Redefining climate regions in the United States of America using satellite remote sensing and machine learning for public health applications

  • Alexander Liss Department of Civil and Environmental Engineering, Tufts University, Medford; Tufts Initiative for Forecasting and Modeling of Infectious Diseases, Medford, United States.
  • Magaly Koch Tufts Initiative for Forecasting and Modeling of Infectious Diseases, Medford; Center for Remote Sensing, Boston University, Boston, United States.
  • Elena N. Naumova | elena.naumova@tufts.edu Department of Civil and Environmental Engineering, Tufts University, Medford; Tufts Initiative for Forecasting and Modeling of Infectious Diseases, Medford, United States.

Abstract

Existing climate classification has not been designed for an efficient handling of public health scenarios. This work aims to design an objective spatial climate regionalization method for assessing health risks in response to extreme weather. Specific climate regions for the conterminous United States of America (USA) were defined using satellite remote sensing (RS) data and compared with the conventional Köppen-Geiger (KG) divisions. Using the nationwide database of hospitalisations among the elderly (≥65 year olds), we examined the utility of a RS-based climate regionalization to assess public health risk due to extreme weather, by comparing the rate of hospitalisations in response to thermal extremes across climatic regions. Satellite image composites from 2002-2012 were aggregated, masked and compiled into a multi-dimensional dataset. The conterminous USA was classified into 8 distinct regions using a stepwise regionalization approach to limit noise and collinearity (LKN), which exhibited a high degree of consistency with the KG regions and a well-defined regional delineation by annual and seasonal temperature and precipitation values. The most populous was a temperate wet region (10.9 million), while the highest rate of hospitalisations due to exposure to heat and cold (9.6 and 17.7 cases per 100,000 persons at risk, respectively) was observed in the relatively warm and humid south-eastern region. RS-based regionalization demonstrates strong potential for assessing the adverse effects of severe weather on human health and for decision support. Its utility in forecasting and mitigating these effects has to be further explored.

Dimensions

Altmetric

PlumX Metrics

Downloads

Download data is not yet available.
Published
2014-12-01
Info
Issue
Section
Original Articles
Keywords:
remote sensing, LKN-regionalization, machine learning, morbidity, climate, classification, decision support, United States of America
Statistics
  • Abstract views: 1971

  • PDF: 716
How to Cite
Liss, A., Koch, M., & Naumova, E. N. (2014). Redefining climate regions in the United States of America using satellite remote sensing and machine learning for public health applications. Geospatial Health, 8(3), S467-S659. https://doi.org/10.4081/gh.2014.294