Baroclinic Wave Simulation Ensemble: a Machine Learning ready dataset

TL;DR

This paper introduces a large dataset of baroclinic wave simulations to study mid-latitude weather systems and cyclones using machine learning.

Contribution

The paper provides a machine learning-ready dataset of 6,388 baroclinic wave simulations with tracked cyclone features.

Findings

The dataset includes raw model outputs and 89 features for each cyclone, including 16 intensity measures.

The distribution of cyclone intensity metrics matches those found in reanalysis and climate model studies.

The 112 failed simulations were analyzed and their computational issues were statistically explained.

Abstract

A large ensemble of 6,500 different baroclinic wave simulations have been run, processed and provided to study extra tropical cyclones and mid-latitudes dynamics. The data were generated using OpenIFS@home, an open science climateprediction.net project allowing the distribution of the computation of the ensemble with the OpenIFS 43R3v2 model. For each simulation, the cyclones were tracked and 89 features -including 16 intensity measures- were extracted. The presented dataset is composed of the raw output of the OpenIFS model for 6,388 of the 6,500 members of the ensemble and the extracted features of the tracked cyclones. The distribution of the minimum mean sea level pressure and the maximum relative vorticity at 850 hPa is plotted to enable comparison with studies that have analysed ETCs in reanalyses and climate model data. The computational failure of the missing 112 ensemble members is statistically assessed and explained. Outside of OpenIFS, the dataset and the associated code and configuration files are available and accessible.

Figures