Field Campaign Decision Making in Atmospheric Science Using an Automated Decision Algorithm
Open Access
- Author:
- Hanlon, Christopher James
- Graduate Program:
- Meteorology
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- March 02, 2015
- Committee Members:
- George Young, Dissertation Advisor/Co-Advisor
Johannes Verlinde, Committee Chair/Co-Chair
Martin Patrick Tingley, Committee Member
Paul Griffin, Committee Member - Keywords:
- decision algorithm
probabilistic forecasting
optimization
dynamic programming
logistic regression
genetic algorithm - Abstract:
- An automated decision algorithm was developed for resource deployment decisions under weather uncertainty for an atmospheric science field campaign. Scientists on the Deep Convective Clouds and Chemistry (DC3) field campaign were tasked with using aircraft to gather in-situ measurements of isolated deep convection over three separate study regions in the United States during spring-summer 2012. The DC3 campaign was budgeted a finite number of flight hours with which to sample convection and was faced with a fixed start date and end date for the field campaign, forcing them to make difficult decisions each day about whether to fly their aircraft or whether to save their flight hours for a more promising future day. To guide decision recommendations, a quantitative definition of atmospheric conditions denoting a “successful” flight and a function defining field campaign utility as a function of “successful” flights were developed through communication with DC3 principal investigators. Utility-maximizing automated decision recommendations were generated using a dynamic programming-based decision algorithm with automated forecasts of the likelihood of “successful” conditions generated by a system employing a logistic regression with parameters tuned by a genetic algorithm. The forecasts generated by the automated forecasting system showed better skill than those produced concurrently by human forecasters, and the decisions generated by the automated decision algorithm would have improved field campaign utility relative to the decisions made by human decision-makers.