1. UK Met Office snow data assimilation
Samantha Pullen
GCW Snow Watch Team – Second Session, June 2016, Columbus, OH.

2. Contents Met Office NWP systems Land surface model
Land surface analysis from satellite
NH snow analysis for global NWP
Snow analysis for UK NWP
Plans for the future
Discussion items
Requirements for NWP
Issues, gaps, GCW priorities

3. Unified model (UM) coupled with JULES land surface model
Met Office NWP systems
Global
17 km, 70 levels, forecast range to 6 days
Main DA hybrid incremental 4D VAR
Land surface DA – NH snow analysis, EKF soil moisture and soil temp, from screen errors in RH and T and ASCAT soil wetness. UK
Variable resolution 4 km down to 1.5 km, 70 levels, forecast range to 36 hours
Main DA incremental 3D VAR
No land surface DA as yet
Unified model (UM) coupled with JULES land surface model
© Crown copyright Met Office

4. Land surface model: JULES
Multi-layer snow module (Essery)
From Best et al., 2011 (doi: /gmd )
Tiled – sub-grid heterogeneity – fluxes for each surface type:
5 Plant functional types:
Broadleaf trees
Needleleaf trees
C3 grass
C4 grass
Shrubs
Plus:
Urban
Inland water
Bare soil
Land ice
Prognostic snow variables:
Snow depth
Snowpack bulk density
Number snow layers
Within layers:
Thickness
Ice mass
Liquid mass
Temperature
Density
Grain size
Operational UKV, global implementation winter 2016.
© Crown copyright Met Office

5. Land Surface Analysis Remote Sensing Observations
Climatologies
Data Assimilation
Increasing obs processing complexity
LAI, Albedo
SST, Sea Ice*
Snow
Soil Moisture & Temp
* From OSTIA
Reconfiguration to different domains
Develop Land Surface DA for different domains
© Crown copyright Met Office

6. Gridbox Fractional Cover (Obs)
06 UTC Analysis
Daily NH snow analysis
NESDIS IMS 4km
Relative resolutions
Gridbox Fractional Cover (Obs)
Relating fc and depth +
Time delay check
T+6 from previous cycle
Snow amount (kgm-2)
Using T+6 from previous day
Remove snow
© Crown copyright Met Office

7. UK snow forecasting
The UK does not experience regular widespread snowfall except in the Highlands of Scotland
Tends to be transient, often wet, shallow, multiple snowfall/melt cycles in one season.
Low frequency, but high impact event – accurate analyses and forecasts of snowfall and lying snow extremely important
Currently no snow observations assimilated in UK model (UKV)
December 2010
Comparison of model vs observed (SYNOP) snow depth shows considerable scatter – potential for improvement to freely-evolving snow amount

8. Snow DA for the UK NWP system
In development….
Data source
Snow depth values
Ground-based obs of snow depth, and state of ground (snow or no snow) from synoptic network
SD where reported
0 m SD from snow-free state of ground reports
0 m SD from snow-free pixels
0.05 m SD from snow-covered pixels where model snow-free
Satellite-derived snow cover from H-SAF (MSG-SEVIRI) daily product
Model first-guess SD
Optimal Interpolation
Snow depth analysis

9. H-SAF and UKV comparisons with SYNOP
H-SAF vs SYNOP agreement well over 90% most of period
Often low coincidence of SYNOP with classified pixels - beware
UKV vs SYNOP agreement high but not as high as H-SAF
Large reduction in agreement rate 17-19th (both comparisons).
Rapidly changing snow cover, timing of obs relative to falling snow, model evolution.
SYNOP too sparse for detailed validation of snow edge.
H-SAF vs SYNOP
UKV vs SYNOP
Overall results
H-SAF closer to ground truth than UKV
Where H-SAF and UKV differ, can infer that UKV errors proportionally greater than H-SAF errors on average
Assimilation of H-SAF into UKV will add value
(Repeated using common set of SYNOP for direct comparison) Ra Ro Ru
H-SAF vs UKV
80.82
6.16
13.05
H-SAF vs SYNOP
89.38 (89.10)
0.64 (0.33)
9.98 (10.57)
UKV vs SYNOP
82.65 (85.64)
4.86 (3.83)
12.49 (10.53)

10. 17th December 2010 UKV underestimated snow
Some (very) preliminary results…
17th December 2010 UKV underestimated snow
Prototype system developed to test assimilation of:
Snow depths from SYNOP reports
Snow cover from H-SAF
Separately at first – can explore sensitivities, ob errors, correlation length scales, spatial densities etc…
© Crown copyright Met Office

11. Using SYNOP snow depth observations
Bg snow amount
Using SYNOP snow depth observations
Snow depths where snow present
Zero depths diagnosed from state of ground reports
Analysis increments
Analysis
BGE correlation length scale 5.5 km - variable
Limits horizontal influence of obs to ~ 50 km
Obs error SD 0.04 m
Bg error SD 0.03 m
© Crown copyright Met Office

12. Using H-SAF snow cover observations
Diagnosed snow depth obs (0.05 m)
All usable pixels
Bg snow amount
Using H-SAF snow cover observations
0.05 m snow depth where H-SAF snow cover and UKV snow-free
Zero depths diagnosed from snow-free H-SAF pixels
Analysis increments
Analysis
BGE correlation length scale 5.5 km - variable
Limits horizontal influence of obs to ~ 50 km
Obs error SD 0.08 m
Bg error SD 0.03 m
© Crown copyright Met Office

13. Longer term...
Station snow depth assimilation in global model
+ dense national networks now available on GTS
+ zero snow reports – better snow line
Use of complementary satellite observations
Wet snow extent from SAR (SEN3APP project)
Snow water equivalent (SWE) from passive microwave (AMSR-2 v2?)
Novel ob sources
WOW (crowd sourced)
GPS receivers
EKF/EnKF? Microwave radiances with snow emission model?
Met Office may adopt LIS (NASA)
Hard for any single (remote-sensed) snow dataset to fulfil requirements for NWP assimilation – best approach may be to exploit the best features of a number of products to use in a complementary way.
© Crown copyright Met Office

14. Coverage - depends on model domain, global/NH common
Requirements for NWP
Continuity - operational robustness, long-term security to justify development work, succession of satellite sources...
Temporal resolution - daily sufficient for snow change timescales. Complementary data sources can have lower frequency
Level of derivation – preferably not assimilation products themselves, e.g. contain some model information (not consistent), contain ground-based obs (not suitable if model already assimilates)
Coverage - depends on model domain, global/NH common
Cloud cover - how extensively does it affect product? High temporal sampling can mitigate to some extent. Multi-sensor approach can allow gap-filling. Is it the only data source?
Errors - well-defined and documented, quality flags disseminated with product. SC 15-20%, SWE 10mm. Has to improve forecast/analysis to be used.
Availability in near-real-time - daily product within half a day, 6-hourly within 3 hours
Spatial resolution - guided by model resolution, doesn’t have to match. Higher resolution allows fractional cover calculation on model grid. Too low, representativity issues.

15. Issues, gaps, GCM priorities (from NWP perspective)
Satellite-derived SWE (nrt, global)
Improved uncertainty – WMO OSCAR Requirements – 20 mm min.
forest
Observations of density, grain size
Snow cover – complementary data sources (forest, clouds)
Highly derived, combined data sources less useful for NWP
Ground-based snow depth
Improved exchange of dense national network data
Routine reporting of zero snow (mandated)
Support for operationally robust, long-term datasets (succession planning between instruments)
© Crown copyright Met Office

16. Questions?
Thank you