A Bayesian Forecast Framework for Climatic Parameters in Geotechnical Modeling

The Thornthwaite Moisture Index (TMI) is commonly used by geotechnical engineers to model soil moisture-climate interaction within the vadose zone. The long-term average (20+ years) TMI has correlated well with unsaturated soil moisture flow parameters but struggles to characterize sporadic and extreme climate events crucial to the performance of civil infrastructure. Short-term TMI, or precipitation data directly, better captures seasonal climate variability and extreme events, although the nonstationary nature of these parameters can hinder their usefulness in civil design. This study presents a Bayesian forecast framework for estimating the seasonal TMI using time-series analysis techniques with a component-wise transitional Markov Chain Monte Carlo (MCMC) approach. The proposed framework is presented using an example study site in San Antonio, Texas, from using 30 years of prior climate date to forecast 10 years of the seasonal TMI from June 2012 to June 2022. Validation and performance studies of the example forecast demonstrate the effectiveness of the proposed framework to capturing seasonal variability and extreme climate events.