Pyemu: A Python-Based Framework for Linear-Based Model Uncertainty Analysis

Friday, 19 December 2014
Jeremy White, USGS Texas Water Science Center, Austin, TX, United States
pyEMU is an open-source python-based framework for model-independent linear-based parameter and predictive uncertainty analysis. The framework is designed to support the analysis of high-dimensional inverse problems that have thousands of parameters and hundreds of thousands of observations. The code is compatible with the PEST and PEST++ software suite, and implements several forms of linear analysis equations, such as Schur's complement for conditional uncertainty propagation and subspace error variance, including a form of error variance analysis of model structural error. These linear analysis equations are the most common and also the most applicable to large-scale environmental models.

Several native python operators (such as multiplication, subtraction, addition, exponentiation) have been overloaded to make equation building more concise as well as to achieve speedup with operations involving diagonal matrices. To help ensure pyEMU is intuitive and easy to use, emphasis was placed on flexibility and concise object instantiation. As a result, several types of arguments can be handled elegantly.