Genetic algorithm method for determining temperature profiles

A genetic algorithm is used to determine atmospheric temperature profiles from infrared observations of ambient carbon dioxide gas in the atmosphere. The observation range is divided into four range bins and the goal is to predict with reasonable accuracy the temperature in each range bin. An overdetermined system of equations corrupted by measurement noise describes the horizontal temperature profile. The use of a classical linear least squares fit to solve the equations results in predicted temperature errors over 30 K for small amounts of measurement noise. To overcome this problem, a genetic algorithm was developed using fitness function correction terms based on the authors' experience with the atmospheric temperature profiles and it was tested on a simulated data system. Excellent results were obtained demonstrating the usefulness of the genetic algorithm approach for solving this type of problems.