Variable Bounds Analysis of a Climate Model Using Software Verification Techniques

Software verification techniques often use some approximation method that identifies the limits of the possible range of values that variables in a computer program can take during execution. Current climate models are complex computer programs that are typically iterated time-step by time-step to predict the next value of the climate-related variables. Since these iterative methods are necessarily computed only for a fixed number of iterations, they are unable to answer many long-range questions that may be posed regarding climate change, for example, whether there are natural fluctuations or whether a tipping point is reached after which there is no return to normal. In a departure from the usual step-by-step climate models, we propose to use software verification techniques to predict absolute bounds, i.e., predict maximum and minimum values that could ever arise according to a climate model. Hence we can answer some of the long-range climate prediction questions that were not possible to answer before. Key–Words: Software Verification, Computing Invariant Values, Datalog, Climate Model