Abstract
In this talk, I present different strategies for evaluating the adequacy of a given model in context of high uncertainties, all based on the idea of iterative coherence testing, i.e., a methodology that anchors the improvement of the models accuracy in an iterative process of solving the discrepancies between the model’s predictions and observations. Based on Le Qu´er´e (2006), we consider three different phases of the development of a model and how iterative conference testing, or “the closing the loop” methodology in G.Smith’s terms, can guide decisions for improving the model for each one of these phases. I argue that iterative testing serves a different purpose and is connected to a different evaluation strategy at each phase of the development of a model, from the adoption of a initial model known to be imperfect to a mature model ready to be confronted to observations. I will illustrate the merits and pitfalls of such an account of iterative testing and model evaluation based on the example of the photochemical models of Titan’s atmosphere developed over the last two decades. Models of Titan’s atmosphere have roughly followed two very different paths: either they have started with a complex chemical model but simplified astrophysical constraints, or with a complex description of the astrophysical conditions but a simplified chemistry. Both, however, failed to reproduce observations, although uncertainties attached to observations made the discordance not so worrying, until the observations of the Cassini-Huygens’ mission were released in 2007. The latter showed that the discrepancies were not dampened, but on the contrary worsen by improved observations: the uncertainty attached to the model were shown to be higher than those related to observations, rendering the comparison between observations and models impossible to interpret. This case has led to an impressive surge of creativity from the astrochemists, who have exploited this opportunity to develop innovative iterative tools designed to enrich and develop their photochemical network to the point where a discordance between their chemical models and the Cassini’s observations could finally be significantly interpreted.
