Periodic Reporting for period 1 - CYCLOPS (Improving Mediterranean CYCLOnes Predictions in Seasonal forecasts with artificial intelligence)
Okres sprawozdawczy: 2022-07-01 do 2024-06-30
Intense cyclones form frequently in the Mediterranean region, with the potential to cause damage to life and property when they hit highly populated coastal areas. Cyclone impacts are caused by the associated strong winds, flash flooding and storm surge. The social and economic impacts are not limited to the Mediterranean area, as cyclones forming in the region can affect Central Europe. While the skill of weather models to forecast such events has dramatically improved over the last decade, the seasonal predictability of Mediterranean cyclones lags behind due to the limitations on horizontal resolution in probabilistic forecasts requiring a large ensemble of simulations. Improving the prediction at a seasonal scale of those extreme events would be of great benefit for society, enabling better disaster risk management and reducing the economic losses they cause. A better prediction of climate extremes would also directly benefit a number of economic sectors such as the insurance and re-insurance industry.
The goal of the CYCLOPS project is to use Artificial Intelligence techniques to enhance the skill of a state-of-the-art seasonal prediction system for predicting Mediterranean cyclones.
The goal of the CYCLOPS project is to use Artificial Intelligence techniques to enhance the skill of a state-of-the-art seasonal prediction system for predicting Mediterranean cyclones.
The first step of the project implementation was be the training of AI algorithm to detect the linkages between
observed cyclones (i.e. detected in a high-resolution reanalysis product such as ERA5) and their large-scale meteo-climatic drivers.
A number of cyclone predictors were tested to identify the most significant ones, including Eady growth rate, SST anomalies, wind shear, atmospheric humidity, vorticity and vertical velocity. Different machine learning algorithms were also implemented, ranging from more complex algorithms such as CNN, which allow for the spatial structure of the drivers to be taken into account, to simple algorithms such as random forest that take as input only spatial aggregates but have the advantage of allowing an easier interpretation of which features were more important for the prediction. Gradient boosting algorithms (XGBOOST) were also found to show good predictive skills.
The second step was the application of the trained machine learning models on the output of the hindcasts of a state-of-the-art operational seasonal forecast system. As expected, the skill of the modelling chain is degraded with respect to the one emerging from the validation on the reanalysis dataset, due to the underlying climate model biases. A number of different metrics of forecast performance (both deterministic and probabilistic). State-of-the art techniques to mitigate the forecast error, such as bias correction were implemented.
observed cyclones (i.e. detected in a high-resolution reanalysis product such as ERA5) and their large-scale meteo-climatic drivers.
A number of cyclone predictors were tested to identify the most significant ones, including Eady growth rate, SST anomalies, wind shear, atmospheric humidity, vorticity and vertical velocity. Different machine learning algorithms were also implemented, ranging from more complex algorithms such as CNN, which allow for the spatial structure of the drivers to be taken into account, to simple algorithms such as random forest that take as input only spatial aggregates but have the advantage of allowing an easier interpretation of which features were more important for the prediction. Gradient boosting algorithms (XGBOOST) were also found to show good predictive skills.
The second step was the application of the trained machine learning models on the output of the hindcasts of a state-of-the-art operational seasonal forecast system. As expected, the skill of the modelling chain is degraded with respect to the one emerging from the validation on the reanalysis dataset, due to the underlying climate model biases. A number of different metrics of forecast performance (both deterministic and probabilistic). State-of-the art techniques to mitigate the forecast error, such as bias correction were implemented.
Results from the project consists in the provision of a method based on a hybrid AI-approach to predict cyclone activity in the Mediterranean at the seasonal scale. The viability of the method was demonstrated in a pseudo-operational setting during the project. In order to maximise the uptake, some further research to further optimize the model skill, and demonstrators in a fully operational setting might be beneficial.