東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Sub-seasonal Forecasting Using Large...

Weyn-Vanhentenryck, Jonathan.

FindBook

Google Book

Amazon

博客來

Sub-seasonal Forecasting Using Large Ensembles of Data-driven Global Weather Prediction Models.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Sub-seasonal Forecasting Using Large Ensembles of Data-driven Global Weather Prediction Models./
作者:	Weyn-Vanhentenryck, Jonathan.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2020,
面頁冊數:	141 p.
附註:	Source: Dissertations Abstracts International, Volume: 82-03, Section: B.
Contained By:	Dissertations Abstracts International82-03B.
標題:	Atmospheric sciences. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28001993
ISBN:	9798664710199

Sub-seasonal Forecasting Using Large Ensembles of Data-driven Global Weather Prediction Models.
Weyn-Vanhentenryck, Jonathan.

Sub-seasonal Forecasting Using Large Ensembles of Data-driven Global Weather Prediction Models. - Ann Arbor : ProQuest Dissertations & Theses, 2020 - 141 p.

Source: Dissertations Abstracts International, Volume: 82-03, Section: B.

Thesis (Ph.D.)--University of Washington, 2020.

This item must not be sold to any third party vendors.

The current state-of-the-art in numerical weather prediction (NWP) is to generate probabilistic forecasts using large ensembles consisting of equally-likely realizations of future weather. Such large ensembles, however, require significant computational resources. I have developed a purely data-driven weather prediction model using convolutional neural networks (CNNs) trained on globally-gridded analysis of the atmosphere. While this model only evolves a small set of key atmospheric variables and does not quite approach the performance of state-of-the-art NWP models, it has a number of desirable properties: 1) by using data remapped to a cubed sphere, our CNN model is a closed system which can be integrated forward indefinitely, 2) our model remains stable indefinitely, producing realistic atmospheric states and even a correct seasonal cycle when allowed to run freely for up to a year, and 3) our model executes extremely quickly, requiring only one tenth of a second for a 1-week forecast on a global 1.5-degree grid. Taking advantage of the efficient computation, I designed a large 320-member ensemble of CNNs using both initial-condition perturbations and stochastic model perturbations yielded by the internal randomness of training multiple CNNs. While the ensemble is under-dispersive, ensemble mean forecasts notably outperform single deterministic data-driven forecasts, but still lag the skill of the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble forecasts. Armed with an efficient large ensemble, I then target predictions on the sub-seasonal-to-seasonal (S2S) time frame, or about 2 weeks to 2 months out, where traditional NWP models struggle due to a lack of information from initial conditions and difficulty outperforming persistence forecasts of slowly-evolving earth system components such as ocean sea surface temperatures. Ensemble mean forecasts of 2-meter temperature and 850-hPa temperature from my CNN ensemble clearly outperform persistence forecasts across the S2S time frame. Evaluating full ensemble probabilistic forecasts using the continuous ranked probability score and the ranked proba- bility skill score, I demonstrate that my CNN ensemble provides nearly universal useful S2S skill relative to persistence and climatology, notably over most land masses instead of just over oceans. My ensemble even compares well with the ECMWF S2S ensemble, matching or exceeding the latter at forecast lead times of weeks 5-6, and particularly excels during the boreal spring and summer months, where the ECMWF ensemble is weakest. While my CNN ensemble shows great promise as an S2S forecasting tool, many opportunities remain to further improve it, especially for its predictions of long-term climate variability including the Madden-Julian Oscillation and the El Nino-Southern Oscillation.

ISBN: 9798664710199Subjects--Topical Terms:

3168354
Atmospheric sciences.
Subjects--Index Terms:

Convolutional neural network

Sub-seasonal Forecasting Using Large Ensembles of Data-driven Global Weather Prediction Models.
LDR:04034nmm a2200361 4500 001 2283385
005 20211029101434.5
008 220723s2020 ||||||||||||||||| ||eng d
020 $a 9798664710199
035 $a (MiAaPQ)AAI28001993
035 $a AAI28001993
040 $a MiAaPQ $c MiAaPQ
100 1 $a Weyn-Vanhentenryck, Jonathan. $3 3562345
245 1 0 $a Sub-seasonal Forecasting Using Large Ensembles of Data-driven Global Weather Prediction Models.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2020
300 $a 141 p.
500 $a Source: Dissertations Abstracts International, Volume: 82-03, Section: B.
500 $a Advisor: Durran, Dale R.
502 $a Thesis (Ph.D.)--University of Washington, 2020.
506 $a This item must not be sold to any third party vendors.
520 $a The current state-of-the-art in numerical weather prediction (NWP) is to generate probabilistic forecasts using large ensembles consisting of equally-likely realizations of future weather. Such large ensembles, however, require significant computational resources. I have developed a purely data-driven weather prediction model using convolutional neural networks (CNNs) trained on globally-gridded analysis of the atmosphere. While this model only evolves a small set of key atmospheric variables and does not quite approach the performance of state-of-the-art NWP models, it has a number of desirable properties: 1) by using data remapped to a cubed sphere, our CNN model is a closed system which can be integrated forward indefinitely, 2) our model remains stable indefinitely, producing realistic atmospheric states and even a correct seasonal cycle when allowed to run freely for up to a year, and 3) our model executes extremely quickly, requiring only one tenth of a second for a 1-week forecast on a global 1.5-degree grid. Taking advantage of the efficient computation, I designed a large 320-member ensemble of CNNs using both initial-condition perturbations and stochastic model perturbations yielded by the internal randomness of training multiple CNNs. While the ensemble is under-dispersive, ensemble mean forecasts notably outperform single deterministic data-driven forecasts, but still lag the skill of the European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble forecasts. Armed with an efficient large ensemble, I then target predictions on the sub-seasonal-to-seasonal (S2S) time frame, or about 2 weeks to 2 months out, where traditional NWP models struggle due to a lack of information from initial conditions and difficulty outperforming persistence forecasts of slowly-evolving earth system components such as ocean sea surface temperatures. Ensemble mean forecasts of 2-meter temperature and 850-hPa temperature from my CNN ensemble clearly outperform persistence forecasts across the S2S time frame. Evaluating full ensemble probabilistic forecasts using the continuous ranked probability score and the ranked proba- bility skill score, I demonstrate that my CNN ensemble provides nearly universal useful S2S skill relative to persistence and climatology, notably over most land masses instead of just over oceans. My ensemble even compares well with the ECMWF S2S ensemble, matching or exceeding the latter at forecast lead times of weeks 5-6, and particularly excels during the boreal spring and summer months, where the ECMWF ensemble is weakest. While my CNN ensemble shows great promise as an S2S forecasting tool, many opportunities remain to further improve it, especially for its predictions of long-term climate variability including the Madden-Julian Oscillation and the El Nino-Southern Oscillation.
590 $a School code: 0250.
650 4 $a Atmospheric sciences. $3 3168354
650 4 $a Artificial intelligence. $3 516317
653 $a Convolutional neural network
653 $a Deep learning
653 $a Numerical weather prediction
653 $a S2S forecasting
653 $a Weather forecasting
690 $a 0725
690 $a 0800
710 2 $a University of Washington. $b Atmospheric Sciences. $3 3174193
773 0 $t Dissertations Abstracts International $g 82-03B.
790 $a 0250
791 $a Ph.D.
792 $a 2020
793 $a English
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28001993