Enabling Reproducibility of Scientific Data Flows with Provenance Equivalence
by Curt Tilmes
Tuesday, February 8, 2011, 11:00am - Tuesday, February 8, 2011, 12:00pm
325 ITE, UMBC
Reproducibility of results is a key tenet of science. Some modern scientific domains, such as Earth Science, have become computationally complicated and, particularly with the advent of higher resolution space based remote sensing platforms, tremendously data intensive. Over the last few decades, these complexities along with the the rapid advancement of the state of the art confound the goal of scientific transparency.
We explore concepts of data identification, organization, equivalence and reproducibility for such data intensive scientific processing. We present a conceptual model useful for describing and representing data provenance suitable for very precise data and processing identification. We present a scheme for creating and maintaining identifiers for precise dataset membership and provenance equivalence at various degrees of granularity and data aggregation.
Application of this model will allow more specific data citations in scientific literature based on large datasets and data provenance equivalence. Our provenance representations will enable independent reproducibility required by scientific transparency. Increasing transparency will contribute to understanding, and ultimately, credibility of scientific results.