DAPyr (pronounced "dapper") is a Python package for running, testing, and experimenting with data assimilation methods using Lorenz models. DAPyr supports quick configuration and running of experiments, allowing one to setup and run data assimilation cycles in as little as two lines. In addition, DAPyr utilizes numbalsoda and numba for model integration, allowing for faster runtimes.

For a in-depth tutorials and explanations, see the documentation.

DAPyr currently supports the following:

1. Lorenz 1963
2. Lorenz 1996
3. Lorenz 2005

1. Ensemble Square Root Filter (Whitake and Hamill, 2002)
2. Localized Particle Filter (Poterjoy, 2022)

1. Linear
2. Quadratic
3. Logarithmic

Pip:

python -m pip install DAPyr

import DAPyr as dap
import numpy as np

# Use the Expt class to initialize an new experiment and calculate all necessary initial states
# Experiment will use Lorenz 63 (model_flag: 0), run for 300 time steps, and use the EnSRF method (expt_flag: 0)
expt = dap.Expt('Basic_L63', {'model_flag': 0, 'expt_flag': 0, 'T': 300})

# Run the experiment using the runDA function
dap.runDA(expt)

# Access RMSE and spread using the rmse and spread attributes
print('Avg. RMSE: {:.3f}'.format(np.mean(expt.rmse)))
print('Avg. Prior Spread: {:.3f}'.format(np.mean(expt.spread[:, 0])))
print('Avg. Posteriod Spread: {:.3f}'.format(np.mean(expt.spread[:, 1])))

#Save the experiment to filesystem using saveExpt, the file will be called it's experiment name
dap.saveExpt('./', expt)

#Load in previous experiments using loadExpt
expt2 = dap.loadExpt('./Basic_L63.expt')

For more examples, see tutorial.ipynb in src.

Please cite this repository if you use DAPyr in your research publications or presentations.

Many thanks to the following for their contributions to DAPyr:

• Henry Santer (LPF Python implementation and many, many bug discoveries)
• Raelynn Wonacott (For advice on speeding up model integration using numba)

The development of this software is supported by the following funding sources:

• NSF CAREER Award #1848363