Climate change is driving rapid changes in environmental conditions and affecting population and species' persistence across spatial and temporal scales. Integrating climate change assessments into biological resource management, such as conserving endangered species, is a substantial challenge, partly due to a mismatch between global climate forecasts and local or regional conservation planning. Here, we demonstrate how outputs of global climate change models can be downscaled to the watershed scale, and then coupled with ecophysiological metrics to assess climate change effects on organisms of conservation concern. We employed models to estimate future water temperatures (2010-2099) under several climate change scenarios within the large heterogeneous San Francisco Estuary. We then assessed the warming effects on the endangered, endemic Delta Smelt, Hypomesus transpacificus, by integrating localized projected water temperatures with thermal sensitivity metrics (tolerance, spawning and maturation windows, and sublethal stress thresholds) across life stages. Lethal temperatures occurred under several scenarios, but sublethal effects resulting from chronic stressful temperatures were more common across the estuary (median >60 days above threshold for >50% locations by the end of the century). Behavioral avoidance of such stressful temperatures would make a large portion of the potential range of Delta Smelt unavailable during the summer and fall. Since Delta Smelt are not likely to migrate to other estuaries, these changes are likely to result in substantial habitat compression. Additionally, the Delta Smelt maturation window was shortened by 18-85 days, revealing cumulative effects of stressful summer and fall temperatures with early initiation of spring spawning that may negatively impact fitness. Our findings highlight the value of integrating sublethal thresholds, life history, and in situ thermal heterogeneity into global change impact assessments. As downscaled climate models are becoming widely available, we conclude that similar assessments at management-relevant scales will improve the scientific basis for resource management decisions.