A Shot of CAFE in the 2021 ASP Summer Colloquium

Share on facebook
Share on twitter
Share on linkedin
Share on email

Nikolaos Mastrantonas | European Centre for Medium-Range Weather Forecasts, Reading, UK.

Pedro Herrera Lormendez | Technische Universitaet Bergakademie Freiberg

Xinjia Hu | Max Planck Institute for the Physics of Complex Systems

___________________________________________________________________________________________

Summer courses are a great experience for PhD students. Learning, sharing knowledge, networking, and participating in active discussions with fellow students and distinguished researchers can boost the enthusiasm, further motivate the participants, and support the development of new ideas. Especially when the summer course is the NCAR ASP Colloquium, it is more than certain that the event will be a very rewarding experience. Even when it takes place online, due to the ongoing pandemic.

This year’s NCAR ASP Colloquium was entitled “The Science of Subseasonal to Seasonal (S2S) Predictions”, a topic of immense importance for our CAFE project. We (Xinjia, Pedro, and Nikos) had the pleasure and honour to participate in this 3-week event. Here we would like to share our experiences and thoughts on the event and present some results of the projects each of us worked on during the colloquium (Yes, this event included, among others, very interesting collaborative research projects 😊).

Initially, let’s give a short history of this year’s ASP Colloquium. As in every year, this event was originally supposed to take place in summer 2020 in Boulder, USA. With the COVID pandemic and the uncertainty surrounding in-person events, the event was postponed until the following summer (2021), with the original intention to take place physically in Boulder. Eventually, as the pandemic continued to persist (definitely longer than the weather and S2S timescales 😔), the event took place in the scheduled time, but remotely, rather than in Boulder. Scheduled time means not only the dates (July-August), but also the time zone (Mountain time; UTC -6)…

And there we were; connecting from our (home) offices in Europe (from evening till late night; 2 a.m. 😉), meeting the other participants and joining the great activities. What a diverse group: people joining from the USA, Latin America, Europe, and Asia. Each of us with our unique experiences, skills, and engaging different points of view, and all of us sharing one common interest: S2S predictions!

Participants at the NCAR ASP Colloquium 2021 sponsored by the NCAR ASP program

The event had a multitude of lectures covering a variety of aspects. We had talks about physical processes, machine learning, the stratosphere, Madden-Julian Oscillation (MJO), El Niño–Southern Oscillation (ENSO), and air-sea coupling. Besides the lectures, a very interesting part of the colloquium was the debate sessions, where each day experts were advocating their (somehow contradicting) opinions on key topics, as for example the usefulness (or not) of Machine Learning compared to physical modelling, and the relative importance of tropical/polar processes in modulating/influencing the midlatitudes’ weather and climate. We also had networking and poster sessions, taking advantage of nice interactive platforms (gathertown) that make the online networking almost as fun and productive as the physical one. The final element of the colloquium was the group projects; each tutorial was carefully chosen and covered important research questions to be explored by the participants. Each of us joined a different project: Xinjia worked on Forecast Verification, Pedro on Weather Regimes, and Nikos on Atmospheric Rivers.

Forecast Verification – Xinjia 

I joined the tutorial with the topic of “S2S verification using climpred”  led by Judith Berner, Abby Jaye and Aaron Spring, We had eight PhD students in our group (Innocent Masukwedza; Irina Statnaia; Jan Wandel; Matt Davison; Paul Buchmann; Pauline Rivoire; and Xinjia Hu), who are all from European research institutes or universities. The tool we were learning and using is a great python package called climpred, which works on prediction hindcast data. Climpred is very powerful because it utilizes xarray to deal with the complicated issue of using prediction hindcasts, namely the several dimensions (initialization, lead time, ensemble member, latitude, longitude, and etc.). 

In our group, we worked on different variables and formed several sub groups. Together with Pauline, we used climpred to estimate prediction skill of precipitation after it is corrected by a post-processing method called quantile mapping. Post-processing is a necessary and useful tool to improve model forecasts for practical applications. Quantile mapping is a very powerful tool for post-processing ensemble model forecasts. We choose this method because precipitation is not normally distributed, and simple bias correction is not very applicable. The theory behind the quantile mapping is to match the cumulative distribution function (CDF) of raw forecasts to the CDF of observations. Therefore we applied the quantile mapping in Climpred to the hindcasts of precipitation in different countries from the ECMWF model. We used several different metrics to estimate the prediction skill. Here I show the continuous ranked probability score (CRPS) as an example. From this figure, we can see that after the correction by quantile mapping, the skill is improved, as the lower the CRPS, the better the skill of the model. 

Comparison of prediction skill of summer average precipitation over the selected region (including Switzerland, France, Germany, Italy and Austria), between the raw model and post-processed model (using quantile mapping) by the metric CRPS (analyzed by Pauline and Xinjia)

Weather Regimes – Pedro

I joined the tutorial entitled “Process-based subseasonal to seasonal diagnostic’s package” along with another four members from the USA (Danni Du, Kyle Lesinger, Kyle Nardi, and Kelsey Malloy). The tutorial was led by Andrew Robertson and Ángel Muñoz from the International Research Institute for Climate and Society (IRI). The sessions were focused on a newly developed python package intended to help understand and apply the science regarding Weather Regimes and their usefulness in S2S predictions. Weather regimes, also known as weather patterns or weather types, are a common tool used to characterize specific weather conditions with persistent large-scale atmospheric circulation patterns over a region. 

We explored the links between these weather regimes and other atmospheric drivers and oscillations (ENSO, Madden-Julian Oscillation, North American Oscillation, Monsoons, atmospheric rivers, etc). Our team studied the North Atlantic and European region where four distinctive patterns were identified. This can be seen in the picture below, where the first row depicts the atmospheric patterns where the red areas represent likely high pressure systems (lack of rain) and blue areas show low pressure dominated areas (higher rainfall). The four weather regimes proved to be strongly interconnected with moisture transport from the ocean towards the continent (second row, green colours) which eventually leads to wet weather over these regions (third row, green colours). Further work is planned to explore the ability of some numerical models in correctly reproducing and predicting such patterns in the S2S predictions which will help us to better understand their relationship with some extreme meteorological events like floods and droughts.

Winter North Atlantic - European weather regimes clustered based on 500 hPa geopotential heights and integrated vertical water vapour (a to d), water vapour transport anomalies linked to each pattern (e to h) and precipitation anomalies associated with the different regimes (i to l). Analysis and figure produced by Danni Du.

Atmospheric Rivers – Nikos

I joined a team with four PhD students from the USA (Alex Mitchell, Deanna Nash, Janak Joshi, and William Scheftic) and we worked on Atmospheric Rivers (ARs) over California. ARs are “rivers in the atmosphere”; relatively long and narrow regions in the sky, transferring large amounts of water moisture. For California in particular, these structures contribute a great amount of the total annual precipitation in very short periods, making them crucial for the water resources, and, at the same time, the formation of extreme rainfalls and floods. Our team focused on the February 2017 period, which was a season with multiple ARs that led to severe rainfall and the failure of the Oroville dam. We analyzed the connections between ARs and different climatic phenomena like ENSO and MJO, the synoptic conditions during February 2017, as well as the predictability of ARs, moisture and precipitation in the sub-seasonal forecasts for that period. 

The below figure refers to the forecasts valid for 7th February, when a major AR occured. It presents the ensemble mean forecast for the Integrated Vapour Transport (IVT) of the ECMWF model at various lead times. Shaded areas refer to the forecast, and contours show the observed situation. As can be noticed, the model is able to predict quite accurately the location and intensity of the IVT, depicting the AR, up to 8 days ahead, with some indication 12 days in advance. The performance of the model in predicting the actual precipitation was rather low for the same period (not shown). Given the high connections between ARs and extreme precipitation over California, the results suggest that the predictability of ARs can serve as a useful proxy for informing about upcoming extreme rainfall at the sub-seasonal time scales. This project was led by Aneesh Subramanian and Mike Deflorio, two very experienced researchers with significant contributions on the topic. Their feedback was of key importance for advancing our work.

Predictability of the Integrated Vapour Transport for 7th February 2017. The results are based on the ensemble mean of the ECMWF model for various lead times. Shaded areas refer to the forecasted data, and contours to the actual observations. The spatial correlations between the forecasted and observed fields are shown on the upper left on each subplot. Analysis and figure produced by William Scheftic.

Concluding remarks

This is our short summary of the ASP 2021 Colloquium. In a nutshell, we gained valuable knowledge, expanded our professional network, and had some nice activities for our evenings and nights 😉. Covid definitely had an impact on this event and is in fact still affecting the course of the colloquium. This time in a positive aspect: The current plan is to conclude the 2021 ASP Colloquium with an in-person event taking place in Boulder in summer 2022. Borrowing terminology from the atmospheric science (e.g heatwaves/droughts) this Colloquium had an onset in 2020, a peak in 2021, and a (forecasted) dissipation in 2022. This gives us a lot of time to continue our contributions to the collaborative projects, and it may even be possible to write up the work in  some peer-reviewed publications 😊 

And… we kept the best for the end: Most of the material of the colloquium is freely available for anyone interested. The lectures and slides can be accessed via Trello boards (board1, board2), and the repositories with the scripts for all projects are available on GitHub. Happy learning everyone 😊 

Share this post

Share on facebook
Share on twitter
Share on linkedin
Share on print
Share on email

More articles