Publications

Journal

2026

1. Value of probabilistic spatiotemporal urban temperature forecasts for building energy purchases in day-ahead market

   Byeongseong Choi, Mario Bergés, and Matteo Pozzi

   Energy and Buildings, 2026

   Abs DOI

   Urban heat significantly affects human comfort and increases energy consumption for space cooling in built environments. Adaptive decision-making regarding space cooling can reduce the economic cost of a building. In particular, accurate weather forecasts enable operators to predict building energy consumption more precisely and make informed decisions. In this paper, we assess the impact of urban weather forecasts on an energy-related decision in a building using Value of Information (VoI) analysis, a framework to rigorously quantify how reducing uncertainty propagates to informed decisions. To achieve this, we integrate two models: (a) a probabilistic spatiotemporal model for temperature forecasting and (b) a probabilistic model for forecasting the power consumption of a building. By coupling these two models, we quantify the propagating uncertainty from temperature forecasts to load forecasts and assess its influence on adaptive energy purchasing strategy in an energy market, specifically in the day-ahead market, which sets electricity prices based on purchasing timing. Our investigation shows that improved temperature forecasts can reduce energy costs by up to 38.9 % in commercial buildings and up to 8.2 % in residential buildings, compared to a scenario that does not incorporate temperature forecasts, while the degree of savings varies significantly depending on the building types. The main contribution of this study is to propose a VoI-based framework that quantifies the benefit of reducing uncertainty in building-environment interactive systems, demonstrated through the practical application of uncertainty-aware urban temperature forecasts for optimizing energy decisions in buildings.

2. Optimal sensor placement in low-cost PM2.5 sensor networks using value of information and multimodal data fusion

   Byeongseong Choi and Michelle A Hummel

   Environmental Research Letters, Jan 2026

   Abs DOI

   Monitoring of air pollutants such as fine particulate matter (PM2.5) supports improved understanding of trends, assessment of exposure, and development of strategies to protect public health. However, determining the optimal placement of pollutant sensors is an ongoing challenge, as different sensor configurations may yield varied insights into regional air quality. In this study, we propose a quantitative framework that integrates a stochastic advection-diffusion (SAD) model with the value of information (VoI) to design a sensor placement strategy for regional PM2.5 monitoring. Predictions from the SAD model are used to compute the VoI based on a cost function that penalizes misclassifications between model-assigned and observed air quality index values. We apply the framework to the Coastal Bend Region of Texas, where increased development is creating new air quality concerns. We compare three weighting strategies: (1) combined weighting, which accounts for both population density and social vulnerability; (2) vulnerability-focused weighting, which emphasizes socially vulnerable communities regardless of population density; and (3) unweighted VoI, which solely focuses on minimizing assignment error without demographic weighting. With a 10-sensor installation, the combined strategy covered areas with 22%–88% higher population density than the other schemes, whereas the vulnerability-focused strategy reduced uncovered social vulnerability by 6%–19%. Prediction accuracy differed by less than 1.5% across strategies. While differences in outcomes are expected given the distinct objectives, the framework adds value by quantifying these trade-offs and enabling a systematic sensor placement strategy that best aligns with policy goals. The proposed framework is scalable to different sensor types and capable of balancing information gain, population exposure, and social vulnerability, providing a versatile tool to guide sensor network design under resource constraints.

3. Cross-entropy-based adaptive concurrent importance sampling for probabilistic seismic loss analysis of spatially distributed multi-state systems

   Byeongseong Choi and Junho Song

   ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, 2026

   Awarded Abs DOI Note

   Editor’s Choice

   Spatially distributed infrastructure systems are crucial assets of a community, and thus assessing their risk under seismic hazards is essential. These assessments should be system-wide, regional, and probabilistic to provide comprehensive insights for decision-makers. Monte Carlo simulation is often used owing to its straightforward application but can be computationally inefficient for rare, high-impact events. While importance sampling can improve computational efficiency by focusing on significant outcomes, it may struggle with computing multiple probabilities simultaneously as it is usually optimized for single-event scenarios. To address these issues, we propose a novel approach termed cross-entropy-based “concurrent” adaptive importance sampling (CE-CAIS). This method efficiently samples multistate systems by concurrently estimating multiple system failure probabilities using a single near-optimal importance sampling density. The effectiveness of CE-CAIS is demonstrated using the Sioux Falls traffic network example consisting of 11 system states and a large-scale regional analysis. In the latter, dimensionality reduction techniques, such as principal component analysis, and the central limit theorem are additionally employed to assess the loss of Shelby County, Tennessee, which includes 305,694 buildings. These results highlight the potential of CE-CAIS in risk assessment of complex systems, paving the way for more effective and scalable approaches to evaluating regional impacts of seismic hazards on spatially distributed infrastructure systems. This study introduces a sampling-based approach to improve the computational efficiency of seismic risk and loss assessment at both the network and regional scales. We propose a cross-entropy-based concurrent adaptive importance sampling (CE-CAIS) method, which constructs an approximate probability density using a Gaussian mixture model, enabling effective performance across multiple system states. CE-CAIS improves efficiency by concurrently estimating the probability of various system states while minimizing variance across all estimates. This makes CE-CAIS particularly effective for evaluating complex systems such as infrastructure networks and regional building losses. In this study, we apply CE-CAIS to assess the seismic risk of a traffic network using the matrix-based system reliability method and to compute exceedance probabilities of regional loss from building damage in Shelby County, Tennessee, incorporating principal component analysis and the central limit theorem. The results demonstrate that CE-CAIS can produce accurate and efficient risk estimates, positioning this method as a promising tool for large-scale seismic risk assessment.

   Notice: The Y-axis label in Fig. 10 should read “Coefficient of Variation” instead of Probability of exceedance. This correction does not affect the results and conclusions.

2025

1. Seismic Structural Response and Loss Estimation for Dense Urban Districts Using Neural Network Parameterized Gaussian Process

   Byeongseong Choi, Sang-ri Yi, and Taeyong Kim

   Earthquake Engineering & Structural Dynamics, 2025

   Abs DOI

   Earthquakes pose a major threat to urban areas, causing fatalities, injuries, and significant economic losses. This study proposes a Gaussian process parametrized by deep neural networks (DNN–GP) as an efficient surrogate for assessing seismic losses of building structures at a regional scale. The framework constructs a GP surrogate using three DNN models that predict the mean, standard deviation, and correlation of structural responses. These models are trained on a seismic demand database developed from a large number of dynamic analyses of structural systems, composed of nonlinear single-degree-of-freedom systems subjected to ground motions with various scaling factors. The DNN–GP model targets assessing an aggregated loss over a cluster of buildings within a limited spatial domain that are subjected to identical ground motion. A key feature of the proposed method is its ability to refine initial estimates from the component DNN models by incorporating reference points, localized nonlinear responses obtained from a selected subset of structures within the considered building portfolio. This study demonstrates the effectiveness of the surrogate model through two numerical studies: (1) prediction of seismic responses for a set of structural systems, and (2) estimation of seismic losses for a building cluster. The results show that the DNN–GP significantly reduces the need for exhaustive nonlinear simulations while maintaining accuracy and quantifying prediction uncertainty. This enables rapid, simulation-informed decision-making using the surrogate model for managing seismic risk and enhancing the resilience of urban infrastructure.

2. Uncertainty quantification of building energy use using a probabilistic spatiotemporal model of urban temperature

   Byeongseong Choi, Mario Bergés, and Matteo Pozzi

   Energy and Buildings, 2025

   Abs DOI

   Urban temperature significantly impacts energy use in the built environment. This paper presents a model coupling approach to quantify how uncertainty in urban temperature, particularly from localized weather effects such as the urban heat island, propagates through building energy systems and affects consumption patterns. The proposed method integrates two models: (1) a probabilistic spatiotemporal model (PSTM) to forecast near-surface temperature at a regional scale with sub-city resolution and (2) a time-series electrical load forecasting model at the building scale. The load forecasting model is calibrated using simulated energy consumption data for representative commercial and residential buildings in the United States. To improve computational efficiency, we also propose combining two uncertainty quantification algorithms: the first-order reliability method and the sequential compounding method, as an alternative to sampling-based approaches. As a numerical case study, we analyze the impact of hot weather in the City of Pittsburgh on energy consumption and the sensitivity of various building types. We also investigate how urban temperatures influence high energy use across different land use types. The numerical experiments on residential building models revealed that high energy consumption was underestimated by 4.2–10.3 %, depending on the building model, when compared to analyses based on temperatures that represent the typical weather of the region. This finding highlights that temperature conditions at a representative weather reference point, e.g., an airport, may underestimate energy use in urban areas, as it fails to account for localized weather such as the urban heat island.

3. Bayesian inference of historical streamflow changes suggests further stress in the Colorado River Basin

   Yuchuan Lai, Byeongseong Choi, and Sujoy B. Roy

   Journal of Hydrology: Regional Studies, 2025

   Abs DOI

   Bayesian inference is used to analyze long-term streamflow in the CRB. Key parameters describing long-term hydroclimatic changes are integrated into a state-space model (SSM). Results from Global Climate Models (GCMs), coupled with downscaling and hydrologic modeling, are used by the SSM to inform these parameters, which are then updated based on historical observations. Multidecadal streamflow projections are made using the SSM and subsequently applied to the Colorado River Simulation System (CRSS) model to examine future river system conditions. New hydrological insights for the region. The SSM projects a long-term decline (in both average and upper and lower confidence intervals) of annual streamflow in the CRB. Compared to some GCM-coupled modeling results, the SSM suggests a modest decrease of annual precipitation in the Upper Basin because of climate change (e.g., 2.3 % lower from the long-term average by 2050), a greater temperature sensitivity (reduction per unit temperature increase) of streamflow, and thus a larger streamflow decline. The observed averaged streamflow during 2000–2023 generally aligns with the decreasing trend in the SSM out-of-sample validations and is projected to become an average condition by ∼2040, along with further declines afterwards. Additional river system modeling from the CRSS suggests, despite extraordinary conservations in recent years, additional efforts may be necessary to sustain regional water supply amid climate change.

4. Spatiotemporal air quality prediction using stochastic advection–diffusion model for multimodal data fusion

   Byeongseong Choi and Michelle Hummel

   Environmental Research Letters, 2025

   Abs DOI

   Particulate matter poses significant risks to respiratory and cardiovascular health. Monitoring ambient particulate matter concentrations can provide information on potential exposures and inform mitigation strategies, but ground-based measurements are sparse. Data fusion approaches that integrate data from multiple sources can complement existing observation networks and reveal insights that single-sensor data might miss to better manage pollutant exposure risks. However, data fusion approaches face multiple challenges, including incompatible measurement units, varying data resolutions, and differing levels of uncertainty. As a result, the optimal method for data fusion remains an open question. Here, we propose a probabilistic spatiotemporal model, based on the stochastic advection–diffusion (SAD) equation, as a data fusion method to process multimodal air quality data to predict hourly concentrations of fine particulate matter (PM2.5). We employ a variational inference method to calibrate the probabilistic model using ground-level observations and the numerical output of two simulation models. We then evaluate the prediction performance of our model for two scenarios: (1) incorporating simulation outputs and ground-level observations from sparse regulatory-grade stations and (2) using ground-level observations from both low-cost and regulatory-grade stations. For the first scenario, the data fusion method reduces prediction error by 14% compared to the nearest regulatory-grade air monitor located 20 km away. For the second scenario, error is reduced by 40% compared to the nearest regulatory-grade monitor and 11% compared to the nearest low-cost sensor located approximately 1 km away. The model captures 78% of observed data within a 75% confidence interval across both scenarios, demonstrating its ability to accurately represent uncertainty. Our findings demonstrate that the proposed SAD model can effectively integrate multimodal data to provide improved prediction of particulate matter concentrations at high spatial resolution. Model outputs can inform individual and community-level decision-making to mitigate air pollutant exposures.

2021

1. Short-term probabilistic forecasting of meso-scale near-surface urban temperature fields

   Byeongseong Choi, Mario Bergés, Elie Bou-Zeid, and Matteo Pozzi

   Environmental Modelling & Software, 2021

   Abs DOI

   This paper introduces a probabilistic approach to spatio-temporal high resolution meso-scale modeling of near-surface temperature and applies it to regions of dimension about 150∼200 km, with 1 km grid spacing and 30-min interval. Our probabilistic approach, based on linear Gaussian models and dimensionality reduction, can accurately forecast short-term temperature fields and serve as a computationally less expensive alternative to physics-based models that necessitate high-performance computing. The probabilistic models here are calibrated from simulations of a physics-based model, the Princeton Urban Canopy Model, coupled to the Weather Research and Forecasting Model (WRF-PUCM). We assess the performance of the calibrated models to forecast short-term near-surface temperature in various cases. In the numerical campaign, our models achieve 0.97–1.13 °C root mean squared error (RMSE) for 24-hours ahead forecast; generating three days of forecast takes between 20 and 170 sec on a single processor (Intel Xeon E5-2690 v4@2.60GHz). Hence, the proposed approach provides predictions at relatively high accuracy and low computational cost.