bivapp 0.2.0

Bivariate polar plots in Python

Bivariate polar plots in Python

What it says on the tin. This repo provides functions for producing bivariate polar plots, a useful graphical analysis tool in air pollution research. This implementation is largely based on the R package openair and Carslaw and Beevers (2013) - these sources also provide some excellent example cases of bivariate polar plots in practice.

What bivapp does and does not provide

bivapp is intended to provide bivariate polar plots similar to the implementation in openair and as described by Carslaw and Beevers (2013). It is not intended to be a full-featured alternative to openair, as that package provides enough features that it is effectively a complete data analysis suite for air pollution studies. Many of openair's features are already available in other popular Python libraries. For example, openair provides a function for calculating Theil-Sen slopes, but scikit-learn and scipy already feature such tools.

bivapp currently also does not support producing windroses. See windrose instead. This may change in the future.

Documentation

At this early stage functions are only self-documented. Proper documentation is planned.

Getting started

Install from PyPI: pip install bivapp.

Note that because the dependency pyGAM has fallen behind on maintenance bivapp depends on specific versions of some common dependencies, so you might want to use it in a dedicated virtual environment for the time being.

Here's an example of a basic plotting setup.

from bivapp.sampledata import ImportOpenairDataExample
import bivapp.plots as bp
import cmcrameri.cm as cm

df = ImportOpenairDataExample()
fig, axs = bp.BivariatePlotGAM(
    df["so2"],
    df["ws"],
    df["wd"],
    pred_res=200,
    positive=True,
    vmin=None,
    vmax=df["so2"].quantile(0.9),
    cmap=cm.batlowK,
    colourbar_label="SO$_2$ [ppbv]",
    masking_method="near",
    near_dist=1,
)
fig.set_figwidth(6)
fig.set_figheight(4.5)

Existing solutions

The openair package for R provides all these features, but is obviously in R and not Python. The topic of bivariate polar plots in Python also pops up occasionally, like here, here, here, and here. Lastly, there is the existing windrose library, but it lacks bivariate polar plots.

Differences from openair

Users should be aware that the implementation of smoothed bivariate polar plots in this library differs from openair. openair uses the mgcv R package to fit a thin-plate spline GAM to smooth their bivariate polar plots. In their implementation, they bin input data by wind direction and speed, split this data into wind vector components, and then fit the GAM to this binned-then-split data. In bivapp there is currently only one method that fits a GAM, BivariatePlotGAM. This method differs from openair's in a few ways:

1. The GAM can be fit to either raw measurements or binned data;
2. If fitting to binned data, the binning is applied after splitting the wind data into vector components rather than before. In other words, where openair bins then splits, we split then bin.
3. Due to differences in GAM libraries (and their documentation), this package does not exactly replicate the thin-plate spline approach. Instead, bivapp uses the pygam package to fit a GAM to a tensor product of the $u$ and $v$ components of the input wind data. Thus, the GAM-smoothed bivariate polar plot in bivapp is not a perfect replication of openair's smoothed plots, but does appear to achieve the same goal of producing a reasonably smoothed plot.

Even with these differences, both packages can produce similar-looking plots that should permit comparable qualitative analyses. In both this package and openair, fitting a GAM serves to smooth the data, allowing for identification of potential emissions sources from noisy data.