Generating daily high-resolution gridded meteorological datasets for New York State from 2017-2018 using two-stage downscaling model
Year: 2022
Type: conference-abstract
Abstract: Background and Aim: Exposure misclassification in weather-related epidemiological studies is always a concern because there are a limited number of weather monitoring sites. This study aimed to build a high-resolution meteorological dataset using two-stage downscaling model and simulate exposures with a highly dense monitor system, New York State (NYS) Mesonet. Methods: High-resolution covariates for terrestrial characteristics, meteorological conditions, and temporal features were linked with NYS Mesonet data from 2017-2018, which contains 126 weather monitor stations (around 27 km x 27 km). These 126 weather stations were split into training set (116 stations) and testing set (10 stations). Machine learning models were used to generate initial meteorological simulations at the first stage. In the second stage, Bayesian spatial-temporal model was used to correct the spatial-correlation residuals from the first stage to get the final high-resolution gridded meteorological datasets (1 km x 1 km). Results: From 2017-2018, the optimal two-stage downscaling model on testing set had R2 = 0.992, RMSE = 0.996 °C, and MAE = 0.703 °C for temperature; R2 = 0.876, RMSE = 4.698 %, and MAE = 3.580 % for RH; R2 = 0.952, RMSE = 1.743 MJ/m2, and MAE = 1.261 MJ/m2 for SR; and R2 = 0.774, RMSE = 3.895 mm, and MAE = 1.603 mm for rainfall. The overall performance of the optimal two-stage downscaling model across training set and testing set had R2 = 0.999 for temperature, R2 = 0.978 for RH, R2 = 0.996 for SR, and R2 = 0.980 for rainfall. Conclusions: This study proposed a two-stage downscaling model by combining machine learning models and Bayesian spatial-temporal models. We generated high-resolution gridded meteorological datasets (1 km x 1 km) which can be used by other researchers in environmental epidemiological studies to reduce exposure misclassification. Keywords: Meteorological exposure, exposure assessment, downscaling model, high resolution
Source: ISEE Conference Abstracts
Language: English
Subfield: Health, Toxicology and Mutagenesis
Field: Environmental Science
Domain: Physical Sciences
SDG: Climate action
Open Access status: diamond
OpenAlex ID: W4317181762