Automated decision algorithm applied to a field experiment with multiple research objectives: The DC3 campaign

Abstract

Automated decision systems have shown the potential to increase data yields from field experiments in atmospheric science. The present paper describes the construction and performance of a flight decision system designed for a case in which investigators pursued multiple, potentially competing objectives. The Deep Convective Clouds and Chemistry (DC3) campaign in 2012 sought in situ airborne measurements of isolated deep convection in three study regions: northeast Colorado, north Alabama, and a larger region extending from central Oklahoma through northwest Texas. As they confronted daily flight launch decisions, campaign investigators sought to achieve two mission objectives that stood in potential tension to each other: to maximize the total amount of data collected while also collecting approximately equal amounts of data from each of the three study regions. Creating an automated decision system involved understanding how investigators would themselves negotiate the trade-offs between these potentially competing goals, and representing those preferences formally using a utility function that served to rank-order the perceived value of alternative data portfolios. The decision system incorporated a custom-built method for generating probabilistic forecasts of isolated deep convection and estimated climatologies calibrated to historical observations. Monte Carlo simulations of alternative future conditions were used to generate flight decision recommendations dynamically consistent with the expected future progress of the campaign. Results show that a strict adherence to the recommendations generated by the automated system would have boosted the data yield of the campaign by between 10 and 57%, depending on the metrics used to score success, while improving portfolio balance.

Introduction to the Deep Convective Clouds and Chemistry (DC3) 2012 Studies

“…Being tested during DC3 field operations were probabilistic forecasts and automated decision algorithms to aid in field campaign decisions. Hanlon, Small, et al () found that automated decision algorithms would have increased the yield of data from the field campaign, while Hanlon, Young, et al () showed that their probabilistic forecasts had predictive skill over climatology for each of the DC3 regions. Another pivotal reason for the success of DC3 was the strong cooperation and interaction between the lead PIs for the project.…”