Impact of Enhanced Atmospheric Motion Vectors on HWRF Hurricane Analyses and Forecasts with Different

Data Assimilation Configurations

Abstract

The impacts of enhanced satellite-derived atmospheric motion vectors (AMVs) on the numerical prediction of intensity changes during Hurricanes Gonzalo (2014) and Joaquin (2015) are examined. Enhanced AMVs benefit from special data processing strategies, and are examined for impact on model forecasts via assimilation experiments by employing the National Centers for Environmental Prediction (NCEP) operational Hurricane Weather Research and Forecasting (HWRF) model using a gridpoint statistical interpolation (GSI)-based ensemble-variational hybrid system. Two different data assimilation (DA) configurations, one with and one without the use of vortex initialization (VI) are compared.

It is found that the assimilation of enhanced AMVs can improve the HWRF track and intensity forecasts of Gonzalo and Joaquin during their intensity change phases. The degree of data impact depends on the DA configuration used. Overall, assimilation of enhanced AMVs in the innermost HWRF domains (hurricane core region) outperforms other DA configurations, both with and without VI, as it results in better track and intensity forecasts. Compared to the experiment with VI, assimilation of enhanced AMVs without VI reveals more notable data impact on the hurricane forecasts. Even in the configuration with VI, assimilation of enhanced AMVs in the inner-core region can mitigate the vortex spin-down by removing unrealistic outflow structure and unfavorable thermodynamic conditions, thus leading to improved intensity forecasts.

1. Introduction

In contrast to the significant improvements in tropical cyclone (TC) track forecasts, only limited progress has been made in TC intensity forecasting in the last two decades (Rogers et al. 2006 and 2013, Rappaport et al. 2009, Gall et al. 2013). Part of the difficulty in forecasting the intensity of TCs originates from deficiencies in the representation of the initial vortices in numerical weather prediction (NWP) models due to the general lack of high-resolution observations within the TC inner-core region (Pu and Braun 2001, Pu et al. 2009, Zhang et al. 2011). Recently, observations of the TC inner-core region and its near environment, obtained via Tail Doppler radar (TDR) data and dropwindsonde observations that have been collected by NOAA aircraft reconnaissance missions, and made their way into NWP systems and have contributed toward significant improvements in Atlantic TC intensity forecasting (Rogers et al. 2006 and 2013, Zhang et al. 2011, Gall et al. 2013, Pu et al. 2016). However, outside the Atlantic basin, these airborne Doppler radars and dropwindsonde data are not routinely available. And even in the Atlantic, there can be large temporal gaps in the data collection in the TC inner-core region.

TCs spend most of their lifetime over the open ocean, where conventional observations are sparse (Hsiao et al. 2010). Because of the lack of routine radar observations and aircraft dropwindsonde observations from reconnaissance missions, satellite radiances and satellite-derived products are important data sources for use in operational data assimilation. Unfortunately, due primarily to limitations of current data assimilation methodologies, most satellite radiances in the TC inner-core and near-environmental regions are not assimilated due to cloud and precipitation contamination, although all-sky data assimilation has become an active research area in recent years (Zhu et al. 2016).  Fortunately, satellite-derived products, especially atmospheric motion vectors (AMVs; Velden et al. 1997, 2005) derived from geostationary satellites, have supplied useful information for improving hurricane forecasting. Previous studies have demonstrated that assimilating these AMVs into NWP models can result in improved analyses and forecasts of TCs and their environment. Langland et al. (2009) and Berger et al. (2011) found that track forecasts in the Naval Operational Global Atmospheric Prediction System (NOGAPS) were improved owing to more accurate representation of the environmental flow when enhanced AMVs were assimilated. Soden et al. (2001) demonstrated that the assimilation of AMVs led to a more accurate representation of the steering flow, thus improving TC track forecasts in the Geophysical Fluid Dynamic Laboratory (GFDL) hurricane model. Pu et al. (2008) also found that the assimilation of GOES-11 rapid-scan AMVs has a positive impact on numerical forecasts of TC intensity and precipitation.

More recently, upgraded AMV processing algorithms and strategies have enabled improvements in the coverage, density and quality of AMVs. These ‘enhanced’ AMV datasets can better capture smaller-scale wind flows and provide information on TC-scale flow fields (Velden et al. 2017).  Wu et al. (2014 and 2015) used an ensemble Kalman filter method to assimilate enhanced AMV data into the mesoscale community Weather Research and Forecasting (WRF) model. They found that initial analyses of TC vortex location, intensity, and structure are all improved along with subsequent forecasts due to the assimilation of enhanced AMVs. Furthermore, Velden et al. (2017) incorporated enhanced AMVs into the National Centers for Environmental Prediction (NCEP) Hurricane Weather Research and Forecasting (HWRF) model, and found an overall modest positive impact on HWRF forecasts, but suggested the magnitude of the impact may be limited by the vortex initialization procedure used and likely the degree to which unbalanced states are allowed to enter the model analyses via the AMV observations.

All of the above studies demonstrate that enhanced AMVs are useful for improving the numerical prediction of TCs. However, most of these studies incorporate the enhanced AMVs into only the storm environment region, not the TC inner core. Therefore, optimal DA approaches to promote the impacts of enhanced inner-core AMVs to improve model intensity forecasts has not yet been examined. Considering the emerging availability of operational high temporal (5-15 minutes) and spatial (~15-25 km) resolution enhanced AMVs, it is essential to investigate the potential impacts on hurricane vortex initialization and intensity forecasts. Moreover, satellite-derived observations such as enhanced AMVs are now being assimilated in the operational NCEP HWRF (Tallapragada et al. 2015). However, the TC inner core AMV increments are not used in favor of a separate vortex initialization (VI) procedure. It is our goal to use a research version of the HWRF model and the NCEP Gridpoint Statistical Interpolation (GSI)-based ensemble-variation hybrid data assimilation (DA) system to further examine the potential impacts of assimilating enhanced AMVs on the HWRF analyses and forecasts, especially mature hurricanes.  Specifically, we will evaluate two different DA strategies one that is similar to current operational approach, in which VI is performed before all other data are assimilated, and an exploratory method that uses the same HWRF DA configuration but omits the use of VI.  Two hurricane cases are examined that had notable intensity changes during their lifecycles.

The paper is organized as follows: an introduction of the HWRF model, the DA system, and descriptions of the hurricane case studies are presented in Section 2. The experiment design and the impacts of enhanced AMV DA on track and intensity forecasts are discussed in Section 3. Section 4 demonstrates the improvements in initial conditions and the changes in analyzed and forecasted storm structures due to the assimilation of enhanced AMVs, and their sensitivities to the different DA strategies. A summary and discussion along with some insights into future work are given in Section 5.

2. HWRF model, hurricane cases, and experimental design

1. HWRF model and initialization

The NCEP HWRF model (Gopalakrishnan et al. 2011, Bao et al. 2012) Version 3.7a (HWRF V3.7a), is used in this study. In brief, the model grid setup and physics options are configured as closely as possible to the 2015 operational HWRF (Tallapragada et al. 2015). The dynamical core used in HWRF is the same as that used in NCEP’s WRF-Nonhydrostatic Mesoscale Model (NMM, Janjic et al. 2010). In HWRF V3.7a, the NMM core is configured with three domains, d01, d02 and d03 with domain sizes of 5900 km x 5900 km, 1500 km x 1500 km, and 800 km x800 km (see Figure 1), and grid resolutions of 18 km, 6 km, and 2 km, respectively.   A suite of advanced physical parameterizations developed for TC applications are employed. These include the Geophysical Fluid Dynamics Laboratory (GFDL) surface-layer parameterization, the Noah Land Surface Model, the modified GFDL short-wave and long-wave radiation scheme, the Ferrier-Aligo microphysical parameterization, the Global Forecast System (GFS) Planetary Boundary Layer (PBL) scheme, and the GFS simplified Arakawa–Schubert (SAS) cumulus scheme. The cumulus parameterization is active only in the 18 km and 6 km resolution domains.

The HWRF initialization adopts a combination of VI with the NCEP GSI-based ensemble-three dimensional variational hybrid data assimilation system (GSI-3DEnVar hybrid DA system, hereafter; Wu et al. 2002, Wang et al. 2013). The VI scheme (Liu et al.2006) performs the relocation, resizing and intensity correction on the vortex using the National Hurricane Center (NHC) tropical cyclone vital statistics (TCVitals) database in order to correct the storm position and intensity approach to the real-time estimation (see details in Tallapragada et al. 2015). After VI, observations are assimilated by the GSI-3DEnVar hybrid system to further improve the initial conditions for the HWRF forecast. Background error covariance of this hybrid system is a combination of the static background error covariance generated by the NMC method (Parrish and Derber, 1992; Wu et al., 2002; Kleist et al., 2009) and a flow-dependent background error covariance derived from the ensemble forecasts (Wang et al., 2013). The ratio of static and flow-dependent error covariance is set to 0.2 and 0.8, respectively, meaning more weight is given to the ensemble background. In this study, the flow-dependent background error covariance is generated from the NCEP operational GFS 80-member ensemble forecast at a resolution of T574 (~23 km).

2. Hurricane cases and observations

Hurricanes Gonzalo (2014) and Joaquin (2015) are selected as study cases. Detailed descriptions of these two cases can be found in Brown (2014) and Berg (2016).  Considering previous studies of the TC intensity changes and data coverage, intensification periods from 0000 UTC 13 October to 1800 UTC 16 October 2014 for Hurricane Gonzalo and from 1800 UTC 27 September to 0600 UTC 02 October 2015 for Hurricane Joaquin are emphasized in this study.

Observations assimilated in the HWRF GSI-3DEnVar hybrid DA system include the clear-sky satellite radiances, satellite-derived winds, and the conational observations from in-site and remote-sensing instruments. The satellite-derived winds here refer to the operational AMVs, but not the enhanced AMV datasets considered in our study. A list of types of satellite and conventional data could be found on the NCEP website (http://www.emc.ncep.noaa.gov/mmb/ data_processing/prepbufr.doc/table_2.htm and Table 18.htm). In addition, NOAA P3 Tail Doppler Radar (TDR) radial winds are also assimilated in the current operational HWRF when they are available.

3. Enhanced AMVs

The novel dataset being assimilated in this study is the enhanced AMVs processed and provided by CIMSS. These datasets are derived from GOES data, and are part of a demonstration for applications to advanced geo satellite imagers now becoming operational (GOES-R/S, Himawari 8/9). The enhanced AMVs are quality controlled before being entered into the HWRF GSI; enhanced AMVs are assimilated only if the quality indicator (QI) is equal to or larger than an empirically determined value of 0.6 (Wu et al. 2014). In addition, enhanced AMVs meeting these QI thresholds but with expected error (EE) values > 4.5 ms^-1 are filtered out unless the AMV is > 25 m.s^-1 and has an attending QI > 0.7.  The QI and EE values are produced during the AMV derivation process and represent internal QC indicators of AMV quality (Velden et al. 2017).

Figures 2a and b show the horizontal distribution of enhanced AMVs over a 20^o x 20^o box around the storm following QC at 1800 UTC 13 October 2014 for Hurricane Gonzalo and at 0000 UTC 29 September 2015 for Hurricane Joaquin, respectively. The datasets are produced at hourly intervals, and all AMVs within a 6-h window (+/- 3-h of analysis time) are included. Overall, the coverage of enhanced AMVs blankets the storms and their environment. In terms of the vertical distribution, AMVs are single-level, and are most often associated with either high or low-level clouds. In the examples in Fig. 2, the enhanced AMVs for Hurricane Gonzalo are concentrated around the storm in the lower troposphere (Fig.2a), while the enhanced AMVs for Hurricane Joaquin are nearly even between the lower level and upper level (Fig.2b;). In addition, the enhanced AMVs are dominant in the upper level (within 400 km of the storm center) for Hurricane Gonzalo (Fig. 2a) but are denser in the lower (upper) levels in the north (south) quadrants for Hurricane Joaquin (Fig. 2b), indicative of the different cloud structures and outflow configurations between the two storms at these times.

3. Data assimilation experiments and results

a. DA with VI

The first group of experiments, referred to as “DA with VI,” is similar to the operational HWRF configuration. A combination of VI with DA is used in each analysis cycle. The DA is performed only on ghost d02 and ghost d03 domains as shown in Figure 1, and the analysis is interpolated into the d02 and d03 domains after DA to form the initial analysis for HWRF within these two forecasts domains. The initial analysis for d01 is downscaled from a GFS analysis, which implies that the DA on HWRF d01 is indirectly performed by GFS DA system. For Hurricane Gonzalo, the HWRF model is initialized at 1800 UTC 12 October 2014 and allowed to spin-up until 0000 UTC 13 October 2014, and then a DA process is performed that is cycled from 0000 UTC to 1800 UTC 13 October 2014 in 6-h windows, for a total run time of 24 hours. A 72-h forecast from 1800 UTC 13 to 1800 UTC 16 October 2014 is then performed to predict intensity changes of Hurricane Gonzalo. For Hurricane Joaquin, the HWRF model is initialized at 1800 UTC 27 September 2015, allowed to spin-up to 0000 UTC 28 September 2015, and the cycled DA is performed from 0000 UTC 28 to 0600 UTC 29 September 2015 in 6-h windows, for a total run time of 36 hours. Then a 72-h forecast from 0600 UTC 29 September to 0600 UTC 02 October 2015 is performed to simulate the period of intensity changes of Hurricane Joaquin. A set of three DA experiments is performed for Hurricanes Gonzalo and Joaquin to examine the effects of incorporating the enhanced AMVs into one or both ghost domains. Details of the experimental design are listed in Table 1.

The track and intensity forecasts from the HWRF simulations are compared against the NHC best track data. Figure 3 shows the time series of track and intensity in terms of minimum sea level pressure (MSLP) and maximum surface wind (MSW) during the 72-h forecast for Hurricane Gonzalo. The track errors in all experiments are similar and small (e.g., within 50 km) over the 72-h forecasts (Fig.3a). A notable spin-down is found in the intensity forecast of Hurricane Gonzalo in terms of both MSLP (Fig.3b) and MSW (Fig.3c) in the first 6-h forecast (until 0000 UTC 14 October 2014) in the VI-CTRL1 simulation (no enhanced AMVs). As a result, the intensity forecast is significantly degraded, leading to an average MSLP error of 15.4 hPa and an average MSW error of 14.2 m s^-1 over the 72-h forecasts. The assimilation of enhanced AMVs leads to significant improvements in the intensity forecasts. Specifically, the vortex spin-down in the first 6-h forecast still exists in the VI-NIN-AMV1 simulation (AMVs only assimilated in the d02 domain); however, the averaged MSLP (MSW) error over the 72-h forecast is reduced by 50% (35%) compared with that in VI-CTRL1 (Fig.3b, c). With the assimilation of enhanced AMVs to also include the innermost domain (d03; inner-core region) in VI-AMV1, the vortex spin-down in the first 6-h forecast is eliminated and significant improvements in the intensity forecasts of Hurricane Gonzalo are evident: the average error of MSLP (MSW) over the 72-h forecasts in VI-AMV1 is reduced by about 80% (40%) compared with that in VI-CTRL1 (Fig.3b, c).

For the Hurricane Joaquin case, the track forecast during its rapid intensification stage was difficult to capture, as most of the operational models reveal large track forecast errors during the real-time forecasting (Berg 2016). Fig.4 illustrates the track and intensity forecast over a 72-h forecast for Hurricane Joaquin. As shown in Fig.4a, large average track errors over the 48-h forecast are present in the VI-CTRL2, VI-NIN-AMV2, and VI-AMV2 experiments (150 km, 163 km and 133 km, respectively). However, VI-AMV2 saw a 10% reduction in track errors compared with VI-CTRL2, indicating that the assimilation of enhanced AMVs in the inner-core region may also improve the track forecast in some cases.

As opposed to Hurricane Gonzalo (Fig.3b, c), the intensity forecast for Hurricane Joaquin (Fig.4b, c) in both the operational HWRF forecast and the simulations in this study did not show spin-down problems. The forecasts do a nice job of capturing the intensity changes of Hurricane Joaquin during the period from 0600 UTC 29 September to 0600 UTC 02 October 2015 (Fig.4b, c). However, the MSLP (MSW) is slightly underestimated in the first 42-h (54-h) forecasts, and is overestimated after the 42-h (54-h) forecasts in VI-CTRL2, leading to an average error of 10.4 hPa (7m s^-1) over all of the 72-h forecasts. The assimilation of enhanced AMVs also leads to positive impacts on the intensity forecasts of Hurricane Joaquin. Specifically, the assimilation of enhanced AMVs only in the intermediate domain (VI-NIN-AMV2) leads to a reduction in the average error of MSLP (MSW) by 10% (21%) relative to VI-CTRL2. Further improvements can be achieved by assimilating the enhanced AMVs in the hurricane inner-core region (VI-AMV2) as the average MSLP (MSW) error in VI-AMV2 is reduced by 21% (35%) relative to that in VI-CTRL2.

b. DA without VI

Previous studies have shown that VI and DA can counteract each other in some cases (Tallapragada et al. 2015). In addition, the VI can also induce gradient imbalances in the initial conditions, leading to vortex spin-down problems in HWRF forecasts in some cases (Pu et al., 2016). For these situations, impacts from DA are hard to interpret clearly. In order to better understand the impacts of enhanced AMVs on HWRF forecasts, and the role of VI in the assimilation of enhanced AMVs, the VI is completely turned off during the DA period in the second group of experiments (NVI experiments) for Hurricanes Gonzalo and Joaquin. Here we refer to these experiments as “DA without VI”. In order to achieve this configuration in the current operational HWRF, some changes need to be made in the DA and initialization processes: First, the background fields in each analysis cycle are from the previous cycle’s 6-h HWRF forecasts directly, and neither TC relocation nor intensity corrections are performed; Second, since the d01 is from the HWRF forecasts rather than the GFS analysis as in the first set of experiments under the operational HWRF scenario, the DA for d01 is turned on to make the initial conditions in d01 become more comparable with the GFS analysis. Thus in this set of experiments, DA is performed on all three forecast domains d01, d02, and d03 (Fig.1). Similar to the operational scenario, for the control run the conventional and all other satellite observations are assimilated only in d01 and d02 (storm environment region), while only conventional observations and tail Doppler radar (TDR) are assimilated in d03 (storm inner core region). Furthermore, the model spin-up and DA period are the same as those in the first group of experiments. Corresponding to those DA with VI experiments, six “DA without VI” (NVI) experiments are also performed (see details in Table 1).

Figure 5 and Figure 6 show the track and intensity forecasts from the NVI experiments. It is apparent that the assimilation of enhanced AMVs leads to positive impacts on both track and intensity forecasts. Notably, the extent of the impacts depends on whether the enhanced AMVs are assimilated in the inner-core region or not. For Hurricane Gonzalo (Fig.5a), the assimilation of enhanced AMVs leads to a reduction of average track errors over the 36-h forecast by 9% in NVI-NIN-AMV1 (52 km) and by 18% in NVI-AMV1 (46 km), compared with that in NVI-CTRL1 (56 km). The average MSLP (MSW) forecast error over the 72-h forecasts is reduced by 12% (15%) in NVI-NIN-AMV1 and by 42% (20%) in NVI-AMV1 (Fig.5b, c). For Hurricane Joaquin (Fig.6a), the initial track deviates from the best track since VI is turned off during the DA period, leading to large track errors at the initial time in all three experiments. The assimilation of enhanced AMVs in the parent and intermediate domains (NVI-NIN-AMV1) leads to little to no impact on either track or intensity forecasts as compared with that in NVI-CTRL2. In contrast, the assimilation of enhanced AMVs (NVI-AMV2) to include the innermost domain leads to significant improvements in both track and intensity forecasts: the average track error over the 48-h forecasts in NVI-AMV2 is decreased by 43% compared to that in NVI-CTRL2 (Fig.6b, c). Also notable is that the forecasted track in NVI-AMV2 moves southwest in the first 48-h forecast, which captures the actual trend in the NHC best track. The average intensity errors of both MSLP and MSW over the 72-h forecasts in NVI-AMV2 are decreased by about 70% compared to those in NVI-CTRL2 (Fig.6b, c).

Overall, the results from the two sets of experiments cannot be directly compared since by necessity there were configuration differences. However, what is clear in these two cases is that the assimilation of enhanced AMVs leads to positive impacts on track and intensity forecasts of Hurricanes Gonzalo and Joaquin, although the magnitudes of the impacts depend on the DA configuration used and varies with the hurricane cases. Among all the experiments, the assimilation of enhanced AMVs that include the TC inner-core region leads to the best intensity forecasts.

4. Impacts of enhanced AMVs on hurricane initialization

a. Impacts on initial analysis

To obtain further insights into the influence of the enhanced AMVs on the initial analyses, especially the hurricane inner-core region, o-a (differences between AMV observations and analyses) and o-b (differences between AMV observations and background fields) are investigated for experiments VI-AMV1 and NVI-AMV1 over the innermost domain (d03 in Fig.1). Although direct comparisons of VI-AMV1 and NVI-AMV1 may be misleading as noted earlier because of configuration differences, it is still informative to examine the relative impacts of the enhanced AMVs in these two scenarios. Figure 7 illustrates a histogram of wind speed departure from observations.  It is readily apparent that the assimilation of enhanced AMVs leads to improvements in the initial conditions fit to observations. Specifically, for NVI-AMV1, the wind speed differences between the analysis and the observations (o-a; Fig. 7a), ranged [-7.1, 5.7] m s^-1, are reduced significantly from the difference between the first guess and the observations (o-b) with a range of [-12.2, 9.5] m s^-1. There are approximately 771 (1312) observations where the differences fall within the ± 0.25 m s^-1 range in o-b (o-a). For VI-AMV1, the o-a (Fig.7b) for wind speed with a range of [-7.5, 7.1] m s^-1 are greatly reduced compared with the o-b at a range of [-8.3, 12.6] m s^-1. There are about 793 (1302) observations where the differences fall within ± 0.25 m s^-1 in o-b (o-a).

In addition, a total of 3,545 (1,513) enhanced AMVs are rejected by the DA system in VI-AM1 (NVI-AMV1), leading to a total number of 6,580, (8,612) enhanced AMVs data are assimilated within the d03 region for VI-AMV1 (NVI-AMV1) at 0000, 0600, 1200, and 1800 UTC 24 October 2014. Despite the other configuration differences between VI-AMV1 and NVI-AMV1, the results here indicate that the VI in VI-AMV1 is also an important factor to increase the data rejection rate in the DA system, and reduce the number of enhanced AMVs assimilated into the DA system. However, the o-a differences in VI-AMV1 have nearly the same distribution as those in NVI-AMV1, implying that the improvements in the initial conditions fit to the observations can still be achieved in VI-AMV1, although the DA system assimilates less number of enhanced AMVs. This can partially explain the positive impact of VI-AMV1 on hurricane forecasts compared with VI-CTRL1.

b. Impacts on vortex spin-down

As mentioned in Section 3, VI can induce vortex spin-down problems in HWRF forecasts in some cases. For Hurricane Gonzalo in this study, initial vortex spin-down is not shown in these experiments without VI (i.e., NVI-CTRL1, NVI-NIN-AMV1, NVI-AMV1) or the VI with the assimilation of enhanced AMVs in the inner-core region (VI-AMV1). However, the spin-down issue appears in the control experiment with VI and the experiments with VI but not assimilating enhanced AMVs in the inner-core region (VI-CTRL1 and VI-NIN-AMV1). These results suggest that the VI could be inducing the initial vortex spin-down in our cases as well, and that the assimilation of enhanced AMVs in the inner-core region may help mitigate the vortex spin-down problem.

To further support this, a diagnosis is conducted into the gradient wind balance, following Pu et al. (2016). Willoughby (1990) and Smith et al. (2009) indicated that the azimuthal-mean tangential circulation of TCs, especially in their inner-core region, is approximately in gradient wind and hydrostatic balance. The gradient wind balance relationship in pressure coordinates can be rewritten as:

F=-g∂z∂r+v2r+f0v               (1)

where

gis gravitational acceleration,

zis geopotential height,

vis the tangential wind speed, and

f0is the Coriolis parameter at the storm center.

Frepresents the net radial force field, which is defined as the difference between the local radial pressure gradient and the sum of the centrifugal and Coriolis forces by (Smith et al. 2009). If =0, the tangential flow is in exact gradient wind balance; if <0, the flow is subgradient; and if >0 it is supergradient. Note that all variables in equations (1) are in azimuthal average framework.

Figure 8 compares radius-height cross-sections of isopleths in the GFS analysis and the vortex generated by HWRF VI (before DA, namely, the result of vortex relocation, size, and intensity correction) at 1800 UTC 13 October 2014. It clearly reveals that the vortex in the GFS analysis satisfies the gradient wind balance in its inner-core region (Fig.8a). However, supergradient winds are present in the inner-core region of the HWRF vortex due to its VI procedure (Fig.8b). In addition, a subgradient wind is also clearly seen in the lower boundary layer of the vortex (Fig.8 a, b). These features depicted by Fig.8a, b are consistent with Smith et al. (2009).

Figure 9 compares radius-height cross-sections of isopleths in VI-CTRL1 and VI-AMV1 at HWRF analysis time (1800 UTC 13 October 2014) and its 12-h forecast (0600 UTC 14 October 2014; after the vortex spin-down). At the analysis time, the supergradient winds appear in the vortex inner-core region in VI-CTRL1 (Fig.9a), while in VI-AMV1, the gradient wind balance is more closely established especially above 500hPa (Fig.9b). For the 12-h forecast shown in Fig.9c, the gradient wind imbalances disappear in the inner-core region above 500 hPa in VI-CTRL1 after the vortex spin-down. Meanwhile, there is no substantial adjustment in gradient wind structure in VI-AMV1 during this 12-h forecast (Fig.9d). Therefore, the results here suggest that the imbalance exists in the HWRF initial vortex in VI-CTRL1, and that this imbalance likely derives from the HWRF VI due to the artificial specification of the vortex structure during its relocation, size and intensity corrections. Data assimilation with enhanced AMVs can reduce this initial vortex imbalance to a great extent.

We further examine the vortex structures responding to the imbalances in the HWRF analyses. Figures 10 illustrates the azimuthally-averaged radial velocity (shading) from the GFS analysis, VI-CTRL, and VI-AMV1 initial conditions at 1800 UTC 13 October 2014. Since the GFS analysis is at a coarser resolution compared with the HWRF high-resolution analysis, we use these comparisons only to verify the general features of the hurricane secondary circulation. It is confirmed that VI-AMV1 analysis (Fig. 10c) leads to a secondary circulation of Hurricane Gonzalo that is close to the main features in GFS analysis (Fig. 10a). Meanwhile, VI-CTRL1(Fig.10b) leads to weaker upper-layer outflow (above 300 hPa) within the hurricane inner-core region than the GFS analysis (Fig.10a) and VI-AMV1 do (Fig.10c). A strong outflow between 500 hPa and 300 hPa within the inner-core region is also shown in VI-CTRL1 (Fig.10b), while these features are not shown in the GFS (Fig.10a) analysis and VI-AMV1 (Fig.10c). In addition, the secondary circulation in VI-CTRL1 is tilted with height, while this is not true in the GFS analysis (Fig.10a) and VI-AMV1 (Fig.10c).

Figure 11 shows the azimuthally-averaged relative humidity (green contour), temperature perturbation (shading), and secondary circulation (vector) from VI-CTRL1 and VI-AMV1 for Hurricane Gonzalo from the 6-h forecast valid at 0000 UTC 14 October 2014. It is seen that VI-CTRL1 (Fig.11a) leads to a weaker secondary circulation, upper level warming, and midlevel relative humidity than VI-AMV1 (Fig.11b). Specifically, the upper level outflow (above 300hPa) in VI-CTRL1 is 1-5 m s^-1 smaller than in VI-AMV1 (Fig.11c). The maximum temperature difference between VI-AMV1 and VI-CTRL1 in the upper level within the hurricane inner-core is above 2 ^oC, and the middle level relative humidity within the hurricane inner-core in VI-CTRL1 is 5-50% smaller than that in VI-AMV1 (Fig.11c). Higher middle level moisture and upper level warming in the hurricane inner-core region have been shown to be essential for rapid intensification (RI) (Malkus and Riehl 1960; Holland 1997; Zhang and Chen 2012; Chen and Zhang 2013). Thus, VI-CTRL1 presents less favorable conditions for RI of Hurricane Gonzalo, compared with VI-AMV1. These results help explain why VI-CTRL1 does not capture the RI of hurricane Gonzalo.

Overall, the vortex spin-down problem in VI-CTRL1 for Hurricane Gonzalo is associated with the imbalances generated by the vortex initialization process in HWRF. Consequently, unrealistic upper-level outflow structure and more unfavorable thermodynamic conditions for the RI of Hurricane Gonzalo is induced, leading to the degradation of HWRF intensity forecasts in VI-CTRL1. The assimilation of enhanced AMVs in the inner-core region is shown to mitigate some of these imbalances in the vortex, and the unrealistic outflow structure and unfavorable thermodynamical conditions are improved. As a result, the vortex spin-down is mostly mitigated, and the intensity forecasts are improved in VI-AMV1.

5. Summary and discussion

In this study, DA experiments and numerical simulations are conducted using the HWRF model and its GSI-based ensemble-variational hybrid DA system to predict the intensification phase of Hurricanes Gonzalo (2014) and Joaquin (2015). Specifically, the impacts on inner-core DA from enhanced AMV datasets (produced from GOES satellites in a demonstration mode by CIMSS at the University of Wisconsin) on the intensity prediction of the two hurricanes are evaluated. Two different DA strategies, one including HWRF operational settings that use a combination of VI with DA in each analysis cycle (DA with VI), and another using all available data (DA without VI), are utilized. Results from these experiments on HWRF initial analyses and forecasts are examined. It is found that in both the DA with VI and DA without VI experiments, the inner-core assimilation of enhanced AMVs leads to promising improvements in track and intensity forecasts, while assimilating the enhanced AMVs outside the inner domain leads to more limited impacts. This indicates that inner-core AMV information could be an important source for HWRF initialization.

For Hurricane Gonzalo, inner-core assimilation of enhanced AMVs alleviated much of the vortex spin-down problem induced by the HWRF VI, leading to significant positive impacts on track and intensity forecasts. For Hurricane Joaquin, the assimilation of enhanced AMVs leads to limited impacts on HWRF forecasts as the control experiment with VI produced realistic intensity forecasts. However, DA without VI shows consistent positive impacts on HWRF analyses and forecasts when enhanced AMVs are assimilated in this case.

It is further found that VI has potential negative impacts on the DA of enhanced AMVs, as the VI before DA alters the first guess and through quality control processes then reduces the actual number of AMV observations assimilated into the DA system. Also, it is shown that the vortex spin-down for Hurricane Gonzalo is associated with imbalances generated by the VI process in HWRF. Consequently, unrealistic upper-level outflow structure and unfavorable thermodynamic conditions for the rapid intensification of Hurricane Gonzalo are induced, leading to the degradation of intensity forecasts. Meanwhile, assimilation of enhanced AMVs in the inner-core region can mitigate the imbalances in the vortex, and the unrealistic outflow structure and unfavorable thermodynamic conditions are improved. As a result, the intensity forecasts are improved.

Results from this study are encouraging but only encompass two hurricane cases. A larger sample of TC cases will be necessary in order to evaluate the consistency of the forecast impacts on the HWRF model provided by enhanced AMV data.  However, insights from this study should be useful for further investigations. It is demonstrated that enhanced AMVs can be successfully incorporated into the analysis of the hurricane inner-core region, leading to improved initial conditions and positive impacts on model track and intensity forecasts.

REFERENCES

Bao, J.-W., S. G. Gopalakrishnan, S. A. Michelson, F.D. Marks, and M. T. Montgomery, 2012: Impact of physics representations in the HWRFX on simulated hurricane structure and pressure–wind relationships. Mon. Wea. Rev., 140, 3278–3299.

Berg, R., 2016: Tropical cyclone report: Hurricane Joaquin, 28 September – 7 October 2015. NOAA/NHC Tech. Rep. AL112015, 36 pp. [Available online at http://www.nhc.noaa.gov/data/tcr/AL112015_Joaquin.pdf]

Berger, H., R. Langland, C. S. Velden, C. A. Reynolds, and P. M. Pauley, 2011: Impact of enhanced satellite-derived atmospheric motion vector observations on numerical tropical cyclone track forecasts in the western North Pacific during TPARC/TCS-08. J. Appl. Meteor. Climatol., 50, 2309–2318.

Brown, D. P., 2014: Tropical cyclone report: Hurricane Gonzalo, 12-19 October 2014. NOAA/NHC Tech. Rep. AL082014, 30 pp. [Available online at http://www.nhc.noaa.gov/data/tcr/AL082014_Gonzalo.pdf]

Chen, H., and D.-L. Zhang, 2013: On the rapid intensification of Hurricane Wilma (2005). Part II: Convective bursts and the upper-level warm core. J. Atmos. Sci.,70, 146–162.

Gopalakrishnan, S. G., F. Marks, X. Zhang, J.-W. Bao, K.-S. Yeh, and R. Atlas, 2011: The experimental HWRF system: A study on the influence of horizontal resolution on the structure and intensity changes in tropical cyclones using an idealized framework. Mon. Wea. Rev., 139, 1762–1784.

Holland, G. J., 1997: The maximum potential intensity of tropical cyclones. J. Atmos. Sci., 54, 2519–2541.

Hsiao, L.-F., C.-S. Liou, T.-C. Yeh, Y.-R. Guo, D.-S. Chen, K.-N. Huang, C.-T. Terng, and J.-H. Chen, 2010: A vortex relocation scheme for tropical cyclone initialization in Advanced Research WRF. Mon. Wea. Rev., 138, 3298–3315.

Janjic, Z. I., R. Gall,and M. E. Pyle, 2010: Scientific documentation for the NMM Solver. NCAR Technical Note NO. NCAR/TN–477+STR, 1-125, 53 pp. (Available from NCAR,447P.O. Box 3000, Boulder, CO 80307)

Kleist, D. T., D. F. Parrish, J. C. Derber, R. Treadon, W.-S. Wu, and S. Lord, 2009: Introduction of the GSI into the NCEP Global Data Assimilation System. Wea. Forecasting, 24, 1691–1705, doi:10.1175/2009WAF2222201.1.

Liu, Q., N. Surgi, S. Lord, W.-S. Wu, S. Parrish, S. Go palakrishnan, J. Waldrop, and J. Gamache, 2006: Hurricane initialization in HWRF model. Preprints, 27th Conference on Hurricanes and Tropical Meteorology, Monterey, CA.

Malkus, J. S., and H. Riehl, 1960: On the dynamics and energy transformations in steady-state hurricanes. Tellus, 12, 1–20.

Parrish, D., and J. Derber, 1992: The National Meteorological Center’s spectral statistical interpolation analysis system. Mon. Wea. Rev., 120, 1747–1763.

Pu, Z, and S. Braun, 2001: Evaluation of bogus vortex techniques with four-dimensional variational data assimilation. Mon. Wea. Rev., 129, 2023-2039.

Pu, Z., X. Li, C. Velden,S. Aberson, W. T. Liu, 2008: Impact of aircraft dropsonde and satellite wind data on the numerical simulation of two landfalling tropical storms during TCSP . Wea. Forecasting 23, 62-79.

Pu, Z., X. Li, and E. J. Zipser, 2009: Diagnosis of the initial and forecast errors in the numerical simulation of rapid intensification of Hurricane Emily. Wea. Forecasting, 24, 1236–1251.

Pu, Z., S. Zhang, M. Tong and V. Tallapragada. 2016: Influence of the self-consistent regional ensemble background error covariance on hurricane inner-core data assimilation with the GSI-based hybrid system for HWRF, J. Atmos. Sci., in press. http://dx.doi.org/10.1175/JAS-D-16-0017.1

Rappaport, E. N., and Coauthors, 2009: Advances and challenges at the National Hurricane Center. Wea. Forecasting, 24, 395–419.

Rogers, R., and Coauthors, 2006: The intensity forecasting experiment: A NOAA multiyear field program for improving tropical cyclone intensity forecasts. Bull. Amer. Meteor. Soc., 87, 1523–1537.

Rogers, R. F., S. Aberson, A. Aksoy, B. Annane, M. Black, J. Cione, N. Dorst, J. Dunion, J. Gamache, S. Goldenberg, S. Gopalakrishnan, J. Kaplan, B. Klotz, S. Lorsolo, F. Marks, S. Murillo, M. Powell, P. Reasor, K. Sellwood, E. Uhlhorn, T. Vukicevic, J. Zhang, and X. Zhang, 2013: NOAA’s Hurricane Intensity Forecasting Experiment: A progress report. Bull. Amer. Meteor. Soc., 94, 859–882.

Smith, R. K., M. T. Montgomery, and N. Van Sang, 2009: Tropical cyclone spin-up revisited. Quart. J. Roy. Meteor. Soc., 135, 1321–1335.

Soden, B. J., C. S. Velden, and R. E. Tuleya, 2001: The impact of satellite winds on experimental GFDL hurricane model forecasts. Mon. Wea. Rev., 129, 835–852.

Tallapragada V., L. Bernardet, M. K. Biswas, I. Ginis, Y. Kwon, Q. Liu, T. Marchok, D. Sheinin, B. Thomas, M. Tong, S. Trahan, W. Wang, R. Yablonsky, X. Zhang, 2015. Hurricane Weather Research and Forecasting (HWRF) model: 2015 scientific documentation. Available online at http://nldr.library.ucar.edu/collections/technotes/TECH-NOTE-000-000-000-893.pdf

Velden, C., C. M. Hayden, S. J. Nieman, W. P. Menzel, S. Wanzong, and J. S. Goerss, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc., 78, 173–195.

Velden, C., J. Daniels, D. Stettner, D. Santek, J. Key, J. Dunion, K. Holmlund, G. Dengel, W. Bresky, and P. Menzel, 2005: Recent Innovations in Deriving Tropospheric Winds from Meteorological Satellites. Bull. Amer. Meteor. Soc., 86, 205–223.

Velden, C., W. Lewis, D. Stettner, W. Bresky and J. Daniels, S. Wanzong, 2017: Assimilation of high-resolution satellite-derived atmospheric motion vectors: Impact on HWRF forecasts of tropical cyclone track and intensity. Mon. Wea. Rev., 145, 1107 – 1125.

Wang, X., D. Parrish, D. Kleist, and J. Whitaker, 2013: GSI  3DVar-based ensemble–variational hybrid data assimilation for NCEP Global Forecast System: Single-resolution experiments. Mon. Wea. Rev., 141, 4098 – 4117.

Willoughby, H. E., 1990: Gradient balance in tropical cyclones. J. Atmos. Sci., 47, 265–274

Wu, T.-C., H. Liu, S. J. Majumdar, C. S. Velden, and J. L. Anderson, 2014: Influence of assimilating satellite-derived atmospheric motion vector observations on numerical analyses and forecasts of tropical cyclone track and intensity. Mon. Wea. Rev., 142, 49–71.

Wu, T.-C., C. S. Velden, S. J. Majumdar, H. Liu, and J. L. Anderson, 2015: Understanding the influence of assimilating subsets of enhanced atmospheric motion vectors on numerical analyses and forecasts of tropical cyclone track and intensity with an ensemble Kalman filter. Mon. Wea. Rev., 143, 2506–2531.

Wu, W. S., R. J. Purser, and D. F. Parrish, 2002: Three-dimensional variational analysis with spatially inhomogeneous covariances. Mon. Wea. Rev., 130, 2905–2916.

Zhang, F., Y. Weng, J. F. Gamache, and F. D. Marks, 2011: Performance of convection- permitting hurricane initialization and prediction during 2008-2010 with ensemble data assimilation of inner-core airborne Doppler radar observations. Geophys. Res. Lett., 38, 481 L15810.

Zhu, Y., E. Liu, R. Mahajan, C. Thomas, D. Groff, P. Van Delst, A. Collard, D. Kleist, R. Treadon and J.C. Derber, 2016. All-sky microwave radiance assimilation in NCEP’s GSI analysis system. Mon. Wea. Rev., 144, 4709-4735.

Zhang D. -L., and H. Chen, 2012: Importance of the upper-level warm core in the rapid intensification of a tropical cyclone. Geophys. Res. Lett., 39, L02806.

Figure Captions

Figure 1 HWRF-simulated sea level pressure (color contours, units hPa) and storm center from NHC best track (black storm sign) at 0600UTC 13 October 2014 for Hurricane Gonzalo.  HWRF model forecast domains, as indicated by d01, d02, and d03, labels and HWRF data assimilation domains, as indicated by ghost d02 (black shaded area), and ghost d03 (pink shaded area) are also indicated. The black symbol indicates the storm center at current time.

Figure 2 Spatial distribution of enhanced AMVs over a 20^o x 20^o box around storm after quality control at (a) 1800 UTC 13 October 2014 for Hurricane Gonzalo, and (b) 0000 UTC 29 September 2015 for Hurricane Joaquin. For illustration purposes, the AMVs are grouped into four layers by their assigned heights: < 250 hPa (red), 250-400 hPa (green), 700-850 hPa (purple), and > 850hPa (blue), with the numbers above the figures indicating the total number of AMVs in each grouped layer. There are no enhanced AMVs observations between 400hPa and 700hPa.

Figure 3 The 72-h HWRF forecast from 18 UTC 13 October, 2014 for Hurricane Gonzalo of (a) track and intensity in terms of (b) minimum mean sea level pressure (hPa) and (c) maximum surface wind (m s^-1) from VI-CTRL1 (red lines), VI-AMV1 (blue lines) and VI-NIN-AMV1 (green lines). The track, minimum central sea level pressure (MSLP), and maximum surface wind (MSW) are verified against NHC best track data (black lines). The colored numbers in (c) denote the averaged track errors and absolute intensity errors for the corresponding experiments as indicated in (b).

Figure 4 The same as Figure 3 but from VI-CTRL2 (red lines), VI-AMV2 (blue lines) and VI-NIN-AMV2 (green lines) for Hurricane Joaquin.

Figure 5 The same as Figure 3 but from NVI-CTRL1 (red lines), NVI-AMV1 (blue lines) and NVI-NIN-AMV1 (green lines) for Hurricane Gonzalo.

Figure 6 The same as Figure 3 but from NVI-CTRL2 (red lines), NVI-AMV2 (blue lines) and NVI-NIN-AMV2 (green lines) for Hurricane Joaquin.

Figure 7 Histogram of wind speed departure from observations for HWRF first-guess winds (o-b, blue lines) and HWRF analyzed winds (o-a, red lines) for (a) NVI-AMV1 and (b) VI-AMV1 experiments. The departure is calculated for the enhanced AMVs data in the HWRF d03 domain from the first-guess and analysis fields at 0000, 0600, 1200 and 1800 UTC 03 October 2014.  The horizontal axis denotes the wind speed departure (m s^-1) and the vertical axis denotes the number of observations.

Figure 8. Radius–height cross-sections of the isopleths of the net radial force per unit mass for (a) GFS and (b) vortex initialization (before data assimilation) at 1800 UTC 13 October 2014. The contour interval is 5 m s⁻¹ h⁻¹, with dashed lines indicating negative values. The zero contour is not plotted. The red lines indicate the radius of maximum wind.

Figure 9.  Same as Fig.8, except for (a, c) VI-CTRL1 and (b, d) VI-AMV1 at 1800 UTC 13 October 2014 (a, b) and 12-h forecasts at 0600 UTC 14 October 2014 (c, d).

Figure 10 The azimuthally averaged radial wind (color contours; units m s^-1) from (a) GFS analysis, (b) VI-CTRL1, and (c) VI-AMV1 for Hurricane Gonzalo at 1800 UTC 13 October 2014.

Figure 11 Azimuthally averaged relative humidity (green contour in 5% interval, only contours above 60% are shown), temperature perturbation (shading), secondary circulation [represented by u-w vectors; u (radial velocity) units m/s; w (vertical velocity), units cm/s] for (a) VI-CTRL1 and (b) NVI-CTRL1 for Hurricane Gonzalo at 0000 UTC 14 October 2014. The azimuthally averaged values are calculated as differences between the temperatures near the simulated circulation center (0-300 km) and the environmental temperature, which is determined by averaging the temperature within the whole of nested domain 3.

Table and Table Caption

Table.1 List of DA experiments.

Figure 1 HWRF-simulated sea level pressure (color contours, units hPa) and storm center from NHC best track (black storm sign) at 0600UTC 13 October 2014 for Hurricane Gonzalo.  HWRF model forecast domains, as indicated by d01, d02, and d03, labels and HWRF data assimilation domains, as indicated by ghost d02 (black shaded area), and ghost d03 (pink shaded area) are also indicated. The black symbol indicates the storm center at current time.

Figure 2 Spatial distribution of enhanced AMVs over a 20^o x 20^o box around storm after quality control at (a) 1800 UTC 13 October 2014 for Hurricane Gonzalo, and (b) 0000 UTC 29 September 2015 for Hurricane Joaquin. For illustration purposes, the AMVs are grouped into four layers by their assigned heights: < 250 hPa (red), 250-400 hPa (green), 700-850 hPa (purple), and > 850hPa (blue), with the numbers above the figures indicating the total number of AMVs in each grouped layer. There are no enhanced AMVs observations between 400hPa and 700hPa.

Figure 3 The 72-h HWRF forecast from 18 UTC 13 October, 2014 for Hurricane Gonzalo of (a) track and intensity in terms of (b) minimum mean sea level pressure (hPa) and (c) maximum surface wind (m s^-1) from VI-CTRL1 (red lines), VI-AMV1 (blue lines) and VI-NIN-AMV1 (green lines). The track, minimum central sea level pressure (MSLP), and maximum surface wind (MSW) are verified against NHC best track data (black lines). The colored numbers in (c) denote the averaged track errors and absolute intensity errors for the corresponding experiments as indicated in (b).

Figure 4 The same as Figure 3 but from VI-CTRL2 (red lines), VI-AMV2 (blue lines) and VI-NIN-AMV2 (green lines) for Hurricane Joaquin.

Figure 5 The same as Figure 3 but from NVI-CTRL1 (red lines), NVI-AMV1 (blue lines) and NVI-NIN-AMV1 (green lines) for Hurricane Gonzalo.

Figure 6 The same as Figure 3 but from NVI-CTRL2 (red lines), NVI-AMV2 (blue lines) and NVI-NIN-AMV2 (green lines) for Hurricane Joaquin.

Figure 7 Histogram of wind speed departure from observations for HWRF first-guess winds (o-b, blue lines) and HWRF analyzed winds (o-a, red lines) for (a) NVI-AMV1 and (b) VI-AMV1 experiments. The departure is calculated for the enhanced AMVs data in the HWRF d03 domain from the first-guess and analysis fields at 0000, 0600, 1200 and 1800 UTC 03 October 2014.  The horizontal axis denotes the wind speed departure (m s^-1) and the vertical axis denotes the number of observations.

Figure 8. Radius–height cross-sections of the isopleths of the net radial force per unit mass for (a) GFS and (b) vortex initialization (before data assimilation) at 1800 UTC 13 October 2014. The contour interval is 5 m s⁻¹ h⁻¹, with dashed lines indicating negative values. The zero contour is not plotted. The red lines indicate the radius of maximum wind.

Figure 9.  Same as Fig.8, except for (a, c) VI-CTRL1 and (b, d) VI-AMV1 at 1800 UTC 13 October 2014 (a, b) and 12-h forecasts at 0600 UTC 14 October 2014 (c, d).

Figure 10 The azimuthally averaged radial wind (color contours; units m s^-1) from (a) GFS analysis, (b) VI-CTRL1, and (c) VI-AMV1 for Hurricane Gonzalo at 1800 UTC 13 October 2014.

Figure 11 Azimuthally averaged relative humidity (green contour in 5% interval, only contours above 60% are shown), temperature perturbation (shading), secondary circulation [represented by u-w vectors; u (radial velocity) units m/s; w (vertical velocity), units cm/s] for (a) VI-CTRL1 and (b) VI-AMV1 for Hurricane Gonzalo at 0000 UTC 14 October 2014. (c) The differences of azimuthally averaged temperature (shading, only positive values are shown), relative humidity (green contour in 5% interval), and radial velocity above 300hPa (red contour in 2m s^-1 contour) between VI-AMV1 and VI-CTRL1. The azimuthally averaged values are calculated as differences between the temperatures near the simulated circulation center (0-300 km) and the environmental temperature, which is determined by averaging the temperature within the whole of domain 3.