Modelling & DA National Centre for Medium Range Weather - - PowerPoint PPT Presentation

Activities and Plans in

Modelling & DA

National Centre for Medium Range Weather Forecasting A-50, Sector 62, NOIDA-201309 www.ncmrwf.gov.in

Major Mandates of NCMRWF

NCMRWF is a Centre of Excellence in Numerical Modelling and Data Assimilation Ø Development and improvement of weather prediction models for IMD to underpin their forecasting capability Ø Development of Data Assimilation (DA) systems for both Global Forecast System (GFS) & Unified Model (UM) Ø Development of a Seamless prediction system based on UM

History of Global Models

Year Model Resolution DA 1989

TL79L19 (ECMWF)

• 1989-1992

R40L18

(COLA)

250 km OI 1992 - 2007 T80L18 T170L28 150 km 75 km 3D-VAR (SSI) 2007 NGFS (T254L64) 50 km 3D-VAR (SSI) 2009 NGFS (T254L64) 50 km 3D-VAR (GSI) 2010 NGFS (T382L64) 35 km 3D-VAR (GSI) 2011-15 2016- NGFS (T574L64) NGFS (T1534L64) 23 km 12 km Hybrid 3D-VAR (GSI) 2012 -15 2016- NCUM (N512L70)

NCUM (N768L70)

25 km 17 km 4D-VAR Hybrid 4D-Var

Model Resolution Upgrades have taken Place Whenever there was a major HPC upgrade

Year Computing Power Model Resolution 1988-91 234 MF 250 km 1992-99 468 MF 150 km 2000-02 1 GF 150 km 2003-05 28.8 GF 150 km 2005-06 500 GF 75 km 2006-10 1 TF 50 km 2010-15 24 TF 25 km 2015 - 350 TF 17 km

Current Operational Data Assimilation Systems

• Hybrid Ensemble Global Data Assimilation System for NCUM
• 44 members – ETKF/4D-VAR – 17 km
• Hybrid Ensemble Global Data Assimilation System for GFS
• 80 members –EnKF/3D-VAR (GSI) – T1534 (SL) ~12km
• Global Ocean DA using NEMO-Var – 0.25 Deg

T1534 (SL) ICs NCMRWF IITM-GEFS INCOIS- LETKF ODA IMD-extended IITM-seasonal IMD-GFS T382 (EL)

Increase in Data Reception at NCMRWF

q Number of observations (conventional and satellite/radar) has increased by orders of magnitude over past 10 years. q Data assimilated at NCMRWF comparable with

• ther International Global modelling centres

Indian Satellite Data Assimilated

• OSCAT- Ocean surface winds
• Megha-Tropiques SAPHIR
• INSAT-3D AMV
• INSAT-3D Sounder Radiance
• GPSIPW

10 20 30 40 50 60 70 1 2 3 4 5 6 1997 2001 2003 2006 2008 2009 2010 2011 2012 2013 2014 2015 2016

FTP (GB/day) GTS (GB/day)

Year

FTP (SAT + RADAR) IMD(GTS)

BUFR Data not being decoded at NCMRWF

1.5 km regional model up to 48 hr forecast 17 km global model up to 10 Day forecast 4 km regional model up to 72 hr forecast Global Ensemble Prediction System – 33 km with 44 members up to 10 Days Coarse resolution coupled model (NEMO+UM) Unified Model at NCMRWF Same Model for Global/Regional/Mesoscale – Seamless model 330 m Delhi Fog Model

Operational/Experimental Models

• NCMRWF Unified Model (N768L70) – since June 2015

ØNCUM (17 km) -10 day forecasts at 00 UTC ØNCUM (17 km) -5 day forecasts at 12 UTC – for RIMES Ø 4-km NCUM-R (with explicit rain processes) running with NCUM-G inputs at 00 UTC for 72 hours – since July 2015 Ø NCMRWF Global EPS – based on NCUM (33 km/44 members) -10 Day Forecasts at 00 UTC (450 nodes -3.5 hrs) Ø 1.5-km NCUM-R (with upgraded physics) - experimental runs for whole Indian domain for 72 hours during Monsoon 2016 (196 nodes; 7 hrs WCT) ØNCUM-DM (330 m) –Daily 36 hr forecasts based on 00 UTC

Day 2 Forecast: Mean Rainfall (10-30 June 2016)

Obs 1.5 km 4.0 km 17 km Global

• Positive Impact of higher resolution and better model physics in better prediction

is seen in the case of 1.5 km model

• Seamless prediction: Same model for global, regional & convective scale predictions

The decrease in the Error can be attributed to:

• increase in the resolution of the model,
• increase in the amount of data being assimilated,
• improvements in data assimilation techniques,
• improvements in model physics/dynamics.

Verification of Day 03 Forecasts against Radiosondes over India Error in 850 hPa winds (m/s) during (2005-2016)

How do we compare with other centres?

Error (RMSE) of 850 hPa winds(m/s) against Radiosondes over Tropics (~75 obs)

1 2 3 4 5 1 2 3 4 5

RMSEV Forecast Days ECMWF NCEP UKMO NCMRWF

Real time Verification System for MoES Medium Range Weather Forecast Models

Monthly Mean 850 hPa Wind RMSE

• ver Tropics (5 Jan - 3 Feb 2017)

Global Monthly Mean 500 hPa ACC of Geopotential Height (5 Jan - 3 Feb 2017)

IMD

Observations

Feedback on

• bservation quality

IC & FCST (Local & Global)

Central/ State GOVT

Sectoral Users (Agriculture , Aviation ….) Public Media

Value added products

IITM IMD IAF SASE

Capacity Building

• n NWP

New Applications

Wind energy Water Cycle …

Linkage of NCMRWF with Various Organizations INCOIS

satellite obs.

ISRO NCMRWF Numerical Modelling of Weather & Climate

IC & FCST

IC First Guess

Ocean and Fishery services

Forecast Products Partners and User Agencies Analysis Fields from Global Model IMD, IITM, INCOIS and SAC Rainfall, Temperature and Winds from Global and Ensemble Models IMD, SWFDP (WMO) Snow and Avalanche Study Establishment/DRDO Bhakra Beas Management Board (Flood Monitoring & Forecasting) Krishna and Bhima-Basin Simulation Division (Maharashtra Govt; flood Forecasting) Wind, Temperature, Rainfall, SW Radiation etc. from Global and Regional

Models

Wind and Solar Energy Sector à National Institute of Wind Energy (Ministry of New and Renewable Energy) Manikaran Power Limited & Energon Power Resources Pvt Ltd Wind, Temperature, Geo- potential Height and Humidity from Regional Model Nuclear Power Corporation of India Ltd. Weather forecast for Nuclear Power Plants (9) (1) Kaiga (4) Kalpakkam (7) Visakhapatnam (2) Trombay (5) Narora (8) Kudankulam (3) Tarapore (6) Rajasthan (9) Kakrapara

Partners and User Agencies of NCMRWF Products

This product is shared with IMD, Agromet & SAC

Recent Developmental Activities

NCUM Activities

• Regional DA – Assimilation system (3D-Var) for 4-km

regional NCUM has been implemented. Successfully assimilated radial wind observations from Indian DWRs

– For calculation of background error, new CVT system has been setup and preparation of background error has started.

• FSO system has been setup and experimented for 20 days
• IMDAA reanalysis (1979-1988) – 42 + 30 months completed

at NCMRWF/Met Office.

• NCUM at 330 m set up over Delhi and is being run daily for

fog prediction

• NGEPS – Impact of ensemble member size (11,22 & 44)

» Impact of Horizontal resolution (33 v/s17 km)

• JULES 1-D & 2-D simulations for INCOMPASS

4 km regional domain

Impact of DWR data in Regional NCUM

• Expt. Name

Observational data used GTS Assimilation of observations received from GTS + Satellite (Surface, Sonde, Aircraft, AIRS, ATOVS, IASI, Satwind) GTS+Radar GTS (GTS+ Satellite) and DWR radial wind

Observed and 24 hr forecast (accumulated) rainfall valid for 03 UTC 27 July 2016 Observed Rainfall 24 hr Forecast (GTS) (GTS + Radar) DWR Data used in DA

Observed and 24 hrs accumulated rainfall for Day-1 valid for 03 UTC 10 July 2016 Observed Rainfall 24 hr Forecast (GTS) 24 hr Forecast (GTS +Radar)

Observed and 24 hrs accumulated rainfall for Day-1 forecast valid for 03 UTC 6 July 2016

Observed Rainfall 24 hr Forecast (GTS) 24 hr Forecast (GTS +Radar)

u-wind

RMSE of 24 hr forecast (valid for 00UTC 27July 2016) (Against Radiosonde Observations)

v-wind

Forecast Sensitivity to Observations (FSO) in NCUM Global system

• Forecast Sensitivity to Observation help us to understand how
• bservations are used by the data assimilation system.

FSO v/s OSE

• FSO quantifies the impact of all assimilated
• bservations on a

selected forecast metric. It shows, if any observation decreases or increases forecast error.

• OSE

(data denial experiments) gives impact of one selected change to the observation system.

• However, FSO (adjoint-based technique) is restricted by the tangent

linear assumption, valid up to 24-48 hours forecast only. NWP centers use FSO routinely to monitor their Data Assimilation and Global Observing System.

• It helps to assess the impact of specific sensors and provide

feedback to data providers. It also helped to improve the Quality Control, Bias Correction, etc.

Forecast Sensitivity to Analysis:

• The calculation of forecast sensitivity to analysis is carried out using two 24

hour forecast trajectories by the NCUM global model, which is initialized with the analysis and background, respectively.

• The background is a prior model forecast (6-hour forecast), and the analysis

is produced by 4D-Var (Hybrid 4D-Var) data assimilation system.

• The forecast error will be calculated with respect to the verifying analysis for

the forecast of analysis and background.

• A forecast error cost function will be defined based on the difference

between those two forecast errors.

• Along the forecast back trajectory with adjoint of model (PF) is able to

calculate the forecast sensitivity to analysis using the forecast error cost function.

Observation Impact: The observation impact is obtained by calculating the inner product of the forecast error sensitivity to observation and the innovation vector (y-H(x)) Forecast Sensitivity to Observation: The second step is to extend the forecast sensitivity to analysis from the grid point provided by the previous step into the observation space using the adjoint of the data assimilation system.

Time-averaged total observation impact on 24 h forecast (13 October 2016 to 09 November 2016)

2.5% 2.6% 4.7% 1.0% 18.1% 4.7% 10.9% 0.0% 2.8% 0.1% 18.2% 4.6% 0.1% 12.7% 0.7% 0.6% 8.9% 7.9%

Aircraft AIRS ASCAT ATMS ATOVS Buoy CrIS GOESCLR GPSRO GroundGPS IASI MTSAPHIR MVIRI SATWIND SEVIRI SHIP SYNOP TEMP

• 3.0
• 2.5
• 2.0
• 1.5
• 1.0
• 0.5

0.0

Impact per day (J/kg)

SYNOP includes :SYNOP & METAR; TEMP includes: TEMP, WINPRO & PILOT; ATOVS includes: AMSU-A, AMSU-B & HIRS; SATWIND includes: GOES, JMA, MSG & IMD (INSAT)

The % values to the left of the bars give the fraction of the total impact

Channel-wise MT-SAPHIR Observation Impact

40.5% 30.9% 15.9% 7.2% 2.1% 3.4%

• 0.25
• 0.20
• 0.15
• 0.10
• 0.05

0.00 S6 S5 S4 S3 S2 S1

Impact per day (J/kg)

The % values to the left of the bars give the fraction of the total impact

MHS/AMSU-B Observation Impact (channel-wise)

21.9% 21.3% 9.3% 4.5% 9.8% 9.3% 3.4% 1.8% 12.1% 9.6% 4.0% 2.2%

NOAA18_3 NOAA19_3 AMSUB_MetOp1_3 AMSUB_MetOp2_3 NOAA18_4 NOAA19_4 AMSUB_MetOp1_4 AMSUB_MetOp2_4 NOAA18_5 NOAA19_5 AMSUB_MetOp1_5 AMSUB_MetOp2_5

• 0.09
• 0.08
• 0.07
• 0.06
• 0.05
• 0.04
• 0.03
• 0.02
• 0.01

0.00

Impact per day (J/kg)

The % values to the left of the bars give the fraction of the total impact

3.6% 2.6% 6.0% 1.0%

MSG GOES JMA INSAT3D

• 1.0
• 0.5

0.0

24.1% 7.4% 45.8%

Impact per day (J/kg)

19.8%

Observation Impact SATWIND: GOES, JMA, MSG & IMD (INSAT)

• The % values to the left of the bars give the fraction of the total impact.
• The % values to the right of the bars give the fraction of SATWIND total impact

INSAT-3D (IMD) Wind (height-wise observation impact)

1.2% 6.8% 8.9% 5.2% 9.1% 11.5% 15.6% 17.6% 12.7%

IMD200-100 IMD300-200 IMD400-300 IMD500-400 IMD600-500 IMD700-600 IMD800-700 IMD900-800 IMD1000-900

• 0.04
• 0.03
• 0.02
• 0.01

0.00

Impact per day (J/kg)

##The % values to the left of the bars give the fraction of the total impact

TC vortex initialization takes input from NCUM analysis:

• U, V, T, Z, RH at all pressure levels
• Surface temperature, surface pressure and MSLP

Vortex initialisation scheme has four steps:

• Filtering of the analyzed circulation from the original

analysis

• Construction of an inner core of cyclone (MSLP only)
• Relocation of inner core to observed position
• Merging of relocated vortex with the large-scale analysis

TC Vortex Initialization

Vortex initialization in NCUM-G and NCUM-R (TC Hudhud, base time 20141009 00UTC)

Track error (km)

Vortex initialization with/without satellite data assimilation (TC Chapala, 29 Oct-2 Nov 2015)

• Verification for forecasts run from 5 ICs starting 00 UTC 29 Oct 2015
• Vortex initialisation with assimilation of satellite data has positive

impact on the track and intensity forecast, landfall time and position error of the TC.

Mean track error (km)

no-assim assim

Mean 10m wind error (m/s)

no-assim assim

IMDAA Reanalysis Domain (coloured region)

Verification of IMDAA Re-analaysis (12 km)

• ver IMDAA domain against RS/RW

(1983, 1985 & 1987)

• IMDAA funded under

Monsoon Mission for 4 years

• Capacity building and

support in DA

• Reanalysis from 1979-

2017 at 12 km resolution

• 10 year reanalysis 1979-

1988 production to be completed by FY

• 29 year reanalysis (1989-

2017) to be completed in next FY

Wind Speed -850 hPa (Jan, 1983) Wind Speed -850 hPa (Jul, 1983) Wind Speed -500 hPa (Jan, 1983) Wind Speed -500 hPa (Jul, 1983)

Wind Speed -850 hPa (Jan, 1985) Wind Speed -850 hPa (Jul, 1985) Wind Speed -500 hPa (Jan, 1985) Wind Speed -500 hPa (Jul, 1985)

Wind Speed -850 hPa (Jan, 1987) Wind Speed -500 hPa (Jan, 1987) Wind Speed -850 hPa (Jul, 1987) Wind Speed -500 hPa (Jul, 1987)

Developments in GFS Analysis

• 4D ensemble Var analysis system with all

sky radiance has been tested successfully.

– Possible to make it operational by 2017 monsoon season

• INSAT-3DR and Scatsat assimilation

schemes have been developed.

Quality of Scatsat data

Scatsat vs with tropical moored buoy observation Wind Direction Wind Speed

Scatsat data accepted and rejected in a typical DA cycle 20161014 06 UTC Mean O-B of u-wind of ascat and Scatsat 13 - 21 Oct 2016

• Impact of Scatsat winds
• n analysis is maximum

just before it attains T- 1.5 intensity (after that cyclone bogus procedure takes place).

• Scatsat winds

assimilated, in the 3-30 km/h range only.

• The impact of Scatsat

data is clearly seen in simulating track & intensity of cyclone

72 hr forecast track of cyclone Kyant based on IC of 25 Oct 2016

INSAT-3DR

• Immediately after launch of INSAT-3D, NCMRWF,

received Imager & Sounder Sensor Response Functions (SRF) and got FRTM coefficients computed for both RTTOV and CRTM. Shared them with SAC/ISRO,

• Started receiving INSAT-3DR data from middle of

November.

• GFS analysis has been modified to assimilate INSAT-

3DR radiance - tested for a day & found ok.

• Further experiments required before operationalization.

Delhi Model (330m) for Visibility Prediction

• Triple Nested- NCUM-G (17km) → NCUM-IND (1.5km) → NCUM-DM (330m)
• Vertical levels: 80 with model top at 38.5km
• Model time step: 12 seconds
• Radiation time step (prognostic/diagnostic): (4 minute/1 minute)
• LBC updation frequency: 15 minutes for DM & 1hr for Regional (1.5km)
• RHcrit=12x0.970, 0.960, 0.950, 3x0.940, 3x0.930, 5x0.920, 5x0.910, 50x0.900
• BL option: Blending with grey zone convection parametrization
• Sub-grid vertical turbulence scheme: Smagorinsky parameter Cs=0.5
• DM model uses leaf-area index (LAI) vegetation of the 1.5 km model
• Orography: 90 m SRTM for both DM & 1.5km – Plans to use 30 m CARTOSAT
• NRSC Land Use land Cover data
• Convective parameterisation is “ turn off “

Configuration Setup of 330 m Delhi Model (DM)

GL-17km NCUM IND-1.5km DM-330m

Sensitivity of Ensemble Member Size

NGEPS: 33-km 11 members + control 22 members + control 44 members + control

TC Roanu (18-22 May 2016)

12 members 23 members 45 members (11 perturbed + 1 control ) (22 perturbed + 1 control) (44 perturbed + 1 control)

Mean Tracks

Direct Position Error

100 200 300 400 500 600 700 6 12 18 24 30 36 42 48 54 60 66 72 78 84 90 96 Direct Position Error (km) CNTL Ens_mean_11 Ens_mean_22 Ens_mean_44

Impact of Model Resolution NGEPS: 33-km v/s 17-km 22 members + control

Day 5 forecast of ensemble mean rainfall of NGEPS valid for 23rd May 2016 33-km (N400) 17-km (N768)

TC Roanu Tracks

33-km (N400) 17-km (N768)

33-km (N400) 17-km (N768)

TC Roanu Strike Probability

Coupled Modelling

• Ocean Initialisation based on NEMOVar now running

in real-time using UKMO observational datasets.

• Coupled Model (UM [60kmL85] & NEMO [25 kmL75])

implemented in Nov 2016.

• The Coupled Model with UM at 25kmL85 will be

used for coupled NWP experiments using Ocean and Atmospheric IC generated at NCMRWF

Total Rainfall Surface Temperature 15th Day Forecast Valid for 20110923

JJAS 2016 Rainfall Verification and Inter-comparison

• IMD-NCMRWF (Sat+Gauge) rainfall analysis (03UTC)
• NCUM and ACCESS-G rainfall forecasts (03UTC)
• NEPS (44 member) ensemble mean rainfall forecasts

(00UTC)

• 0.25° x 0.25 ° grid spacing
• Verification over India (8-38N/68-98E) (Land only)

NCUM, NEPS, and ACCESS show similar behaviour Rainfall peaks over- estimated over central India, under- estimated over Bay

• f Bengal

Drying from Day-1 to Day-5, especially in ACCESS NCUM NEPS ACCESS Day-1 Day-5 Day-3

Global model performance for 2016 monsoon season

Extreme rains (near tails) underestimated by NEPS and ACCESS West coast heavy rains underestimated NCUM overestimated heavy rains over central India and east coast but underestimated over BoB

Observed and Model Forecast Maximum Rainfall (cm/day) (JJAS 2016)

NCUM NEPS ACCESS Day-1 Day-5 Day-3

Global model performance for 2016 monsoon season

Odd Ratio measures the ratio of the odds of making a hit to the odds

• f making a false alarm.

Range: 0 to ∞ Values > 1 denotes skill Poor skill over core monsoon region NCUM NEPS ACCESS Day-1 Day-5 Day-3

Global model performance for 2016 monsoon season

Odds ratio

Performance Improvements in Day-3 forecasts

• f heavy rainfall (2014 v/s 2016)

Major improvement* in NCUM frequency of heavy rain. Slight improvement in ACCESS biases. Major improvements* in ETS for moderate rain, both models. Heavy rain remains difficult.

*Caveat: JJA in 2014, JJAS in 2016 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.28 2.56 5.12

Frequency bias Daily rain threshold (cm/day)

NCUM JJA2014 NCUM JJAS2016 ACCESS JJA2014 ACCESS JJAS2016 0.00 0.05 0.10 0.15 0.20 0.25 0.30 1.28 2.56 5.12

Equitable threat score Daily rain threshold (cm/day)

NCUM JJA2014 NCUM JJAS2016 ACCESS JJA2014 ACCESS JJAS2016

Rainfall diurnal cycle (Domain averaged rain (mm))

Central India Western Ghats (78E–83E, 20N-25N) (65E-68E, 18N-28N)

Observed Rain (black) Ver 8.5 (red ) Ver 10.2 (blue) Global Regional Regional model with moisture conservation is showing better match in hourly rainfall amount and in-phase diurnal cycle with the observed rain, especially over central India. Without moisture conservation there is notable over prediction Global model has out of phase relationship in rainfall diurnal cycle. No change with model upgradation as there is no physics changes

Tropical Cyclone Verification 2016

• NCUM-G v/s NGEPS

TC ‘Roanu’

TC ‘Kyant’

Observed and Forecast track for CS Kyant

TC ‘Nada’

TC Vardah

Average DPE

50 100 150 200 250 300 350 400 450 12 24 36 48 60 72 84 NCUM NEPS

Average Direct Position Error during 2016

(4 cases of Roanu, Kyant, Nada and Vardah)

Future Plans

Ø To increase the global model resolution to 12 km. Ø To implement very high resolution regional model of 1.5-km resolution

• ver Indian region for prediction of high impact weather.

Ø Operational implementation of a high resolution (4 km) regional data assimilation system which will have the capability to assimilate Radar and other high resolution datasets

Ø Upgrade the horizontal resolution of NCMRWF-GEPS initially to 17 km in 2017 and to 12 km by March 2019 (with 22 or 44 members)

Ø To implement a high resolution (25 km) atmosphere-ocean coupled modeling system- “ Coupled NWP Model” for week-2 forecasts Ø Develop new application areas (Wind/Solar energy, Water resource management Ø Develop and improve new products (Visibility, Fog, Dust etc.)

Thank You

• Visualization of NCUM data through Earth-Wind framework and D3.js