Impact of SAPHIR radiances on the simulationof tropical cyclones over the Bay of Bengal usingNCMRWF hybrid-4DVAR assimilation andforecast system

DEVANIL CHOUDHURY1,2, ANKUR GUPTA

1,*, S INDIRA RANI1

and JOHN P GEORGE1

1National Centre for Medium Range Weather Forecasting (NCMRWF), Ministry of Earth Sciences (MoES),A-50, Institutional Area, Sector 62, Noida 201 309, India.2Center for Monsoon System Research, Institute of Atmospheric Physics, Chinese Academy of Sciences,Beijing 100 029, People’s Republic of China.*Corresponding author. e-mail: guptaankur1583@gmail.com ankur@ncmrwf.gov.in

MS received 12 March 2020; revised 4 June 2020; accepted 9 July 2020

Observing System Experiments (OSEs) were conducted to analyze the impact of assimilation ofMegha-Tropique’s (MT) Sounder for Probing Vertical ProBles of Humidity (SAPHIR) radiances on thesimulation of tracks and intensity of three tropical cyclones (Kyant, Vardah, and Maarutha) formed overthe Bay of Bengal during 2016–2017 North Indian Ocean cyclone period. National Centre for MediumRange Weather Forecast (NCMRWF) UniBed Model (NCUM) Hybrid-4DVAR assimilation and forecastsystem was used for the OSEs. Assimilation of SAPHIR radiances produced an improvement of 9% and12%, respectively, in the cyclones’ central sea level pressure (CSLP) and the maximum sustained wind(MSW), while an improvement of 38% was seen in the cyclone tracks within the forecast lead time of 120hrs. Initial assessment shows that the improvement in the cyclone intensity is due to the assimilation ofthe unique surface peaking channel of SAPHIR (channel-6), whereas the improvement in the cyclonetrack is due to the assimilation remaining Bve channels of SAPHIR. Thus, the assimilation of SAPHIRradiances in the NCUM system showed improvement in both intensity and track of the cyclones over theBay of Bengal; however, more cyclone cases over different ocean basins have to be analyzed to make arobust conclusion. This study speciBes the importance of similar microwave humidity instruments in thesame frequency range for the detailed exploration of cyclone track and structure.

Keywords. SAPHIR; hybrid-4DVAR; data assimilation; tropical cyclone.

1. Introduction

Tropical cyclone (TC) forecasts over the NorthIndian Ocean are increasingly dependent on theNumerical Weather Prediction (NWP) models inrecent years. The numerical forecasts of TCs inmost ocean basins have improved considerably in

the last few decades, attributed mainly to anincrease in model resolution and better represen-tation of convection and microphysics (Gentry andLackmann 2010) in addition to improvements inobserving system and data assimilation methods.The simulation of TC’s tracks is highly sensitiveto initial conditions (Singh and Mandal 2014;

J. Earth Syst. Sci. (2020) 129:209 � Indian Academy of Scienceshttps://doi.org/10.1007/s12040-020-01473-2 (0123456789().,-volV)(0123456789().,-volV)

Greeshma et al. 2015), small perturbations in theinitial conditions can lead to large errors in fore-casted tracks. The importance of better represen-tation of humidity proBle in the initial conditionsfor TC forecasts has been addressed in severalstudies (Chu et al. 2008; Wu et al. 2012). Thesestudies showed that the TCs intensify rapidly inthe presence of higher environmental humidity,thus a potential predictor for rapid intensiBcation.Similarly, higher diabatic heating in the outer rain-bands of TCs in a relatively humid environment isshown to enhance the spatial extent of cyclones(Wang 2009).Inadequate representation of humidity in the

initial state can cause uncertainties in the forecast.By improving the moisture Beld in the analysis,improvement in the track and intensity forecast ofTCs can be expected (Fabry and Sun 2010).Radiosondes, which are the most reliable sources ofatmospheric moistures, are primarily available incontinental regions. Satellite observations providereasonable estimates of atmospheric humidityinformation. EAorts on the assimilation of satelliteradiances (including humidity information) andretrievals from instruments sensitive to atmospherichumidity have a rich heritage. The positive impact ofthe assimilation of Infrared Atmospheric SoundingInterferometer (IASI) observations on forecastingtrack and minimum sea-level pressure (MSLP) hasbeen reported by Xu et al. (2013). Reale et al. (2009)have shown improvement in track prediction due tothe assimilation of Atmospheric Infrared Sounder(AIRS) radiance and temperature retrievals.Unlike the satellite observations in infrared

wavelengths, which are highly contaminated byclouds and precipitation, microwave-based retrie-vals provide important humidity information inall-sky conditions except for optically thick orprecipitating clouds. Several microwave-basedsatellite instruments provide measurements ofatmospheric humidity soundings, includingAdvanced Microwave Sounding Unit-B (AMSU-B), now replaced by Microwave Humidity Sounder(MHS); Special Sensor Microwave Imager/Sounder(SSMI/S); Advanced Technology MicrowaveSounder (ATMS); Advanced Microwave ScanningRadiometer (AMSR); etc. Liu et al. (2012) repor-ted improvement in the ambient Belds along withtracks and intensity due to the assimilation ofAMSU radiances. In the sensitivity studies,Greeshma et al. (2015) showed that assimilation ofAMSU radiances improved the structural featuresof eight Bay of Bengal (BOB) cyclones.

Sounder for Probing Vertical ProBles ofHumidity (SAPHIR) onboard Megha-Tropiques(MT) satellite is a six-channel microwave humiditysounder. The MT orbit is unique with an orbitalinclination of 208 compared to the equatorial planeand a relatively high orbit of 867 km altitude. Thisallows the satellite to have a high temporal sam-pling of 2–5 overpasses per day over a given loca-tion over the Tropics. One of the uniquecharacteristics of SAPHIR is its lowest peakingchannel (channel-6, S6) compared to other micro-wave instruments operating in the same frequencyrange. Rani et al. (2015) reported the positiveimpact of SAPHIR on the assimilation of hyper-spectral radiance. Rani et al. (2016a) showed thatthe 183.31±11 GHz channel (S6) extends the ver-tical coverage of SAPHIR, relative to ATMS andMHS and modiBes the imager driven humidityincrements compared to other microwave soun-ders. Improvement in moisture analyses and fore-casts due to the assimilation of SAPHIR radiancewith the Weather Research and Forecast (WRF)model is reported by Singh et al. (2013). Ingestionof SAPHIR radiance in the Meteo-France globalmodel ARPEGE resulted in systematic improve-ments in the humidity distribution (Chambon et al.2015). A positive impact in the prediction of rela-tive humidity is also seen by assimilating cloudclear SAPHIR radiance into GFS (Global ForecastSystem) assimilation system used at NationalCentre for Medium Range Weather Forecasting(NCMRWF) (Singh et al. 2015). Potential forall-sky simulation of the SAPHIR radiance is repor-ted by Madhulata et al. (2017). Compared to similarinstruments likeMHSandATMS, the assimilation ofSAPHIR radiance shows considerable improvementsin the forecast of moisture, temperature, and winds(Kumar et al. 2018). Doherty et al. (2018) reportedthat SAPHIR oAers improved performance oversimilar instruments, and further improvement whenassimilated in conjunction with observations fromthe microwave imager AMSR-2.However, few studies are available so far on the

impact of assimilation of SAPHIR on the TC simula-tion. Dhanya et al. (2016) reported substantialimprovement in track forecast of TCs Thane andPhailin, but deterioration for short-lived TC Nilam.Similarly, Singh and Prasad (2017) found mostly posi-tive butmarginal improvement inTCHelen and Leharforecasts due to the assimilation of SAPHIR radiances.BothDhanya et al. (2016) andSingh andPrasad (2017)investigated the impact of SAPHIR radiances in the3D-variational assimilation framework.

209 Page 2 of 13 J. Earth Syst. Sci. (2020) 129:209

In this study, an attempt has been made toanalyze the impact of SAPHIR data assimilationon the simulation of TCs over the BoB in theNCUM hybrid-4DVAR framework in whichCow-dependent ensemble covariances are added tothe traditional incremental 4DVar using linear andadjoint models. Observing System Experiments(OSEs) are designed to analyze the impact ofSAPHIR radiances on the simulation of three TCs,viz., Kyant, Vardah and Maarutha formed over theBOB during 2016–2017. Since the intensity of thecyclone is highly inCuenced by the lower tomid-tropospheric humidity (Wang 2009; Wu et al.2012), separate experiments were carried out to seethe impact of the SAPHIR lower peaking channel(S6) during the TC Vardah. A brief description ofthe TCs selected in this study is provided in sec-tion 2. Section 3 describes the characteristics ofSAPHIR instrument followed by the description ofthe data assimilation and forecasting system insection 4. The design of the OSEs is briefed insection 5. The main Bndings from the study andthe discussions are included in section 6, followedby conclusions in section 7.

2. Selected tropical cyclones

Three significant TCs formed over the BOB duringthe North Indian Ocean Cyclone period of2016–2017 are selected in this study. They are: (1)Kyant (21–28 October 2016), (2) Vardah (6–19December 2016), and (3) Maarutha (15–17 April2017). Table 1 describes the characteristics of thesecyclones. Further details are available in the bul-letin of Regional Specialized Meteorological Centre(RSMC) Tropical Cyclones, New Delhi (http://www.rsmcnewdelhi.imd.gov.in).

3. SAPHIR instrument

Satellites, in the low inclination orbits, like MTprovide valuable information on humidity over theTropics compared to those in the sun-synchronized

orbit. Centered around 183.31 GHz, the SAPHIRinstrument in the MT satellite operates in sixnarrow water vapour absorption bands with ahorizontal resolution of 10 km at nadir. The watervapour absorption line centered around 183.31GHz is high enough to proBle the atmospherichumidity from the surface to 12 km. To get theinformation from the surface to the top of theatmosphere, the SAPHIR is designed with largebandwidth (183.31 ± 11 GHz) without adding aspeciBc window channel. SAPHIR instrumentdetails are provided in table 2.Satellite radiances are bias corrected before

using in the assimilation systems. In order tomonitor the biases in the observations, measuredsatellite radiances are compared with their equiv-alents computed from a short-term forecasts oranalysis estimate of the atmospheric state using aradiative transfer (RT) model. Many assumptionsare made in this type of bias correction. It isassumed that observed satellite radiances are freefrom calibration errors, the radiative transfermodel is accurate, and the short-term forecastprovided by NWP model is free from systematicerrors. However, these assumptions are not alwaysvalid. Biases vary with time (both diurnal andseasonal), geography or air mass, scan position ofsatellite instrument, and the position of the satel-lite around its orbit (Bell et al. 2008; Lu et al. 2011;Doherty et al. 2015; Rani et al. 2015, 2016b).Details of the SAPHIR radiances bias correction isavailable in Rani et al. (2015, 2016b) and showedapplying bias correction shifted the mean of theinnovation (Observation–Background) distribu-tion of SAPHIR radiances towards zero, ensuringthe eAectiveness of bias correction.

4. Data assimilation and forecast system

The NCMRWF UniBed Model (NCUM) is anadapted version of the UM system of the UK MetODce. The main components of the NCUMassimilation and forecast system are the

Table 1. Details of tropical cyclones considered in the study.

Cyclone Period Landfall Category MSW (m/sec) CSLP (hPa)

Kyant 21–27 October, 2016 Dissipated over west

central Bay of Bengal

Cyclonic storm 23.6 988

Vardah 7–13 December, 2016 Chennai, Tamil Nadu Very severe cyclonic

storm

36.1 975

Maarutha 15–17 April, 2017 Sandoway, Myanmar Cyclonic storm 20.8 996

J. Earth Syst. Sci. (2020) 129:209 Page 3 of 13 209

observation processing system (OPS), a hybridfour-dimensional variational assimilation system(Hybrid-4DVAR), surface analysis preparation(SURF), and the forecast model (UM). The OPSdoes processing and quality control of observationand produce the required inputs for the Hybrid-4DVAR assimilation system. The NCUM Hybrid-4DVAR assimilation system uses Cow-dependentday-to-day varying forecast error covariances fromthe NCMRWF’s 44 members Ensemble PredictionSystem (EPS) along with the climatologicalcovariances using linear and adjoint models (Lor-enc 2003). The operational EPS at NCMRWF iscalled as NCMRWF Ensemble Prediction System(NEPS, Sarkar et al. 2016). The NEPS has a hor-izontal resolution of approximately 33 km and 70vertical levels and 45 ensemble members (44 per-turbations and one control). The EnsembleTransform Kalmon Filter system (ETKF) providesthe initial conditions for NEPS members, andgenerates global perturbations for wind, tempera-ture, humidity and pressure Belds. Six hourlyintermittent assimilation cycles centered on 00, 06,12 and 18 UTCs are produced routinely usingNCUM assimilation system. Further details of theNCUM data assimilation system can be found inGeorge et al. (2016) and Kumar et al. (2018). Dif-ferent types of observations assimilated in thesimulation of TCs along with SAPHIR observa-tions are listed in table 3.SURF is the surface data assimilation system,

which produces surface analysis (snow, SST, sea-ice and soil moisture) for initializing the modelsurface conditions. The NCUM Land Data Assim-ilation System (LDAS) uses a 3D-Var assimilationsystem to produce an accurate screen level analysisof temperature and humidity increments in everyassimilation cycles. A soil moisture analysis isproduced using an Extended Kalman Filter (EKF)based land data assimilation system (de Rosnayet al. 2013). A key component of the EKF is the

calculation of the Jacobians of the observationoperator. The Jacobians are estimated using fore-casts of the land surface model JULES (Joint UKLand Environment Simulator) with perturbed ini-tial conditions. Further details are available inLodh et al. (2016).The forecast model used in this study is UM

version 10.2. The NCUM has non-hydrostaticdeep atmosphere equations with semi-implicit,semi-Lagrangian numerical schemes. The modelincludes a comprehensive set of parameterizationschemes for surface, boundary layer, convection,radiation, etc. The model has latitude–longitudehorizontal grid with Arakawa C staggering andterrain-following hybrid height vertical coordinatewith Charney–Phillip staggering. The horizontalresolution of the NCUM global model used in thisstudy is 17 km and has 70 vertical levels (N768L70) from the surface to 80 km height. More detailsof the NCUM forecast model system can be foundin Rajagopal et al. (2012) and Rakhi et al. (2016).The schematic diagram of the complete NCUMassimilation system is given in Bgure 1. The detailsof the model conBguration used in the presentstudy are given in table 4.

5. Design of observing system experiments(OSEs)

Two sets of experiments for the OSEs are carriedout: the control run (CNTL) in which all theobservations listed in table 3 are assimilated andthe data denial experiment (EXPT) in which allthe observations except SAPHIR are assimilated.Four data assimilation cycles covering 24 hrs(centered on 00, 06, 12 and 18 UTCs) were runbefore the start of the simulations of both CNTLand EXPT experiments for all three selected TCs.Thus, two initial conditions for each TC, one withSAPHIR observations and the other without, were

Table 2. SpeciBcations of the SAPHIR sensor.

Channel

Central

frequency

Bandwidth

(MHz) Polarization

Pressure levels

(hPa)

S1 183.31 ± 0.20 200 H *250–100

S2 183.31 ± 1.10 350 H *400–250

S3 183.31 ± 2.80 500 H *550–400

S4 183.31 ± 4.20 700 H *700–550

S5 183.31 ± 6.20 1200 H *850–700

S6 183.31 ± 11.0 2000 H *1000–850

209 Page 4 of 13 J. Earth Syst. Sci. (2020) 129:209

obtained. The assimilation window starts 24 hrsbefore the Brst reported time of the depressionstage of the storm for TCs Kyant and Maarutha.Since Vardah was the long-lasted TC; the initialassimilation cycle was chosen to be within 5 days ofits landfall. Hence, initial conditions were selected

at 00 UTC of 21 October 2016 (Kyant), 18 UTC of7 December 2016 (Vardah), and 00 UTC of 15April 2017 (Maarutha).Figure 2 shows the SAPHIR coverage plot over

India and the surrounding oceanic regions (shownis the brightness temperature from channel-6)

Table 3. Observations used in the OSEs.

Type of observation Description Assimilated variables

AIRCRAFT Upper-air wind and temperature from aircraft Wind and temperature

AIRS Atmospheric infrared sounder of AQUA satellite Brightness temperature

AMSR-2 Advanced microwave scanning radiometer 2 onboard

GCOM-W satellite

Brightness temperature

ASCAT Advanced scatterometer in Metop A and B Wind

ATMS Advanced technology microwave sounder in NPP satellite Brightness temperature

ATOVS AMSU-A, AMSU-B/MHS from NOAA-18 and 19, Metop-A and B Brightness temperature

BUOY Surface synoptic observation over oceans from buoy Sea level pressure, wind,

temperature and

humidity

CrIS Cross-track infrared sensor in NPP satellite Brightness temperature

MWHS Microwave humidity sounder onboard FY-3C Brightness temperature

GOES Cloud clear imager radiance from GOES E and W Brightness temperature

GPSRO Global positioning system radio occultation Bending angle

Ground GPS Ground based GPS observations Zenith total delay

IASI Infrared atmospheric sounding interferometer from Metop A and B Brightness temperature

SAPHIR SAPHIR microwave radiance onboard Megha-Tropique satellite Brightness temperature

Satellite winds Atmospheric motion vectors from various geostationary and

polar orbiting satellites

Winds

SEVIRI Spinning enhanced visible and infrared imager instrument in

METEOSAT 9 and 10

Brightness temperature

SHIP Surface synoptic observation from ship Sea level pressure, wind,

temperature and humidity

SSMIS Special Sensor Microwave Imager/Sounder onboard Defense

Meteorological Satellite Program (DMSP-F17)

Brightness temperature

SYNOP + METAR

+AWS

Surface synoptic observation from land Sea level pressure, wind,

temperature and humidity

TEMP + DROP +

PILOT+WINPRO

+ VAD Winds

Radiosonde observations, upper-air wind proBle from a pilot

balloons and wind proBles, VAD wind observation from

RADARSs

Pressure, wind, temperature

and humidity

Figure 1. Schematic diagram of the NCUM data assimilation and forecast system.

J. Earth Syst. Sci. (2020) 129:209 Page 5 of 13 209

during the selected four assimilation cycles ofabove three cyclones. First column in Bgure 2 isthe SAPHIR passes during the cyclone Kyant.Figure 2(a–d) shows the passes during 06, 12 and18 UTCs of 20 October 2016 and 00 UTC of 21October 2016. It is noted that no SAPHIR passover the Indian Ocean region during the 18 UTCof 20 October 2016 and 00 UTC of 21 October2016. Five-day forecasts of cyclone Kyant wasprepared from the 00 UTC of 21 October 2016initial condition. Second column in Bgure 2 issimilar to the Brst column, but the SAPHIR pas-ses during the four assimilation cycles of cycloneVardah. Figure 2(e–h) shows the SAPHIR passesduring 00, 06, 12 and 18 UTCs of 7 December2016. It is noted that during 00 and 18 UTCs of 7December 2016, there was no SAPHIR pass overthe cyclone Vardah. The initial condition for the

day-5 forecast for cyclone Vardah is preparedbased on the 18 UTC of 7 December 2016. Thelast column in Bgure 2 shows the SAPHIR passesduring the four assimilation cycles of cycloneMaarutha. Figure 2(i–l) shows the SAPHIR pas-ses during 06, 12 and 18 UTCs of 14 April 2017and 00 UTC of 15 April 2017. During the assim-ilation cycles of cyclone Maarutha, only duringthe 12 UTC of 14 April 2017, there was noSAPHIR pass over the cyclone region. Five-dayforecasts of cyclone Maarutha was prepared basedon the initial condition from the 00 UTC of 15April 2017.The channel 6 (S6) of SAPHIR is distinct from

other similar instruments. Rani et al. (2016a)showed that the weighting function of 183.31 ± 11GHz channel (S6) extends the vertical coverage ofSAPHIR, relative to other similar microwaveinstruments and thus provides an additional ‘clean’183 GHz sounding channel which has only veryweak sensitivity to surface emission. To investigatethe impact of SAPHIR S6 on the simulation ofTCs, another OSE is conducted for TC Vardahonly.The simulated tracks have been estimated using

National Centre for Environmental Prediction’s(NCEP) vortex tracker (Marchok 2002) and thetracks are compared with the India Meteorological

Table 4. The NCUM model conBguration.

Model version 10.2

Resolution 17 km

Grid 1536 9 1152

Model levels 70

Forecast length 120 hrs

Time step 7.5 min

Radiation time step 1 hr

Figure 2. SAPHIR passes over India and surrounding oceanic regions during the selected four assimilation cycles of the cyclonesKyant (a–d), Vardah (e–h) and Maarutha (i–l).

209 Page 6 of 13 J. Earth Syst. Sci. (2020) 129:209

Department’s (IMD) best track dataset (Mohapa-tra et al. 2012). The track errors are estimated asdirect positional errors (DPE in km) and intensityerrors are estimated as root mean square errors(RMSE) of maximum sustained winds (MSW inm/s) and central sea level pressure (CSLP in hPa).The biases in each of these forecasted parametersare estimated as the difference between the IMDobserved value and the corresponding simulatedvalue (IMD observation � Model simulation). ToBnd the impact of assimilating SAPHIR radianceon the simulated tracks and intensity of TCs, thepercentage reduction in RMSE of each parameteris computed as normalized impact (NI):

NI ¼ RMSEEXPT � RMSECNTL

RMSECNTL� 100: ð1Þ

A positive/negative value of NI suggests theRMSE in the EXPT is higher/lower than that inthe CNTL, and hence a positive/negative value ofNI indicates the positive/negative impact ofSAPHIR radiances assimilation in the simulationof cyclones selected in this study. All metrics areaveraged over 24 hrs-time period. Five-day trackand intensity errors are reported for TC Vardahand Kyant, whereas two-day errors are reported forTC Maarutha because of its short lifespan.

6. Results

6.1 OSEs with all SAPHIR channels

Figure 3 shows the difference in the analysisincrement of speciBc humidity (g/kg) at the lowestmodel level between CNTL and EXPT for thethree cyclones. Figure 3(a–c), respectively, repre-sents the difference in the analysis increments ofspeciBc humidity for Kyant valid at 00 UTC of 21October 2016, Vardah valid at 18 UTC of 7December 2016 and Maarutha valid at 00 UTC of15 April 2017, the assimilation cycle from whichthe 5-day forecasts were computed. It is noticedfrom Bgure 3, that except for Maarutha (Bgure 3c),assimilation of SAPHIR radiances produced amoist model atmosphere, positive difference in theanalysis increment. As mentioned above, sinceMaarutha was a short-lived cyclone, even if therewas SAPHIR pass over the cyclone area (Bgure 2l)during the assimilation cycle of 00 UTC of 15 April2017, the data impact was not seen in the analysis.This can be due to the presence of the cyclonecharacteristics in the ambient Cow and hence the

data assimilation did not make noticeable changesto the background. For Kyant (Bgure 3a) andVardah (Bgure 3b), though there was no explicitSAPHIR passes over the cyclone area (Bgure 2dand h), the information assimilated in the previouscycles and hence in the background Belds produceda moist environment for further enhancement ofthe cyclone.Figure 4(a) shows the spatial distribution of sea

level pressure (SLP) and the 10-m winds in theCNTL and Bgure 4(b) shows the difference in SLPand 10-m wind speed between CNTL and EXPTfor Vardah valid at 18 UTC of 7 December 2016.The observed position of the cyclone (IMD bestposition) is marked as a circle with extended arms,while TC position estimated from simulations areshown in a solid circle (EXPT) and circle with a

Figure 3. Differences in the speciBc humidity increments(g/kg) in the lowest model level between CNTL and EXPTfor (a) Kyant (20161021T00), (b) Vardah (20161207T18),(c) Maarutha (20170415T00) (negative differences are shownas dashed contours).

J. Earth Syst. Sci. (2020) 129:209 Page 7 of 13 209

central dot (CNTL). The difference in the SLPbetween CNTL and EXPT over the cyclone area isnegative (Bgure 4b) and this indicates SAPHIRassimilation produced a comparatively lower pres-sure. Also it can be seen that the vector wind dif-ference is also large, leading to the strong windsdue to the assimilation of SAPHIR radiances.From Bgure 4(a and b), it is also seen that in bothexperiments, the simulated initial position of theTC is different from the observed one. Similardifferences in the simulated initial position of othertwo cyclones were also noticed, but not shown herefor brevity. Though the assimilation of SAPHIRradiances produced moist and stormy atmosphereconducive for cyclone, the difference in the simu-lated initial position of the cyclone from the IMDbest position could contribute to the error in thesimulated cyclone tracks.

6.1.1 Track and intensity errors

The track and intensity of the TCs are estimated interms of DPE and RMSEs of both MSW andCSLP. These values are calculated for both CNTLand EXPT and the impact of SAPHIR radianceassimilation is estimated through NI.Figure 5 shows the simulated tracks from CNTL

and EXPT, along with the best tracks from theIMD. Figure 5(a) is the tracks simulated fromCNTL and EXPT along with the IMD best trackfor cyclone Kyant, and Bgure 5(b and c) are thesame for the cyclones Vardah and Maarutha. Theinitial locations of all the storms in the analysis

differ from the IMD observed locations. However,further 6-hrly location, track, and directions weresimulated considerably well by NCUM. The simu-lation of TC Kyant is particularly noticeable(Bgure 5a). The Kyant started as a depression on21st September 2016 and headed fast towards thenortheast in the next couple of days. It slowlymoved northward on day-3 before steering fasteastward on day-4 and 5. Both CNTL and EXPTsimulated all these features including sharpre-curving on day-3. We notice that the predictedtracks up to day-3 for TC Kyant and Vardah andday-1 for Maarutha were similar in both CNTL andEXPT. For the case of cyclones, Kyant and Vardah(Bgure 5a and b), the CNTL was close to the IMDbest track beyond day-3, while for Maarutha(Bgure 5c), during land fall CNTL become closer toIMD best track.The 24-hrly averaged errors in track positions for

all the cyclones for both experiments are shown inBgure 6(a) and the RMSE of CSLP and MSW arepresented in Bgure 6(b and c), respectively. The6-hrly errors (not shown) are also analyzed and donot give any qualitative different interpretations as24-hrly averaged errors. For all three TCs, anexperiment in which SAPHIR radiance is included,i.e., CNTL, has better track forecasts compared toEXPT. A larger error (83–150 km) is observed inforecasted tracks in day-1 simulations of bothCNTLand EXPT. The track errors generally increase withsubsequent forecast days. The track errors fromboth experiments are close to each other up to day-3and begin to show differences beyond day-3.

Figure 4. Spatial distribution of (a) sea level pressure (hPa, contour), 10 m winds (m/s, vectors) from CNTL, and (b) thedifference between CNTL and EXPT for the cyclone Vardah at 18.00 UTC of 7 December 2016. MSLP differences are shaded over± 1 hPa with dashed lines surrounding negative values. The IMD cyclone position is indicated by a circle with extended arms,while the cyclone position estimated from simulations are shown with a circle (EXPT) and circle with a central dot (CNTL).

209 Page 8 of 13 J. Earth Syst. Sci. (2020) 129:209

The model simulates much stronger stormsagainst the IMD observations. This is noticed inboth CNTL and EXPT mean errors (not shown).For all the TCs, the error in CSLP suggests anoverestimation of storm intensiBcation in all fore-cast days. The MSW errors also indicated themodel’s tendency to simulate stronger winds than

observations. Interestingly, a sharp decrease inRMSE of MSW (Bgure 6c) is observed on the day-2forecast for all the TCs. Beyond day-2, an increasein intensity errors is noticed except for the case of aday-5 error for TC Kyant. This is coherent with asharp reduction in 24-hr averaged day-5 track errorfor Kyant (Bgure 5a). The RMSE of both CSLPand MSW is similar in both CNTL and EXPT.This suggests that SAPHIR humidity observationshave a marginal impact on winds and pressureBelds near the TC’s core area.

Figure 5. Simulated trajectories of tropical cyclones(a) Kyant, (b) Vardah, and (c) Maarutha. IMD best tracksare shown in grey, black lines show track forecasts for CNTL(solid line) and EXPT (dashed line). The simulations areinitialized at: 00 UTC, 21st Oct 2016 (Kyant); 18 UTC, 7thDec 2016 (Vardah); and 00 UTC, 15th April 2017 (Maarutha).

Figure 6. Daily averaged (a) track errors (DPE), (b) RMSEof CSLP, and (c) RMSE of MSW, for Maarutha (9), Vardah(diamond), Kyant (triangle). CNTL (EXPT) is shown in black(blue). The composite values over all cyclones are shown inblack (CNTL) and grey (EXPT) lines. Track errors are shownuntil the landfall (5 days for Vardah and 2 days for Maarutha)and 5 days for Kyant till depression stage. Solid lines depictcomposites of errors over all three cyclones.

J. Earth Syst. Sci. (2020) 129:209 Page 9 of 13 209

The 24 hrs averaged composites of DPE, andRMSE of CSLP and MSW for all three TCs aretabulated in table 5. The average DPE in day-1 isslightly higher in the CNTL compared to EXPT;however, day-2 onwards the average DPEdecreased in the CNTL throughout the forecastlength as seen from table 5. This can be attributedto the retained memory of the model due to theassimilation of SAPHIR radiances. Similar to DPE,the RMSE in the MSW was also same for both theexperiments in day-1, while the CNTL showed aslight improvement in the subsequent forecast daysas seen from table 5. Unlike DPE and the RMSE inthe MSW, the RMSE in the CSLP was less in theCNTL in day-1 forecast, whereas for the subse-quent forecasts the RMSE in the CSLP remainsalmost same in both CNTL and EXPT. The 24-hrsaveraged composites of NI of DPE, and RMSE ofCSLP and MSW are tabulated in table 6. It can beseen from table 6 that SAPHIR assimilation pro-duced improvements of 2.4%, 2.0%, 12.2% and38.3%, respectively, in the cyclone tracks in day-1,day-2, day-4 and day-5; however, a deterioration inthe predicted track (�3.4%) due to the assimila-tion of SAPHIR radiances is noticed in day-3forecast. The impact SAPHIR assimilation in theCSLP was higher (28.3%) in the day-1 forecast, butthe impact was negative in day-2 and day-4 fore-casts as seen from table 6. Unlike the NI in trackand CSLP, the NI in MSW showed positive impactof SAPHIR throughout the forecast length com-pared to the EXPT. Thus the DPE, CSLP, MSW,

and cyclone track were improved due to theassimilation of SAPHIR radiances.

6.2 Channel-6 OSE

The channel S6 of SAPHIR is unique compared toother similar instruments and it peaks lower thanthe lowest peaking channel of MHS (Rani et al.2016a). A separate OSE was conducted to assessthe impact of S6, if any, on the track and intensityof the TCs. Hereafter, this OSE is read as S6-EXPT. In this experiment along with otherobservations, only S6 of SAPHIR was assimilatedin the case of TC Vardah. Similar to the other twoOSEs, CNTL and EXPT, four data assimilationcycles are performed to produce the analysis validat 18 UTC of 7 December 2016, and from thisinitial condition 5-day forecasts were prepared.

6.2.1 Track and intensity errors

The errors in track positions at different forecastlead times for TC Vardah are shown in Bgure 7(a).The initial track errors for all the experiments(CNTL, EXPT, S6-EXPT) are about 150 km. Thetrack errors are around 200 km by day-3 in allexperiments and increased sharply beyond day-3,particularly for channel-6. Day-4 and day-5 trackerrors are as high as 390.93 and 563.92 km,respectively, in S6-EXPT. It can be noticed thatS6-EXPT track errors are larger than both CNTL

Table 5. Composite 24-hrly averaged DPE (km) and RMSE of MSW (m/s) and CSLP(hPa).

Forecast

days

DPE

(CNTL)

DPE

(EXPT)

MSW

(CNTL)

MSW

(EXPT)

CSLP

(CNTL)

CSLP

(EXPT)

1 120.0 119.8 7.0 7.0 1.4 1.8

2 156.0 159.1 2.6 2.9 1.7 1.7

3 163.7 158.2 6.6 6.9 4.7 5.1

4 233.4 262.0 9.3 10.1 9.0 8.9

5 198.0 273.8 9.0 9.2 11.8 11.8

Table 6. Composite normalized impact (NI).

Forecast

days

Impact in

track (%)

Impact in

CSLP (%)

Impact in

MSW (%)

1 2.4 28.3 0.6

2 2.0 �1.3 12.7

3 �3.4 9.0 4.8

4 12.2 �1.9 8.8

5 38.3 0.7 2.8

209 Page 10 of 13 J. Earth Syst. Sci. (2020) 129:209

and EXPT and indicate deterioration of trackforecasts due to the assimilation of S6 alone.The intensity errors are presented as RMSE of

CSLP and MSW Bgure 7(b and c), respectively. Inall the three experiments, intensity errors reduced

on day-2 and increased thereafter. MinimumRMSEs of CSLP and MSW of around 2 hPa and2.5 m/s, respectively, are found in all experimentson day-2. Maximum RMSEs of about 9 m/s and14 hPa in MSW and CSLP, respectively, areobserved in S6-EXPT in day-5 forecast, which aresmaller than other experiments. Interestingly inS6-EXPT, the RMSE in both CSLP and MSWreduced due to the assimilation of S6 radiancesalone compared to the CNTL, where all theSAPHIR channels are assimilated. This can beattributed to the better representation of lowerhumidity in the S6-EXPT as compared to thecombined vertical proBle of humidity provided byall the SAPHIR channels.

7. Conclusions

OSEs were designed to investigate the impact ofSAPHIR radiance assimilation in the simulationof track and intensity of three selected cyclones(Kyant, Vardah and Maarutha) formed over theBoB during 2016–2017 using the NCUM Hybrid-4DVAR data assimilation and forecast system,which uses the day-to-day error covariances fromthe NCMRWF Ensemble Prediction System.Assimilation of all the SAPHIR channel radiancesproduced improvement in the track, and intensityof the simulated cyclones with respect to the IMDobserved best track. An average reduction of upto 9% and 12% RMSE in the CSLP and MSW,respectively, and 38% in DPE for the TCs areachieved within a forecast lead time of 120 hrsdue to the assimilation of SAPHIR radiances.Since the channel-6 of SAPHIR, which peaks inthe lower altitudes, is unique compared to similarinstruments, a separate OSE which assimilatedonly the channel-6 radiances from SAPHIR alongwith other observations was carried out only forcyclone Vardah. Initial assessment shows thatthe SAPHIR channel-6 assimilation improves theintensity of TC simulation, while deteriorate thetrack simulation compared to the assimilation ofall SAPHIR channels. In general, SAPHIRassimilation improved the track and intensitysimulation of TCs. More case studies should becarried out to conBrm the impact of SAPHIRchannel-6 in the TC intensity prediction. Thisstudy brought out the importance of similarmicrowave humidity instruments in the samefrequency range for the detailed exploration ofcyclone track and structure.

Figure 7. Daily averaged (a) track errors (DPE) (b) RMSE ofCSLP and (c) RMSE of MSW for CNTL (black circle), EXPT(grey cross) and S6-EXPT (light grey triangles) for cycloneVardah.

J. Earth Syst. Sci. (2020) 129:209 Page 11 of 13 209

Acknowledgements

The authors would like to acknowledge Head,National Centre for Medium Range Weather Fore-casting for providing consistent support. Authorsalso would like to thank colleagues of theNCMRWF. We are thankful to the Indian SpaceResearch Organization for providing real-timeMegha-Tropique’s Sounder for Probing VerticalProBles of Humidity data and India MeteorologicalDepartment for providing the cyclone’s best tracks.

Author statement

AG conducted the OSE experiments and extractedtrack locations. AG and DC generated the Bgures.DC conceptualized this work and analyzed theresults. SIR and JPG helped in interpreting theresults. All authors contributed to drafts andrevisions.

References

Bell W, English S J, Candy B, Hilton F, Atkinson N, SwadleyS, Baker, N, Bormann N and Kazumori M 2008 Theassimilation of SSMIS radiances in numerical weatherprediction models; IEEE Trans. Geosci. Remote Sens. 46884–900.

Chambon P, Meunier L F, Guillaume F, Piriou J M, Roca Rand Mahfouf J F 2015 Investigating the impact of thewater-vapour sounding observations from SAPHIR onboard Megha-Tropiques for the ARPEGE global model;Quart. J. Roy. Meteorol. Soc. 141(690) 1769–1779,https://doi.org/10.1002/qj.2478.

Chu K, Xiao Q, Tan Z M and Gu J 2011 A forecast sensitivitystudy on the intensity change of Typhoon Sinlaku (2008);J. Geophys. Res.: Atmos. 116(D22), https://doi.org/10.1029/2011JD016127.

de Rosnay P, Drusch M, Vasiljevic D, Balsamo G, Albergel Cand Isaksen L 2013 A simpliBed extended Kalman Blter forthe global operational soil moisture analysis at ECMWF;Quart. J. Roy. Meteor. Soc. 139(674) 1199–1213, https://doi.org/10.1002/qj.2023.

Dhanya M, Gopalakrishnan D, Chandrasekar A, Singh S Kand Prasad V S 2016 The impact of assimilating Megha-Tropiques SAPHIR radiances in the simulation of tropicalcyclones over the Bay of Bengal using the WRF model;Int. J. Remote Sens. 37(13) 3086–3103, https://doi.org/10.1080/01431161.2016.1194542.

Doherty A, Atkinson N, Bell W and Smith A 2015 Anassessment of data from the advanced technology micro-wave sounder at the Met ODce; Adv. Meteorol., Article ID956920, https://doi.org/10.1155/2015/956920.

Doherty A, Indira Rani S, Newman S and Bell W 2018BeneBts of assimilating SAPHIR observations on analysisand forecasts of tropical Belds in the Met ODce Global

Model; Quart. J. Roy. Meteorol. Soc., https://doi.org/10.1002/qj.3258.

Fabry F and Sun J 2010 For how long should what data beassimilated for the mesoscale forecasting of convection andwhy?Part I:On the propagation of initial condition errors andtheir implications for data assimilation; Mon. Weather Rev.138(1) 242–255, https://doi.org/10.1175/2009MWR2883.1.

Gentry M S and Lackmann G M 2010 Sensitivity of simulatedtropical cyclone structure and intensity to horizontalresolution; Mon. Weather Rev. 138(3) 688–704, https://doi.org/10.1175/2009MWR2976.1.

George J P, Rani S I, Jayakumar A, Mohandas S, Mallick S,Lodh A, Rakhi R, Sreevathsa M N and Rajagopal E N 2016NCUM data assimilation system; NCMRWF TechnicalReport, NMRF/TR/01/2016.

Greeshma M M, Srinivas C V, Yesubabu V, Naidu C V,Baskaran R and Venkatraman B 2015 Impact of local dataassimilation on tropical cyclone predictions over the Bay ofBengal using the ARW model; Ann. Geophys. 33(7),https://doi.org/10.5194/angeo-33-805-2015.

Kumar S, Indira Rani S, George J P and Rajagopal E N2018 Megha-Tropiques SAPHIR radiances in a hybrid4D-Var data assimilation system: Study of forecastimpact; Quart. J. Roy. Meteorol. Soc., https://doi.org/10.1002/qj.3251.

Liu Z, Schwartz C S, Snyder C and Ha S Y 2012 Impact ofassimilating AMSU-A radiances on forecasts of 2008Atlantic tropical cyclones initialized with a limited-areaensemble Kalman Blter; Mon. Weather Rev. 140(12)4017–4034, https://doi.org/10.1175/MWR-D-12-00083.1.

Lodh A, John G P and Rajagopal E N 2016 Extended KalmanBlter based land data assimilation system for soil moistureanalysis at NCMRWF; NCMRWF Technical ReportNMRF/TR/06/2016.

Lorenc A C 2003 Modelling of error covariances by fourdimensional variational data assimilation; Quart. J. Roy.Metereol. Soc. 129 3167–3182, https://doi.org/10.1256/qj.02.131.

Lu Q, Bell W, Bauer P, Bormann N and Peubey C 2011 Anevaluation of FY-3A satellite data for numerical weatherprediction; Quart. J. Roy. Meteorol. Soc. 137 1298–1311,https://doi.org/10.1002/qj.834.

Madhulatha A, George J P and Rajagopal E N 2017 All-skyradiance simulation of Megha-Tropiques SAPHIR micro-wave sensor using multiple scattering radiative transfermodel for data assimilation applications; J. Earth Syst. Sci.126(2) 24, https://doi.org/10.1007/s12040-017-0805-3.

Marchok T P 2002 How the NCEP tropical cyclone trackerworks. In Preprints, 25th Conf. on Hurricanes and TropicalMeteorology, San Diego, CA, Am. Meteor. Soc. P (Vol. 1).

Mohapatra M, Bandyopadhyay B K and Tyagi A 2012 Besttrack parameters of tropical cyclones over the North IndianOcean: A review; Nat. Hazards 63(3) 1285–1317, https://doi.org/10.1007/s11069-011-9935-0.

Rajagopal E N, Iyengar G R, George J P, Gupta M D,Mohandas S, Siddharth R, Gupta A, Chourasia M, PrasadV S, Aditi S K and Ashish A 2012 Implementation of uniBedmodel based analysis-forecast system at NCMRWF;NMRF/TR/2/2012.

RakhiR, JayakumarA, SreevathsaMNandRajagopal EN 2016Implementation and up-gradation of NCUM in BhaskaraHPC; NCMRWF Technical Report, NMRF/TR/03/2016.

209 Page 12 of 13 J. Earth Syst. Sci. (2020) 129:209

Rani S I, Doherty A, Atkinson N, Bell W, Newman S,Renshaw R, George J P and Rajagopal E N 2015 Impact ofMegha-Tropiques’ SAPHIR humidity proBles in the UniBedModel Analysis and Forecast System; In: 20th InternationalTOVS Study Conference (ITSC-20) (https://cimss.ssec.wisc.edu/itwg/itsc/itsc20/papers/1˙10˙rani˙paper.pdf).

Rani S I, Bell W, Doherty A M and Newman S 2016a Aninvestigation of the impact of the assimilation of M-TSAPHIR data in the Met ODce data assimilation system.NWPSAF report of visiting scientistmission NWP˙VS16˙03.Exeter, UK. Document NWPSAF-MO-VS-055. Available at:https://nwpsaf.eu/vs˙reports/nwpsaf-mo-vs-055.pdf.

Rani S I, Doherty A, Atkinson N, Bell W, Newman S,Renshaw R, George J P and Rajagopal E N 2016b EAect ofnew radiance observations on numerical weather predictionmodels; SPIE News room, 10.1117/2.1201607.006533.

Reale O, Lau W K, Susskind J, Brin E, Liu E, Riishojgaard LP, Fuentes M and Rosenberg R 2009 AIRS impact on theanalysis and forecast track of tropical cyclone Nargis in aglobal data assimilation and forecasting system; Geophys.Res. Lett. 36(6), https://doi.org/10.1029/2008GL037122.

Sarkar A, Chakraborty P, John P G and Rajagopal E N 2016Implementation of uniBed model based ensemble predictionsystem at NCMRWF (NEPS); NMRF/TR/02/2016.

Singh R, Ojha S P, Kishtawal C M and Pal P K 2013 Qualityassessment and assimilation of Megha-Tropiques SAPHIRradiances into WRF assimilation system; J. Geophys. Res.:

Atmos. 118(13) 6957–6969, https://doi.org/10.1002/jgrd.50502.

Singh S K and Prasad V S 2017 Impact of Megha-TropiquesSAPHIR radiances in T574L64 global data assimilationand forecasting system at NCMRWF; Int. J. Remote Sens.38(16) 4587–4610, https://doi.org/10.1080/01431161.2017.1323279.

Singh K S and Mandal M 2014 Impact of initial condition onprediction of Bay of Bengal Cyclone’Viyaru’ – A case study;Int. J. Comput. Appl. 94(10).

Singh S K, Prasad V S, Rajagopal E N and Basu S 2015Assimilation of cloud clear radiance of Megha-TropiquesSAPHIR in NCMRWF GFS (T574L64); In: 20th Interna-tional TOVS study conference (ITSC-20).

Wang Y 2009 How do outer spiral rainbands aAect tropicalcyclone structure and intensity? J. Atmos. Sci. 66(5)1250–1273, https://doi.org/10.1175/2008JAS2737.1.

Wu L, Su H, Fovell R G, Wang B, Shen J T, Kahn B H,Hristova-Veleva S M, Lambrigtsen B H, Fetzer E J andJiang J H 2012 Relationship of environmental relativehumidity with North Atlantic tropical cyclone intensity andintensiBcation rate; Geophys. Res. Lett. 39(20), https://doi.org/10.1029/2012GL053546.

Xu D, Liu Z, Huang X Y, Min J and Wang H 2013 Impact ofassimilating IASI radiance observations on forecasts of twotropical cyclones; Meteorol. Atmos. Phys. 122(1–2) 1–8,https://doi.org/10.1007/s00703-013-0276-2.

Corresponding editor: KAVIRAJAN RAJENDRAN

J. Earth Syst. Sci. (2020) 129:209 Page 13 of 13 209