develop more accurate SDMs that provide actionable information to support stakeholders making climate-informed decisions, strengthening the resilience of ecosystems, resources, and resource users to global climate change.