README.md

In siliusmv/inlaBGEV: Spatial block maxima modelling with the bGEV distribution using R-INLA

This repository contains the code used for creating all the results in the paper Modelling Sub-daily Precipitation Extremes with the Blended Generalised Extreme Value Distribution, which performs spatial Bayesian hierarchical modelling of extreme sub-daily precipitation in Norway using the bGEV distribution.

Installation

You can install the package directly from github

devtools::install_github("siliusmv/inlaBGEV")

Scripts

There are a number of scripts available in the exec/ folder. These are described below.

Univariate data exploration

In the script local-gev-estiation.R, the method of probability weighted moments is used for univariate modelling of yearly maximum precipitation at all the available weather stations. Exploratory analysis is performed to look for trends in the data.

Examination of priors

In the script pc-priors.R we compute PC-priors for the tail parameter of the GEV, bGEV and GP distributions and compare them for different penalty parameters.

Cross validation

The script cross-validation.R performs k-fold cross-validation using the scaled threshold weighted CRPS.

Estimation of return levels

The script BGEV-estimation.R is used for modelling short-term precipitation maxima and estimating return levels in a map.

Simulation studies

The script simulation-study-univariate-MLE.R contains simulation studies for comparing the performance of the bGEV with that of the GEV.

Visualisation of all weather stations

The script visualise-data.R simply plots the location of all weather stations on a map.

Data

This package contains three data-sets, saved under the data/ folder. Those are observations, estimated_sd and prediction_grid. observations is an sf-object containing all yearly precipitation maxima for periods of 1, 3, 6, 12 and 24 hours. The object also contains coordinates for all stations and the values of the available explanatory variables. estimated_sd is a data.frame that contains the standard deviation of all observations larger than the 99% quantile from all available weather stations. The prediction_grid object is an sf-object containing coordinates and explanatory variables for all locations in a 1x1 km^2 grid over the south of Norway.

R package versions

The version of all R packages used for running this code can be found in the file renv.lock

siliusmv/inlaBGEV documentation built on Dec. 5, 2022, 5:23 a.m.