Using Ecosystem Dynamics to Evolve Skillful Forecasts of Convection

Evolutionary programming (EP) as a concept was first proposed by computer scientists in the 1960s and began seeing application to real-world problems in engineering in the 1990s. Conceptually, EP can be thought of as a process by which the principles of evolution are applied to a problem in order to develop algorithmic solutions. In that sense, all that is required is a well-defined problem, a clear measure of the success or fitness of a solution, and a mechanism for producing the next generation of solutions from the previous one, based on this information. In the case of weather forecast data, the author has applied this concept in the form of using a set of conditional equations with a set of adjustable variable inputs, operators, and coefficients. This approach has been shown to produce skillful deterministic and probabilistic forecasts for a range of forecast problems including temperature, precipitation amount and phase, and tropical cyclone intensity. Here, the author presents a further development of this technique in which ecosystem dynamics (co-evolution in a predator-prey context) are employed to drive further improvements in forecast skill. The technique is demonstrated in the context of the operational Autonowcaster forecast tool.