Authors: Ernesto Munoz, Emily Becker, Sheri Mickelson  

The need to process large amounts of data from climate ensembles presents an opportunity to develop efficient software that handles multiple realizations. Large ensemble efforts, such as the North American Multi-Model Ensemble and the CESM Large Ensemble Project, produce and make available enormous amounts of climate data. In this talk we reflect from a scientist perspective on the process of developing a workflow with Python to handle multiple model realizations and quantify statistics from the ensemble. We consider software design aspects that a scientist may not normally consider such as: testing, efficiency, reproducibility and extensibility. One of the end products we target is Python code to compute ensemble statistics such as: the ensemble mean, the multi-model ensemble mean, anomalies, anomaly correlation and the root-mean-square error. The presentation will provide perspectives and workflow models on how to achieve these goals.