634 WEATHER AND FORECASTING VOLUME 26

Real-Time Adaptive Observation Guidance Using Singular Vectors for Typhoon Jangmi (200815) in T-PARC 2008

HYUN MEE KIM,SUNG-MIN KIM, AND BYOUNG-JOO JUNG Atmospheric Predictability and Data Assimilation Laboratory, Department of Atmospheric Sciences, Yonsei University, Seoul, South Korea

(Manuscript received 17 August 2010, in final form 17 March 2011)

ABSTRACT

In this study, structures of real-time adaptive observation guidance provided by Yonsei University (YSU) in South Korea during The Observing System Research and Predictability Experiment (THORPEX)-Pacific Asian Regional Campaign (T-PARC) are presented and compared with those of no-lead-time adaptive ob- servation guidance recalculated as well as other adaptive observation guidance for a (Jangmi 200815). During the T-PARC period, real-time dry total energy (TE) singular vectors (SVs) based on the fifth- generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) and the corresponding tangent linear and adjoint models with a Lanczos algorithm are provided by YSU to help determine sensitive regions for targeted observations. While YSU provided the real-time TESV guid- ance based on a mesoscale model, other institutes provided real-time TESV guidance based on global models. The overall features of the real-time MM5 TESVs were similar to those generated from global models, showing influences from tropical cyclones, midlatitude troughs, and subtropical ridges. TESV structures are very sen- sitive to verification region and forecast lead time. If a more accurate basic-state trajectory with no lead time is used, more accurate TESVs, which yield more accurate determinations of sensitive regions for targeted ob- servations, may be calculated. The results of this study may imply that reducing forecast lead time is an im- portant component to obtaining better sensitivity guidance for real-time targeted observation operations.

1. Introduction observations, sensitive regions for future forecasts had to be determined in advance. The Observing System Research and Predictability Methods for determining sensitive regions that in- Experiment (THORPEX), a World Weather Research fluence forecasts verified in the future are called tar- Program (WWRP) of the World Meteorological Orga- geted observation strategies (or adaptive observation nization (WMO), began in 2003 and is scheduled to end guidance). Various organizations have developed real- in 2012. From 1 August to 6 October 2008, the THORPEX- time adaptive observation guidance that suggests pos- Pacific Asian Regional Campaign (T-PARC) was imple- sible target regions for adaptive observations to improve mented in the western North Pacific to improve short- to tropical cyclone track forecasts (Kim et al. 2008). The medium-range typhoon forecasts. During the T-PARC real-time guidance employed for T-PARC includes total period, multiple types of in situ and remotely sensed non- energy singular vector (TESV; Peng and Reynolds 2006; regular observations were deployed in this area (Elsberry Reynolds et al. 2010) and adjoint sensitivity (Amerault and and Harr 2008). In particular, intensive dropsonde ob- Doyle 2009; Reynolds et al. 2010) guidance, both used by servations using aircraft were performed, following the the Naval Research Laboratory (NRL); TESV guidance, typhoon life cycle from tropical convection to extratropical used by the European Centre for Medium-Range Weather transition. To perform these aircraft-based dropsonde Forecasts (ECMWF; Buizza et al. 2007), the Mete- orological Administration (JMA; Yamaguchi et al. 2009), and Yonsei University (YSU; Kim and Jung 2009b,a); adjoint-derived sensitivity steering vector (ADSSV) guid- Corresponding author address: Hyun Mee Kim, Dept. of Atmo- spheric Sciences, Yonsei University, Shinchon-dong 134, Seodaemun- ance, used by National University (Wu et al. 2007); ku, Seoul 120-749, South Korea. an ensemble transform Kalman filter (ETKF), used by E-mail: [email protected] the University of Miami (Majumdar et al. 2006) and the

DOI: 10.1175/WAF-D-10-05013.1

Ó 2011 American Meteorological Society Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 635

Met Office (Bowler et al. 2008); and analysis of ensemble observation guidance and no-lead-time TESVs. The pur- deep-layer mean (DLM) wind variance, used by the Na- pose of this paper is to describe the real-time adaptive ob- tional Oceanic and Atmospheric Administration (NOAA; servation guidance produced to support the T-PARC field Aberson 2003). These sensitivity products for adaptive program, and compare those real-time products with no- observations were collected during the T-PARC period lead-time TESVs. To compare TESVs provided from by the Earth Observing Laboratory (EOL) of the National several institute during T-PARC, the effects of physics Center for Atmospheric Research (NCAR), the Data and norms on TESV structures are also discussed. The Targeting System (DTS) of ECMWF, and the JMA mathematical SV formulation, experimental framework, T-PARC Web site, and were used to determine the target and real-time procedures are presented in section 2. The regions for adaptive observations of typhoons. real-time adaptive observation guidance provided by Using the real-time guidance provided by the organi- YSU and by other institutes during T-PARC and the zations named above, flight plans were determined for recalculated TESV guidance are presented and com- individual typhoons. Once tropical convections that might pared in section 3. Section 4 contains a summary and develop into a typhoon were observed, meetings were discussion. held almost every day through the extinction of the ty- phoon to plan flight paths that cover the sensitive regions 2. Mathematical formulation and experimental suggested by various strategies. With the exception of framework YSU in Korea, which provided real-time TESV guid- a. Total energy norm SVs ance based on a mesoscale model, the other institutes provided the real-time TESV guidance based on global The calculation of SVs involves selecting an initial models. The impacts of real-time targeting guidance dur- disturbance subject to the constraints that the initial ing T-PARC have not yet been fully reported, but the disturbance has unit amplitude in a specified norm and TESV guidance provided by YSU showed a positive effect evolves to have a maximum amplitude in a specified on tropical cyclone (TC) track forecasts (Jung et al. 2010). norm after some finite optimization time, t 5 topt. In this In this study, structures of real-time adaptive obser- study, the initial and final norms were the dry total en- vation guidance provided by YSU during T-PARC are ergy (TE), defined by Zou et al. (1997) and Kim and presented and compared with those of other adaptive Jung (2009b,a) as

" # ððð 2 2 1 2 2 2 g 2 1 2 Ed 5 u9 1 y9 1 w9 1 u9 1 p9 dy dx ds, (1) s,x,y 2 Nu rcs

where Ed is the dry TE in a nonhydrostatic model; u9, y9, the verification region, M is the tangent linear model and w9 are the zonal, meridional, and vertical wind per- (TLM) of the nonlinear model, C is the dry TE norm, turbations, respectively; u9 is the potential temperature and x9(0) is the initial perturbation state vector. In (2), perturbation; p9 is the pressure perturbation; N, u, r,and the state vector at the initial time evolves linearly. By cs are the Brunt–Va¨isa¨la¨ frequency, potential tempera- defining the local projection operator, the amplitude of ture, density, and speed of sound at the reference level, the state vector with norm C at the optimization time is respectively; and x, y,ands denote zonal, meridional, and maximized over a specific region. The maximum ratio is vertical coordinates, respectively. While the first three realized when x9(0) is the leading SV of the TLM M for terms on the right-hand side (rhs) of (1) are associated the C norm; that is, x9(0) satisfies with the kinetic energy (KE), the final two terms in the rhs of (1) are associated with the potential energy (PE). MTPTCPMx9(0) 5 l2Cx9(0). (3) The ratio of the final and initial perturbation ampli- tudes is the Rayleigh quotient l2 (the amplification factor) defined as The generalized eigenvalue problem in (3) can be reduced to an ordinary eigenvalue problem by left- hPMx9(0), CPMx9(0)i multiplying both sides of (3) by the inverse of the square l2 5 , (2) hx9(0), Cx9(0)i root of C (Zou et al. 1997). A Lanczos-type algorithm (e.g., Ehrendorfer and Errico 1995; Kim 2003; Kim et al. where the inner product is denoted by h,i, P is a local 2004; Kim and Jung 2009b,a) can then be used to solve projection operator that zeros out the state vector outside for x9(0). The Lanczos algorithm is an iterative algorithm

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 636 WEATHER AND FORECASTING VOLUME 26 ! 3 l2 composite 5 n n Sij å 2 Sij, (5) n51 l1 2 2 where l1 and ln are singular values for the first and nth n TESVs, respectively, and Sij denotes the nth vertically integrated TESV energy field. b. Model and SV configurations To calculate SVs in real time, this study used the fifth- generation Pennsylvania State University–National Cen- ter for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5) adjoint modeling system (Zou et al. 1997) and a Lanczos algorithm. The model domains were cen- tered at 258N and 1258E, 308Nand1268E, and 368Nand 1258E for verification regions corresponding to Taiwan, Japan, and Korea, respectively (Fig. 1), with 120-km horizontal grid spacing in a 50 3 50 domain and 14 evenly spaced vertical sigma levels from the surface to 50 hPa. The model initial and lateral boundary conditions used for real-time calculation were obtained from the National Centers for Environmental Prediction (NCEP) Global FIG. 1. Model domains (solid lines) and verification regions (dot- Forecast System (GFS; 18318 global grid). For re- ted lines) for Taiwan (red), Japan (purple), and Korea (blue). Real- calculated SVs with more accurate basic states, the NCEP time TESV sensitivity products using MM5 were generated for the final analysis (FNL; 18318 global grid) was used for the Taiwanese verification region of 188–308N, 1188–1328E(188–308N, 1178–1408E) from 0000 UTC 11 May to 0000 UTC 3 Aug 2008 model initial and lateral boundary conditions. Physical (from 0000 UTC 4 Aug to 0000 UTC 26 Dec 2008); the Japanese parameterizations used for the nonlinear basic-state in- verification region of 208–408N, 1208–1508Efrom0000UTC18 tegrations included the Grell convective scheme, a bulk Apr to 0000 UTC 26 Dec 2008; and the Korean verification region aerodynamic formulation of the planetary boundary of 308–428N, 1188–1328E from 0000 UTC 4 Jun to 0000 UTC 26 layer, a simple radiational cooling scheme, horizontal Dec 2008. and vertical diffusion, dry convective adjustment, and the explicit treatment of cloudwater,rain,snow,andice. The same physical parameterizations were used in the that is an adaptation of power methods to find eigen- TLM and adjoint model integrations, except that the TLM values and eigenvectors of a square matrix or the singular- and adjoint model considered the effects of moisture value decomposition of a rectangular matrix, and is using a large-scale scheme rather than the particularly useful for finding decompositions of very Grell convective scheme and an explicit treatment of cloud large sparse matrices (Golub and Van Loan 1996). water, rain, snow, and ice. These configurations of non- Similar to Langland et al. (2002), the vertically in- linear and adjoint model integrations were used by Kim tegrated TESV energy is calculated to combine each and Jung (2009b,a) to consider moist physics effects in component of the TESV with a different unit into a linear integrations of SV calculations. As mentioned in single two-dimensional TESV field with a unit of en- section 2a, the dry TE norm was used for the calculation ergy (J kg21): of real-time SVs to reduce the computation time, which was constrained by the real-time configuration as shown Sij 5 å x9(0)ijk Cx9(0)ijk, (4) in Fig. 2. k c. Experimental setup and procedures where i, j, and k denote grid points in the x, y, and z directions, respectively. To provide real-time adaptive observation guidance To investigate the overall properties of the first to during the T-PARC 2008 period, SVs and adjoint-based third TESVs, the composite of the first to third TESVs forecast sensitivities using the MM5 and its tangent was calculated as in Kim and Jung (2009a). The formula linear and adjoint models were calculated at YSU for of the vertically integrated energy composite of the first several typhoons that have occurred since 2006 (e.g., to third TESVs is Kim and Jung 2006; Kim et al. 2008; Kim and Jung

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 637

FIG. 2. A schematic diagram of the preparation procedure for MM5 real-time TESV guidance at YSU.

2009b,a). Based on these prior experiences, real-time trajectories in real-time calculations on TESV charac- MM5 SVs were calculated for the period from spring teristics. In contrast to the real-time TESVs, the TESVs to winter 2008 for each of three verification regions are recalculated for 0 h of lead time and 48 h of optimi- (Japan, Taiwan, and Korea; see Fig. 1), and were pro- zation time. The recalculated TESVs are based on the vided to the ECMWF T-PARC DTS and JMA T-PARC 48 h of model trajectory from Ta to Ty, integrated from Web site during the T-PARC period to help determine the initial conditions at Ta as shown in Fig. 2. targeting regions. The dry TE norm SVs were calculated in real time for 48 h of lead time and 48 h of optimization time. Real- time TESVs were calculated based on the 48 h of model trajectory between Ta and Ty, which is integrated from Ti with 48 h of lead time (Fig. 2). Real-time TESVs were perturbations at Ta that grew most rapidly from Ta to Ty. Compared to TESVs based on model integrations start- ing from Ta, the real-time TESVs based on model tra- jectories from Ta to Ty that are initialized from Ti use more uncertain model trajectories. This defect of real- time TESVs may lead to unrealistic sensitive regions, but TESVs should follow the real-time constraints as do other strategies. The start time of the real-time TESV calculations was 0000 UTC. Between 6 and 6.5 h were required to down- load the NCEP GFS data for the initial and boundary conditions for the MM5 model, the subsequent TESV calculation and postprocessing took about 10 h, using a single processor of a 16-processor Linux PC cluster for each target region. The real-time TESVs were then uploaded to the ECMWF T-PARC DTS and JMA T-PARC Web FIG. 3. The Regional Specialized Meteorological Center (RSMC)- site. Flight decisions were made by participating scien- Tokyo Typhoon Center best track (black line with 1 symbols), 48-h MM5 forecast track (red line with circles) with 48-h lead time, and tists at about 2200 to 0000 UTC. 48-h MM5 forecast track (blue line with squares) with 0-h lead

The dry TESVs are recalculated after the T-PARC time, for TC Jangmi (200815). Green circles denote Ti, Ta, and Ty in period to identify the impacts of uncertain model the best track. Each symbol is plotted at 6-h intervals.

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 638 WEATHER AND FORECASTING VOLUME 26

23 21 FIG. 4. Vertically integrated energy distribution of the real-time TESV (10 Jkg , colors with varied scales) for the Taiwan verifi- cation region (box) along with 500-hPa geopotential height (contours, interval of 50 m) at 0 h (0000 UTC 27 Sep 2008) for the (a) first, (b) second, (c) third, and (d) composite TESVs, and at 48 h (0000 UTC 29 Sep 2008) for the (e) first, (f) second, (g) third, and (h) composite TESVs. The thick solid line over the center of TC Jangmi denotes the 1002-hPa MSLP. The numbers at the bottom right of (a)–(c) denote the amplification factors associated with each TESV.

3. Results sensitivities north of the TC in the East Sea (Fig. 4a). The sensitive regions are closely associated with a. Real-time TESV products for the Taiwan the midlatitude trough. The second TESV has overall verification region structures that are very similar to those of the leading The vertically integrated TESV energy is provided TESV; however, the largest sensitivities, consisting of as real-time guidance to represent the general features elongated structures from the southeast coast of China of the sensitivities of typhoons. The real-time TESV to the right-front quadrant of TC Jangmi, are much guidance products for TC Jangmi (200815), one of the closer to the TC (Fig. 4b) than are those of the leading typhoons intensively observed during T-PARC, are TESV. The third TESV had major sensitivities south- shown in this study. The observed and simulated tracks west of the Korean Peninsula (Fig. 4c). The amplifica- of TC Jangmi are shown in Fig. 3. For this specific ex- tion factors of the first, second, and third SVs are 52.3, ample, the model simulation started at 0000 UTC 24.0, and 15.4, respectively, showing large growth of the

25 September (Ti), and the real-time TESV is calcu- initial SVs associated with the TC. Because of contribu- lated based on the model trajectory from 0000 UTC tions from the first TESV, the composite TESV showed

27 September (Ta) to 0000 UTC 29 (Ty)September.TC vertically integrated energy structures that are very Jangmi formed at 0000 UTC 24 September east of similar to those of the first TESV (Fig. 4d). The first, the Philippines, curved northeastward after 0000 UTC second, and composite evolved TESVs after 48 h have 29 September, and dissipated at 0000 UTC 1 October the largest sensitivities over the TC center (Figs. 4e, 4f, 2008. and 4h). The first and third evolved TESVs have other The vertically integrated energy of the first, second, sensitive regions on the boundaries of the Taiwan ver- and third TESVs, as well as the composite of the first to ification region, and the sensitivities northeast of the TC third TESVs, along with the 500 hPa geopotential center for the first (third) TESV are much smaller height at the initial and final times, are shown in Fig. 4. (larger) in magnitude than those over the TC center For the leading TESV, the initial TESV shows large (Figs. 4e and 4g). The second TESV has a secondary

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 639

FIG. 5. As in Fig. 4, but for (top) the 100–400-hPa layer along with the layer-average potential vorticity (PV) of the 200–500-hPa layer [contour interval of 1 PV unit (PVU), where 1 PVU = 1026 m2 s21 Kkg21] at (a) 0 h (0000 UTC 27 Sep 2008) and (d) 48 h (0000 UTC 29 Sep 2008); (middle) the 400–750-hPa layer along with the layer-average PV of 500-hPa geopotential height (contour interval of 50 m) at (b) 0 and (e) 48 h; and (bottom) the 750 hPa–MSLP layer along with the MSLP (contour interval of 4 hPa) at (c) 0 and (f) 48 h.

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 640 WEATHER AND FORECASTING VOLUME 26

of the troposphere (Fig. 5). The TESV energy in the lower part (Fig. 5c) is associated with the sensitive region to the north of the typhoon in Fig. 4a, from southeast China to the East China Sea, whereas the TESV energies in the middle and upper regions are as- sociated with the midlatitude trough (Figs. 5a and 5b). The largest contribution to the evolved TESV energy is from the lower part of the troposphere (Fig. 5f). The secondary maximum of the evolved TESV energy is located in the region between the midlatitude trough and the subtropical high to the east of the TC (Figs. 5e and 5f), indicating that the initial TESVs are associated with large-scale background systems as well as with the TC itself. The large sensitivities in the midlatitudes are located in the lower and middle parts of the tropo- sphere, consistent with the results of Kim and Jung (2009b). Vertical profiles of the first, second, third, and composite TESV energy are shown in Fig. 6. The maxima of the leading TESV for both the TE and KE at the initial time (Fig. 6a) are located in the lower troposphere. In contrast, the leading TESV for the PE has a peak at the upper boundary, with much smaller contributions throughout the troposphere (Fig. 6a). The leading TESV at the final time (Fig. 6e) increased in amplitude by at least an order of magnitude and has a maximum near the lower tro- posphere. The PE has a smaller contribution and a more uniform structure, which also holds for the second and third TESVs. Major and minor peaks in the second TESV for both the TE and KE at the initial and final times are different from those of the leading TESV at the corresponding times, showing large TE and KE near the mid- to upper troposphere at the initial time (Fig. 6b) and near the upper troposphere at the final time (Fig. 6f). The PE of the second TESV also has a maximum at the upper boundary at the initial time (Fig. 6b) and a uniform distribution at the final time (Fig. 6f). The maximum of the third TESV for the TE and KE at the initial time (Fig. 6c) is located in the lower troposphere, FIG. 6. Vertical energy distributions of the real-time TESV (J kg21: TE, closed circles; KE, open circles; PE, open squares) at and the maxima of the third TESV for the TE and KE at 0 h for the (a) first, (b) second, (c) third, and (d) composite TESVs, the final time (Fig. 6g) are located in the lower and upper and at 48 h in the Taiwanese verification region for the (e) first, (f) troposphere. Because of the largest contribution of the second, (g) third, and (h) composite TESVs. Note the different leading TESV to the composite TESV, the composite and magnitudes along the abscissas of the evolved TESVs after 48 h. The leading TESVs have very similar vertical structures (Figs. ordinate represents the vertical level (pressure) and the abscissa denotes the TESV energy (J kg21). 6a and 6d). As shown in Kim et al. (2008) and Kim and Jung (2009b,a), the KE of the initial TESV is dominant, except at the upper boundary, and the KE of the final maximum far to the northeast of the TC center (Fig. 4f). TESV is also dominant throughout the troposphere. These variations in evolved TESVs imply that the initial Unlike the leading TESV, the second and third TESVs TESVs are associated not only with the TC itself but also show upward energy propagation during the evolution, with the midlatitude trough. which implies that the second and third TESVs may be The largest contributions to the vertically integrated more closely associated with extratropical systems (e.g., leading TESV energy in Fig. 4a are from the lower part Palmer et al. 1998; Morgan 2001).

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 641

FIG. 7. As in Fig. 4, but for the Korean verification region (box).The thick solid line over the center of TC Jangmi denotes 1000 hPa MSLP. b. Real-time TESV products for the Korean northwest of the TC (Figs. 7a–d). In addition, amplifi- verification region cation factors of 17.9, 14.0, and 12.6 for the first, second, and third SVs (Figs. 7a–c) are much smaller than those In addition to the two fixed verification regions of for the Taiwan verification region in Fig. 4, implying the Taiwan and Japan, another fixed verification region smaller growth of the initial SVs, which are mostly as- centered on the Korean Peninsula is used for calculat- sociated with the midlatitude weather systems. The first ing real-time TESV guidance at YSU. The purpose of TESVs in the upper, mid-, and lower levels at the initial this fixed region is to assess the sensitive regions for and final times show large sensitivities primarily near weather systems affecting the Korean Peninsula during the midlatitude trough (Fig. 8). Only the lower-level the T-PARC period, including typhoons. Because no evolved TESV shows a large sensitivity structure over typhoon activities occurred over the Korean Peninsula the TC center, with other maxima in the extratropics during the T-PARC period, the sensitive regions asso- (Fig. 8f), indicating that initial lower TESV structures are ciated with TCs cannot be determined, but other in- associated with several weather systems including the teresting features are captured, as shown in Fig. 7.1 First, typhoon. because the TC center is outside of the verification re- Vertical profiles of the first, second, third, and com- gion, the evolved TESV structures have several maxima posite TESV energy for the Korean verification region over the TC center, in the verification region, and east are shown in Fig. 9. The maxima of the leading and third of the verification region (Figs. 7e, 7g, and 7h), which TESVs for both the TE and KE at the initial time (Figs. implies that the initial TESVs are associated with sev- 9a and 9c) are located in the lower troposphere. Major eral weather systems including the typhoon, as in- peaks of the second TESV for both TE and KE at the dicated in Reynolds et al. (2009). Unlike the initial initial and final times are different from those of the TESVs for the Taiwan verification region shown in Fig. leading TESV at the corresponding times, showing large 4, the initial TESVs are located north to the far TE and KE in the midtroposphere at the initial time (Fig. 9b). At the final time, all the evolved TESVs (Figs. 9e–g) have increased in amplitude by at least an order 1 The model simulation and the real-time TESV calculation in of magnitude and have attained maxima in the mid- this section are based on the same time frame as in section 3a. troposphere. The PE makes a smaller contribution and

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 642 WEATHER AND FORECASTING VOLUME 26

FIG. 8. As in Fig. 5, but for the Korean verification region (box). The thick solid line over the center of TC Jangmi denotes the 1000-hPa MSLP.

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 643

c. Comparison with TESV products with 0-h lead time

Unlike the case for the real-time calculation, the model simulation started at 0000 UTC 27 September

(Ta), and the 0-h lead-time TESV for the Taiwan veri- fication region is calculated based on the model trajec-

tory from 0000 UTC 27 September (Ta) to 0000 UTC 29 September (Ty). Because there is no lead time, the basic-state trajectory used for TESV calculation is closer to the real atmospheric state. The vertically integrated energy distributions of the first, second, third, and composite of the first to third TESVs superposed on the 500-hPa geopotential height at the initial and final times are shown in Fig. 10. For the leading TESV, the initial TESV has large sensitivities north of the TC in the East China Sea and in the right– rear quadrant of the TC (Fig. 10a). The sensitive regions are closely associated with the midlatitude trough and the inflow region between the TC and the subtropical ridge. These structures are similar to the real-time TESVs from the ECMWF and NRL (Figs. 13b and 13c), in- dicating that the basic-state trajectory for the TESVs at this time may be closer to the real-time basic-state trajectories from the ECMWF and NRL. Compared to Fig. 4a, the subtropical ridge for this case is extended westward (Fig. 10a), which indicates that the subtropical ridge may not be well represented in real time. The second TESV shows the largest sensitivities in the inflow region between the TC and the subtropical ridge (Fig. 10b), and the third TESV has major sensitivities north of the TC near the midlatitude trough (Fig. 10c). The amplification factors of 62.7, 42.2, and 22.5 for the first, second, and third SVs (Figs. 10a–c) are larger than those for the real- time SVs in Fig. 4, implying the initial SVs are more closely associated with the TC itself. All of the evolved TESVs after 48 h have their largest sensitivities over the TC center (Figs. 10e, 10g, and 10h), except for the second

FIG. 9. As in Fig. 6, but for the Korean verification region. evolved TESV, whose largest sensitivities appear in the middle of the TC and the subtropical ridge (Fig. 10f). Compared to the real-time TESVs in Fig. 4, the structures exhibits a more uniform structure for the first and third of the evolved TESVs are more concentrated in the TC TESVs and shows a peak at the mid- to upper tropo- itself, implying that the initial TESVs with 0-h lead time sphere for the second TESV. The contribution of PE for are associated with the TC itself and experience a rela- all the TESVs increased at the final time. The KE and tively small influence from the midlatitude system. PE of the second evolved TESV show comparable The largest contributions to the vertically integrated magnitudes. Even though the KE of the initial TESV is leading TESV energy, as shown in Fig. 10a, are from the dominant, the contribution of PE is much larger than mid- to lower troposphere (Fig. 11). The TESV energy that for the Taiwan verification region, indicating that in the lower part (Fig. 11c) is associated with the sensi- TESV structures that maximize energy in the Korean tive region to the north and northwest of the typhoon in verification region, located far north of the Taiwan Fig. 10a. The TESV energies in the middle and upper verification region, detect midlatitude weather systems parts are associated with both midlatitude and inflow instead of the TC. regions (Figs. 11a and 11b), with the largest amplitude

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 644 WEATHER AND FORECASTING VOLUME 26

FIG. 10. As in Fig. 4, but for the 0-h lead-time TESV. The thick solid line over the center of TC Jangmi denotes the 996-hPa MSLP. being in the inflow regions for the upper TESVs (Fig. the TE and KE at the initial time (Fig. 12c) is located in 11a). The largest contribution to the evolved TESV the lower troposphere, whereas the maxima of the third energy is also from the mid- to lower troposphere (Figs. TESV for the TE and KE at the final time (Fig. 12g) are 11e and 11f). The secondary maximum of the evolved located in the upper troposphere and mid- to lower TESV energy in the upper troposphere is located in the troposphere. The PE of the third TESV exhibits a uni- outflow region between the midlatitude trough and the form distribution at the initial and final times (Figs. 12c subtropical high to the far northeast of the TC (Fig. 11d), and 12g). Even though the contribution of the leading but the amplitude is much smaller than that in the mid- TESV is the largest, the composite TESV differs from to lower troposphere. As indicated in Fig. 10, the the leading TESV because the magnitude of the second evolved TESV is located over the TC, which indicates TESV in the upper troposphere is much larger than that that the initial TESVs are associated with the TC itself. of the leading TESV (Figs. 12d and 12h). As discussed in Vertical profiles for the first, second, and third TESV the case of the real-time TESVs, the KE of the initial energy are shown in Fig. 12. While the maximum of the TESV is dominant, except at the upper boundary, and leading TESV for both the TE and KE at the initial the KE of the final TESV is dominant throughout the time (Fig. 12a) is located in the mid- to lower tropo- troposphere. Compared to the leading and third sphere, the leading TESV for the PE makes a much TESVs, the second TESV shows quite different ver- smaller contribution throughout the troposphere. The tical energy structures, of which large sensitivities in leading TESV at the final time (Fig. 12e) increased in the upper troposphere are associated with the in- amplitude by at least one order of magnitude and has teraction between the TC and the subtropical ridge. maxima in the mid- to lower troposphere. The PE has The mechanism underlying these large sensitivities in a smaller contribution and a more uniform structure, as the upper troposphere in the right-rear quadrant of the is also the case for the third TESV. Major peaks of the TC will be discussed in a future paper on TC Man-Yi second TESV for both TE and KE at the initial and final (200704). times are located in the upper troposphere (Figs. 12b d. Comparison with other sensitivity products and 12f). The PE of the second TESV has a maximum at the upper boundary at the initial time (Fig. 12b), and at Figure 13 shows several real-time sensitivity guidance the upper troposphere at the final time (Fig. 12f), similar examples provided by many organizations to the to the TE and KE. The maximum of the third TESV for ECMWF T-PARC DTS for TC Jangmi. While the overall

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 645

FIG. 11. As in Fig. 5, but for the 0-h lead-time TESV. The thick solid line over the center of TC Jangmi denotes the 996-hPa MSLP.

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 646 WEATHER AND FORECASTING VOLUME 26

(2010). The JMA TESV shows two large sensitivities in the northwest and northeast of the TC, with both maxima denoting the influences of midlatitude weather systems (Fig. 13d). The different structures of the JMA TESVs may be due to the model physics and normal configura- tions used for the TESV calculations (Komori et al. 2009). Komori and Kadowaki (2010) also showed that TESV structures of depend on the model resolutions. Compared to the TESVs, the guidance pro- vided by the ETKF method used by two organizations shows similar structures, with large sensitivities over the TC center (Figs. 13e and 13f). As reported in Majumdar et al. (2006), sensitive regions tend to be similar when using the same method with different models. Even though the overall features of the TESV guid- ance are similar, their detailed structures are different. Kim and Jung (2009a) showed that the TESV structures and evolutions of Typhoon Usagi (200705) highly de- pend on the moist physics and norms used to calculate the TESVs. To investigate the relative influence of physics, norms, and forecast lead time of the basic-state trajectory to the detailed TESV structures shown in Figs. 13a–d, additional experiments with different physics and norms used to calculate the real-time and 0-h lead-time MM5 TESVs are performed. Figure 14a (Fig. 14e) denotes the real-time initial (final) dry norm TESVs with dry physics for TLM and adjoint model integrations. Figure 14b (Fig. 14f) denotes the real-time initial (final) moist norm2 TESVs with the large-scale precipitation as the moist physics for TLM and adjoint model integrations as in section 2b. Because the SV in Fig. 14b is calculated by use of the moist TE norm, the TESV in Fig. 14b shows a larger amplification factor than that in Fig. 14a. However, the TESV structures in Figs. 14a and 14b are similar with those in Figs. 4a and 13a. Figure 14c (Fig. 14g) denotes the initial (final) dry norm TESVs with 0-h lead-time model trajectories and FIG. 12. As in Fig. 6, but for the 0-h lead-time TESV. dry physics for TLM and adjoint model integrations. Figure 14d (Fig. 14h) denotes the initial (final) moist norm TESVs with 0-h lead-time model trajectories and features of the TESV guidance products from YSU, the large-scale precipitation as the moist physics for the ECMWF, and NRL are similar (Figs. 13a–c), the TESV TLM and adjoint model integrations as in section 2b. from JMA shows different structures (Fig. 13d). The Similar to the real-time TESVs, the TESV in Fig. 14d largest sensitivities of the YSU TESV are elongated from shows a larger amplification factor than that in Fig. 14c the northwest of the TC to the right circle of the TC, as due to the moist physics and moist norm effects. Com- shown in Fig. 4a, indicating the influence of a midlatitude pared to the TESVs in Figs. 14a and 14b, the TESVs in trough (Fig. 13a). However, the TESVs of the ECMWF Figs. 14c and 14d are much more similar to those in Figs. (NRL) show the largest sensitivities on the right half-cir- 13b and 13c, indicating that the basic-state trajectory for cle (right-rear quadrant; see Figs. 13b and 13c), indicating a substantial influence from the subtropical high to the right of the TC. In particular, the NRL TESV shows large 2 The moist TE norm in this study is calculated as in Kim and sensitivities in the inflow region, as in Reynolds et al. Jung (2009a).

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 647

FIG. 13. Sensitivity guidance for the Taiwan verification region (box) with MSLP (hPa) at 0000 UTC 27 Sep 2008 for TC Jangmi: (a) YSU MM5 TESV, (b) ECMWF TESV, (c) NRL TESV, (d) JMA TESV, (e) UKMO ETKF, and (f) University of Miami–NCEP ETKF. (Courtesy of ECMWF T-PARC DTS.)

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC 648 WEATHER AND FORECASTING VOLUME 26

23 21 FIG. 14. Vertically integrated energy distribution of TESV (10 Jkg , colors with varied scales) for the Taiwanese verification region (box) with 500-hPa geopotential height (contours, interval of 50 m), for the real-time leading TESV at (a) 0 and (e) 48 h with the dry norm and dry physics and at (b) 0 and (f) 48 h with the moist norm and large-scale precipitation as the moist physics for the TLM and adjoint model integrations, for the 0-h lead-time leading TESV at (c) 0 and (g) 48 h with the dry norm and dry physics and at (d) 0 and (h) 48 h with the moist norm and large-scale precipitation as the moist physics for the TLM and adjoint model integrations. the TESVs at this time may be closer to the real-time The overall features of the real-time MM5 TESVs are basic-state trajectories from ECMWF and NRL. How- consistent with those found in previous works (e.g., Peng ever, those sensitivities east of the TC center are more and Reynolds 2006; Chen et al. 2009; Kim and Jung prominent in Figs. 10a and 14d than in Fig. 14c, indicating 2009b), showing the influences of TCs, midlatitude that those sensitivities to the right of the TC center be- troughs, and subtropical ridges. Unlike an extratropical come clearer by using large-scale precipitation physics cyclone system in which potential (kinetic) energy is for TLM and adjoint model integrations with a moist TE generally dominant for the initial (evolved) TESV, ki- norm. Therefore, for this case, the structures of TESVs netic energy was dominant for the initial and evolved are more sensitive to forecast lead time than to the TESVs of TC Jangmi, as indicated in Kim and Jung physics and norm configurations used to calculate the (2009b). Careful determination of the verification region TESVs. While the short forecast lead time is a necessary was an important component for detecting sensitive re- condition for detecting sensitivities to the right of the TC gions associated with the TC considered. The real-time center, the moist physics and norms are sufficient con- TESVs generated in a mesoscale model were generally ditions for showing strong sensitivities there. similar to those generated from global models. As in- dicated in Majumdar et al. (2006), TESV methods using different numerical models (regional or global) pro- 4. Summary and discussion duced similar results, whereas TESV and ETKF guid- During the T-PARC period, real-time TESVs were ance differed significantly. provided by Yonsei University to help determine sen- Structures of TESVs are sensitive to forecast lead sitive regions for targeted observations. In this study, time; given more accurate trajectory information, more structures of real-time adaptive observation guidance accurate TESVs would yield more accurate sensitive provided by Yonsei University in South Korea during regions for targeted observations that may be calcu- T-PARC were presented and compared with those of lated. For the case in this study, structures of TESVs other adaptive observation guidance and no-lead-time depend more on the forecast lead time than the physics TESVs recalculated after T-PARC. and norm configurations used for TESV calculations.

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC OCTOBER 2011 K I M E T A L . 649

There are limitations in real-time operations due to the ——, and B.-J. Jung, 2006: Adjoint-based forecast sensitivities of lengths of the lead time. If different observational methods Typhoon Rusa. Geophys. Res. Lett., 33, L21813, doi:10.1029/ (such as satellite remote sensing observation) are used for 2006GL027289. ——, and ——, 2009a: Influence of moist physics and norms on real-time targeting operations, then lead time can be sig- singular vectors for a tropical cyclone. Mon. Wea. Rev., 137, nificantly reduced, paving the way for more accurate sen- 525–543. sitivity guidance for targeted observations. ——, and ——, 2009b: Singular vector structure and evolution of a recurving tropical cyclone. Mon. Wea. Rev., 137, 505–524. Acknowledgments. The authors wish to thank two ——, M. Morgan, and R. E. Morss, 2004: Evolution of analysis anonymous reviewers for their valuable comments. error and adjoint-based sensitivities: Implications for adaptive This study was supported by the Korea Meteorological observations. J. Atmos. Sci., 61, 795–812. ——, B.-J. Jung, Y.-H. Kim, and H.-S. Lee, 2008: Adaptive ob- Administration Research and Development Program un- servation guidance applied to Typhoon Rusa: Implications der Grant CATER 2011-2211. The authors appreciate Dr. for THORPEX-PARC 2008. Asia-Pac. J. Atmos. Sci., 44, Prates at ECMWF and Mr. Yamaguchi at JMA for sup- 297–312. porting the real-time MM5 TESV display for the T-PARC Komori, T., and T. Kadowaki, 2010: Resolution dependence Web sites. Real-time sensitivity products of ECMWF of singular vectors computed for Typhoon Sinlaku. SOLA, 6, 45–48. T-PARC DTS used in this study have been obtained ——, and Coauthors, 2009: JMA singular vector guidance for from the ECMWF Data Server. T-PARC 2008. Extended Abstracts, Fourth Japan–China–Korea Joint Conf. on Meteorology, Tsukuba, Japan, Meteorological REFERENCES Society of Japan, S5-03. [Available online at http://wwwsoc.nii. ac.jp/msj/jckjc09/JCKJC09-Abstract_Collection.pdf.] Aberson, S. D., 2003: Targeted observations to improve opera- Langland, R. H., M. Shapiro, and R. Gelaro, 2002: Initial condition tional tropical cyclone forecast guidance. Mon. Wea. Rev., 131, sensitivity and error growth in forecasts of the 25 January 2000 1613–1628. East Coast snowstorm. Mon. Wea. Rev., 130, 957–974. Amerault, C. M., and J. Doyle, 2009: Applications of the COAMPS Majumdar, S. J., S. D. Aberson, C. H. Bishop, R. Buizza, M. S. adjoint model. Preprints, 23rd Conf. on Weather Analysis and Peng, and C. A. Reynolds, 2006: A comparison of adaptive Forecasting/19th Conf. on Numerical Weather Prediction, observing guidance for Atlantic tropical cyclones. Mon. Wea. Omaha, NE, Amer. Meteor. Soc., 14A.4. [Available online at Rev., 134, 2354–2372. http://ams.confex.com/ams/23WAF19NWP/techprogram/paper_ Morgan, M., 2001: A potential vorticity and wave activity diagnosis 152970.htm.] of optimal perturbation evolution. J. Atmos. Sci., 58, 2518– Bowler, N. E., A. Arribas, K. R. Mylne, K. B. Robertson, and S. E. 2544. Beare, 2008: The MOGREPS short-range ensemble pre- Palmer, T. N., R. Gelaro, J. Barkmeijer, and R. Buizza, 1998: diction system. Quart. J. Roy. Meteor. Soc., 134, 703–722. Singular vectors, metrics, and adaptive observations. J. Atmos. Buizza, R., C. Cardinali, G. Kelly, and J.-N. Thepaut, 2007: The Sci., 55, 633–653. value of observations. II: The value of observations located in Peng, M. S., and C. A. Reynolds, 2006: Sensitivity of tropical cy- singular-vector-based target areas. Quart. J. Roy. Meteor. Soc., clone forecasts as revealed by singular vectors. J. Atmos. Sci., 133, 1817–1832. 63, 2508–2528. Chen, J. H., M. S. Peng, C. A. Reynolds, and C. C. Wu, 2009: In- Reynolds, C. A., M. S. Peng, and J.-H. Chen, 2009: Recurving terpretation of tropical cyclone forecast sensitivity from the tropical cyclones: Singular vector sensitivity and downstream singular vector perspective. J. Atmos. Sci., 66, 3383–3400. impacts. Mon. Wea. Rev., 137, 1320–1337. Ehrendorfer, M., and R. M. Errico, 1995: Mesoscale predictability ——, J. D. Doyle, R. M. Hodur, and H. Jin, 2010: Naval Research and the spectrum of optimal perturbations. J. Atmos. Sci., 52, Laboratory multiscale targeting guidance for T-PARC and 3475–3500. TCS-08. Wea. Forecasting, 25, 526–544. Elsberry, R. L., and P. A. Harr, 2008: Tropical cyclone structure Wu, C. C., J.-H. Chen, P.-H. Lin, and K.-H. Chou, 2007: Targeted (TCS08) field experiment science basis, observational plat- observations of tropical cyclone movement based on the adjoint- forms, and strategy. Asia-Pac. J. Atmos. Sci., 44, 209–231. derived sensitivity steering vector. J. Atmos. Sci., 64, 2611– Golub, G. H., and F. Van Loan, 1996: Matrix Computations. The 2626. Johns Hopkins University Press, 728 pp. Yamaguchi, M., T. Iriguchi, T. Nakazawa, and C.-C. Wu, 2009: An Jung, B.-J., H. M. Kim, Y.-H. Kim, E.-H. Jeon, and K.-H. Kim, observing system experiment for Typhoon Conson (2004) 2010: Observation system experiments for Typhoon Jangmi using a singular vector method and DOTSTAR data. Mon. (200815) observed during T-PARC. Asia-Pac. J. Atmos. Sci., Wea. Rev., 137, 2801–2816. 46, 305–316. Zou, X., F. Vandenberghe, M. Pondeca, and Y.-H. Kuo, 1997: In- Kim, H. M., 2003: A computation of adjoint-based sensitivities in troduction to adjoint techniques and the MM5 adjoint modeling a quasigeostrophic model. Korean J. Atmos. Sci., 6, 71–83. system. NCAR Tech. Note NCAR/TN-435STR, 110 pp.

Unauthenticated | Downloaded 10/10/21 06:22 PM UTC