scholarly journals Boosting performance in machine learning of geophysical flows via scale separation

2020 ◽  
Author(s):  
Davide Faranda ◽  
Mathieu Vrac ◽  
Pascal Yiou ◽  
Flavio Maria Emanuele Pons ◽  
Adnane Hamid ◽  
...  
Keyword(s):  
Machine Learning ◽  
Neural Networks ◽  
Sea Level ◽  
Recurrent Neural Networks ◽  
Multiple Scales ◽  
Geophysical Flows ◽  
Small Scale ◽  
Pressure Data ◽  
Sea Level Pressure ◽  
Level Pressure

Abstract. Recent advances in statistical and machine learning have opened the possibility to forecast the behavior of chaotic systems using recurrent neural networks. In this article we investigate the applicability of such a framework to geophysical flows, known to involve multiple scales in length, time and energy and to feature intermittency. We show that both multiscale dynamics and intermittency introduce severe limitations on the applicability of recurrent neural networks, both for short-term forecasts, as well as for the reconstruction of the underlying attractor. We suggest that possible strategies to overcome such limitations should be based on separating the smooth large-scale dynamics from the intermittent/small-scale features. We test these ideas on global sea-level pressure data for the past 40 years, a proxy of the atmospheric circulation dynamics. Better short and long term forecasts of sea-level pressure data can be obtained with an optimal choice of spatial coarse grain and time filtering.

scholarly journals Enhancing geophysical flow machine learning performance via scale separation

2021 ◽  
Vol 28 (3) ◽  
pp. 423-443
Author(s):  
Davide Faranda ◽  
Mathieu Vrac ◽  
Pascal Yiou ◽  
Flavio Maria Emanuele Pons ◽  
Adnane Hamid ◽  
...  
Keyword(s):  
Machine Learning ◽  
Neural Networks ◽  
Sea Level ◽  
Recurrent Neural Networks ◽  
Multiple Scales ◽  
Learning Performance ◽  
Small Scale ◽  
Pressure Data ◽  
Sea Level Pressure ◽  
Level Pressure

Abstract. Recent advances in statistical and machine learning have opened the possibility of forecasting the behaviour of chaotic systems using recurrent neural networks. In this article we investigate the applicability of such a framework to geophysical flows, known to involve multiple scales in length, time and energy and to feature intermittency. We show that both multiscale dynamics and intermittency introduce severe limitations to the applicability of recurrent neural networks, both for short-term forecasts as well as for the reconstruction of the underlying attractor. We suggest that possible strategies to overcome such limitations should be based on separating the smooth large-scale dynamics from the intermittent/small-scale features. We test these ideas on global sea-level pressure data for the past 40 years, a proxy of the atmospheric circulation dynamics. Better short- and long-term forecasts of sea-level pressure data can be obtained with an optimal choice of spatial coarse graining and time filtering.

scholarly journals The Application of Barnes Filter to Positioning the Center of Landed Tropical Cyclone in Numerical Models

2018 ◽  
Vol 2018 ◽  
pp. 1-9 ◽  
Author(s):  
Haibo Zou ◽  
Shanshan Wu ◽  
Xueting Yi ◽  
Nan Wu
Keyword(s):  
Numerical Model ◽  
Tropical Cyclone ◽  
Sea Level ◽  
Numerical Models ◽  
Small Scale ◽  
Model Output ◽  
Sea Level Pressure ◽  
Level Pressure ◽  
Distance Difference

After a tropical cyclone (TC) making landfall, the numerical model output sea level pressure (SLP) presents many small-scale perturbations which significantly influence the positioning of the TC center. To fix the problem, Barnes filter with weighting parameters C=2500 and G=0.35 is used to remove these perturbations. A case study of TC Fung-Wong which landed China in 2008 shows that Barnes filter not only cleanly removes these perturbations, but also well preserves the TC signals. Meanwhile, the centers (track) obtained from SLP processed with Barnes filter are much closer to the observations than that from SLP without Barnes filter. Based on the distance difference (DD) between the TC center determined by SLP with/without Barnes filter and observation, statistics analysis of 12 TCs which landed China during 2005–2015 shows that in most cases (about 85%) the DDs are small (between −30 km and 30 km), while in a few cases (about 15%) the DDs are large (greater than 30 km even 70 km). This further verifies that the TC centers identified from SLP with Barnes filter are more accurate compared to that directly obtained from model output SLP. Moreover, the TC track identified with Barnes filter is much smoother than that without Barnes filter.

scholarly journals Polar Mesoscale Cyclones in the Northeast Atlantic: Comparing Climatologies from ERA-40 and Satellite Imagery

2006 ◽  
Vol 134 (5) ◽  
pp. 1518-1533 ◽  
Author(s):  
Alan Condron ◽  
Grant R. Bigg ◽  
Ian A. Renfrew
Keyword(s):  
Sea Level ◽  
Satellite Imagery ◽  
Detection Algorithm ◽  
Ocean Modeling ◽  
Small Scale ◽  
Sea Level Pressure ◽  
Mean Sea Level ◽  
Northeast Atlantic ◽  
Level Pressure ◽  
Mesoscale Cyclones

Abstract Polar mesoscale cyclones over the subarctic are thought to be an important component of the coupled atmosphere–ocean climate system. However, the relatively small scale of these features presents some concern as to their representation in the meteorological reanalysis datasets that are commonly used to drive ocean models. Here polar mesocyclones are detected in the 40-Year European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis dataset (ERA-40) in mean sea level pressure and 500-hPa geopotential height, using an automated cyclone detection algorithm. The results are compared to polar mesocyclones detected in satellite imagery over the northeast Atlantic, for the period October 1993–September 1995. Similar trends in monthly cyclone numbers and a similar spatial distribution are found. However, there is a bias in the size of cyclones detected in the reanalysis. Up to 80% of cyclones larger than 500 km are detected in MSL pressure, but this hit rate decreases, approximately linearly, to ∼40% for 250-km-scale cyclones and to ∼20% for 100-km-scale cyclones. Consequently a substantial component of the associated air–sea fluxes may be missing from the reanalysis, presenting a serious shortcoming when using such reanalysis data for ocean modeling simulations. Eight maxima in cyclone density are apparent in the mean sea level pressure, clustered around synoptic observing stations in the northeast Atlantic. They are likely spurious, and a result of unidentified shortcomings in the ERA-40 data assimilation procedure.

Sign in / Sign up

Export Citation Format

Share Document