NEDAS : Next-generation Ensemble Data Assimilation System
Introduction
Data assimilation (DA) combines information from model forecasts and observations to obtain the best estimate of a dynamical system. To improve the prediction skill of the Earth-system models, DA algorithms face two main challenges: 1) the increasing size of model state and observations demands computationally efficient algorithms; 2) the nonlinearity in error growth mechanisms and in state-observation relation requires more sophisticated algorithms.
NEDAS provides a light-weight solution for developing new DA algorithms for Earth-system models. To address the first challenge, parallel computation is implemented with the mpi4py package to ensure scalability to large-dimensional problems; scientific computation packages, such as numpy, along with the numba.jit compilation technology, ensure computational efficiency. As for the second challenge, NEDAS employs a modular design that separates the DA workflow into managable parts, which can be upgraded with new approaches to tackle with the nonlinear problems. Thanks to the rich Python ecosystem for scientific computing and machine learning, NEDAS provides a flexible platform for rapid prototyping of innovative DA methods.
NEDAS now offers a collection of DA algorithms for benchmarking and intercomparison, including the serial approaches that assimilate one observation at a time (similar to DART), and the batch assimilation approaches (similar to the LETKF in PDAF). Its intuitive user interfaces and interoperability with other DA software enable early testing of new DA algorithms, well before committing resources to full-scale operational implementation.
Getting started
Design
User interfaces
API documentation