Prediction |

[Under Review] Impact of Temporally-Varying Geographic Predictors on Long-Term Air Pollution Exposure Estimation for Epidemiology

Sat, 01 Nov 2025 00:00:00 +0000

[Manuscript in Preparation] Correlation of Air Pollution and Traffic Volume: Comparison between Road Density and Traffic Volume Estimates

Thu, 01 Sep 2022 00:00:00 +0000

Modeling of Spatio-temporal changes of Urban Sprawl in Jeju-island: Using CA (Cellular Automata) and ARD (Automatic Rule Detection)

Mon, 24 May 2021 00:00:00 +0000

Air Pollution Predictive Modeling

Sun, 27 Dec 2020 00:00:00 +0000

Summary

This project aims to predict air pollution concentrations at locations across South Korea where direct measurements are unavailable. Using the Universal Kriging (UK) model, we estimate the annual concentrations of key air pollutants—PM10, PM2.5, O3, and NO2. The UK model leverages observed pollutant concentrations at monitoring sites and incorporates spatial and environmental characteristics represented by geographic variables, along with spatial autocorrelation, to predict concentrations at unmonitored locations. The observational data consists of annual air pollution measurements collected from air quality monitoring stations nationwide. Additionally, the model integrates 320 geographic variables, derived from spatial data processing pipelines, which capture factors related to air pollution levels, such as road networks, elevation, population density, land use, and vegetation index (NDVI). To manage the high dimensionality of these variables, we employ Partial Least Squares (PLS) regression to reduce the dataset to 2–3 primary predictive components, which serve as inputs for the UK model. This dimensionality reduction optimizes model performance by focusing on the most relevant predictors. In many cases, our team utilizes this model to estimate air pollution exposure for cohort study participants based on their residential locations. This process involves an initial geocoding step, where participants’ address information is converted into geographic coordinates. Subsequently, approximately 320 geographic variables are calculated for each residential location. These variables are then reduced to key components via PLS, which are used as inputs in the UK model to estimate pollutant concentrations for individual residences. Through this project, we enhance the precision of individual exposure assessments by estimating air pollution concentrations directly at individuals’ residential locations rather than relying on the concentrations from nearby monitoring sites. This approach enables more accurate evaluations of personal exposure and contributes to more precise health effect assessments related to air pollution exposure.

Land Cover Change Prediction Modeling

Wed, 13 Nov 2019 00:00:00 +0000

Abstract

This project aims to develop an algorithm that can accurately predict land cover changes across large regions that include both urban and rural areas. Existing land cover prediction models are typically tailored to urban characteristics, which limits their effectiveness in regions where urban and rural areas are mixed. To address this, our project seeks to create a new predictive algorithm that incorporates the unique features of both urban and rural environments. The first step involves using multi-temporal satellite imagery to create land cover maps at various time points. We then perform a time-series analysis to identify patterns of land cover change over time. Based on this data, we are developing the BPLE (Bayesian-based Probability Land cover Estimation) algorithm, which is designed to detect and predict land cover changes while accounting for the different characteristics of urban and rural regions. This algorithm applies Bayesian probability theory to capture the spatiotemporal regularities in land cover change, enabling us to forecast future changes. To validate the accuracy and utility of our model, we will compare its predictions with actual observed changes in both urban and rural areas. By providing a precise tool for predicting land cover change, this project aims to support a range of applications, including national land management, natural disaster risk assessment, and water cycle management. The resulting model will offer a versatile approach for studying and managing large, mixed urban-rural regions.