To classify events into convective or non-convective storms based on analysis of evolution of wind speed, temperature, station pressure and other meteorlogical variables.

This code uses sktime¹ to perform a supervised classification of the time series of weather variables. A training event set is developed from visual classification of wind gust events over 90 km/h at high-quality stations. The full event set comprises all storm events with a maximum wind gust in excess of 60 km/h. The approach is based on previous concepts from Spassiani and Mason (2021)² and Cook (2023)³.

Quality flags are used to eliminate spurious gusts, but there still remain some anomalous events (instrument spikes).

Data from this analysis is used to create a geodatabase for publication and use in GIS applications.

This work has been supported by Queensland Fire and Emergency Services.

1-minute weather observations from all available weather stations across Australia.

Variables required:

• gust wind speed
• wind speed
• wind direction
• temperature
• dew point
• station pressure
• rainfall
• quality flags for all variables

Original source data can be requested from the Bureau of Meteorology, Climate Data Services team,

• python 3.9
• jupyter
• numpy
• matplotlib
• pandas
• pytz
• lxml
• prov
• seaborn
• sktime-all-extras
• gitpython
• fiona
• geopandas
• cartopy
• scikit-learn
• metpy
• pycodestyle

1. Build the conda environment
2. Install additional code from https://github.com/GeoscienceAustralia/nhi-pylib
3. Add the path to the additional code to the PYTHONPATH environment variable

1. extractStationDetails.py - extract station details from the raw data. Creates a geojson file of station locations used in subsequent scripts
2. extractStationData.py - extract all events from the raw data. This should be executed twice, initially with a threshold of 90 km/h and again with a threshold of 60 km/h. The outputs for each execution need to be stored in different folders. Users will need to check the path to the original source files (OriginDir in the configuration files.)
3. selectHQStations.ipynb - use this to select high-quality stations that will form the training dataset for the ML classification process
4. classifyGustEvents.py - classifies all daily maximum wind gusts using El Rafei et al. (2023)⁴
5. ClassifyEvents.ipynb - interactive notebook to visually classify storms with maximum gust > 90 km/h at high-quality stations
6. classifyTimeSeries.py - use ML approach in sktime to classify all storm events (> 60 km/h) ⁵
7. convertStormCounts.py - convert classified storms to counts of storm type at each station
8. joinStormRatesToStationDetails.py - Join storm rate data to station details and store output in GeoJSON file
9. joinStormClassToMaxDailyData.py - Join storm classification data to daily maximum gust data
10. joinDailyWeatherObservations.py- Join the daily weather observations to the storm data

1. AnalyseClassifiedStorms.ipynb - interactive notebook to compare this classification against the El Rafei et al.⁴ gust classification
2. analyseStormEventTypes.ipynb - interactive notebook to examine the full classified storm event set, e.g. median and 99th percentile wind gusts for each storm type, seasonal distribution of storm type, comparison against other metrics (e.g. gust ratio, emergence). Still a work in progress.

Link to eCat record for the data

Add a licence