Using Climate Model Decadal Predictions and Statistical Extrapolations of Location-Specific Near-Term Temperature and Precipitation for Infrastructure Engineering

Abstract

Location-specific, near-term (≤10 years) temperature and precipitation conditions are informative and important in many infrastructure engineering activities. Two alternative approaches for obtaining near-term future climate change conditions—decadal predictions from global climate models (GCMs) in the Decadal Climate Prediction Project and observation-based extrapolations from a statistical forecasting model (autoregressive integrated moving average, ARIMA model)—were used and assessed for engineering applications. The climate model predictions and statistical extrapolations were obtained for 20 US cities across climate regions and estimated as four annual temperature and precipitation indices (e.g., annual heating/cooling degree days and maximum 1-day precipitation) and as joint distributions of annual temperature and precipitation changes commonly used in engineering studies. Quantitative assessments suggest that the two approaches generally provide comparable results and can be more accurate than common baseline methods of using historical data and assuming climate stationarity. Year-to-year climate variability can lead to large errors for both approaches, suggesting that obtaining probabilistic predictions or extrapolations is important and informative to quantify uncertainty. Applying climate model predictions in engineering currently involves significant time and effort for tasks such as downloading and processing large GCM prediction files. To facilitate the design, construction, and operation of infrastructure projects with near-term climate change information, observation-based statistical extrapolations of location-specific climate data provide a computationally and procedurally efficient option for engineering applications. Ensuring the resilience and reliability of infrastructure amid continuously evolving climate change requires obtaining and implementing future climate information in engineering practice. Two alternative approaches of obtaining regional climate information were assessed for near-term (≤10 years) engineering applications: climate-model-based predictions and local-historical-data-based extrapolations. Climate model predictions were acquired from the recently developed prediction results based on state-of-the-art, process-based global climate models. The observation-based approach directly extends the historical trend exhibited in local temperature and precipitation data. Several temperature and precipitation variables—commonly used to inform and facilitate engineering designs and analyses—were calculated using these two approaches and assessed with respect to accuracy and accessibility. In general, the two approaches have comparable accuracy and are more accurate (especially in recent years) than using long-term average historical conditions, and both approaches provide a practical means to quantify uncertainty from climate variability, which can be important in a near-term timeframe. The use of climate-model-based predictions in engineering is currently restrained by large prediction files and multiple processing procedures involved. The observation-based approach can serve as an efficient option to obtain and apply near-term regional climate information in infrastructure engineering.