Abstract: This paper reviews methods that have been used to evaluate global climate simulations and to downscale global climate scenarios for the assessment of climate impacts on hydrologic systems in the Pacific Northwest, USA. The approach described has been developed to facilitate integrated assessment research in support of regional resource management. Global climate model scenarios are evaluated and selected based on historic 20 th Century simulations. A statistical downscaling method is then applied to produce a regional data set. To facilitate the use of climate projections in hydrologic assessment, additional statistical mapping may be applied to generate synthetic station time series. Finally, results are presented from a regional climate model that indicate important differences in the regional climate response from what is captured by global models and statistical downscaling. 1. Introduction Some of the most important anticipated impacts of climate change are expressed through hydrologic processes such as streamflow, snowpack, and flooding. Modeling these impacts requires high- resolution regional data for future scenarios of temperature and precipitation. The science of climate change at global and regional scales is quite advanced and climate simulations are typically downscaled to as fine as 10-50 km grids or to station locations. While there remains significant research to be done to fully understand climate dynamics at these scales and to bolster confidence in future scenarios, the current climate modeling is adequate for many applications in hydrology. A principal challenge is linking global climate simulations to existing computational tools and institutional mechanisms within an integrated assessment. For example, under global climate change, system impact assessment is complicated by the constantly shifting underlying climate trends within large year-to-year variability (Arnell, 1996). The analysis of water resource systems and their reliability, yield, and specific event frequency, generally assumes a static state that can be described statistically using a time series of historic events and depends on using the observed record of the past to estimate the probability of future events. The observed record is assumed to be statistically stationary so that all events are equally probable and these probabilities are assumed to carry into the future. Typically, climate projections are based on transient simulations from multiple projected emissions scenarios and climate models. While this approach can generate a large number of projections based on various models and emissions scenarios, it does not correspond well to the current approach in resource management. This paper reviews methods developed by the Climate Impacts Group (CIG) at the University of Washington for integrated assessment of climate change impacts in the Pacific Northwest, United States. This research focuses on four diverse yet connected natural systems of the Pacific Northwest (fresh water, forests, salmon and coasts) and the socioeconomic and/or political systems associated with each. Hydrologic processes are central to the climate impacts in all sectors; thus, downscaling climate scenarios for hydrologic simulations forms the basis for quantitative analyses. Many of the approaches we have developed are based on empirical corrections to simulated climate data. These corrections are based on a relationship between the observed statistics of a parameter and the simulation of that parameter for equivalent climate conditions. This relationship is then used to correct the simulation of that parameter for future climate conditions. In its simplest form, that relationship could be a simple perturbation to correct a bias. In the quantile mapping, however, the full probability distribution is taken into account. For example, the temperature simulated by a given model for present-day conditions at a given location may be 5°C too cold compared with observations. For the future climate, one would add 5°C to all values simulated at that location to correct this bias. The bias may be simply a lapse-rate correction for unresolved topography or it may stem from a deficiency in the