Rydén (2025) Statistical analysis of monthly precipitation in Sweden using the Tweedie distribution

Identification

• Journal: Theoretical and Applied Climatology
• Year: 2025
• Date: 2025-09-17
• Authors: Jesper Rydén
• DOI: 10.1007/s00704-025-05792-6

Research Groups

• Swedish University of Agricultural Sciences, Uppsala, Sweden

Short Summary

This study applies the Tweedie distribution to model monthly precipitation across 14 sites in Sweden, extending previous research primarily focused on Australia. The findings demonstrate that the Tweedie distribution is a relevant model for Northern European precipitation, revealing seasonal and geographical patterns in its index parameter and similarities to Australian rainfall characteristics.

Objective

• To examine monthly precipitation at various locations in Sweden using the Tweedie distribution.
• To study the seasonal and geographical variability of the Tweedie distribution's index parameter.
• To classify precipitation stations based on their estimated monthly index parameters using hierarchical clustering.

Study Configuration

• Spatial Scale: 14 meteorological stations across Sweden, covering four Köppen–Geiger climate zones (temperate oceanic, temperate continental, subarctic, and tundra).
• Temporal Scale: Monthly precipitation data from 1900–2019 (1900–2009 for one station), analyzed on a monthly basis.

Methodology and Data

• Models used:
  • Tweedie distribution, fitted using maximum-likelihood techniques (profile likelihood) via the R package tweedie.
  • Hierarchical clustering, employing Fréchet distance and Autocorrelation Function (ACF) dissimilarity index via the R package TSclust.
• Data sources:
  • Daily precipitation data (in millimeters) retrieved from the Swedish Meteorological and Hydrological Institute (SMHI).
  • Monthly total precipitation derived from daily observations.

Main Results

• The overall mean of the Tweedie distribution's index parameter (p) across all sites was 1.56.
• The monthly mean values of the index parameter p showed a peak in early autumn, with a general concave shape over the year.
• A positive and statistically significant correlation (r = 0.62, p = 0.02) was found between annual median index parameter values and latitude. Latitude was a significant covariate for the index parameter, but elevation was not.
• The proportion of zero precipitation days was consistently higher than zero across all months and sites, peaking in May.
• Hierarchical clustering results, particularly with the Fréchet dissimilarity index for two clusters, indicated that southern stations tended to group together. Similarities were noted with Australian studies regarding the magnitude of the index parameter (around 1.6) and the influence of coastlines on higher values.

Contributions

• This study extends the application of the Tweedie distribution for precipitation modeling from Australia to Northern Europe (Sweden), providing a first account of its suitability in this region.
• It offers novel insights into the seasonal and geographical variability of the Tweedie distribution's index parameter in a diverse climate.
• It introduces a clustering analysis of precipitation stations based on the estimated index parameter, contributing to empirical climate classification methods.

Funding

• Swedish University of Agricultural Sciences
• Faculty of Natural Resources and Agricultural Sciences (Swedish University of Agricultural Sciences)

Citation

@article{Rydén2025Statistical,
  author = {Rydén, Jesper},
  title = {Statistical analysis of monthly precipitation in Sweden using the Tweedie distribution},
  journal = {Theoretical and Applied Climatology},
  year = {2025},
  doi = {10.1007/s00704-025-05792-6},
  url = {https://doi.org/10.1007/s00704-025-05792-6}
}