Changes for document Machine Learning

From version 71.1
edited by s2s_wiki
on 2020/05/17 10:09
To version 72.1
edited by s2s_wiki
on 2020/12/04 00:37
Change comment: There is no comment for this version

Content changes

... ... @@ -92,3 +92,17 @@
92 92 (% style="background-color:#e9e9e9" %)
93 93 |=(% style="font-weight: normal;" %)//10. ECMWF (From Dr. Michel Rixen. Updated on April 13, 2020)//
94 94 Machine learning seminars:
95 +
96 +**S2S machine Learning competition:**
97 +
98 +
99 +Artificial Intelligence (AI) or Machine Learning (ML) methods for weather forecasting have recently generated a huge interest in the research community. These methods could be used to improve data assimilation methods, physical parameterizations or the post-processing of model outputs. Research on using AI/ML methods as an alternative to dynamical models is also ongoing. The newly formed WMO Research Board has identified Artificial Intelligence (AI) as a key research topic in weather and climate science for the upcoming years. The World Meteorological Organization (WMO) Science & Innovation Department, in collaboration with the Services and Infrastructure, has encouraged holding an open competition to explore new services based on AI methods and applied to the WWRP/WCRP Sub-seasonal to Seasonal S2S project database. Following this recommendation, the WWRP/WCRP S2S Project is planning to organize an Artificial Intelligence (AI)/Machine Learning (ML) competition in 2021. The innovation coming out of this competition will support the goals and actions areas of the S2S and WWRP implementation plans as well as the WCRP strategic plan.
100 +
101 +The main goal of the competition is to encourage the use of AI/ML tools to extract valuable information from the S2S database. The S2S database contains a huge amount of data (more than 100 TBs) which makes it a potentially powerful resource for AI/ML methods to explore possibilities of improving current S2S forecasts through, for instance, improved bias correction and multi-model combination. The competition should provide us more insight on the potential benefit of S2S/AI methods for S2S prediction.
102 +
103 +The current proposal for this competition is to provide the “best possible” forecast of 2-metre temperature and precipitation, at forecast lead times of weeks 3-4 using bi-weekly averages. The forecast domain will be global (on the 1.5-degree spatial grid resolution of the S2S database but limited to land gridpoints) and the forecasts must be issued as tercile probabilities. The verification will be performed using the Ranked Probability Skill Score (RPSS) on 3 domains: Northern Exratropics, tropics and Southern Extratropics. The verification data will come from CPC 2-metre temperature and gridded data. The created software, code, documentation and results will be required to be open source and open access.
104 +
105 +It is envisaged to have 2 rounds. During the first round, hindcasts from all the Thursdays of a given year (e.g. 2020) will be produced. The benchmark will be the ECMWF hindcasts after simple calibration. No data more recent than the forecasts start date should be used. During the second round, the most highly ranked teams from round 1, will compete on real-time forecasts. The competition will be open to AI/ML methods using data from the S2S database, but it will also be open to AI/ML methods using other types of input data, such as large climate model ensembles or reanalysis data. There will be a monetary prize from WMO for the winners.
106 +
107 +The competition is planned to take place in 2021 and will be advertised via the S2S mailing list. Depending on the platform which will be used to run the competition, some of the aspects of the competition, as described above, may be modified.
108 +
This wiki is licensed under a Creative Commons 2.0 license
XWiki Enterprise 6.2.2 - Documentation