Periodic Reporting for period 1 - CausalBoost (Using causal discovery algorithms to boost subseasonal to seasonal forecast skill of Mediterranean rainfall)
Periodo di rendicontazione: 2020-03-01 al 2022-02-28
The Mediterranean region (MED) is a hotspot of anthropogenic climate change and impacts are probably already felt today; recent heatwaves and persistent droughts have led to crop failures, wild fires and water shortages, causing large economic losses. Climate models robustly project further warming and drying of the region, putting it at risk of desertification. The particular vulnerability of this water-limited region to climatic changes has created an urgent need for reliable forecasts of rainfall on sub seasonal to seasonal (S2S) timescales, i.e. 2 weeks up to a season ahead. This S2S time-range is particularly crucial, as the prediction lead time is long enough to implement adaptation measures, and short enough to be of immediate relevance for decision makers.
However, predictions on lead-times beyond approximately 10 days fall into the so-called “weather-climate prediction gap”, with operational forecast models only providing marginal skill. The reasons for this are a range of fundamental challenges, including a limited causal understanding of the underlying sources of predictability.
This project aims to improve S2S forecasts of MED rainfall by taking an innovative, interdisciplinary approach that combines causal discovery algorithms with operational forecast models. This will overcome current limitations of conventional statistical methods to identify relevant sources of predictability and to evaluate modelled teleconnection processes.
The objectives of this project are
(i) to identify key S2S drivers of MED rainfall,
(ii) to systematically evaluate them in forecast models,
(iii) to derive process-based bias corrections to
(iv) boost forecast skill.
However, predictions on lead-times beyond approximately 10 days fall into the so-called “weather-climate prediction gap”, with operational forecast models only providing marginal skill. The reasons for this are a range of fundamental challenges, including a limited causal understanding of the underlying sources of predictability.
This project aims to improve S2S forecasts of MED rainfall by taking an innovative, interdisciplinary approach that combines causal discovery algorithms with operational forecast models. This will overcome current limitations of conventional statistical methods to identify relevant sources of predictability and to evaluate modelled teleconnection processes.
The objectives of this project are
(i) to identify key S2S drivers of MED rainfall,
(ii) to systematically evaluate them in forecast models,
(iii) to derive process-based bias corrections to
(iv) boost forecast skill.
During the fellowship, a focus was set on identifying and quantifying the remote and local sources of S2S predictability of MED winter rainfall variability and extreme events and to evaluate their representation in climate models, which can pave the way for statistical post-processing. Different lines of work were carried out towards the achievement of these goals. In particular a focus was set on the methodological development and application of causality tools to address these issues.
Guided by the research objectives, the project is organized around four WPS.
- Work Package 1 (Driver Detection in Observations)
- Work package 2 (Process-based model evaluation)
- Work package 3 (Statistical post-processing)
- Work package 4 (Dissemination and Communication)
At the end of the fellowship, several research results were achieved which can be grouped into two main parts:
1) Achievements in understanding the drivers of Mediterranean precipitation, including its representation in climate models.
This includes in particular an improved understanding of the role of the stratospheric polar vortex in affection weather and climate in the Mediterranean (as well as other parts of Europe). Moreover, the role of tropical drivers such as the Madden Julien Oscillation, the Quasi-Biennial Oscillation and the El Nino Southern Oscillation have been assessed. These processes have been evaluated using observation/reanalysis data and using seasonal and sub-seasonal hindcast data provided by the European Centre for Medium-range Weather forecasts (ECMWF).
2) Achievements in the development, application and promotion of causality tools to study drivers of regional weather and climate extremes including their use for statistical post-processing
This includes fundamental progress on introducing the concept of causal inference theory to quantify teleconnection pathways to the wider climate science community. The related publication was accompanied by the release of python code on github on the use of causal inference methods. Moreover, this includes the application and development of causal discovery algorithms to study large-scale drivers of regional weather and climate extremes.
Guided by the research objectives, the project is organized around four WPS.
- Work Package 1 (Driver Detection in Observations)
- Work package 2 (Process-based model evaluation)
- Work package 3 (Statistical post-processing)
- Work package 4 (Dissemination and Communication)
At the end of the fellowship, several research results were achieved which can be grouped into two main parts:
1) Achievements in understanding the drivers of Mediterranean precipitation, including its representation in climate models.
This includes in particular an improved understanding of the role of the stratospheric polar vortex in affection weather and climate in the Mediterranean (as well as other parts of Europe). Moreover, the role of tropical drivers such as the Madden Julien Oscillation, the Quasi-Biennial Oscillation and the El Nino Southern Oscillation have been assessed. These processes have been evaluated using observation/reanalysis data and using seasonal and sub-seasonal hindcast data provided by the European Centre for Medium-range Weather forecasts (ECMWF).
2) Achievements in the development, application and promotion of causality tools to study drivers of regional weather and climate extremes including their use for statistical post-processing
This includes fundamental progress on introducing the concept of causal inference theory to quantify teleconnection pathways to the wider climate science community. The related publication was accompanied by the release of python code on github on the use of causal inference methods. Moreover, this includes the application and development of causal discovery algorithms to study large-scale drivers of regional weather and climate extremes.
On the S2S timescale, many high-impact events such as droughts and heat waves take place, which are, however, not well captured either by weather or by climate models. Forecasts on this timescale are extremely challenging since the memory of the atmospheric initial conditions is mostly already lost, but large-scale oceanic boundary conditions only play a small role. There is potential for skilful S2S predictions by considering relatively slow moving components of the climate system that interact with the atmosphere, like sea-surface temperatures, sea-ice or soil-moisture. However, the signal of these processes is small compared to internal atmospheric variability and the exact causal pathways and physical mechanisms are poorly understood and likely not accurately captured by climate models
The proposed research will develop a new framework to improve weather and climate forecasts based on causality tools. This will allow to improve subseasonal to seasonal predictions of Mediterranean rainfall by statistically post-processing dynamical forecasts models through empirical information provided by different teleconnection drivers.
Next to academic researchers, there are several socio-economic sectors which can benefit from the research results
1) operational forecasting centres and their users can be directly informed about research findings.
2) Policy and decision makers will profit from the improved regional forecasts provided. This will help to increase awareness of climate risks and allows for better adaptation strategies.
3) the public will be informed about climate change in general as well as particular regional climate risks.
The proposed research will develop a new framework to improve weather and climate forecasts based on causality tools. This will allow to improve subseasonal to seasonal predictions of Mediterranean rainfall by statistically post-processing dynamical forecasts models through empirical information provided by different teleconnection drivers.
Next to academic researchers, there are several socio-economic sectors which can benefit from the research results
1) operational forecasting centres and their users can be directly informed about research findings.
2) Policy and decision makers will profit from the improved regional forecasts provided. This will help to increase awareness of climate risks and allows for better adaptation strategies.
3) the public will be informed about climate change in general as well as particular regional climate risks.