This project was aimed at providing the U.S. Department of Defense (DoD) with a comprehensive analysis of the uncertainty associated with generating climate projections at the regional scale that can be used by stakeholders and decision makers to quantify and plan for the impacts of future climate change at specific locations. The merits and limitations of commonly used downscaling models, ranging from simple to complex, were compared, and their appropriateness for application at installation scales was evaluated. Downscaled climate projections were generated at selected DoD installations using dynamic and statistical methods with an emphasis on generating probability distributions of climate variables and their associated uncertainties. The sites selection and selection of variables and parameters for downscaling was based on a comprehensive understanding of the current and projected roles that weather and climate play in operating, maintaining, and planning DoD facilities and installations.

Using interviews with DoD installation stakeholders, researchers have identified climate variables and key vulnerabilities of interest in the initial phase of this project. Researchers generated high-resolution climate projections for the North American continent by combining state-of-the-art dynamical and statistical downscaling models with quality-controlled observations and the latest simulations from global models of the Earth’s climate system. Expected changes in climate variables were evaluated by analyzing these downscaled climate model products at the relevant spatial scales for selected DoD built installations.

Researchers conducted a survey of perceptions regarding climate change and its potential impacts on specific DoD installations. Precipitation changes (particularly at the distribution extremes) at the installations emerged as a primary concern of the participants. The survey results and the priorities identified by the impact assessment community were used to identify a set of climate variables for this downscaling study. Evaluation of different datasets for precipitation and temperature over the model domain resulted in selecting the PRISM dataset as the most accurate data for precipitation and temperature, especially in topographically complex regions. An extensive evaluation of model bias for dynamically downscaled model products during the historical period was generated with observational data over different regions of the CONUS. Researchers established that the 12-km model resolution and the model setup (parameters, nudging, and spin-up) led to a decrease in model bias as compared to coarser-resolution models, and added value as compared to a method that purely depends on spatial interpolation from a coarser grid. This was especially true when calculating the diurnal variability and extremes of temperature and precipitation.

One of the primary findings of this study was that the accuracy of both relative errors and extreme values was highly dependent on the region being analyzed and the boundary conditions used to drive the simulation. Similarly, knowing the model rank for relative errors from the climatology does not represent how that model performed in extreme climate cases. This could be a result of several different factors. First, the configurations for each climate run are different. While adjusting the initial boundary, conditions can be beneficial in many situations, bias correction and nudging were not always an improvement compared to the reference data. In addition, the results show that many variables have the largest errors for surface variables in the wettest and driest regions of the CONUS. High-precipitation regions, such as the Southeast, yielded higher errors because of the dominance of convective processes in these regions, which was challenging to predict at this resolution. Similarly, drier regions have been shown to have greater errors or biases due to small-scale processes that were hard to capture using downscaling techniques. The ensemble’s ability to capture these historical uncertainties using different reference data was important for future projections across this domain.

The most striking results from the study were that for 2085–2095, the model projections show temperature changes of 5–7°C for summer compared to 1995–2004 and a change of >7°C over northern Canada and Alaska for the winter months. In summer, the projections show a widespread summertime precipitation increase (with precipitation up to 60% higher than present- day average values) throughout much of Canada, Alaska, and the southwestern United States, while the winter experiences lower precipitation than at present over the Southwest and the Southern Great Plains, with precipitation 40–60% lower than present averages. Researchers also estimated the increase in precipitation and temperature in the projections. The model projections indicate 3–5 additional days with precipitation >20 mm/day over the eastern United States, Alaska, and Canada, and ~1 additional day over the western United States. The number of days with precipitation >40 mm/day also increased, especially over the eastern United States and the Cascade Range, with ~2 more days than the present averages. The model projection indicated >60 additional days/year with daily maximum temperature >90°F over the Great Plains and most of the eastern United States (except over the southern mountain ranges). Over the Rockies, the Cascade Range, Alaska, and Canada, <20 additional days had daily maximum temperatures >90°F. Nearly all of the CONUS and Canada are projected to experience a decrease of >20 frost days/year in 2085–2094 in the RCP8.5 scenario, especially over the West Mountain sub region, which is projected to have >60 fewer frost days/year. The projected changes in extreme temperature show significant elevation dependence, the reasons for which need further investigation.

This project generated the knowledge base, data sets, and tools needed for making preliminary assessments of vulnerabilities due to climate change (e.g., severe-event probabilities), expected changes in operating parameters such as heating and cooling needs, and other potential challenges for DoD installations and range management. Specifically, by evaluating the value added and the appropriateness of downscaled projections to the size of a DoD installations, researchers significantly advanced the appropriate use of downscaled climate projections. By generating downscaled projections and the associated uncertainty at specific locations, they provided an important resource for DoD to use in planning to adjust to changing local environments that may affect DoD facilities and installations. The methodology, observational data sets, and model products produced during the course of the project were available as an easily accessible database for future use by DoD. An illustrative example developed will be made available in a separate document for DoD use. A geographic information system (GIS) application for further exploration of the data at selected installations will also be provided. Together, these products will meet the ultimate goal of assisting DoD in developing informed policies when confronting future change.