Tuesday, Aug 5: 10:30 AM - 12:20 PM
4114
Contributed Papers
Music City Center
Room: CC-202C
Main Sponsor
Section on Statistics and the Environment
Presentations
Estimating probabilistic property damage from extreme storm events is crucial for assessing the vulnerability and risk to property and populations exposed to weather-related hazards. High-wind storm events may interact with exposed communities and infrastructure in urban and rural settings in intricate ways, making the dependence between hazard intensity and the expected losses challenging to determine. This complexity may differ across the spectrum of low to high values in the variable's domain, exhibiting distinct tail dependencies. This research employs copulas -a popular method for multivariate probability estimation- to address these intricate relationships. We examine various copula models to determine joint and conditional probabilities of property damage resulting from extreme wind events. Our study also investigates arbitrary upper and lower tail dependencies through a multivariate non-Gaussian correlation technique, applied to several Illinois' locations after merging storm events and property damage data from multiple sources for records consistency. Results demonstrate the importance of considering arbitrary tail dependence for probabilistic damage assessments.
Keywords
Copulas
Extreme weather events
Property damage
non-Gaussian dependence
Tail dependence
Global climate models (GCMs) are essential tools for simulating the complex interactions among atmospheric, terrestrial, and oceanic processes, yet they frequently exhibit systematic biases, especially in precipitation estimates. Often, GCMs overestimate rainfall frequency and fail to capture extreme events accurately. Although Quantile Delta Mapping (QDM) is widely used for bias correction, its reliance on empirical distributions can lead to instability at high quantiles due to limited data. To overcome these challenges, we propose an enhanced QDM approach that integrates spatial basis functions to adjust data from multiple observational sites simultaneously while incorporating a generalized Pareto distribution to model extreme precipitation tails. By borrowing information across locations, our method reduces high-quantile instability and improves bias correction. We demonstrate the effectiveness of our approach using daily precipitation projections from ECMWF Reanalysis v5 (ERA5) over Taiwan during the summer.
Keywords
extreme precipitation
generalized Pareto distribution
Co-Author(s)
Nan-Jung Hsu, Institute of Statistics and Data Science, National Tsing Hua University
Hsin-Cheng Huang, Academia Sinica
First Author
Bing-Ru Jhou, Institute of Statistics and Data Science, National Tsing Hua University
Presenting Author
Bing-Ru Jhou, Institute of Statistics and Data Science, National Tsing Hua University
Rainfall is the most important factor in the hydrological system, and it plays a vital role in managing and planning water resources and associated issues. Very few studies have been done to study extreme precipitation by machine learning or deep learning technique. It has been widely proven that predictive model using deep learning has outperformed significantly than other previously existing methods. In this project, we will implement long short-term memory (LSTM) to develop a model to analyze and forecast extreme rainfall events in the US. LSTM is a special type of RNN with memory structures for learning long‐term information. Historically observed daily rainfall data from 1950 to 2022 for the USA obtained from the Historical Climatology Network (HCN) will be used for analysis. After implementing some filtering criteria for each rain gauge station, a total of 1108 rain gauge stations. The performance of the model has been evaluated based on the criterions RMSE, MAPE, and R to identify the potential best model among proposed models. Our results show that the developed model explains more than 95% of variability and can predict extreme precipitation events of the USA.
Keywords
Extreme rainfall events
LSTM
Climatology
National forest inventory (NFI) data are often costly to collect, which inhibits efforts to estimate parameters of interest for small spatial, temporal, or biophysical domains. Traditionally, design-based estimators of forest parameters are used to estimate status of forest metrics of interest, but are unreliable for small areas where data are sparse. Further, direct estimates are often unavailable when sample sizes are especially small. Direct estimate missingness precludes use of traditional small area estimation (SAE) estimators such as Fay–Herriot type models. Here we propose a spatio-temporal SAE model that efficiently uses sparse NFI data to estimate status and trends for forest parameters. The proposed model bypasses the use of direct estimates, and instead uses sampling unit measurements along with auxiliary data including remotely sensed percent tree canopy cover. We provide an analysis of real forest carbon NFI data from the United States Forest Service Forest Inventory and Analysis program over 14 years across the contiguous US, and conduct a simulation study to assess bias, coverage, and model accuracy.
Keywords
National Forest Inventory
Small Area Estimation
Bayesian
Spatio-temporal
Co-Author(s)
Andrew Finley, Michigan State University
Paul May
Grant Domke, USDA Forest Service, Northern Research Station, St. Paul, MN, USA.
Hans-Erik Andersen, USDA Forest Service, Pacific Northwest Research Station, Seattle, WA, USA.
George Gaines
Arne Nothdurft, Department of Forest and Soil Sciences, University of Natural Resources and Life Sciences, Vienna, A
Sudipto Banerjee, University of California Los Angeles
First Author
Elliot Shannon
Presenting Author
Elliot Shannon
Air pollution, specifically Particulate Matter (PM2.5), disproportionately impacts low-income and high-minority communities in Chicago, where limited fine-spatial resolution data exacerbates the problem. Poor outdoor air quality is strongly linked to indoor air pollution, yet gaps remain in understanding this relationship due to inconsistent monitoring. This research addresses these gaps by optimizing sensor placement to enhance the accuracy and coverage of low-cost air quality monitors already established within the sensor network. Using Chicago-specific data, we prioritize matching indoor sensor placements with existing outdoor sensors to ensure consistency across data sources. Additionally, we strategically pair outdoor and indoor sensor locations to optimize spatial diversity. We model the spatial field using a Gaussian Process, incorporating demographic factors and adverse health outcomes. This approach not only improves data quality but also informs future sensor deployments in underserved areas, contributing to more equitable environmental monitoring and a deeper understanding of pollution-related health disparities.
Keywords
Spatial Optimization
Low-Cost Sensor Network
Gaussian Process
Environment
Co-Author
Tuyen Tran, Loyola University of Chicago Mathematics and Statistics
First Author
Mena Whalen, Loyola University Chicago
Presenting Author
Mena Whalen, Loyola University Chicago
Economic impact from extreme storm surges and rainfall under tropical cyclones (TCs) can be enormous, making these hazards of great interest. Sea level rise (SLR) under high greenhouse gas (GHG) emissions can greatly increase the surge-rainfall hazard. However, human intervention can reduce SLR and greatly reduce the future joint hazard. Focusing on Miami, Florida, the National Centers for Environmental Prediction (NCEP) Reanalysis historical joint return period (JRP) is 438 years, and the Max Planck Institute Earth System Model (MPI) projected JRP without SLR adjustment is 113 years, increasing the hazard by 4-fold. With SLR adjustment, however, by 2100 the MPI-projected JRP becomes 4.39 years, a shocking 100-fold increase. Optimistic JRP projections are examined with intervention to GHG emissions by scaling down SLR projections. Scaling values of 2/3, 1/2, 1/3, 1/4, and 1/8 increase projected JRPs to 10, 19, 43, 58, and 97 years, respectively. We examine locations along the Gulf and East coast, visualizing results with an interactive map of JRPs. Future research can focus on how human intervention can reduce future extreme climatology, sea level rise, and subsequent hazards.
Keywords
Sea-Level Rise
Joint Hazard
Storm Surge
Rainfall
Tropical Cyclone (TC)
Extreme Climatology
The Saffir-Simpson Hurricane Wind Scale (SSHWS) sometimes falls short of accurately communicating the hazards associated with a hurricane. While some have proposed that a "Category 6" be added to this scale, we argue that a panel of z-scores -- among the simplest of statistics -- should be used instead. The hazards we assessed in our demonstration were wind speed (which is the only one measured by the SSHWS), storm surge, rainfall, and tornado occurrence. We examine four storms, each of a different SSHWS category, that are well-known for a specific hazard they presented. We show that our multivariate hazard assessment highlights those hazards, providing a portrait of the storms that is more comprehensive than the SSHWS.
Keywords
z-score
Chebyshev's rule
Saffir-Simpson Hurricane Wind Scale
hurricane hazards