Impact of Vortex Initialization with Satellite Data

Remote Sensing in Earth Systems Sciences https://doi.org/10.1007/s41976-018-0003-3

ORIGINAL PAPER

Impact of Vortex Initialization with Satellite Data Assimilation for Predicting Track and Intensity of Tropical Cyclone Chapala Vivek Singh 1 & A. Routray 1 & Devajyoti Dutta 1 & John P. George 1 Received: 22 June 2018 / Revised: 1 October 2018 / Accepted: 10 October 2018 # Springer Nature Switzerland AG 2018

Abstract The present study evaluates the impact of vortex initialization (VI) with assimilation of satellite observations in the National Centre for Medium Range Weather Forecasting-UK Met Office Unified (NCUM) global Model for prediction of extremely severe cyclonic storm (ESCS) Chapala formed over the Arabian Sea (AS) during 28 October–03 November 2015. For this purpose, two numerical experiments viz. Vort-Gts (assimilation of only Global Telecommunication System (GTS) observations with VI) and Vort-Rad (assimilation of GTS plus satellite observations with VI) are conducted. The model is run with a total of five different initial conditions (ICs) through 24 h cycles. The standard metrics of cyclone track errors such as Direct Position Errors (DPEs), CrossTrack Errors (CTEs), and Along-Track Errors (ATEs) are computed and found to have improved by 47, 40, and 57%, respectively in Vort-Rad over Vort-Gts. The intensity and land fall (LF) errors are significantly reduced in Vort-Rad in comparison with the VortGts. The statistical skill scores such as equitable threat score (ETS) and bias are better represented in the Vort-Rad. Keywords Vortex initialization . NCUM global model . GTS and satellite observation . Tropical cyclone

1 Introduction Tropical cyclones (TCs) are hazardous weather phenomenon causing extensive ravage to life and property along the coastal areas of the world. Owing to the strong destructive impact, there is an immense need of accurate prediction of TC track and intensity to reduce TC disasters. Many researchers through different approaches have reported the improvement in TC track and intensity prediction in various numerical weather prediction (NWP) models. Improving trend of TC prediction for 15 years (1991–2003) have been found for the sophisticated NWP models [46]. The upgraded physics scheme and higher resolution of models have found that a positive impact on TC track and intensity prediction occurred over various basins [7, 18, 23, 30, 31, 45]. Adopting the improved and state-of-the-art newer data assimilation techniques have proven focal for TC prediction [22, 29, 40, ]. The assimilation of radiance and Doppler Weather Radar (DWR) data in the

* A. Routray [email protected] 1

National Centre for Medium Range Weather Forecasting (NCMRWF), A-50, Sector-62, Noida, UP, India

Weather Research Forecasting (WRF) model has shown improvements in skill of TC prediction of the model for the Northern Indian Ocean (NIO) region [39, 45]. The assimilation of multi-satellite and conventional data sets has reported a gain in skill in the WRF model for TC forecasting [47, 48]. The TC analyses and forecasts are improved in NWP models using many other satellite-derived observations [10, 11, 17, 20, 54, 55]. Apart from these approaches, physical initialization of precipitation rates is also known to have a positive impact on hurricane intensity forecast in the WRF model [25]. The accuracy in TC prediction depends upon better representation of initial vortex position in the NWP models. Mostly, the larger-scale environmental steering flow governs the track of the TCs. On the other hand, the TC intensity depends upon many factors such as smaller-scale internal dynamics, moist convection, humidity, sea surface temperature (SST), etc. However, despite the remotely sensed modern satellite observations, the lack of observational network near the core of these data sparse convective systems limits the forecast skill of the dynamical models [1, 3, 5]. So, in another approach, the errors in TC track and intensity prediction are reduced with the creation of synthetic bogus vortex in the initial condition of models [19, 21, 26, 50]. The objective of the present study is to quantify the impact of vortex initialization on assimilation of satellite data in

Remote Sens Earth Syst Sci

NCMRWF UK Met office Unified global Model (NCUM; [41]) for prediction of track and intensity of TC BChapala.^ The following Section 2 deals with a detailed description of the NCUM global model. Section 3 presents details of numerical experiments and data and methodology adopted in the study. Synoptic features of the TC have been discussed in Section 4. Results and discussions have been presented in Section 5. The summary and conclusion of the study have been presented in Section 6.

2 Model Description In the present study, two numerical experiments (Vort-Gts and Vort-Rad) are conducted using the NCUM global model. The horizontal resolution of the model is approximately 17 km near mid-latitudes with a total of 70 vertical levels. The numerical scheme of the model is semi-implicit and semi-Lagrangian, having rotated latitude–longitude horizontal grid with Arakawa-C staggering. The physics option of the model is based on Global Atmosphere (GA) version 6.1 and Global Land (GL) 6.1 [51]. The scheme derives convection physics based on mass flux convection scheme with an increase of entrainment rate for deep convection [16]. The Planetary Boundary Layer (PBL) scheme of the model is of first order non-local with the reduction of turbulent mixing for stably stratified boundary layers [28]. The model has a time step of 7.5 min. It incorporates land surface process based on Joint UK Land Environment Simulator (JULES; [6, 12]). The model’s dynamical core, i.e., Endgame (Even Newer Dynamics for General Atmospheric Modeling of the Environment) is designed to allow the code to be switchable between various options, i.e., from non-hydrostatic to hydrostatic, deepatmosphere to shallow-atmosphere formulations [49, 51]. A detailed description of the model can be found in [43, 44].

3 Numerical Experiments and Methodology The present study assesses the impact of Vortex Initialization (VI) on assimilation of satellite data in the NCUM global model for simulating the track and intensity of ESCS Chapala. Two numerical experiments such as Vort-Gts and Vort-Rad are conducted using the NCUM global model for this purpose. In the Vort-Gts, only GTS observations are assimilated with the Vortex Initialization. The second experiment (Vort-Rad) is carried out by assimilating the GTS and satellite observation data in the presence of VI. The model is run with a total of five different initial conditions (29th October 2015 to 03rd November 2015) for both the experiments through 24-h cycles (Table 1). The variational data assimilation system uses an incremental four-dimensional scheme, 4DVAR [42]. This method of data assimilation uses a linear Perturbation Forecast (PF)

model to generate time-varying increments over the assimilation window [4]. The 4DVAR data assimilation technique is utilized to prepare the analyses for each IC run in both the experiments. The detailed list of GTS and satellite observations used in the assimilation cycle is presented in Table 2. The vortex initialization scheme utilized in the present study has been adopted from Centre for Australian Weather and Climate Research (CAWCR; [13]). The scheme is based on three parameter profile of TCs [9]. It uses observed values of TCs’ central pressure, maximum wind speed, and radius of outer closed isobar to completely create parameter sets required to specify vortex structure (i.e., latitude, longitude, central pressure, radius of outer closed isobar (roci) radius of maximum wind speed (rm), and radius of gale-force wind (r34). The main equation sets utilized in the scheme are reproduced from Davidson et al. [13]. The tangential wind profile at the surface can be written in terms of non-dimensional radial coordinate system Bs^ as: vðsÞ ¼ vm S α expð1−S α Þ

ð3:1Þ

where s = r/rm, r represents the radius. B푣m^ denotes the axis symmetric maximum wind speed at rm. In the above equation, Bα^ is a parameter, which is always positive and defined as below: 2:0 S ≤1 α¼ ð3:2Þ for ∈ð0:40; 0:72Þ S > 1 The gradient wind equation in terms of non-dimensional coordinate can be written as: dpðsÞ v2 ð s Þ ð3:3Þ ¼ ρrm fvðsÞ þ ds rm s where Bρ^ represents Bair density^ in M.K.S. unit. Similarly, the pressure deficit (dpc) at center can be written as: dpc ¼ ρvm expð1Þ½ f rm I1 þ vm expð1ÞI 2

ð3:4Þ

In this pressure deficit equation, the value of integrals is defined as below: 0

0

I 1 ¼ ∫∞ S α expð−S α Þds; I 2 ¼ ∫∞ S 2α−1 expð−2S α Þds

ð3:5Þ

For the relocation of storm to the observed location, the vortex initialization scheme uses only synthetic MSLP observations in 4DVAR. This bogus MSLP observation helps in defining the inner core circulations of the storm. The VI scheme first filters the analyzed circulation from the original analysis present. In the next step, it locates the storm to observed position after constructing an inner core. In the last, the VI scheme merges the inner core circulation with the largescale analysis at roci.

Remote Sens Earth Syst Sci Table 1

Details of the model simulations and observed landfall time of TC

S. no.

TC name (intensity)

Duration of TC (total no of days)

1

Chapala (ESCS)

28 Oct−04 Nov 2015 (8)

Simulation period in 24-h intervals, (no. of initial conditions) 29 Oct−02 Nov 2015 (5)

Region

Observed landfall, place name, timing, and lat-lon

Arabian Sea (AS)

Between 0100 and 0200 UTC, 03 Nov. 2015 (Yemen; 14.1 N/48.65° E)

ESCS extremely severe cyclonic storm)

The hydrological features associated with precipitation such as Total Precipitable Water (TPW) and Moisture Transport (MT) has been calculated in the present study. The mathematical formulation of TPW can be expressed as: TPW ¼

1 ρw g

0

∫ps qðpÞ:dp

ð3:6Þ

where Bρw’^ is the water density, Bg^ is acceleration due to gravity, q(p) is mixing ratio of water vapor, p is pressure, and ps is surface pressure. The NCEP TC tracker is used to extract the modelpredicted TC position (Latitude and Longitude) for both the experiments [32]. The model-predicted TC tracks are verified with the India Meteorological Department (IMD) provided best track observations of TC. Standard TC error metrics such as Direct Position Error (DPE), Along-Track Error (ATE), and Cross-Track Error [14] have been computed for both the numerical experiments. Intensity forecast of TC has been calculated in terms of Mean Absolute Error (AE) of 10 m wind and MSLP in the present study. The model’s skill for rainfall prediction is quantitatively illustrated in terms of equitable threat score (ETS) and bias in both the experiments. These skill scores can be defined as follows (http://www.cawcr.gov.au/ projects/verification/): ETS ¼ Hits−Hitsrandom =ðHits þ Misses þ False alarms−Hitsrandom Þ

where Hitsrandom = (Hits + False alarms) (Hits + Misses)/Total. Similarly: Bias ¼ ðHits þ False alarmsÞ=ðHits þ MissesÞ:

Table 2 cycle

Observations from GTS and satellite used in the assimilation

S. no. GTS observations

Satellite observations

1 2 3 4 5 6 7.

NOAA (4-channels, 15, 16, 18, 19) GPSRO (COSMIC, GRAS, GRACE) GOES (E&W) METEOSAT (2 channels, 7 and 10) METOP (BA^&^B^) AQUA AMSU(BA^&^B^)

SYNOP PILOT AWS BUOY ARGO RADIO/RAWINSONDE AIRCRAFT

The ETS and bias are also commonly known as Gilbert skill score and frequency bias, respectively. A model giving perfect forecast assumes a value B1^ of ETS and bias. However, the ETS and bias values higher/lower than B1^ illustrate that the model notably over-/underpredicted the forecast field.

4 Synoptic Features and Life Cycle of TC Chapala Figure 1a–c depicts the observed track and INSAT 3-D satellite imagery of TC Chapala. The TC was an extremely severe cyclonic storm (ESCS) which formed over the Arabian Sea (AS) from 28th October to 4th November 2015(Fig. 1a). Historically, the cyclonic storm Chapala was unique in the sense that it was the first severe cyclonic system which crossed the Yemen coast after the severe cyclonic storm of May 1960. The intensity of the TC was so high (maximum wind speed~ 215 km/hr) that IMD coined the term BExtremely Severe Cyclonic Storm^ first time for categorization of TCs over the NIO region. Also, it had a very long track and life time (~ 2248 km and ~ 7 days) over the AS, which makes TC Chapala distinct from other general TCs forming over the basin. These were the primary reasons due to which TC Chapala was chosen for the present study. TC Chapala intensified into depression and deep depression in the morning and evening hours of 28th October 2015 respectively. During early morning time of 29th October 2015, the deep depression further intensified to cyclonic storm. Due to the favorable environmental features such as high sea surface temperature (SST) and low vertical wind shear, the system very rapidly intensified into a severe cyclonic storm (SCS) and a very severe cyclonic storm (VSCS) in the evening and midnight of 29th October 2015 respectively and a strong eye wall clearly seen in the satellite imagery (Fig. 1b). It further upgraded into an ESCS in the morning of 30th October. The storm moved further west due to the presence of anti-cyclonic circulation over the north-west of the system. Despite its interaction with land and dry air intrusion, the system was able to maintain its intensity till 1st November 2015 due to the presence of low vertical wind shear in the


Fig. 1 a Observed track; b INSAT-3D satellite imagery of the TC valid at 29th October, 2015 and c for 3rd November, 2015, respectively

west and west-southwest of the system. Further, the system started weakening and crossed the Yemen Coast as a VSCS around 0100–0200 UTC of 3rd November 2015 (Fig 1c). The synoptic features associated with the TC can be found in I M D ’s R S M C r e p o r t o f y e a r 2 0 1 6 ( h t t p : / / w w w. rsmcnewdelhi.imd.gov.in/images/pdf/publications/annualrsmc-report/RSMC-2015.pdf). A brief description of the TC is presented in Table 1.

5 Results and Discussion Two numerical experiments (Vort-Gts and Vort-Rad) are conducted to assess the impact of Vortex Initialization on assimilation of satellite observation data in the NCUM global

model. The ESSC Chapala is used as a test case for this purpose. Both of the experiments are initialized through 4D-VAR assimilation along with the bogus scheme. The following sections deal with the model simulated results and discussions.

5.1 TC Analysis through GPM Radar Satellite-based imageries provide information about the inner core convection of TCs and hence assist in accurate track and intensity prediction of TCs [8]. TC Chapala developed in the very warm waters of the Arabian Sea to the west of India on October 28, 2015. Figure 2a–c shows rainfall derived from GPM’s Microwave Imager (GMI) and Dual-Frequency Precipitation Radar (DPR) instruments valid for 00 UTC of 29th, 31st Oct. and 2nd Nov. 2015, respectively. The GPM


Fig. 2 Global Precipitation Measurement (GPM) radar analysis as based on 00 UTC of a 29th October; b 31st October, 2015; and c 2nd November, 2015, respectively (courtesy: National Aeronautics and Space Administration (NASA))

core observatory satellite had good views of the TC on 29th Oct 2015 at 0332 and 1456 UTC, respectively (Fig. 2a). GPM’s rainfall from the first pass (Fig. 2a) shows that Chapala was close to hurricane intensity with the location of a developing eye as clearly shown by GMI. By the second pass (Fig. 2b), Chapala’s maximum sustained winds were estimated at 65 kt, making it a category 1 on the Saffir-Simpson hurricane wind scale. Light to moderate precipitation around the small TC was observed by the GMI. Rainfall near the center was measured falling at a rate of slightly more than 28 mm per hour with the first pass and 31.9 mm per hour in the second examination. The 3-D vertical structure of feeder

bands and other scattered storms around Chapala’s center is shown in these views using from GPM’s Ka and Ku bands radar (DPR) reflectivity data. The radar swath is shown in a lighter shade of blue. With the first pass, the GPM radar measured the rain over the eastern side of the storm with the rain rate around 64 mm per hour from the feeder band. The storm top heights were measured by GPM’s radar reaching the altitudes up to 15 km in the feeder band. Generally, most of the storm top heights were much lower. Thus, the GPM’s microwave imager and dual-frequency precipitation radar instruments showed the intensity and location of precipitation within the intensifying TC.


5.2 Vertical Cross Section of Vorticity

5.3 Wind

Mid-level vortices have been found to play a very crucial role in the intensification and weakening of tropical cyclones [27]. The analysis of the vertical cross section of vorticity depicts local rotational behavior of TC. Figure 3a, b illustrates longitudinally averaged (at center of TC) vertical cross section (latitude and pressure) of vorticity (10−5 s−1, shaded positive and contours negative) for both the numerical experiments. The analysis is valid for 00 UTC of 29th October 2015. From the figure, it is seen that both the numerical experiments are able to capture the vertical rotational structure of the TC (i.e., low-level convergence and upper level divergence). However, the structural differences in the vorticity field can be seen between the two analyses. Nonetheless, it is noticed that the Vort-Rad (Fig. 3b) has simulated the better features in comparison with the Vort-Gts (Fig. 3a). The maximum value of positive vorticity (shaded) of Vort-Gts is lower (25–30 × 10 −5 s −1 ) in comparison with the Vort-Rad (40–45 × 10−5 s−1). Strong higher values of vorticity are depicted in the lower troposphere in the Vort-Rad as compared to the Vort-Gts. A strong positive vorticity is advected beyond 500–400 hPa in both the experiments causing strong divergence aloft at the upper atmosphere around 200 hPa. However, these features are clearly seen in the Vort-rad. The subsidence of the ambient air is also seen towards both sides of the TC center in both the experiments. It is clear that the Vort-Rad depicted a strong well-defined in-flowing cyclone at model initial time. The improvements in the vorticity field in the Vort-Rad may be primarily due to the additional assimilation of satellite observations in the model in comparison with the Vort-Gts.

Various aspects of two-dimensional surface wind field structures give information about storm position, duration, and intensity [24]. Figure 4a, b depicts 850 hPa wind analyses of TC based on 00 UTC of 29th Oct 2015 for Vort-Gts and VortRad, respectively. It is clear that Vort-Rad captures a more intense and well-structured cyclonic circulation (Fig. 4b) in comparison with the Vort-Gts (Fig. 4a). The magnitude of wind in the Vort-Rad is higher approximately by 5 m/s near the storm center. It is seen that the storm position from the Vort-Rad lies close to the observed position (marked BX^ in the figure) in comparison with the Vort-Gts. Therefore, comparison of wind vector fields illustrate that the wind in both the analyses could capture the signature of the TC at model initial time. However, the Vort-Rad reproduced more organized and intense evolution of the storm than the Vort-Gts.

Fig. 3 Vertical cross section of vorticity (10−5 s−1) at model initial time for a Vort-Gts and b VortRad valid at 00 UTC of 29th Oct, 2015

5.4 Vertical Velocity A cross section profile of vertical velocity can be used to identify the areas of deepest convection in TC [2]. Figure 5a, b depicts the longitude-height cross section of the vertical velocity of TC at the ESCS phase (00 UTC of 31st Oct 2015 as per IMD) for both the numerical experiments. A clear difference in the vertical velocity distribution is seen in both the simulations. In the Vort-Gts, the distribution of maximum vertical motion (vary up to 220 cm/s) is approximately seen in 400 to 200 hPa (Fig. 5a). On the other hand, the simulated maximum vertical velocity from Vort-Rad (220–260 cm/s) in the pressure level 700 to 300 hPa is higher than the Vort-Gts (Fig. 5b). These features indicate the presence of deep convection and warm core in the Vort-Rad compared to the Vort-

Remote Sens Earth Syst Sci Fig. 4 Model simulated wind and magnitude (shaded, m/s) at 850 hPa for a Vort-Gts and b VortRad valid at 00 UTC of 29th Oct, 2015. BX^ mark near the storm center represents the observed position of TC

a

b

5

Gts. It is also seen that the vertical velocity is simulated to a wider horizontal extent in the Vort-Gts; however, a welldefined region of the vertical velocity is seen in the VortRad. Therefore, the separation of the eye and eye wall region is clearly seen in the Vort-Rad. Hence, the Vort-Rad captures a better structure of TC’s vertical velocity in comparison with the Vort-Gts.

5.5 Total Precipitable Water and Moisture Transport Balancing of the frictional dissipation in TCs is carried out through energy from the environment and ocean in the form Fig. 5 Longitude-height cross section of vertical wind (cm/s) based on 00 UTC of 31st Oct, 2015 (mature stage of TC) for a Vort-Gts and b Vort-Rad, respectively

10

15

20

25

30

35

40

of latent heat (moisture). This moisture transport is the primary force to drive TC formation and intensification [15, 38]. This ultimate energy source of TCs determines the precipitation distribution around them. So, the study of hydrological features such as moisture transport (MT) and total precipitable water (TPW) could better help in understanding the TC intensification. In the present study, these two intensity diagnosing metrics are investigated for both the experiments during mature phase of the TC (Fig. 6a, b). It is seen that the Vort-Rad reported a larger amount of MT (approx 250–320 kg m−1 s−1, Fig. 6b) near the storm center than the Vort-Gts (220– 260 kg m−1 s−1, Fig. 6a). This larger amount of moisture influx


in the Vort-Rad might have increased the specific humidity (or equivalent potential temperature) near the TC center. The increased amount of moisture influx contributes in intense total precipitable water in the Vort-Rad (70–80 kg/m2) near the TC center as compared to the Vort-Gts (60–65 kg/m2). It is also seen that the horizontal spread of TPW of the range of 50– 65 kg/m2 is more in the Vort-Rad as compared to the Vort-Gts. Thus, the Vort-Rad depicted the progressive penetration of moisture flux near the TC center and represented the higher and more realistic values of TPW.

5.6 Track Forecast of TCs The accurate track forecast of TCs is of immense importance as it saves lives and reduces economic losses. This necessitates the post-storm track verification of cyclones predicted by NWP models. Operational agencies across the globe evaluate track forecasting skill of their models with an objective to further improve model performance in the future [34, 36]. For both the numerical experiments, the model simulated and IMD observed best tracks along with DPEs are depicted in Fig. 7a–j. From the figure, it is found that the model is able to realistically predict the tracks for all the ICs in both the experiments. However, in comparison with the Vort-Rad, the Vort-Gts tracks show slightly diverging behavior from the observed tracks for all the ICs. The most noticeable deviation of the VortGts tracks is noticed in Fig. 7c, g, in which the tracks are more displaced towards north-ward and south-ward with respect to the observed tracks. However, the pattern is significantly improved in the Vort-Rad as the tracks are more correlated with the corresponding observed tracks in both the ICs. Similar discrepancy is also noticed in the Fig. 6 Model-simulated moisture transport (kg m−1 s−1; wind vector) and integrated total precipitable water (kg/m2; shaded) for a Vort-Gts and b VortRad based on 00 UTC of 31st Oct, 2015 (at mature stage of TC)

Vort-Gts (Fig. 7i) as compared to the Vort-Rad. It is also clearly seen that for all the ICs, the Vort-Rad simulated tracks are matched reasonably well with the IMD best track in comparison with the Vort-Gts. The model is more skillful for TC tack prediction for Vort-Rad as compared to Vort-Gts in all the ICs. As per the above discussion, the tracks predicted by VortRad are in good agreement with the corresponding observed tracks; therefore, the corresponding DPEs from the Vort-Rad are noticeably less in comparison with the Vort-Gts in Fig. 7b, d, f, h and j, respectively. In these figures, the DPEs are illustrating an increasing trend with the increase in the lead time in both the numerical experiments. However, the DPEs from Vort-Rad are significantly reduced in all forecast hours with all ICs as compared to the Vort-Gts. The initial position errors are reduced reasonably well in the Vort-Rad in all the simulations as compared to the Vort-Gts, hence, the forecast skill of the Vort-Rad is improved. This gain in skill of the model during initial times may be attributed to the synthetic bogus initialization scheme along with the assimilation of radiance observations. Figure 8a–c illustrates the mean values of DPEs, CTEs, and ATEs along with gain in skill (percentage line) of Vort-Rad over Vort-Gts. It is noticed that the mean initial position error of the storms is 20 km in the Vort-Gts. The error is reduced reasonably well in the Vort-Rad (10 km). It is obvious that both the experiments illustrate an increasing trend of mean errors with the lead in forecast length. However, the mean errors are lesser in the Vort-Rad in comparison with the Vort-Gts. The range of variation of mean DPEs is approximately from 20 to 175 km in the Vort-Gts (Fig. 8a); the errors are significantly reduced in the Vort-Rad (11 to 108 km). The gain in skill of errors from Vort-Rad varies from 35 to 61%

Remote Sens Earth Syst Sci Fig. 7 a IMD best estimated track and model simulated tracks and b corresponding DPEs from VortGts and Vort-Rad experiments from IC 00 UTC 29th Oct, 2015. c–d, e–f, and g–h are same as a–b but for ICs based on 00 UTC of 30th and 31st Oct and 1st and 2nd Nov, 2015, respectively

over Vort-Gts. The gain in skill of mean error averaged over all the forecasts is 47%. The mean ATEs and CTEs of Vort-Gts and Vort-Rad are presented in Fig. 8b, c, respectively. It is found that the mean ATEs are negative at all forecast lengths for both the experiments (Fig. 8b). The negative ATE values suggested that the model predicted tracks are slower in comparison with the observed track. The mean ATE values are increased progressively with the increase of the forecast period. However, the values of mean ATEs in the Vort-Rad (− 10 to − 103 km) are lesser in magnitude as compared to Vort-Gts values (− 38 to − 126 km). The gain in skill of Vort-Rad over Vort-Gts varies from 18 to 73% throughout the forecast period. The maximum gain is noticed in the 12- and 72-h forecast. The gain in skill is reduced gradually with the increase of the forecast length. Similarly, the mean CTEs (Fig. 8c) are also negative in both the experiments. This result indicates that the

predicted tracks are always left of the observed tracks. The mean CTEs are following similar trend as of the mean ATEs throughout the forecast period. The mean ATEs from Vort-Gts varies from the range − 36 to −88 km are higher than the VortRad (− 22 to − 50 km). The gain in skill of Vort-Rad over VortGts varies from 30 to 45% throughout the forecast period. The overall mean of gain in skill of CTE and ATE are 40 and 57%, respectively.

5.7 Landfall Position and Time Errors of TCs Apart from track, accurate prediction of LF position and time of TCs is also required to avoid cyclone-related disasters well in advance. Mohapatra et al. [35] evaluated the performance of cyclone LF predicting skill of the operational model of IMD during the years 2003–2013 (total 11 years). Similar

Remote Sens Earth Syst Sci Fig. 8 a Mean DPEs (km), b mean ATEs, and c mean CTEs as well as gain in skill (%; line)

LF position and time verification of TC Chapala is also carried out for both the numerical experiments. The landfall (LF) position (km) and time errors (hours) with gain in skill of Vort-Rad over Vort-Gts is presented in Table 3. From the Table, it is found that the model is not able to predict the landfall position and time correctly in both the experiments. However, the LF time and position errors are improved in the Vort-Rad in comparison with the Vort-Gts. It is seen that the LF time errors of both the experiments are negative for most of the ICs, indicating that the model-predicted LF time is delayed from the observed LF time. The Vort-Rad is exactly predicting the LF time for 30th Oct. and 02nd Nov,,2015 ICs, respectively. However, the Vort-Gts could not predict the correct LF time for any of the ICs. It shows maximum positive (06 h) and negative (12 h) LF time error in the ICs based on 30th October and 01st November 2015, respectively. The LF position errors in the Vort-Gts (99 to 178 km) are higher in comparison with the Vort-Rad (35–147 km). The improvement in the LF position errors of Vort-Rad over Vort-Gts ranges from 17 to 65%. Thus, the Vort-Rad is more skillful in comparison with the Vort-Gts for predicting LF time and position of the TCs.

well [53]. The TC intensity prediction skill of NWP models can be quantified in terms of time averaged absolute error (AE) of mean sea level pressure (MSLP) and 10 m maximum sustained winds [43, 44]. Figure 9a, b depicts AE of MSLP and 10 m maximum sustained winds for TC Chapala obtained from both the numerical experiments. From the figure, it is found that both the experiments could not capture the intensity of TC significantly. However, the mean intensity errors are improved in the Vort-Rad in comparison with the Vort-Gts. It is seen that the mean AE values of MSLP and MSW are reasonably less in the Vort-Rad in comparison with the Vort-Gts throughout the forecast hours. The variation in range of AE of MSLP (Fig. 9a) for Vort-Gts and Vort-Rad is varying from the range of 8 to13.8 hPa and 7.9 to11.7 hPa, respectively. Similarly, the AE values of MSW follow nearly the same pattern as by the AE of MSLP. The mean AE of MSW (7.8 to13.2 m/s) is reasonably less in the Vort-Rad as compared to the Vort-Gts (8.2 to 15.4 m/s). The root mean square error (RMSE) of AE of MSLP and MSW for Vort-Rad are reduced by 03 hPa and 02 m/s than the Vort-Gts. Hence, the Vort-Rad is more skillful for intensity prediction of TC in comparison with the Vort-Gts.

5.8 Intensity Forecast of TCs 5.9 Skill of Rainfall Simulation The internal dynamics and moist convective processes are prime factors affecting the intensity forecast of the TC. Due to the lack of inner core observations, most of the models are generally unable to represent the TC intensity

Equitable threat score (ETS) and bias calculation are among one of the most widely accepted skill score metrics for model quantitative precipitation forecast verification. [33, 52]. For

Remote Sens Earth Syst Sci Table 3

Landfall (LF) errors from Vort-Gts and Vort-Rad simulations

Different ICs (00 UTC)

Observed LF time

LF position errors (km)

LF time errors (hours)

Vort-Gts

VORT-RAD (skill in %)

Vort-Gts

VORT-RAD

No landfall prediction 178

No landfall prediction 147 (17)

No landfall prediction + 06 (1900 UTC 02 Nov.)

No landfall prediction 00 (0200 UTC 03 Nov.)

31 Oct. 2015

199

142 (29)

− 06 (0800 UTC 03 Nov.)

− 06 (0800 UTC 03 Nov.)

01 Nov. 2015

186

101 (46)

02 Nov. 2015

99

35 (65)

− 12 (1400 UTC 03 Nov.) − 06 (0800 UTC 03 Nov.)

− 06 (0800 UTC 03 Nov.) 00 (0200 UTC 03 Nov.)

29 Oct. 2015 30 Oct. 2015

Between 0100 and 0200 UTC, 03 Nov. 2015

+/– sign represents ahead/delay in time

both the experiments, the quantitative analyses of rainfall based on these skill scores with different threshold of rainfall (1, 2, 3, 4, 5, 6, and 7 cm) are carried out to study the impact of VI on satellite data assimilation in the NCUM global model. The gridded IMD-NCMRWF merged satellite-gauge rainfall dataset of 0.5° resolution [37] is used for calculation of the ETS and bias. The ETS and bias for day 1 and day 2 are depicted in Fig. 10a–d, respectively. From Fig. 10a, b, it is seen that ETS values are deteriorated in Vort-Gts in comparison with the Vort-Rad at all thresholds for day 1 and day 2 rainfall. The rainfall is underpredicted in both the experiments, while the rainfall is relatively well simulated in the Vort-Rad than the Vort-Gts. The ETS values from the Vort-Rad are closer to 1. Similarly, the bias values are improved in the Vort-Rad in comparison with the Vort-Gts for day 1 (Fig. 10c) and day 2 (Fig. 10d). It infers that the Vort-Rad depicts less bias and improves the rainfall forecast in comparison with the Vort-Gts for both day 1 and day 2.

Fig. 9 a Mean absolute error (AE) of MSLP and b MSW with forecast lead time

6 Summary and Conclusions Accurate forecast of TC track and intensity well in advance is highly needed over the NIO region. Despite satellite reconnaissance and other improvements in TC forecasting techniques, TC prediction is still a major challenge for operational meteorological agencies. The objective of the study is to evaluate the impact of the vortex initialization on assimilation of satellite data in the NCUM global model for simulation of intensity and track prediction of TC Chapala which formed over the Arabian Sea. For this purpose, two numerical experiments, namely Vort-Gts and Vort-Rad, were conducted. The position and intensity errors of TCs are estimated with respect to the IMD observed datasets. The main findings of the study are summarized as below. The vertical structure of the storm is well defined in the Vort-Rad as compared to the Vort-Gts at model initial time.

Remote Sens Earth Syst Sci Fig. 10 ETS of rainfall with different thresholds for a day 1 and b day 2. c–d are same as a–b but for bias, respectively

The low level convergence and upper level divergence is well featured in the Vort-Rad. The magnitude of wind in the Vort-Rad is higher by ~ 5 m/s near the storm center in comparison with the Vort-Gts. The well-organized structure and position of the storm are precisely captured in the Vort-Rad. The vertical velocity simulated in the Vort-Rad during TC’s mature phase is confined to a comparatively narrower and well-defined region than in the Vort-Gts. The Vort-Rad simulated vertical wind is stronger and distributed up to higher altitude, reporting the presence of deeper convection and warmer core. The features are not clearly depicted in the Vort-Gts. Similarly, the hydrological features such as the TPW and MT are well simulated in the Vort-Rad in comparison with the Vort-Gts. The track of the storm is well simulated in the Vort-Rad with all the ICs; however, the Vort-Gts tracks diverged from the observed tracks. Corresponding DPEs are higher in the Vort-Gts as compared to the Vort-Rad. The initial position error of the storm is reduced in the Vort-Rad in comparison with the Vort-Gts. The mean DPEs, CTEs, and ATEs are improved by 47, 40, and 57% respectively in the Vort-Rad than the Vort-Gts. The landfall position and time errors are improved in the Vort-Rad.

The intensity of the storm is well simulated in the Vort-Rad in comparison with the Vort-Gts. The RMSE of MSLP and MSW from the Vort-Rad are reduced by 03 hPa and 02 m/s than the Vort-Gts. The statistical skill scores viz. ETS and bias for day 1 and day 2 are improved in the Vort-Rad experiment. The results suggested that the VI with assimilation of satellite observation data in the NCUM global model has a positive impact. However, to better understand the effects of Vortex Initialization along with satellite data, more case studies over the NIO region are required using the NCUM global model. Acknowledgements The authors acknowledge Dr. Noel E. Davidson of the Centre for Australian Weather and Climate Research Melbourne, Victoria, Australia, for his immense help in the implementation and use of the vortex initialization scheme at NCMRWF. Thanks are due to IMD for providing the best track of the TC to validate the model simulations.

References 1.

Abersn SD, Franklin JL (1999) Impact on hurricane track and intensity forecasts of GPS dropwindsonde observations from the firstseason flights of the NOAA Gulfstream-IV jet aircraft. Bull Am Meteorol Soc 80:421–442

Remote Sens Earth Syst Sci 2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

Black ML, Burpee RW, Marks FD Jr (1996) Vertical motion characteristics of tropical cyclones determined with airborne Doppler radial velocities. J Atmos Sci 53(13):1887–1909 Burpee RW et al (1994) Real-time guidance provided by NOAA’s Hurricane Research Division to forecasting during Emily of 1993. Bull Am Meteorol Soc 75:1765–1783 Ballard SP, Li Z, Simonin D, Caron J-F (2016) Performance of 4DVar NWP based nowcasting of precipitation at the Met Office for summer 2012. Q J R Meteorol Soc 142:472–487 Burpee RW, Aberson SD, Franklin JL, Lord SL, Tuleya RE (1996) The impact of Omega dropwindsondes on operational hurricane track forecast models. Bull Am Meteorol Soc 77:925–933 Best MJ, Pryor M, Clark DB, Rooney GG, Essery R, Ménard CB, Edwards JM, Hendry MA, Porson A, Gedney N, Mercado LM (2011) The Joint UK Land Environment Simulator (JULES), model description–part 1: energy and water fluxes. Geosci Model Dev 4(3):677–699 Bhaskar Rao, D.V., Hari Prasad, D. and Srinivas, D., 2009. Impact of horizontal resolution and the advantages of the nested domains approach in the prediction of tropical cyclone intensification and movement. J. Geophys. Res. Atmos 114:1–24 Cecil DJ, Zipser EJ (1999) Relationships between tropical cyclone intensity and satellite-based indicators of inner core convection: 85GHz ice scattering signature and lightning. Mon Weather Rev 127(1):103–123 Chan JC, Williams RT (1987) Analytical and numerical studies of the beta-effect in tropical cyclone motion. Part I: zero mean flow. J Atmos Sci 44(9):1257–1265 Chen SH (2007) The impact of assimilating SSM/I and QuikSCAT satellite winds on Hurricane Isidore simulations. Mon Weather Rev 135(2):549–566 Chen SH, Vandenberghe F, Petty GW, Bresch JF (2006) Application of SSM/I satellite data to a hurricane simulation. Q J R Meteorol Soc 130:801–825 Clark DB, Mercado LM, Sitch S, Jones CD, Gedney N, Best MJ, Pryor M, Rooney GG, Essery RLH, Blyth E, Boucher O (2011) The Joint UK Land Environment Simulator (JULES), model description–part 2: carbon fluxes and vegetation dynamics. Geosci Model Dev 4(3):701–722 Davidson NE, Xiao Y, Ma Y, Weber HC, Sun X, Rikus LJ, Kepert JD, Steinle PX, Dietachmayer GS, Lok CC, Fraser J (2014) ACCESS-TC: vortex specification, 4DVAR initialization, verification, and structure diagnostics. Mon Weather Rev 142(3):1265– 1289 Fiorino M, Goerss JS, Jensen JJ, Harrison EJ Jr (1993) An evaluation of the real-time tropical cyclone forecast skill of the Navy Operational Global Atmospheric Prediction System in the western North Pacific. Weather Forecast 8(1):3–24 Fritz C, Wang Z (2014) Water vapor budget in a developing tropical cyclone and its implication for tropical cyclone formation. J Atmos Sci 71(11):4321–4332 Gregory D, Rowntree PR (1990) A mass flux convection scheme with representation of cloud ensemble characteristics and stabilitydependent closure. Mon Weather Rev 118(7):1483–1506 Harasti PR, McAdie CJ, Dodge PP, Lee W-C, Tuttle J, Murillo ST, Marks FD (2004) Real-time implementation of single-doppler radar analysis methods for tropical cyclones: algorithm improvements and use with WSR-88D display data. Weather Forecast 19(2): 219–239 Heming JT (2010) The impact of resolution on Met Office model predictions of tropical cyclone track and intensity. In AMS 29th Conference on Hurricanes and Tropical Meteorology, Tucson, Arizona Heming JT, Chan JCL, Radford AM (1995) A new scheme for the initialisation of tropical cyclones in the UK Meteorological Office global model. Meteorol Appl 2(2):171–184

20.

Hou AY, Zhang SQ, Reale O (2004) Variational continuous assimilation of TMI and SSM/I rain rates: impact on GEOS-3 hurricane analyses and forecasts. Mon Weather Rev 132(8):2094–2109 21. Hsiao LF, Liou CS, Yeh TC, Guo YR, Chen DS, Huang KN, Terng CT, Chen JH (2010) A vortex relocation scheme for tropical cyclone initialization in advanced research WRF. Mon Weather Rev 138(8):3298–3315 22. Huang XY, Xiao Q, Barker DM, Zhang X, Michalakes J, Huang W, Henderson T, Bray J, Chen Y, Ma Z, Dudhia J (2009) Fourdimensional variational data assimilation for WRF: formulation and preliminary results. Mon Weather Rev 137(1):299–314 23. Jin H, Peng MS, Jin Y, Doyle JD (2014) An evaluation of the impact of horizontal resolution on tropical cyclone predictions using COAMPS-TC. Weather Forecast 29(2):252–270 24. Kossin JP, Knaff JA, Berger HI, Herndon DC, Cram TA, Velden CS, Murnane RJ, Hawkins JD (2007) Estimating hurricane wind structure in the absence of aircraft reconnaissance. Weather Forecast 22(1):89–101 25. Krishnamurti TN, Pattnaik S, Rao DB (2007) Mesoscale moisture initialization for monsoon and hurricane forecasts. Mon Weather Rev 135(7):2716–2736 26. Kurihara Y, Bender MA, Ross RJ (1993) An initialization scheme of hurricane models by vortex specification. Mon Weather Rev 121(7):2030–2045 27. Kutty G, Gohil K (2017) The role of mid-level vortex in the intensification and weakening of tropical cyclones. J Earth Syst Sci 126(7):94 28. Lock AP, Brown AR, Bush MR, Martin GM, Smith RNB (2000) A new boundary layer mixing scheme. Part I: scheme description and single-column model tests. Mon Weather Rev 128(9):3187–3199 29. Lorenc AC, Rawlins F (2005) Why does 4D-Var beat 3D-Var? Q J R Meteorol Soc 131(613):3247–3257 30. Mahala BK, Mohanty PK, Nayak BK (2015) Impact of microphysics schemes in the simulation of cyclone Phailin using WRF model. Procedia Eng 116:655–662 31. Mandal M, Mohanty UC, Potty KVJ, Sarkar A (2003) Impact of horizontal resolution on prediction of tropical cyclones over Bay of Bengal using a regional weather prediction model. Proc. Indian Acad. Sci. (Earth Planet. Sci.),112(1), pp 79–94 32. Marchok TP (2002) How the NCEP tropical cyclone tracker works. In Preprints, 25th Conf. on Hurricanes and Tropical Meteorology, San Diego, CA, Amer. Meteor. Soc. P (Vol. 1) 33. Mason I (1989) Dependence of the critical success index on sample climate and threshold probability. Aust Meteorol Mag 37:75–81 34. McAdie CJ, Lawrence MB (2000) Improvements in tropical cyclone track forecasting in the Atlantic basin, 197–1998. Bull Am Meteorol Soc 81(5):989–998 35. Mohapatra M, Nayak DP, Sharma M, Sharma RP, bandopadhyay BK (2015) Evaluation of official tropical cyclone landfall forecast issued by India Meteorological Department. J Earth Syst Sci 124(4):861–874 36. Mohapatra M, Nayak DP, Sharma RP, bandopadhyay BK (2013) Evaluation of official tropical cyclone track forecast over North Indian Ocean issued by India Meteorological Department. J Earth Syst Sci 122(3):589–601 37. Mitra AK, Bohra AK, Rajeevan MN, Krishnamurti TN (2009) Daily Indian precipitation analysis formed from a merge of raingauge data with the TRMM TMPA satellite-derived rainfall estimates. J Meteorol Soc Jpn Ser II 87:265–279 38. Montgomery MT, Farrell BF (1993) Tropical cyclone formation. J Atmos Sci 50(2):285–310 39. Osuri KK, Mohanty UC, Routray A, Niyogi D (2015) Improved prediction of Bay of Bengal tropical cyclones through assimilation of Doppler weather radar observations. Mon Weather Rev 143(11): 4533–4560

Remote Sens Earth Syst Sci 40.

Pu Z, Li X, Sun J (2009) Impact of airborne Doppler radar data assimilation on the numerical simulation of intensity changes of Hurricane Dennis near a landfall. J Atmos Sci 66(11):3351–3365 41. Rajagopal EN, Iyengar GR, George JP, Das Gupta M, Mohandas S, Siddharth R, Gupta A, Chourasia M, Prasad VS, Aditi SK, Ashish A (2012) Implementation of unified model based analysis-forecast system at NCMRWF. NMRF/TR/2/2012, 45p 42. Rawlins F, Ballard SP, Bovis KJ, Clayton AM, Li D, Inverarity GW, Lorenc AC, Payne TJ (2007) The Met Office global fourdimensional variational data assimilation scheme. Q J R Meteorol Soc 133(623):347–362 43. Routray A, Singh V, George JP, Mohandas S, Rajagopal EN (2017a) Simulation of Tropical Cyclones over Bay of Bengal with NCMRWF Regional Unified Model. Pure Appl. Geophys 174: 1101–1119 44. Routray A, Singh V, Singh H, Dutta D, George JP, Rakhi R (2017b) Evaluation of Different Versions of NCUM Global Model for Simulation of Track and Intensity of Tropical Cyclones over Bay of Bengal. Dyn. Atmos. Oceans 78:71–88 45. Routray A, Mohanty UC, Osuri KK, Kar SC, Niyogi D (2016) Impact of satellite radiance data on simulations of Bay of Bengal tropical cyclones using the WRF-3DVAR modeling system. IEEE Trans Geosci Remote Sens 54(4):2285–2303 46. Sakai R, Yamaguchi M (2005) The WGNE intercomparison of tropical cyclone track forecasts by operational global models. CAS/JSC WGNE Research Activities in Atmospheric and Oceanic Modeling 35:0207–0208 47. Singh R, Kishtawal CM, Pal PK, Joshi PC (2011) Assimilation of the multisatellite data into the WRF model for track and intensity simulation of the Indian Ocean tropical cyclones. Meteorog Atmos Phys 111(3–4):103–119

48.

Srinivas CV, Yesubabu V, Venkatesan R, Ramarkrishna SSVS (2010) Impact of assimilation of conventional and satellite meteorological observations on the numerical simulation of a Bay of Bengal tropical cyclone of November 2008 near Tamilnadu using WRF model. Meteorog Atmos Phys 110(1–2):19–44 49. Staniforth A, White A, Wood N (2003) Analysis of semiLagrangian trajectory computations. Q J R Meteorol Soc 129(591):2065–2085 50. Thu TV, Krishnamurti TN (1992) Vortex initialization for typhoon track prediction. Meteorog Atmos Phys 47(2):117–126 51. Walters D, Boutle I, Brooks M, Melvin T, Stratton R, Vosper S, Wells H, Williams K, Wood N, Allen T, Bushell A (2017) The Met Office Unified Model Global Atmosphere 6.0/6.1 and JULES Global Land 6.0/6.1 configurations. Geosci Model Dev 10(4): 1487–1520 52. Wang CC (2014) On the calculation and correction of equitable threat score for model quantitative precipitation forecasts for small verification areas: the example of Taiwan. Weather Forecast 29(4): 788–798 53. Zhang F (2016) Data Assimilation and Predictability of Tropical Cyclones, Advanced Numerical Modeling and Data Assimilation Techniques for Tropical Cyclone Prediction, edited by U.C. Mohanty and Sundararaman G. Gopalakrishnan, Published by Springer, pp 331–360 54. Zhaoxia PU, Wei-Kuo T (2004) Mesoscale assimilation of TMI rainfall data with 4DVAR: sensitivity studies. J Meteorol Soc Jpn Ser II 82(5):1389–1397 55. Zhu T, Zhang DL, Weng F (2002) Impact of the advanced microwave sounding unit measurements on hurricane prediction. Mon Weather Rev 130(10):2416–2432