Predicting future climate conditions is crucial for improving community resilience and engineering adaptations against climate change. Underestimated climate change can induce managers to adopt insufficient mitigation measures, while overestimated climate conditions can lead to higher costs and less efficiency in engineering adaptations. Substantial progress has been made in the scientific community to improve climate models and climate model projections (Eyring et al. 2016), although these projections were not developed specifically for any practical applications and there is a substantial barrier in translating climate model results into engineering practice. Climate model projections are affected by the deep uncertainty with different sources such as natural variability, model uncertainty, and scenario uncertainty (Hawkins and Sutton 2009). At the same time, more and more climate model projections are becoming available (Deser et al. 2020), with new, recently-developed future climate scenarios (O’Neill et al. 2017). Translating these climate model projections into actionable information for engineering is an urgent and critical task to perform. The main objective of this study is to assess climate model projection uncertainty with statistical time series analysis and Bayesian inference for engineering decision-making and infrastructure planning. This work aimed at assessing the performance of different climate models based on historical observation data, reducing model projection uncertainty by selecting best-performed models, and providing improved future climate information. The dependence of future climate projection uncertainty on the availability of observation data can be assessed, allowing the adoption of more flexible engineering planning schemes. While the overall objective is to facilitate the analyses of regional climate for forecasting the rate of occurrence with extreme events, some preliminary results with the analyses on the global average temperature series are available and are presented. The overall approach of applying the time series analysis and Bayesian inference is expected to be applicable for regional climate assessment. This work utilized a state space model (SSM) approach to model time series of temperature anomaly, integrating physics-based parameters. Using statistical time series models serves as a useful alternative for forecasting the time series of climate variables according to the previous work (Lai and Dzombak 2021). Consistent with a simplified global energy balance 11 model, the physical-parameter-based SSM provides a parametric form to evaluate the available climate model simulations with a set of parameter values for individual models, as presented in Figure 1 as an illustration. Both the number of global climate models (currently 36 models are used) and the number of simulation runs from each model are relatively large (up to more than 100 simulation runs for one future scenario), facilitating the analyses using SSM. The evaluation of the model performance and the estimation of future climate projections follow a Bayesian model averaging approach. Under the same parametric form of SSM, the estimated parameter values from individual climate models are used to calculate the marginal likelihood for these models based on historical observation data and this marginal likelihood is combined with the sets of posterior parameter values (across all models) to obtain climate projections. The SSM approach can assess the performance of climate models and provide improved climate projections for engineering applications. With the framework of the Bayesian model averaging and the integration of physical parameters, the best performed climate models or the sets of corresponding parameter values can be selected to reduce both the projection uncertainty for the variable of interest and the uncertainty with respect to the individual physical parameters. An example is provided in Figure 2 as an evaluation result for the projection uncertainty with different availability of historical data. With this application of physical-parameter-based technique and further development in modeling of regional climate and other climate variables, the SSM is expected to be a useful approach to facilitate assessing projection uncertainty and to help improve engineering decision-making by combining with other techniques such as those presented in Pozzi et al. (2017).

Track 3: Assessing Climate Change Uncertainty for
  Infrastructure Planning with Physical‐Parameter‐
  Based State‐Space Models and Bayesian
  Inference

To lean into ASCE’s sustainability efforts to mitigate environmental effects and use resources wisely, ASCE’s International Conferences on Sustainable Infrastructure (ICSI) 2021 will be held as a virtual event. The dates of ICSI 2021 will be December 6-10th.

Powered by the ICSI Steering Committee based in Houston, the 2021 event will bring together leaders to discuss leveraging sustainable infrastructure for resilient communities.  The program will offer content from across the globe, but have several highlights on sustainability initiatives in Houston, Texas. We hope you will join us virtually for plenary sessions, technical discussions, along with opportunities for Q&A and networking, connecting with virtual exhibitors, and to earn professional development hours.  

[Paper Proceedings](https://ascelibrary.org/doi/book/10.1061/9780784483879)

Conference DOI: https://doi.org/10.26226/m.63208d58cb3e8e001271fcc3

Track 3: Assessing Climate Change Uncertainty for Infrastructure Planning with Physical‐Parameter‐ Based State‐Space Models and Bayesian Inference