Uncertainty at scale: how CS hinders climate research

Computer science is a powerful tool for enabling data-driven advances in global ecology and conservation. However the amplification cuts two ways, as mechanisation can also com- pound problems inherent with just how uncertain [13] anything to do with natural ecosystems are! Species habitat datasets are uncertain, local observations are uncertain, the resulting inferences about species distributions are uncertain, side-effects from interventions are uncertain; conservation action has evolved to take this into account [17]. Computer science when applied without consideration of these factors can amplify the uncertainty by running ever-larger datasets through increasingly complex data pipelines and algorithms, all built upon wobbly foundations. What exact version of the dataset is being used? What exact version of the dataset did you use? What assumptions went into generating that dataset? What libraries, system dependencies and environment variables were used to calculate the results?

In this talk, we first segment sources of uncertainty across ecological data sources (§1) and computation over them (§2), and then reflect how these uncertainties impact ecological research and how we might cleanly bound the uncertainty for future conservation research (§3).