- xwiki:Main.WebHome
- Phase2
- Machine Learning
Changes for document Machine Learning
From version 61.1
edited by s2s_wiki
on 2020/04/14 17:19
on 2020/04/14 17:19
To version 62.1
edited by s2s_wiki
on 2020/04/14 17:19
on 2020/04/14 17:19
Change comment: There is no comment for this version
Content changes
| ... | ... | @@ -76,11 +76,11 @@ |
| 76 | 76 | The current work of National Center of Scientific Research “Demokritos” (NCSRD) with respect to Machine Learning (ML) and Seasonal to Subseasonal (S2S) is based on the temporal and spatial enhancement of S2S predictions with Deep Learning approaches. More specifically, NCSRD locally produces a S2S prediction for Greece (at very high spatial resolution of 5x5 km2) downscaled from a European wide (at 20x20 km2 grid resolution) domain. The simulations are forced by the Climate Forecast System (CFS) model from the National Centers for Environmental Prediction (NCEP) in addition to existing datasets from the S2S database. In order to effectively correct the error of the simulation result, a deep learning approach is tested based on a combination of Convolutional neural networks (CNN) and Recurrent Neural Network (RNN) architectures concerning the space and time domains respectively, focusing on Greece, thus enhancing the accuracy and predictability of longer S2S simulations. |
| 77 | 77 | |
| 78 | 78 | (% style="background-color:#f9f9f9" %) |
| 79 | -|=(% style="font-weight: normal;" %)// |
|
| 79 | +|=(% style="font-weight: normal;" %)//9. ETH Zurich (From Prof. Daniela Domeisen. Updated on April 13, 2020)// | |
| 80 | 80 | Our ongoing project is a collaboration between ETH and the Swiss Data Science Center (SDSC), exploring the subseasonal predictability of stratospheric extreme events using data science methods. |
| 81 | 81 | The upper atmosphere, i.e. the stratosphere at about 12 – 50km above the Earth’s surface, provides increased predictability to Europe after extreme stratospheric events, so-called Sudden Stratospheric Warming (SSW) events. These events can provide skill over Europe for up to several weeks to months, with persistently colder than usual weather over Northern and central Europe. SSW events themselves are currently only possible to predict several days in advance. An extended prediction of SSW events would therefore significantly benefit forecasts at the surface. It is therefore crucial to understand the predictability of the stratosphere itself. |
| 82 | 82 | The main objectives of this project are the use of reanalysis data and the S2S prediction database to extract novel insights from this data using data science tools. A first step will be an improved classification of stratospheric events, allowing for a flexible definition that includes the predictability aspects of these events. For instance, we are building new representations of the polar vortex using non-linear dimension reduction techniques that can later be used in unsupervised clustering algorithms. In a second step, this project aims to classify remote predictors of long-term weather variability. In particular, known predictors for stratospheric and tropospheric variability will be evaluated using data science methods and possible new predictors will be identified. This knowledge is expected to lead to an improved predictability of the weather over Europe on weekly to monthly timescales. |
| 83 | 83 | |
| 84 | 84 | (% style="background-color:#e9e9e9" %) |
| 85 | -|=(% style="font-weight: normal;" %)// |
|
| 85 | +|=(% style="font-weight: normal;" %)//10. ECMWF (From Dr. Michel Rixen. Updated on April 13, 2020)// | |
| 86 | 86 | Machine learning seminars: https://www.ecmwf.int/en/learning/workshops/machine-learning-seminar-series |