climtrends 0.0.2

A utility for Bayesian trend regression with a variety of statistical models

ClimTrends

ClimTrends is a python package aimed at making it easy to calculate linear trends in a variety of statistical models. The current implementation includes:

• Normal Distribution - trend in mean
• Poisson Distribution - trend in mean
• Exponential Distribution - trend in mean
• Gamma Distribution - trend in mean and standard deviation
• GEV Distribution - trend in mean

The module utilizes datetime-like objects as the input time value, which makes it easy to interoperate with data from netCDF4 and xarray. All models use a Bayesian framework (using the emcee package) and assume a uniform prior on all model parameters. This module is object-oriented and designed to be easily extendable for regressions of other distributions. To do so, one needs to sub-class the ClimTrendModel class and implement a few required routines; see TrendNormalModel.py or other Trend*Model.py files for examples.

Getting started

# dates - a set of input dates
# data - corresponding data from those dates

import climtrends
import numpy as np

# initialize the MCMC model
linear_model = climtrends.TrendNormalModel(dates, data)

# run the sampler
linear_model.run_mcmc_sampler(num_samples = 1000)

# get samples of the slopes
slopes = linear_model.get_mean_trend_samples()

# get the 5th and 95th percentile slopes
slopes_5 = np.percentile(slopes, 5)
slopes_95 = np.percentile(slopes, 95)

Known Issues

• The model assumes that input data are vectors; things will likely break if not.
• Probably other issues exist. This code is tested and is verified to work in some base cases, but it is still in alpha stage and has not been tested across a range of settings.