LOTUS Regression



The code is tested on Python versions 2.7, 3.4, 3.5, 3.6, 3.7, 3.8 and should work on any of them, however we recommend using the Anaconda python distribution which by default comes with Python version 3.8. Some of the example code may also be incompatible with Python 2.7

A minimal set of requirements are the Python packages numpy, scipy, pandas, statsmodels, xarray, requests, appdirs, netcdf4, dask which should be automatically be installed upon installing the package, if you are using anaconda you can verify/install that these packages are available by running:

conda install numpy scipy pandas xarray statsmodels requests appdirs netcdf4 dask

How To Install

The code can be installed by running:

pip install LOTUS_regression -f https://arg.usask.ca/wheels/

What’s New in Version 0.7.1

  • Updated predictors to the end of 2020

  • Added an option include_monthly_fits to estimate seasonal trends

  • Added an option return_raw_results to return the raw statsmodel regression results

What’s New in Version 0.5.1

  • Updated the documentation and code to be compatible with the newest version of pandas.

What’s New in Version 0.5

The largest change has been the addition of preconstructed finalized baseline predictor files. These three files can be loaded in with:

from LOTUS_regression.predictors import load_data

predictors_pwlt = load_data('pred_baseline_pwlt.csv')
predictors_ilt = load_data('pred_baseline_ilt.csv')
predictors_eesc = load_data('pred_baseline_eesc.csv')

All of the satellite dataset examples have been modified to use the pred_baseline_pwlt.csv file. A detailed description of the predictors contained in the three files can be found on the new technical note page, which in addition also describes the regression algorithm.

Other Changes

  • Predictors module has been updated to include ILT terms

  • GISS AOD loader has been modified to extend the AOD to the current time based on the last non-zero value

  • If linear/EESC terms are included in the first stage of the two-step ILT they are not subtracted when calculating the residuals for the second step (two-step ILT is mostly deprecated in favor of a one step ILT but is still available for testing purposes)


Some of the examples may require additional packages for loading and handling of data, to obtain all of the packages necessary to run the examples you may run:

conda install matplotlib netcdf4 dask hdf5 hdf4 jpeg=8d

Satellite Datasets

Here the piecewise linear trends are calculated for different satellite datasets. The GOZCARDS dataset is not deseasonalized, so seasonal components are added to the predictors prior to fitting.

Other Functionality

Here are several examples which go through some advanced functionality of the regression package. The predictors example shows how different predictors can be created for the regression, the SingleBin example runs the regression for a single bin without using any of the higher level wrappers, and the Linear Components example shows how the linear components of the fit can be changed.