An operational regional ice-ocean prediction system (RIOPS) at 4-5km resolution in the Arctic F. Dupont, J.-F. Lemieux, G. Smith, F. Roy, C. Beaudoin, Y. Lu, S. Higginson, J. Lei, J. Xu, F. Davidson, G. Garric, R. Bourdalle-Badie and other CONCEPTS collaborators

Dept of National Defence Outline of the talk Background Domain configuration Forced runs of the forecasting components (NEMO-CICE overall performances) RIOPS system (+spectral nudging) RIOPS skills (ice against RIPS and ocean against GIOPS) RIOPS added value (tides, landfast ice, coastal processes)

1. Background -DFO short-term forecasting needs: -SAR (link to coast guards), spills and other Lagrangian applications (dead whales, ballast waters...etc), pollution in marine protected areas; -DFO long-term prospective/analysis needs: -State of the Ocean Report (annual): temperature evolution; -fisheries management (bottom temperature); -EC short-term forecasting needs: -ice concentration, ice pressure, Icebergs drift, spills and other emergency response management; - Improving over RIPS (sea-ice alone model with a simple slab ocean); -EC seasonal forecasting needs: -ice concentration and ice thickness. -DND: state of the ice-ocean in general ECCC (Environment and Climate Changes Canada), DFO (Fisheries and Oceans) and DND (national Defense) common goals

2. Domain configuration

Domain presentation: 1580x1817x procs Domain is as in RIPS2.0 (32 procs). Extracted from ORCA12 (Mercator) with the north fold stitched back. Regional CONCEPTS domain (CREG). Resolution is maximum near the artificial pole over northern Canada at 1.8 km and minimum along the Atlantic northern boundary (8.2km) Covers part of North Atlantic (27N), the whole Arctic Ocean. Typical location of the north fold on the ORCA family grid

First Rossby radius of deformation Red: good resolution for eddies Blue- yellow: eddies under- resolved

3. Forced runs

~4 X 6-year forced simulations Run H02 Run H03-H05Run H06Run H05f Vertical turbulence 1.5 Gaspar Vertical turbulence k- epsilon Vertical turbulence 1.5 Gaspar LIM2 1 ice category VP 20 pseudo- iterations CICE (los Alamos) 10 ice thickness cat EVP 900 sub- iterations CICE (los Alamos) 10 ice thickness cat EVP 900 sub- iterations CICE (los Alamos) 10 ice thickness cat EVP 900 sub- iterations No tides Active tidesNo tides Atmospheric forcing from 33km CGRF + GEWEX correction on SW/LW radiation (Smith et al. 2013, QJRMS) Our reference run for RIOPS!

Total ice area and volume in the Arctic H05-H05f: change in physics of ocean mixing causes a slight loss of ice H06-H05: introduction of tides and increase in ice resistance to deformation causes a more pronounced loss

September (minimum) total ice area and volume in the Arctic H05-H05f: change in physics of ocean mixing causes a slight loss of ice H06-H05: introduction of tides and increase in ice resistance to deformation causes a more pronounced loss

Ice thickness in Fall 2007 relative to ICESat ~Config GIOPS!! ~Config RIOPS!!

cm/s Obs-based NSIDC product (neg. bias) ORCA025Run H05f cm/s Run H05ORCA12Run H06 Very large ice-velocity in CREG12 (small improvements in H04-H05 by increasing ice-ocean drag and decreasing ice-atm drag), also increasing Cf reduces the ice-velocity positive bias in CREG12 ~Config GIOPS!! ~Config RIOPS!!

Ice velocity bias relative to IABP (international Arctic Buoy Program) Validation of the idea that the 1.5 Turbulence scheme (H05f) can help reducing the velocity bias. Bias is closer to zero, although more positive by the end of the period

What you get from the 3D model (CREG12-H05f) averaged laterally over the southern BG. The heat in the Pacific Summer Water quickly dies off, the heat of the Atlantic layer diffuses up.

4. RIOPS: Replacing RIPS

Other approaches not tested in CREG12 but in CREG025 that are implemented in RIOPS: -Grounded landfast ice represented by a basal stress parametrization (Lemieux et al. 2015) -Increase in shear and tension resistance (Lemieux et al., in preparation) improves the representation of land-locked ice (another form of landfast ice). Increase in tensile stress Increase in shear resistance Ice ridge touching the ground Principal ice stress diagram

Choice of model parameter for RIOPS RIOPS is based on CREG12 hindcast 05f where the vertical physics of GIOPS is used (TKE 1 equation versus k-eps). Tides are included. Ndte=900 in CICE (better convergence for the ice velocity solver), basal stress activated, eccentricity=1.8, resistance to tension=5%. RIPSA inserted with amin=0.1 and amax=0.83. Continuous cycle (with tides) where we update the reference solution (GIOPS-A) every day to which the model is nudged For the final cycles, RIOPS-PA is run for 7 days at a time (saves time in queue!), Therefore RIOPS-F is run for 48h every 7 days.

A 24h run from previous restart is nudged to the GIOPS analysis using spectral nudging (Thompson et al., 2006) forecast 48h forecast forecast 6Z forecast forecast 48h forecast RIOPS Pseudo- Analysis (RIOPS-PA) RIOPS Forecast 4 x day (RIOPS-F) RIOPS Prediction System 12Z 18Z 3 Components: RIPS 3DVar ice analysis, Pseudo-Analysis and 48hr Forecasts

Spectral nudging implementation in NEMO (Thompson et al., 2006) dT/dt final = (dT/dt) adv+diff+thermo + 1/tau Where <> means spatial and time filter (RIOPS only uses the spatial filter= 20 Shapiro passes, time filter is disabled). The same is done for salinity.

Comparison of velocity on GIOPSGIOPS interpolated on CREG12 RIOPS Use of spectral nudging (in space) towards GIOPS with timescale of 1 day

5. RIOPS evaluations For the final cycles, RIOPS-PA is run for 7 days at a time (saves time in queue!), Therefore RIOPS-F is run for 48h every 7 days. Ice evaluations against RIPS Ocean evaluations against GIOPS Tides

Total ice area and volume in CREG12 domain from RIPSA, RIOPS- PA and GIOPS- A+F The sea-saw features are related to correction by the analysis to the model ice concentration. i.e., both GIOPS and RIOPS melts too much ice in general (too much SW radiation), but growth looks fine

DGLA scores against RIPS-A (2.2) valid at 00Z for 48h persistence of RIPS-A (2.2), RIPS- 2.2-F (48h), RIOPS-F (38h), GIOPS-F (48h). DGLA definition: where the analysis changes by more than 10% over the lead time, the difference between model-analysis is computed. Bias and RMS are then derived for the whole region (here whole domain). Ice concentration error metric 1: DGLA error (against own analysis) RIPS shows a positive bias during winter (too much ice growth) relative to GIOPS and RIOPS (and our analysis)

Metric 2: IMS scores valid at 00Z for 48h persistence of RIPS-A (2.2), RIPS-A (3DVar), RIPS-2.2-F (48h), RIOPS-F (48h), GIOPS-F (48h). RIOPS forecast skills roughly equivalent to that of RIPS, slightly larger bias in Oct- Dec (but closer to your analysis), Better PCT than GIOPS in melt period. IMS (Ice Mapping Service) from NIC (U.S.) provides a ice/no ice field at 4km. A contingency table is derived using an ice concentration threshold from which one can derive: -PCT=(Hit ice+Hit water)/all -Frequency bias=(Hit ice+False alarm)/(Hit ice+miss) IMS iceIMS no ice Forecast iceHit iceFalse alarm Forecast no iceMissHit Water

Ice velocity comparison against IABP data (IABP=International Arctic Buoy Program) RIOPS is slightly more negatively biased than GIOPS or RIPS but the standard deviation is much improved. This is probably due to the increased in resistance to shear and tension. bias=  ( ǁ V m ǁ - ǁ V o ǁ )/n RMSE=sqrt(  ( ǁ V m -V o ǁ 2 )/n) Values are in m/s GIOPSRIPSRIOPS bias RMSE

Ice velocity comparison against IABP data RIPSRIOPS

Conclusions so far in forecast skills: closer to 3D-Var RIPS analysis than RIPS (no deterioration) slightly more negatively biased in ice-velocity but better RMS (negative bias can be fixed) close to GIOPS in general except in SLA

6. RIOPS goodies (things not in RIPS) Forecasts its own currents (+tides) Full 3D ocean (T&S) Landfast ice representation Tides in ice

Summer (July) 2014 tidal ellipse (M2) in the ice Winter (January) 2015 tidal ellipse (M2) in the ice Note the dead zone due to the landfast ice parametriza tion of Lemieux et al. (2015)

Landfast ice detection from mean ice speed over 7 days from 3- hourly averaged output. RIOPS is the more realistic of the two. RIOPSRIPS

Conclusions Evaluations: RIOPS better than RIPS for DGLA metric RIPS slightly better for IMS metrics but RIOPS closer to CMC analysis For ice drift, RIOPS has better RMS but poorer bias Advantages and new products: RIOPS has tides (impact on small-scale ice motion, requested by CIS, coastal oceanography) Landfast ice is better represented RIOPS provides down-scaled ocean features (better coastal water masses expected, but no evaluation done) for DFO, CIS, DND, EER. Milestone toward coupling with RDPS/REPS Better positioned for an operational support to the YOPP field experiment

Users and dissemination Users: CIS: ice forecasts (METAREA), 3D currents for iceberg drift DFO: 3D T&S and currents EC: 3D currents for EER, R&D toward coupling, input to surface wave model, OBCs for coastal systems DND: 3D T,S,P for speed of sound, general output for operations in North Atlantic & Arctic. Industry: 3D currents for iceberg drift (e.g: Hybernia) NOAA/NCEP: ftp push (?) Dissemination: internally (CIS, EC, EER), through pegasus (DFO), DND: possible push to Gagetown, industry through Datamart