maxklr/extremeweather

Introduction to Extreme Value Theory applied to Extreme Weather Events

Statistics of Extreme Weather

We use Extreme Value Theory (EVT) to study the statistics of extreme weather events (eg max daily rainfall).

Example: Extreme Rainfall in New York (Maximum likelihood approach)

We use a Maximum Likelihood Estimation (MLE) approach to model hourly rainfall in New York. Weather data is from single weather station.

Example notebook: Maximum Likelihood Estimation

Example: Bayesian Approach for Extreme Rainfall Data

In this notebook we use a Bayesian Approach (implemented in PyMC) to model extreme weather events. We start with a univariate study (which parallels the MLE approach) for a single weather station. Next we train a spatial model, by incorporating data from many (nearby) weather stations, using a Gaussian processes as prior distribution for the GEV model parameters.

Example notebook: Bayesian Approach with Gaussian Process

PyData 2023 Talk

References

• An Introduction to Statistical Modeling of Extreme Values - Coles (2001)
• Statistical Modeling of Spatial Extremes - Davison, Padoan and Ribatet (2012)
• Generalized Extreme Value Distribution in PyMC - Caprani (2021)
• Modeling spatial data with Gaussian processes in PyMC - Paz (2022)