Status of Diva online as VRE application

Transcription

1 A. Barth, C. Troupin, S. Watelet & J.-M. Beckers GHER - ULiège Status of Diva online as VRE application SeaDataCloud 1st Plenary meeting

2 WP WP11.2 WP11.3 WP11.5 development of DIVA online (VRE) Produce standard climatological data products for the global ocean and European Seas basins Development of new products Training on data products generation

3 Diva: from in situ data to gridded fields

4 divand: generalised, n-dimensional interpolation : Octave/MATLAB 2016: Julia faster, better, stronger

5 User interfaces: Jupyter notebooks and WPS

6 Notebooks: interactive computational environments Notebooks combine: 1 code fragments that can be executed, 2 text for the description of the application and 3 figures illustrating the data or the results.

7 Notebooks: interactive computational environments Notebooks combine: 1 code fragments that can be executed, 2 text for the description of the application and 3 figures illustrating the data or the results. Digital Playground Data Story Telling Computational Narratives

8 Notebooks: interactive computational environments Notebooks combine: 1 code fragments that can be executed, 2 text for the description of the application and 3 figures illustrating the data or the results. Interactive notebooks: Sharing the code, Nature (2014) interactive-notebooks-sharing-the-code

9 divand in a notebook Example online

10 Notebooks in other initiatives

11 divand in the VRE with jupyterhub WP Management of multiple instances of the single-user Jupyter notebook server Demo: (deployed at CINECA)

12 (Brand) New product Velocity field from HF radar

13 H2020 SeaDataCloud call: emphasis on coastal data WP11.3 Data: SOCIB HF radar in the Ibiza Channel

14 Ibiza Channel, data from SOCIB

15 2 antennas located in Ibiza and Formentera

16 Each antenna provide radial currents

17 Each antenna provide radial currents

18 Total velocities are derived on a regular grid

19 New product: currents hypothetical measurement analyzed field Analysis of radial currents to derive total currents Observation operator links the radial currents of the different radar sites

20 Formulation: couple velocity components Norm : φ 2 = Ω (α 2 φ : φ + α 1 φ φ + α 0 φ 2 ) dω Cost function: J( u) = u 2 + v 2 + N ( u ı p ı u rı ) 2 u = (u, v) p ı = normalized vector pointing toward the correspond HF radar site of the ı-th radial observation u rı ı=1 ϵ 2 ı

21 Coastline as a boundary condition ( u n = 0) Cost function (OFF) J bc ( u) = 1 ϵ 2 bc Ω ( u n) 2 ds

22 Coastline as a boundary condition ( u n = 0) Cost function (ON) J bc ( u) = 1 ϵ 2 ( u n) 2 ds bc Ω

23 Low horizontal divergence of currents ( n = 0) Cost function (OFF) J div ( u) = 1 ϵ 2 div Ω ( u) 2 dx

24 Low horizontal divergence of currents ( n = 0) Cost function (ON) J div ( u) = 1 ϵ 2 ( u) 2 dx div Ω

25 3D analysis: longitude, latitude and time Include the data the hour before and after Temporal correlation length Coriolis force Coriolis force and geostrophically balanced mean flow u t v t = fv g η x = fu g η y f = Coriolis frequency η = sea surface elevation

26 Cross validation In 30 current maps with the best coverage, some data points are marked as missing (for both sites)

27 Test cases: more constrains (physics) included Case 2D 2D_bc 2D_iv 3D 3D_Coriolis 3D_Coriolis_geo Description classical 2D-analysis (longitude, latitude) as 2D, but with boundary conditions as 2D, but imposing small horizontal divergence 3D-analysis (longitude, latitude, time) 3D-analysis with the Coriolis force 3D-analysis with the Coriolis force and the surface pressure gradient

28 Skill score S(Case) = 1 MSE(Case) MSE(2D) The 2D case is the base-line for computing the relative improvement MSE(C) is the mean square error (relative to the cross-validation dataset) If S = 0: reconstruction as good/bad as the base-line If S = 1: reconstruction matches perfectly the validation dataset.

29 Comparison: increased skill with more constrains Case RMS Skill Optimal score parameter(s) 2D ϵ 2 = D_bc ϵ 2 =0.0001, ϵ 2 bc=10 2D_div ϵ 2 =9.799e-05, ϵ 2 div=2.778e+08 3D ϵ 2 =0.1219, lent=6904 3D_Coriolis ϵ 2 =5.673e-05, ϵ 2 Cor=9.207e-05 3D_Coriolis_geo ϵ 2 =5.37e-05, ϵ 2 Cor=5.65e-05, ratio=26.46

30 Proposed training activities WP11.5 Diva workshop: 2 6 April 2018, Liège, Belgium

31 Proposed training activities WP11.5 Diva workshop: 2 6 April 2018, Liège, Belgium Diva pre-workshop: 18 October 2017, Athens, Greece (4PM?)

32 Questions?

33 Questions? K n,m (r) = c n,m (2π) n 2 2(1 m) r 2 n 2 = c n,m (2π) n 2 2(m 1) r 4 n 2 = where 0 0 J n 2 (kr)k n J n 4 (kr)k n π(m 1) c n 2,m 1 Kn 2,m 1 (r) c n,m n is the dimension m is the highest derivative K n,m is the Kernel ( d dk 1 (1 + k 2 ) m 1 k (1 + k 2 dk ) m 1 ) dk J ν (r) is the Bessel function of first kind or order ν

34 Questions?