- Owl (the bird)
- *Owl is a bird designed to detect both day-time and night-time temperature extremes, and their drivers. *
Owls are nocturnal creatures with exceptional long-distance vision and the ability to rotate their heads up to 270 degrees. Much like the owl, this prototype can "see" far and wide; it is applicable to all temperature extremes (day and night), and connects the past drivers of extremes to projections of the future.
Specifically, the Owl prototype (1) detects extremes from a given temperature time series input, (2) performs a feature selection on a range of potential predictors and (3) identifies storylines in future projections of these extremes.
Here we describe the three working parts of the Owl:
(1) Heatwave Index Detection Here, the Owl follows the widely used definition of heatwaves: temperatures exceed the 90th percentile for 3 days or longer (e.g Russo et al., 2015). Heatwaves are typically detected using daily maximum 2m temperature, although minimum temperature can be used for night-time heatwaves. The first role of the Owl is to detect the heatwave occurrence in the target time series, achieved by: - calculating the 90th percentile (threshold) - identifying the occurrences of exceedence of the threshold - outputting daily time series of heatwaves
The detection code allows the user to choose a range of parameters, such as the climatology period over which the treshold is calculated, the minimum duration of the HWs (e.g. 3 days as above). It also allows for the detection of heatwaves in any (daily) temperature time series, and is demonstrated here for the ERA5 reanalysis (Hersbach et al., 2021). The examples correspond to the area-averaged temperatures over the Po Valley, Italy.
Fig 1: HW occurrence over Po Valley 1950-2022 in ERA5.
(2) Feature Selection The second role of the Owl is to employ a feature selection algorithm to detect the drivers, from a list of potential predictors, of the HW index produced in step 1. The potential predictors provided in the prototype are area-averages of clusters identfied by a k-means clusters of the following variables: precipitation, 2m temperature, mean sea level pressure, geopotential height, soil moisture, outgoing long-wave radiation, sea ice cover, sea surface temperature.
The feature selection used here is a machine learning classifier wrapped in an optimization algorithm which detects the optimal combinations of predictors (i.e. those which provide the best skill in recreating the HW time series). The HW target data and predictors are split into training and testing periods. The ML classifier selectes combinations of potential predictors to recreate the test data, providing a measure of skill (i.e. F1-score). Then, the Coral Reef Optimization algorithm (Salcedo-Sanz et al., 20; Perez-Aracil et al.,) works to select the optimal combination (i.e. to increase the skill).
The output is a list of predictors which contribute to recreating the HW time series, providing information also on the relevant time lags for each selected predictor.
Fig 2: Feature Selection: optimal predictors and lags used to recreate HW index in Figure 1.
- Storylines
The last role of the Owl is the description of the future evolution of HW, based on the CMIP6 climate simulations. In particular, the HW occurrence is calculated for a wide range of models, considering different members and Shared Socio-economic Pathways and using “current-climate” baseline (CWS14.2, see D5.2). Similarly, the candidate predictors are calculated as above, according to the variables availability of each model. For each simulation, the relevant drivers of HW are identified running the CRO on the corresponding CWS14.2.
The list of drivers selected on ERA5 are used as benchmarks and are employed to construct storylines. These inspect the evolution of HW indices in future climate, putting constraints on the simultaneous changes of relevant drivers. The module checks, for an indicated pair of drivers, which CMIP6 simulations selected both of them during the feature selection. These simulations are then divided between those that, in a future-climate period (CWS15, see D5.2), have lower or larger changes than the multi-model mean, generating 4 groups of models, i.e. storylines. For each combination of driver impact a distribution of projected changes of HW occurrence and frequency in future climate is provided.
Learn more about owl in its official documentation at https://owl.readthedocs.io.
Submit bug reports, questions and feature requests at https://github.com/climateintelligence/owl/issues
You can find information about contributing in our Developer Guide.
Please use bumpversion to release a new version.
- Free software: Apache Software License 2.0
- Documentation: https://owl.readthedocs.io.
This package was created with Cookiecutter and the bird-house/cookiecutter-birdhouse project template.