Model Development Activities at ESSO-NCMRWF

1
Model Development Activities at ESSO-NCMRWF

• E N Rajagopal

2
Unified Model at NCMRWF (NCUM) Same Model for
Global/Regional/Mesoscale! seamless model
1.5 km grid up to 48 hr forecast
12 km grid up to 48 hr forecast
25 km global grid up to 168 hr forecast
3
Outline of Talk

• Current Status
• New Developments
• Model Diagnostics Evaluation
• Future Plans

4
Status of Observations at NCMRWF
5
Observations Assimilated in NCUM
Observation file Observation details
Surface.obstore Land SYNOP, Ship SYNOP, Mobile, AWS, BUOY
Sonde.obstore TEMP (Land Ship), PILOT, DROPSONDE, Wind Profilers
Aircraft.obstore AIREP, AMDAR
Satwind.obstore GOES, Meteosat, MTSat, INSAT-3D, MODIS, MetOp NOAA Satellites
Scatwind.obstore ASCAT
GPSRO.obstore Bending angle from GPS satellites (COSMIC, GRAS)
GOESClr.obstore GOES Imager Radiance (Clear)
ATOVS.bufr MetOp NOAA satellites (including HRPT data)
IASI.bufr MetOp
AIRS.bufr AQUA
6
Current Status

• NCUM DA -4D-Var operational
• NCUM 25 km/L70 operational
• 3D-Var Surface Analysis operational from July 2014

7
New Developments

• Generation of SST and Sea-ice for NCUM from high
  resolution (5 km) OSTIA SST
• Generation of snow analysis for NCUM from imssnow
  dataset (4 km)
• Implementation of 3D-Var surface analysis in NCUM
  using SYNOP data (Temp Humidity)
• Implementation of Nudging Scheme for Surface Soil
  Moisture in NCUM
• Assimilation of surface soil moisture derived
  from ASCAT in NCUM

8
New Developments

• Attained capability to create ancillary files for
  various resolutions of NCUM using CAP utility.
• Use of NRSC/ISRO derived LuLc from IRS-P6
  satellite over South Asia and adjoining region in
  NCUM.
• Assimilation of INSAT-3D AMVs in NCUM from 1st
  January 2015
• Efforts are going on to ingest INSAT-3D CSBT and
  Megha-Tropiques SAPHIR radiances (under IMDAA
  projects young scientist training) in NCUM

9
New Developments

• Implementation and testing of 1.5 km Nested model
• 17 km global model implemented and tested
• Migration to the next generation UM environment
  based on Rose/cylc has been accomplished
• Mirroring of UM shared repository available
  through cloud
• Sensitivity study with convection in UM
• Diagnostic assessment of monsoon behavior in
  Coupled UM on sub-seasonal scale (Training under
  UoR NMM project)
• New Products
• Dust forecasts from NCUM
• Visibility forecasts from NCUM

10
New Developments
11
Soil moisture assimilation scheme based nudging
technique is operational from July 2014
12
ASCAT Surface Soil Wetness in the assimilation
system
ASCAT surface soil wetness observations in the
assimilation System (1-15 Sept 2014)
ASCAT observations used in assimilation system at
NCMRWF (a typical day)
00 UTC
06 UTC
12 UTC
18 UTC
13
Monthly Mean Surface Level (0-10 cm) Soil
Moisture (November-2014)
NCUM
UKMO
14
RMSE () of Surface Level Soil Moisture against
AMSR2 Obs November 2014
NCUM
UKMO
15
Verification of UM Surface Soil Moisture
Analysis over India (Monsoon -2013)
Soil moisture analysis is able to capture large
variabilities seen in the in-situ observations
IMD soil moisture observations are not used
in the analysis
16
High Resolution (1.5 km) Regional Modelling
17

• The high resolution regional model at 1.5 km
  resolution is embedded within a coarser
  resolution global model (25 km).
• Both global and regional models are setup using
  latest version of UM8.5-GA6.1.
• NASAs 90 metre SRTM topographic data is used to
  generate the regional models orography.

18
NCUM-GLOBAL NCUM-REGIONAL
Governing Equations Complete equation (Non-hydrostatic) Deep atmosphere (Model top at 80 km) Complete equation (Non-hydrostatic) Deep atmosphere (Model top at 80 km)
Horiz. Resolution (N-S x E-W) N512 25km (0.234x 0.352) 1.5km (0.01350.0135)
Vertical Layers L70 L70
Forecast Length 10 days (240 hours) 3 days (72 hours)
Model Time Step 600 sec 50 sec
IC/ Data Assimilation 4DVAR Downscaling from global initial condition
Spatial Discretization Finite Difference method Finite Difference method
Time Integration /Advection Semi-implicit Semi-Lagrangian scheme Semi-implicit Semi-Lagrangian scheme
Radiation Process Spectral band radiation (general 2-stream) Spectral band radiation (general 2-stream)
Surface Process JULES land-surface scheme JULES land-surface scheme
PBL Process JULES Revised PBL JULES Revised PBL
Convection Process Turbulence and mass flux convection Convection in UM becomes less active when the area of grid box is decreased (high resolution). The CAPE timescale is increased reducing the activity of the parameterized convection. Turbulence and mass flux convection Convection in UM becomes less active when the area of grid box is decreased (high resolution). The CAPE timescale is increased reducing the activity of the parameterized convection.
Microphysics Improved mixed-phase scheme based on Wilson and Ballard (1999) Improved mixed-phase scheme based on Wilson and Ballard (1999)
Gravity Wave Drag Gravity Wave Drag due to orography (GWD) Gravity Wave Drag due to orography (GWD)
Surface Boundary Condition Climatology or SURF (Surface analysis) Climatology or SURF (Surface analysis)
Operation Frequency Once daily (00 UTC) Once daily (00 UTC)
6hour D.A. cycle Four times daily (00/06/12/18 UTC) Four times daily (00/06/12/18 UTC)
19
Nested regional model at 1.5 km resolution has
been successfully implemented and run for 3 days
for Gujarat, Madhya Pradesh, Odisha, JK and
Delhi domains

• Madhya Pradesh (700 x 450) IC 4th August 2014
• Wall Clock Time for 3 day forecast 5.5
  hours (8 nodes IBM-p6)
• Gujarat (600 x 450) IC 28th July 2014
• Wall Clock Time for 3 day forecast 4 hours
  (8 nodes IBM-p6)

20
Himalayan Orography (km)
SRTM data is at 90 metre resolution and GLOBE
data is at 1 km resolution
21
JK (3-5 Sep 2014)
22
1.5 km
Global
Obs
Day-1
Day-2
Day-3
23
Orography (km) over MP
24
Obs
1.5 km
Global
Day-1
MP 5-7 Aug 2014
Day-2
Day-3
25
Rainfall 06 Aug 2014
26
Land Use Land Cover data

• NCMRWF Unified Model (NCUM) uses the
  climatological 18 class IGBP LuLc dataset to
  derive nine surface types for the JULES land
  surface scheme.
• The IGBP dataset was derived from AVHRR data
  covering the period between April 1992 and March
  1993 and provides data at 30 arc-second (1km)
  resolution globally
• The climatological LuLc data are replaced with
  the NRSC/ISRO derived LuLc from IRS-P6 satellite
  over South Asia and adjoining region.
• AWiFS sensor data of IRS-P6 satellite during 2012
  to 2013 was used to derive the 18 IGBP surface
  types with a resolution of 30 sec (1 km)

27
Surface Types (IGBP v/s JULES)
Merged 18 surface types (NCMRWF)
9 surface types for JULES
Broadleaf trees
Needleleaf trees
C3 (temperate) grass
C4 (tropical) grass
Shrubs
Urban
Inland water
Bare soil
Land ice
18 surface types from NRSC over India (30 arc sec data) AWiFS, IRS P6, 2012-13 period
18 surface types from IGBP (30 arc sec data) AVHRR, 1992-93 period.
Input to JULES land surface scheme in UM
28
LuLc (UKMO NRSC)
29
Surface Type Fraction
Bare soil fraction
IGBP
Urban tile fraction
NRSC
NRSC
IGBP

• NRSC data shows recent changes in urban, forest
  and bare soil tiles.

30
Impact of land use/land cover - JK Rainfall
Results shows an improvement of regional rainfall
pattern with the use of new realistic land use
land cover data from ISRO NRSC.
Average rainfall over (74.5-78 E 33-36.5 N) Average rainfall over (74.5-78 E 33-36.5 N) Average rainfall over (74.5-78 E 33-36.5 N)
Observation (NCMRWF- IMD) NCUM (ISRO NRSC) NCUM (IGBP)
19.26 mm 11.15 mm 8.90 mm
31
Sensitivity Studies with NCUM Convection
Active monsoon spell in 2013 - 72-hr fcst from
NCUM (75 km) Entrainment rate increased by 25
OBS Control
Entrainment (25)
Arabian sea (65-74oE,15-23oN)
Central India (71-89oE, 17-27oN)

• Results
• Total rainfall (t72) from Entrainment (25)
  shows better correlation with observed rainfall.
• Control shows more frequency of deeper clouds in
  Arabian sea compared to Entrainment(25)

Bay of Bengal (85-100oE, 10-20oN)
3hrly averaged OLR count of Kalpana, Control,
Entrainment
32
Impact of better physics in coupled model (GA2.0
v/s GA3.0)
GA3.0 has reduced rainfall biases
33
NEMO Ocean Model simulated SST MLD (Apr-Sept)
SST Bias
Annual cycle of MLD (m)
with chlorophyll
without chlorophyll
Clim with chloro without
chloro

• The reduction of 0.5 C in SST bias and 10m in MLD
  bias is observed in the experiment
• Use of real time chlorophyll observations from
  OCM for ocean initialization would provide
  improvements

34
Model Verification against Analysis
35
Global ACC 500 hPa Z (Jan 2015)
Inter-comparison of models at NCMRWF
36
Wind RMSE 850 hPa (Tropics)
37
Model performance during the monsoon season-2014
38
Rainfall Verification Aug-Sept 2014

• Model Forecast Daily Rainfall (cm/day)
• NCUM NGFS
• Observed Daily Rainfall (cm/day)
• IMD-NCMRWF Merged Sat Gauge
• 0.5 x 0.5 grid resolution
• Continuous type gridded Verification statistics
  using Model Evaluation Tools
• 0.5 x 0.5 grids over Indian region (8-38 N,
  68-98 E).

39
Mean Error (8-38 N, 68-98 E)
NGFS shows higher ME at higher lead times
40
RMSE (8-38 N, 68-98 E)
RMSE magnifies the large errors in the
isolated cases (rare events).
41
Rainfall Verification NCUM, UKMO and ACCESS-G

• JJAS Verification of rainfall forecasts
• Mean monsoon rainfall
• Mean and extreme rain cases
• Verification scores for extremes (tails)
• Flooding in Srinagar

42
Forecasts overestimate the Rainfall along the
gangetic plains Average rainfall along the west
coast and NE India seem realistic. Rainfall
along west coast is drying up in NCUM
43
POD Fraction of observed yes events predicted
correctly.
Higher POD in NCUM,ACCESS-G and UKMO ACCESS-G has
highest POD
44
FAR What fraction of predicted yes events did
not realize??
Higher FAR in ACCESS-G
45
ETS How well did the forecast "yes" events
correspond to the observed "yes" events
(accounting for hits due to chance)?

• UKMO has higher ETS for lower thresholds
• ACCESS-G has higher ETS for higher thresholds

46
Synoptic System
47
Synoptic System
48
Srinagar Rainfall (4th Sept 2014)
Peak CC
RMSE Obs 269mm UKMO
207mm .37 19.2mm NCUM
169mm .20 18.7mm ACCESS-G 119mm
.20 18.8mm
49
All models fail to capture the peak rainfall
amounts along the west coast
Rainfall peaks over central India captured by UKMO
50
ETS tells how the forecast yes events
correspond to observed yes events (accounting
for random hits)
POD tells what fraction of the observed "yes"
events were correctly forecast
BIAS (frequency bias) tells how the forecast
frequency of yes events compare with observed
frequency of yes events
FAR Fraction of predicted events that did not
occur
ETS POD scores are very low for high rainfall
thresholds. Lower rain thresholds over forecast
(BIASgt1) Higher rain thresholds under forecast
(BIASlt1)
51
Extreme Dependency family of scores
Contingency Table Contingency Table Contingency Table Contingency Table Contingency Table
Observed Total
Yes No
Forecast Yes hits False alarms Forecast Yes
No misses Correct negatives Forecast no
Total Observed Yes Observed No Total
Extreme Dependency Score
Extreme Dependence Index
Symmetric Extremal Dependence Index

This family of scores tell what is the
association between observed and forecast rare
events.
52
EDS, SEDI and EDI all range from -1 to 1 0
indicating no skill and 1 indicating perfect
skill.

• Standard scores fail to show the differences in
  the scores near the tails
• Extreme Dependency score is able to bring out the
  difference in model performance for higher
  rainfall thresholds

53
Summary

• JJA Mean Rainfall
• NCUM Day-1 to Day-5 Drying
• Forecast skill (ETS) reasonable for lower
  rainfall thresholds
• Frequency bias over forecasting at lower
  thresholds and under forecasting at higher
  thresholds.
• JJA Maximum Rainfall
• Rainfall over central India (UKMO realistic),
  ACCESS-G and NCUM underestimate
• NCUM Day-1 to Day-5 Drying
• Rainfall along the west coast reducing in NCUM
• EDS, EDI and SEDI
• Extreme dependency family of scores highlight
  relative skill at higher thresholds.
• UKMO forecasts have relatively better skill in
  predicting the extremes.

54

• Assessment of GC2 (ENDGameGA6.0)
• NWP rainfall outputs over the Indian monsoon
  region against NCMRWF Merged Satellite Gauge
  (NMSG) daily observed rainfall dataset.
• Test runs of (i) Old N512 (ii) GC2 N512 and (iii)
  GC2 N768
• Study period Day-1 to Day-6 forecasts,
• 6-July to 15 September 2012
  (72 days).
• All model data interpolated to 0.5x0.5 grid

55
Mean daily rainfall (mm) from 6 Jul-15 Sep 2012
Top Old N512 ?
Middle GC2 N512 ?
Bottom GC2 N768 ?
56
Mean daily rainfall (mm) from 6 Jul-15 Sep 2012
Top Old N512 ?
Middle GC2 N512 ?
Bottom GC2 N768 ?
57
Summary of GC2 Evaluation

• GC2 NWP N512 N768 perform marginally better
  over the Indian monsoon region.
• GC2 captures synoptic scale rainfall variability
  better
• GC2 shows better demarcation (lower rainfall
  region) between high rainfall Monsoon Trough
  and foothills of Himalayas

58
TC Prediction from Regional NCUM
59
TC Name (Intensity) Simulation period in 24-h intervals Obs. Landfall (LF) No. of Forecast
Hudhud (VSCS) 00UTC of 08 - 13 October 2014 06 UTC 12 Oct. 2014 (Visakhapatnam) 04
Lehar (VSCS) 00 UTC of 24 - 29 November 2013 08 UTC 28 Nov. 2013 (Machilipatnam) 04
Phailin (VSCS) 00UTC of 09 - 13 October 2013 17 UTC 12 Oct. 2013 (Gopalpur) 03
60
Direct Position Error (DPE)
61
Direct Position Error (DPE)
62
Landfall (LF) errors in NCUM and Regional UM
TCs Name Different ICs (00 UTC) Obs. LF time LF Errors (km) LF Errors (km) of Improvement
TCs Name Different ICs (00 UTC) Obs. LF time NCUM Reg_UM of Improvement
Hudhud (October 2014 IC08 06 UTC 12 Oct. 2014 (Visakhapatnam) 307.64 129.42 57.93
Hudhud (October 2014 IC09 06 UTC 12 Oct. 2014 (Visakhapatnam) 224.7 185.48 17.45
Hudhud (October 2014 IC10 06 UTC 12 Oct. 2014 (Visakhapatnam) 168.78 103.5 38.67
Hudhud (October 2014 IC11 06 UTC 12 Oct. 2014 (Visakhapatnam) 67.55 62.61 7.31
Lehar (November 2013 IC24 08 UTC 28 Nov. 2013 (Machilipatnam) 578.8 NO --
Lehar (November 2013 IC25 08 UTC 28 Nov. 2013 (Machilipatnam) 458.13 434.44 5.17
Lehar (November 2013 IC26 08 UTC 28 Nov. 2013 (Machilipatnam) 329.38 266.65 19.04
Lehar (November 2013 IC27 08 UTC 28 Nov. 2013 (Machilipatnam) 111.13 72.33 34.91
Phailin (October 2013) IC09 17 UTC 12 Oct. 2013 (Gopalpur) 83.97 34.97 58.35
Phailin (October 2013) IC10 17 UTC 12 Oct. 2013 (Gopalpur) 43.36 34.97 19.34
Phailin (October 2013) IC11 17 UTC 12 Oct. 2013 (Gopalpur) 38.51 15.28 60.32
63
Future Plans
64
Future Plans NCUM DA

• Maximize the use of observations in the
  assimilation system, especially the Indian
  observations
• Efforts are in the final stages to include the
  INSAT-3D sounder and imager as well as
  Megha-Tropiques SAPHIR radiances
• MTSAT imager radiance data in NCUM
• Improvement of the DA system
• Move towards hybrid 4D-Var DA based on 44 member
  ensemble (ETKF) system
• A high resolution regional 4D-Var assimilation
  system will be implemented.
•  Observation Sensitivity Studies
• The tools to study the Forecast Sensitivity to
  Observation (FSO) has been implemented. This
  will help to identify the impact of different
  observations being used in the NCUM system.
• OSE OSSE studies with INDCOMPASS

65
Future Plans -NCUM

• Resolution of global deterministic model to be
  increased to 17 km this year on the new HPC
• Evaluation of Regional NCUM at 4 km/1.5 km
• Move towards high resolution ensemble forecasting
  with more ensemble members (33km (global)/44
  members)
• Incorporate better land surface data
  (land-use/land-cover, vegetation, soil moisture,
  soil temperature etc.) over Indian region with
  support from NRSC/ISRO
• Land surface DA based on Extended Kalman Filter

66
Coupled Modelling-Plans

• Implementation of a higher resolution coupled
  model (Atmos75kmL85 Ocean 25kmL75)
• Implementation of NEMO-Var Ocean Data
  Assimilation (25kmL75)

67
Involvement in NMM Projects

• CAWCR Rainfall verification (CRA) 1 Trained
• Met Office IMDAA 1 scientist visiting MO
  (Sept-Mar)
• UoR 1 Scientist visited during Sept-Dec 2014
• Imperial College Wind Energy
• TERI - Diurnal Variation of model rainfall
  (NCUM)
• FSU/IISc. - GFS Error

68
New HPC will be commissioned Soon 350 TF 1038
compute nodes Thank You
69
Mission Targets

• To implement the Unified Model (NCUM) at 25 km at
  NCMRWF. The resolution to be subsequently
  increased to 17 km/12 km.
• To implement regional version of NCUM at 12
  km/4-km/1.5-km resolution over Indian monsoon
  region for high impact weather.
• To implement 4-D VAR system and develop
  capability for assimilating data/radiances from
  upcoming Indian Satellites and DWRs
• To implement a high resolution Ensemble
  Prediction System (EPS) based on NCUM. - NGEPS
• To implement a NCUM based atmosphere ocean
  coupled modeling system- Coupled NWP Model
  for week-2 forecasts