1. Introduction
Tropical cyclones (TCs) are among the deadliest and costliest natural disasters. Of the billion-dollar weather and climate disasters that have afflicted the United States since 1980, tropical cyclones rank first in terms of both fatalities and cost to the economy (Smith 2020). Improving upon track and intensity forecasts is thus of great societal value.
Over the last couple decades, the average track errors of National Hurricane Center (NHC) 72-h hurricane forecasts have more than halved, whereas the 72-h forecasts of maximum surface wind speed have seen less improvement over the same time period (Cangialosi 2020). The relatively slow improvement in TC intensity forecasting has been attributed to a number of difficulties. These difficulties include the challenges associated with accurately simulating and predicting small-scale nonlinear features such as moist convection over the hurricane inner-core region (e.g., Hendricks et al. 2004; Krishnamurti et al. 2005; Montgomery et al. 2006; Rotunno et al. 2009; Zhang and Weng 2015; Christophersen et al. 2017), model physics deficiencies regarding TC inner-core dynamics (e.g., Bao 2016; Zhang et al. 2017; Zhang 2018; Rajeswari et al. 2020), the lack of sufficient TC inner-core observations (e.g., Pu et al. 2016), and our incomplete understanding of air–sea interactions at high wind speeds (e.g., Green and Zhang 2014; Andreas et al. 2015; Chen and Zhang 2019; Nystrom et al. 2020; Chen et al. 2021). TC intensity forecast errors are also greatly influenced by initial intensity and structure errors within the TC inner-core vortex, which makes prediction of rapid intensification especially challenging (Emanuel and Zhang 2016). In addition to minimizing initial intensity and structure errors, it is also important to properly initialize inner-core moisture (Emanuel and Zhang 2017), something that is not easily observed.
To minimize initialization errors in the TC inner-core vortex, a number of studies have used data assimilation (DA) systems to assimilate either ground-based (e.g., Osuri et al. 2015; Zhu et al. 2015; Shen et al. 2016, 2020) or airborne (e.g., Aberson et al. 2015; Zhang and Weng 2015; Tong et al. 2018; Nystrom and Zhang 2019) Doppler radar observations. Collectively, these studies show that assimilating Doppler radar radial velocities reduces both track and intensity errors relative to assimilating only conventional GTS observations. These improvements are due to better representation of TC initial position and inner-core dynamic structures.
Despite the improvements brought to TC forecasts from assimilating Doppler radar radial velocities, there are some challenges associated with their assimilation. For example, because the radar observations come in high quantities and a resolution significantly higher than the model grid used, substantial data thinning and quality control must be implemented (Zhang et al. 2009). Moreover, Doppler radar observations over the ocean lack continuous spatial and temporal coverage.
In contrast, observations from the new generation geostationary satellites (e.g., GOES-16 and Himawari-8) have continuous coverage. The benefits of assimilating satellite all-sky infrared (IR) observations on convection-permitting analysis and prediction of TCs were first explored in the pilot work of Zhang et al. (2016). Motivated by the strong ensemble correlations between simulated water vapor channel brightness temperature (BT) and model state variables (e.g., moisture and winds), Zhang et al. (2016) assimilated BTs into an ensemble of convection-permitting simulations of Hurricane Karl using an ensemble Kalman filter (EnKF). They found that assimilating all-sky IR BTs improved both the analyzed thermodynamic fields and the intensity forecast of Hurricane Karl. Subsequent studies (e.g., Honda et al. 2018; Minamide and Zhang 2018; F. Zhang et al. 2019) have further highlighted the benefits of assimilating all-sky IR BTs. These more recent studies have shown that TC structure (both inner core and outer rainbands) is better captured when all-sky IR BTs are assimilated. These improved cloud and moisture structures, combined with a more resilient TC vortex, resulted in better intensity forecasts.
Despite the benefits of assimilating all-sky IR BTs, at the time of writing, all-sky IR BTs have yet to be assimilated operationally (Geer et al. 2019) due to inherent difficulties. One problem with satellite observations is that they tend to be biased relative to their model equivalents and removing these biases is not straightforward (Eyre 2016; Otkin et al. 2018). Additionally, models have trouble predicting the exact location and intensity of clouds, which leads to highly non-Gaussian errors that are introduced into DA systems that assume Gaussian error distributions (Geer and Bauer 2011; Chan et al. 2020a). These mismatches between observed and simulated cloud scenes also lead to large representativeness errors that must be accounted for (Minamide and Zhang 2017). Finally, the relationship between cloudy-sky BTs and atmospheric states is highly nonlinear (Bauer et al. 2011; Geer and Bauer 2011), which makes construction of an accurate forward model challenging.
Given the inherent strengths and weaknesses of assimilating either Doppler radar radial velocities or all-sky IR BTs, it is surprising that no studies to date have assimilated both. In this study, we investigate the potential improvements brought to TC analyses and forecasts by leveraging the benefits of assimilating GOES-16 all-sky IR BTs as well as NOAA P-3 tail Doppler radar (TDR) radial velocities. We use the case of Hurricane Dorian (2019), a powerful storm whose rapid intensification was poorly modeled by most operational centers (Avila et al. 2020), as a testbed for this study. This storm is also selected for our study due to the abundance of TDR observations available both before and during its rapid intensification.
2. Data and methodology
In this section we provide a brief summary of the life of Hurricane Dorian. We then introduce the DA system and forecast model utilized for this study. After describing the observations used, we conclude by illustrating the experimental design.
a. Description of Hurricane Dorian
Dorian formed from a tropical wave originating off the west coast of Africa and became a named storm on 25 August 2019 as it drifted westward toward the Lesser Antilles. Over the next few days Dorian remained a tropical storm as it traversed the islands of the Lesser Antilles. The system thereafter curved to the north, putting it in a more favorable environment for development. Hurricane Dorian then began to rapidly intensify north of the Virgin Islands on 30 August. Hurricane Dorian eventually made landfall on 1 September, ravaging the Bahamas with 1-min sustained winds of 185 mph (Avila 2019) and making it one of the strongest Atlantic hurricanes at time of landfall.
b. The DA and forecast system
This study used The Pennsylvania State University ensemble Kalman filter (PSU WRF-EnKF) system, which is based on the ensemble square root filter (EnSRF) proposed by Whitaker and Hamill (2002). The PSU WRF-EnKF system was first used in an observing system experiment (OSE) framework by Meng and Zhang (2008). In its current form, the PSU WRF-EnKF system performs the data assimilation in joint state-observation space, which is described by Eqs. (1) and (2) of Chan et al. (2020b). To assimilate BTs in this study, the Community Radiative Transfer Model (CRTM; Han et al. 2006, 2007; Weng 2007) release 2.1.3 was used to convert from model state space to observation space. We used 60 ensemble members and applied 75% relaxation to prior perturbations (Zhang et al. 2004) in order to prevent filter divergence.
The WRF Model for the cycling EnKF system was configured similar to the real-time PSU WRF-EnKF TC analysis and prediction system (Zhang and Weng 2015), with three two-way nested domains having horizontal grid spacings of 27, 9, and 3 km. The nested domain with 9-km grid spacing was 2700 km × 2700 km, whereas the nested domain with 3-km grid spacing was 900 km × 900 km. Domain configuration and parameterization schemes were nearly identical to F. Zhang et al. (2019). All three domains had 43 vertical levels with model top at 10 hPa. The Tiedtke (1989) cumulus parameterization scheme was only applied to the coarsest (27-km) domain whereas all three domains made use of the WRF single-moment 6-class mixed-phase microphysics scheme (WSM6; Hong and Lim 2006), the Yonsei University planetary boundary layer scheme (Hong et al. 2006), and the Rapid Radiative Transfer Model (RRTM) longwave and shortwave radiation schemes (Iacono et al. 2008). Surface fluxes of sensible and latent heat, as well as momentum, were parameterized via Eqs. (11)–(13) of Green and Zhang (2013).
c. The observations
Observations assimilated in this study fell into three categories: conventional observations from the Global Telecommunications System (GTS), TDR radial velocities collected by the NOAA P-3 aircraft, and IR BTs from channel 8 of the GOES-16 Advanced Baseline Imager (ABI). Conventional GTS observations were assimilated in all three domains whereas TDR and IR observations were assimilated in the 3-km domain (D3) only.
The conventional GTS observations assimilated in this study include hurricane position and intensity (HPI) from the “TC Vitals” database, as well as radiosondes, dropsondes, and METAR data. To eliminate the effects of spurious long-distance correlations, the Gaspari and Cohn (1999) fifth-order piecewise polynomial was applied to localize ensemble covariance. The localization radius of influence (ROI) in the vertical was 43 vertical levels for all observations. Consistent with F. Zhang et al. (2019), the horizontal ROI was 300 km for HPI and METAR and 90 km for radiosondes and dropsondes. During assimilation, observation errors for these observables were those specified by version 3.6.1 of the WRFDA package. For quality control purposes, METAR and sounding observations were discarded if the absolute value of the observation increment (also known as the “innovation”) was greater than 5 times the observation error.
The TDR observations assimilated in this study were obtained from the NOAA Hurricane Research Division (HRD). These observations were first preprocessed by HRD using the superobservation procedure outlined in Weng and Zhang (2012, 2016). These superobservation radar data have a spacing of 5 km in the radial direction and 5° in the azimuthal direction. Further thinning to every tenth superobservation resulted in an average distance between TDR superobservations of approximately 23 km. When assimilating TDR superobservations, Gaspari and Cohn (1999) covariance localization was employed together with the successive covariance localization (SCL) procedure described in Zhang et al. (2009). As per SCL, large-scale features were first corrected by randomly selecting a ninth of the thinned TDR superobservations and assimilating them with an ROI of 405 km. Smaller-scale features were then corrected by randomly assimilating two ninths of the thinned superobservations with an ROI of 135 km and the remaining superobservations with an ROI of 45 km. The vertical ROI for all superobservations was 43 vertical levels. Consistent with Zhang et al. (2009), the observation error for all TDR superobservations was assumed to be 3 m s−1. As with the conventional GTS observations, a TDR superobservation was discarded if the absolute value of the observation increment was more than 5 times the observation error.
The GOES-16 ABI all-sky IR BTs assimilated in this study came from channel 8 (6.2-μm wavelength), which is sensitive to upper-tropospheric water vapor (Schmit et al. 2005). The raw BT data have a horizontal resolution of approximately 2 km. As in F. Zhang et al. (2019), a form of SCL was performed on the IR BTs such that BT observations thinned every 12 km had a 30-km horizontal ROI and observations thinned every 18 km had a 200-km horizontal ROI. No vertical localization was applied when assimilating BT observations. As mentioned in the introduction, assimilating all-sky IR BTs can sometimes lead to large representativeness errors. To deal with this problem, we employed adaptive observation error inflation (AOEI). AOEI is a method that was introduced by Minamide and Zhang (2017) to adaptively adjust observation errors to limit erroneous analysis increments associated with representativeness errors. Note that this study did not reject any IR observations because the observation error is adaptively adjusted by using AOEI. In addition, we applied adaptive background error inflation (ABEI; Minamide and Zhang 2019) to deal with situations in which the model erroneously predicted clear skies despite cloudy observations. This method uses an empirically derived spatially varying multiplicative inflation factor that inflates the state vector components associated with erroneously predicted clear sky regions within each member of the prior ensemble. However, it does not create cloud particles at the analysis step. Instead, it increases the likelihood of developing clouds by increasing the ensemble spread. Following Y. Zhang et al. (2019), bias correction was not performed on the GOES-16 IR BTs used in this study. Previous studies (e.g., Zhang et al. 2018; Y. Zhang et al. 2019) have shown that, based on the statistics of innovations throughout EnKF cycling, there is no significant bias in our system. We confirmed this for the current study as well (figure not shown).
We based the setup of the experiments in this study on the setup of the PSU WRF-EnKF real-time TC forecast system described in Zhang and Weng (2015) and F. Zhang et al. (2019). Data thinning and localization of IR BTs and TDR observations in the PSU system have been extensively tuned over several years in order to achieve a statistically improved TC intensity forecast result in the real-time system (e.g., Zhang et al. 2011; Weng and Zhang 2012; Zhang and Weng 2015; Zhang et al. 2016; Minamide and Zhang 2017, 2018; F. Zhang et al. 2019). Hence, we used the same setup for the experiments in this study. Although it is possible that the experimental results in this study may be further improved by fine-tuning of data thinning and localization, doing so is beyond the scope of this study.
d. The experimental design
A schematic of the experimental design of this study is provided in Fig. 1. The design of these experiments was to mimic near-operational cycling DA systems (similar to the configuration of the real-time PSU WRF-EnKF TC analysis and prediction system) that assimilate different observation types. The initial and boundary conditions for this study came from the National Centers for Environmental Prediction’s (NCEP) Global Forecast System (GFS) analysis at 0000 UTC 27 August. A 60-member ensemble was initiated at this time by adding perturbations to the GFS analysis using WRFDA’s CV3 background error covariance (Barker et al. 2004). At that time, Dorian was still a weak tropical storm with a maximum sustained wind speed of 45 kt (1 kt ≈ 0.51 m s−1) and a minimum central pressure of 1005 hPa. We then integrated the ensemble with WRF for 12 h to develop flow-dependent ensemble statistics before assimilating the first observations at 1200 UTC 27 August. Because no TDR observations were available at that time, the ensemble became the background for the GTS Only and GTS+IR experiments. The GTS Only experiment assimilated only the conventional observations whereas the GTS+IR experiment assimilated the IR BTs after the conventional observations. These two cycling experiments continued assimilating observations every hour up to and including 0000 UTC 29 August. There was a total of 4455 conventional GTS observations available within domain 3 of the GTS Only experiment over the course of the 37 cycles. This translates to an average of approximately 120 observations per cycle. The number of conventional GTS observations available in any particular cycle varied widely, with the number being as high as 456 at the 2300 UTC 28 August cycle and as low as 1 (the HPI observation) at the 0900 UTC 28 August cycle. The number of IR BTs assimilated in each cycle was approximately 8000.
Overview of the experimental design. The names of the hourly cycling experiments indicate the order in which each observation type was assimilated. Black “X” marks indicate the times at which deterministic forecasts were initialized from the analysis means of the experiments, and gray circles indicate hours that TDR observations were assimilated. Note that the GTS+TDR experiment was initialized from the background of the GTS Only experiment at 2200 UTC 27 Aug whereas the experiments that assimilated all observation types were initialized from the background of the GTS+IR experiment at 2200 UTC 27 Aug.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
TDR observations were available starting 2200 UTC 27 August. At that time, the GTS+TDR experiment was initialized from the background of the GTS Only experiment. This experiment assimilated TDR observations after conventional GTS observations and was cycled forward to 0000 UTC 29 August. Note that, from 0300 UTC 28 August through 2000 UTC 28 August, TDR observations were not available. During that time, only conventional GTS observations were assimilated each hour in the GTS+TDR experiment. In addition to the GTS+TDR experiment, two more experiments were conducted that assimilated all three types of observations: GTS+TDR+IR and GTS+IR+TDR. The only difference between these last two experiments is the order in which observations were assimilated. Both experiments were initialized at 2200 UTC 27 August from the background of the GTS+IR experiment and cycled forward to 0000 UTC 29 August. During the window when TDR observations were not available, both experiments assimilated conventional GTS followed by IR BT observations each hour. An average of 2381 TDR observations were assimilated per cycle during the nine cycles when TDR observations were available, and an average of 1 TDR observation per cycle was rejected. These values did not vary across experiments.
As a metric of assessing the value of assimilating these observations, 5-day (120-h) convection-allowing deterministic forecasts were initialized from the analysis mean of each experiment at the times indicated by the black “X” marks in Fig. 1. These deterministic forecasts utilized the same WRF physics options described earlier in this section.
3. Results
This section is divided into two parts. The first part shows the impacts of assimilating all-sky IR BTs on the analyses and forecasts of Hurricane Dorian prior to the first TDR cycle, whereas the second part shows the impacts of simultaneously assimilating TDR and IR observations.
a. Impacts of IR BT assimilation
Before looking at the impacts of simultaneous assimilation of TDR radial velocity and IR BT observations, we first discuss the impacts of assimilating all-sky IR BTs on the analyses and forecasts of Hurricane Dorian prior to the first TDR cycle. We begin by examining deterministic forecasts of Dorian’s track. All deterministic forecasts initialized from the first cycle’s analyses show large track errors (Fig. 2a1). The first cycle’s track forecast in GTS Only is the closest to the best track, followed by that of the GTS+IR. Note that the first cycle’s track forecast in the GTS+IR is comparable to that of HWRF and OFCL. By 1800 UTC 27 August, continued cycling of data assimilation has reduced track errors of both GTS Only and GTS+IR (Fig. 2a2). At this time, the GTS+IR track is comparable to the GTS Only experiment and slightly better than both HWRF and OFCL.
Deterministic forecasts of (a) track and (b) intensity initialized from the EnKF analysis means of the (left) first (1200 UTC 27 Aug) and (right) seventh (1800 UTC 27 Aug) cycles of the experiments that did not assimilate TDR. Large circles on track plots denote the first hour of each day. For comparison, best track values from the NHC HURDAT2 database, as well as the operational HWRF and NHC official forecasts, are shown.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
In terms of intensity, the GTS+IR experiment produces higher maximum wind speeds than the GTS Only experiment after only one cycle (Fig. 2b1). However, both experiments, as well as HWRF and OFCL, fail to capture the rapid intensification of Hurricane Dorian with forecasts starting at 1200 UTC 27 August. Continued cycling of observations leads to dramatically improved intensity forecasts starting at 1800 UTC 27 August (Fig. 2b2). Although the GTS+IR intensity forecast initialized at this time does not predict the category 5 status Dorian would become, this forecast closely mirrors the rapid intensification from 1800 UTC 27 August to 1800 UTC 30 August seen in the best track values. In contrast, the GTS Only experiment, as well as HWRF, exhibit a delayed intensification relative to the GTS+IR experiment, while OFCL misses the intensification altogether.
To understand why the GTS+IR experiment produced an improved intensity forecast starting at 1800 UTC 27 August, we turn our attention to the analyzed and forecasted wind fields and cloud structures. Figure 3 illustrates that the initial primary circulation of the GTS+IR experiment (Fig. 3b1) is stronger, broader, and deeper than the GTS Only experiment (Fig. 3a1). As this stronger circulation is integrated in time, it intensifies faster than the GTS Only circulation (Figs. 3a2,a3 compared to Figs. 3b2,b3). Additionally, the radius of maximum wind (RMW) of the GTS+IR experiment shows a faster contraction, especially below 4 km, during the first 48 h of the forecast. This contraction is noticeable in the GTS Only experiment but is less pronounced.
Azimuthally averaged tangential velocities of the EnKF analysis mean at the time of the seventh cycle (1800 UTC 27 Aug) for the (a1) GTS Only and (b1) GTS+IR experiments, as well as the (a2),(b2) 24-h and (a3),(b3) 48-h deterministic forecasts initialized from those analyses. The radius of maximum wind is denoted by the black dotted line.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
The GTS+IR experiment not only produces a stronger primary circulation, but also a stronger secondary circulation. This can be seen in Fig. 4. At analysis time, the GTS+IR experiment displays a much stronger outflow at upper levels. Low-level inflow in the GTS+IR analysis is also slightly stronger than the GTS Only analysis in the region centered around 50 km from the TC center. When forecasts are initialized from these analyses, the GTS+IR forecast of low-level inflow grows relative to the GTS Only forecast. This low-level inflow advects angular momentum from the outer regions to the TC inner-core region, which helps to spin up the vortex. Consequently, the GTS+IR experiment better captures the rapid intensification of the storm than the GTS Only experiment.
As in Fig. 3, but for azimuthally averaged radial velocities. Black contours show azimuthally averaged vertical velocities (cm s−1), with negative values dashed. Note that a Gaussian smoother was applied to the analysis vertical velocities with a smoothing length scale of 6 km in the horizontal and 0.5 km in the vertical. The Gaussian smoothing length scale for forecasted vertical velocities was 9 km in the horizontal and 0.75 km in the vertical.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
In addition to stronger primary and secondary circulations, the GTS+IR experiment develops more convective activity than the GTS Only experiment (Fig. 5). At analysis time, the azimuthally averaged reflectivities of each experiment are fairly similar; however, the GTS+IR analysis has a rainband that extends from 125 to 200 km from the TC center. From the analysis to the 24-h forecast, there are large drop-offs in reflectivity in both experiments, which we explain shortly. The key feature of Fig. 5 is the difference between experiments from the 24- to 48-h forecasts. As the analyses are integrated forward in the forecasts, the GTS+IR forecast develops areas of higher reflectivity that expand outward and deepen. The enhanced convection in the GTS+IR forecast, compared to the GTS Only forecast, is consistent with the stronger vertical motions illustrated in Fig. 4.
As in Fig. 3, but for azimuthally averaged reflectivity.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
The large drops in reflectivity from the analysis to the 24-h forecasts in both experiments is a result of integrating the ensemble mean to make the forecast. Because clouds and precipitation are tied to the vertical velocity field, and thus near-surface convergence, one would expect the evolution of reflectivity to mirror that of the near-surface convergence field. As it turns out, the near-surface convergence fields of the analyses (not shown) are quite weak with very isolated small-scale maxima and minima, the result of averaging dislocated near-surface convergence fields from ensemble member to ensemble member. Consequently, when the ensemble means are integrated forward to make forecasts, the clouds that exist in the analyses quickly die off. The forecasts then gradually develop clouds and precipitation over time as they reconstruct the small-scale convergence patterns. Notice that the forecast reconstructs the convection sooner in the GTS+IR experiment than in the GTS Only experiment. This is because the small-scale features of the convergence fields in the individual GTS+IR ensemble members are stronger and less dislocated than in the GTS Only experiment, by virtue of the assimilation of all-sky IR BTs.
To see the impact of assimilating IR BTs on the simulated cloud structures, we compare simulated channel-10 BTs to those that were observed. GOES-16 ABI channel 10 (7.3-μm wavelength), which is sensitive to lower-tropospheric water vapor (Schmit et al. 2005), is treated as an independent observation for verification in Fig. 6. At analysis time, the GTS Only experiment (Fig. 6a1) fails to capture the structure of the clouds. The GTS+IR experiment (Fig. 6b1) does display many of the small-scale convective features seen in the observations (Fig. 6c1). Integrating these analyses forward in time, the GTS+IR cloud structures evolve more similarly to the observations than those of the GTS Only experiment. After 54 h of integration, the GTS+IR forecast better captures the large area of deep inner-core convection, as well as the primary spiral rainband, than the GTS Only forecast (Figs. 6a4,b4,c4). The trouble the GTS Only experiment has in producing and sustaining clouds is quantified by its inner-core warm bias (Figs. 6a2–a4). This is not as problematic in the GTS+IR experiment, which has an inner-core bias that is cold to slightly warm (Figs. 6b2–b4). Ultimately, the GTS+IR experiment better captures the cloud structures than the GTS Only experiment.
Comparison of the simulated channel-10 BTs of the EnKF analysis mean at the time of the seventh cycle (1800 UTC 27 Aug) for the (a1) GTS Only and (b1) GTS+IR experiments to (c1) observations. Comparisons of the (a2),(b2) 24-h; (a3),(b3) 48-h; and (a4),(b4) 54-h deterministic forecasts initialized from those analyses are also made to (c2)–(c4) observations. Bias values reported in rows (a) and (b) were calculated for the simulated BTs within 300 km of the TC center, with negative values indicating simulated BTs that are lower than observed.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
To further assess the intensity and structure of simulated surface wind fields, we compare the simulated surface winds to those retrieved by NOAA/HRD’s Stepped Frequency Microwave Radiometer (SFMR). This instrument, which is flown aboard the P-3 aircraft, has a downward-pointing antenna that passively measures microwave radiation emitted from the ocean surface. Together with retrievals of sea surface temperature, its measurements support retrieval of surface wind speeds accurate to within 1 m s−1 at speeds above 15 m s−1 (Uhlhorn and Black 2003). A comparison of these retrieved surface winds to the surface winds of the deterministic forecasts initialized at 1800 UTC 27 August is shown in Fig. 7a for various lead times. These rather unconventional times were chosen because they correspond to times when the observations within the inner core were available. Additionally, the analysis from which these forecasts were initialized is not shown since SFMR retrievals at that time are only available on the outer fringes of the storm. Figure 7a shows that the GTS+IR forecast better captures the intensification of Dorian than the GTS Only experiment. As the GTS+IR forecast is integrated, it develops a classic Rankine vortex-like structure of a developed TC, with the eyewall maxima more closely matching the intensity of the observations than the GTS Only forecast. Although the positions of the eyewall maxima and minimum deviated a bit from those of the observations, the inner-core wind structure of the GTS+IR forecast is noticeably more accurate than that of the GTS Only forecast.
Observed and simulated (top) surface and (bottom) 3-km wind speeds along the line segments within each inset. The (a1),(b1) 30-h; (a2),(b2) 54-h; and (a3),(b3) 78-h deterministic forecasts were initialized from the 1800 UTC 27 Aug EnKF analysis mean of the GTS Only and GTS+IR experiments. Observed surface winds were retrieved by the SFMR instrument aboard NOAA aircraft. Observed 3-km wind speeds were measured by flight-level probes aboard NOAA aircraft.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
Not only does the GTS+IR forecast produce a more accurate inner-core surface wind structure, but it also better captures the inner-core wind structure in the lower troposphere. This can be seen in Fig. 7b, which compares the simulated 3-km inner-core wind speeds to those observed by probes aboard NOAA P-3 aircraft. It is important to stress that neither these observations, nor the SFMR-retrieved observations, were assimilated in any experiment. Although the eyewall maxima of the GTS+IR forecast are often too far apart relative to the observations, it is apparent that the GTS+IR forecast better captures the intensity of the eyewall maxima than the GTS Only forecast.
In summary, relative to an experiment that does not assimilate all-sky IR BTs, hourly assimilation of all-sky IR BTs for 6 h results in an analysis that has stronger primary and secondary circulations, enhanced convection, and better-defined cloud structures. This translates to a deterministic forecast that better captures the rapid intensification and inner-core winds of Hurricane Dorian. These results are consistent with the recent literature on other TCs (Zhang et al. 2016; Honda et al. 2018; Minamide and Zhang 2018; F. Zhang et al. 2019).
b. Combined impacts of TDR radial velocity and IR BT assimilation
Now we present the impacts of simultaneously assimilating TDR radial velocities and all-sky IR BTs. As in section 3a, we first present the results of deterministic forecasts and then elaborate on the physical reasons for differences among them.
1) Deterministic forecasts initialized after first TDR cycle
Figure 8 shows 5 of the 12 deterministic forecast sets that were initialized during or after the first TDR cycle (22 UTC 27 August). Inspecting the track forecasts (Figs. 8a2,a3), it is clear that track errors of the operational HWRF and NHC OFCL forecasts, as well as all experiments, diverge from the best track values in the last two days of the 5-day forecasts. Aside from that, both the HWRF and OFCL forecasts initialized at 0000 and 1200 UTC 28 August overestimate the forward translational speed of Dorian, resulting in an erroneously forecasted Florida landfall (Figs. 8a2,a3). In contrast, forecasts initialized at 0000 UTC 28 August from this study’s experiments better capture the forward translational speed (Fig. 8a2), with hints of the northward curve beginning to appear at the end of forecast day 5. All in all, both the 0000 and 1200 UTC 28 August track forecasts initialized from our experiments better match the best track over the first three days of the forecasts than the operational guidance.
As in Fig. 2, but for deterministic forecasts initialized during and after the first phase of TDR cycling.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
When we compare forecasts of the maximum surface wind speed to operational guidance, we see that at 0000 (Fig. 8b2) and 1200 UTC (Fig. 8b3) 28 August OFCL fails to predict the rapid intensification of Dorian, whereas HWRF at those times predicts a delayed rapid intensification. At these forecast times our experiments better capture the rapid intensification of Dorian than operational guidance. Consequently, they are able to predict peak intensities that are closer to the best track values (Figs. 8b2,b3).
As a side note, our forecasts initialized at 2100 (Fig. 8b4) and 2300 UTC (Fig. 8b5) 28 August best capture the best track peak intensity. At both of these times, all experiments, except the GTS Only, predict Dorian to reach category 5 intensity.
At this point we turn our attention to the average track and intensity errors of the deterministic forecasts. The forecasts initialized at 1200 and 1800 UTC 27 August (Fig. 2) are excluded from this analysis because only two (GTS Only and GTS+IR) of the five experiments produced forecasts at those times due to TDR observations not yet being available. For the remaining 5-day deterministic forecasts, the mean absolute errors (MAEs) of track and maximum surface wind speed relative to the HURDAT2 best track values were calculated as a function of lead time (Fig. 9). From a lead time of 24 h and greater, average track errors of HWRF are always higher than for our five experiments, whereas average track errors of OFCL are comparable to our GTS Only experiment (Fig. 9a2). In terms of intensity, HWRF average errors are comparable to or higher than our GTS Only experiment whereas OFCL has a higher average intensity error than all our experiments for lead times over 40 h (Fig. 9b2). These results show that our forecasts of track and intensity outperform the operational HWRF and OFCL forecasts at lead times of 2 days or more.
Mean absolute errors (MAEs) relative to the HURDAT2 best track values of (a1),(a2) track and (b1),(b2) maximum surface wind speed from deterministic forecasts as a function of lead time. Forecasts initialized prior to 2200 UTC 27 Aug were excluded from the calculations in order to make a fair comparison among experiments. For column 1 the averages were taken over all 12 forecasts denoted by a black “X” in Fig. 1 that were initialized at 2200 UTC 27 Aug or later. For column 2 the averages were taken over the five forecasts denoted by a black “X” in Fig. 1 that were initialized when HWRF and OFCL produced forecasts.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
When we compare the average track errors of our experiments to one another (Fig. 9a1), we find that the GTS+TDR experiment outperforms all other experiments from a lead time of 40 h and longer. More specifically, the average MAE of track for the GTS+TDR experiment for lead times greater than 48 h is 25% lower than that of the GTS+IR+TDR experiment. Furthermore, the Student’s t test (Gosset 1908) indicates at greater than 95% confidence that the GTS+TDR experiment produces a lower average track error than any other forecast at lead times of 48 to 72 h.
Despite the improved track forecasts, the GTS+TDR experiment did not have the lowest average intensity errors at any lead time (Fig. 9b1). Average intensity errors for 48 h and longer lead times were consistently lower for experiments that assimilated IR observations than for experiments that did not. At lead times of 48 to 72 h, the GTS+IR+TDR experiment had the lowest average intensity errors. The improvement of this experiment at those times over the GTS Only and GTS+TDR experiments is statistically significant with confidence values always greater than 92%. When comparing the GTS+IR+TDR experiment to the other experiments that assimilated IR observations, we found the average intensity errors at lead times of 48–72 h to be 37% lower than the GTS+IR and 27% lower than the GTS+TDR+IR experiment. These improvements in the intensity forecast at 48–72-h lead times are statistically significant at confidence values greater than 90% for many of those hours. In summary, average intensity errors of deterministic forecasts initialized from experiments that assimilated IR observations were lowest for all lead times of 48 h and longer, with the greatest improvements coming from the GTS+IR+TDR experiment for lead times from 48 to 72 h. These results demonstrate that assimilating TDR observations along with IR observations does not degrade the intensity forecasts but rather has the potential to improve them.
The order in which observations are assimilated within a cycle has an impact on the forecast. In this study, both the average track and intensity errors are sensitive to that order. We discuss the impacts of simultaneously assimilating TDR and IR observations on the structure of the analyzed and forecasted wind fields in section 3b(2).
2) Impacts on analyzed and forecasted wind fields
In this subsection, we present the impacts of assimilating TDR and IR observations on the analyzed and forecasted wind fields to shed some light on the physical impacts of their combined assimilation on intensity forecasts.
We begin by discussing cross sections of the azimuthally averaged tangential (Fig. 10) and radial (Fig. 11) winds for all five experiments at the time of the first TDR cycle. Note that the GTS+TDR experiment was initialized from the background of the GTS Only experiment whereas the GTS+TDR+IR and GTS+IR+TDR experiments were initialized from the background of the GTS+IR experiment. Additionally, there was a total of 247 conventional GTS observations assimilated within domain 3 during this cycle—one HPI, six METAR (T, U, V at 2 locations), and 240 soundings (T, U, V, Q at 10 levels at six locations). Comparing Figs. 10a1 and 10b1 reveals that the primary circulation in the background of the GTS+IR experiment is substantially stronger than that of the GTS Only experiment. Furthermore, the RMW is smaller in the GTS+IR experiment. These primary circulation and RMW differences are consistent with what we showed for the 1800 UTC 27 August cycle (Fig. 3).
Azimuthally averaged tangential velocities for the first TDR cycle (2200 UTC 27 Aug) of the EnKF (left) background mean, (center) analysis mean, and (right) increment of the (a) GTS Only, (b) GTS+IR, (c) GTS+TDR, (d) GTS+TDR+IR, and (e) GTS+IR+TDR experiments. The radius of maximum wind is denoted by the black dotted line.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
As in Fig. 10, but for azimuthally averaged radial velocities. Black contours show azimuthally averaged vertical velocities (cm s−1). Note that a Gaussian smoother was applied to the vertical velocity with a smoothing length scale of 6 km in the horizontal and 0.5 km in the vertical.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
Inspecting the primary and secondary circulations of the GTS+IR experiment at this cycle, we find that the assimilation of conventional GTS and IR observations results in little to no adjustment of the primary circulation (Fig. 10b3); however, the secondary circulation is impacted (Fig. 11b3). The low-level inflow is increased and the upper-level outflow is decreased within the inner core (Fig. 11b3). These substantial adjustments to the secondary circulation are likely because cloud patterns are dynamically correlated with patterns of convergence and divergence.
When we assimilate both conventional GTS and TDR observations at the first TDR cycle, the increments (Fig. 10c3) show a strengthening of the primary circulation like that of the GTS Only experiment but this time extending to the mid- and upper levels of the storm. This is not surprising, considering that the TDR observations within the inner core of the storm extended to altitudes as high as 10 km. Assimilation of conventional GTS and TDR observations also strengthens the secondary circulation (Fig. 11c3). This strengthening of the secondary circulation likely helps to spin up the vortex in the GTS+TDR experiment. These results show that assimilation of TDR radial velocities adjusts both the primary and secondary circulations of the TC.
Figures 10d3 and 10e3 show that both the GTS+TDR+IR and GTS+IR+TDR experiments have strong negative increments in the tangential velocity centered around 50-km radius. In the region within 25-km radius, the increments are small or slightly positive in the lower troposphere (Figs. 10d3,e3). This pattern indicates that the addition of TDR observations to the already strong GTS+IR background weakens the vortex in the middle and upper troposphere around 50-km radius. At the same time, TDR observations also tighten the vortex in the lower troposphere, which can be seen in the RMW migrating to approximately 25 km in the lowest 2 km (Figs. 10d2,e2). Note that the increments in the GTS+TDR+IR experiment are not exactly the same as for the GTS+IR+TDR experiment. This shows that the order of assimilation with a serial EnKF makes a difference in the primary circulation. While both experiments act to tighten the lower TC vortex, they do so in different ways. For instance, the GTS+TDR+IR experiment produces much stronger negative increments centered on 50-km radius and small positive increments within 25-km radius, whereas the GTS+IR+TDR produces weaker negative increments centered on 50-km radius but produces slightly stronger positive increments within 25-km radius. These differences in the increments explain the differences in the analyzed primary circulations of the GTS+TDR+IR and GTS+IR+TDR experiments (Figs. 10d2,e2). In conclusion, simultaneously assimilating TDR and IR observations results in tightening of the primary circulation in the lower troposphere and weakening of the primary circulation in the middle and upper troposphere relative to the GTS+IR experiment.
Next, we investigate the impacts of simultaneous assimilation of TDR and IR observations on the secondary circulation of Dorian (Figs. 11d,e). Earlier, we showed that the assimilation of IR BTs impacts the secondary circulation in the inner core in such a way as to increase low-level inflow and decrease upper-level outflow (Fig. 11b3). That pattern is also seen in the GTS+TDR+IR and GTS+IR+TDR experiments; however, the pattern is weakened at upper levels and enhanced at lower levels relative to the GTS+IR experiment. This shows that assimilation of TDR radial velocities simultaneously with IR BTs impacts the secondary circulation and does so in a manner that counteracts the impacts of the IR BTs in the upper troposphere. The impacts of the order of assimilation on the secondary circulation are relatively small compared to the impacts on the primary circulation.
To emphasize the previous points, we plot in Fig. 12 the differences between the EnKF analysis means at 2200 UTC 27 August of the azimuthally averaged tangential and radial velocities for the experiments that had the same background at the start of the first TDR cycle. Figure 12a1 shows that the primary circulation of the GTS+TDR analysis is stronger above 4 km and slightly weaker below 4 km than the GTS Only analysis, whereas Fig. 12a2 shows that the secondary circulation of the GTS+TDR analysis is clearly stronger than the GTS Only analysis at all levels.
Experiment differences at 2200 UTC 27 Aug between the EnKF analysis mean (a1)–(d1) azimuthally averaged tangential velocities and (a2)–(d2) azimuthally averaged radial velocities for those experiments that had the same background mean at the start of the first TDR cycle.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
Comparing the differences among the primary circulations of the experiments that assimilated IR (Figs. 12b1–d1), we find once again that simultaneously assimilating TDR and IR observations reduces the overall intensity of the vortex relative to the GTS+IR analysis at 50-km radius. This effect is more dramatic when TDR is assimilated before IR (cf. Fig. 12c1 to Fig. 12b1). The net impact is a primary circulation that is stronger at radii centered around 50 km for the GTS+IR+TDR analysis (Fig. 12d1). When we compare the differences among the secondary circulations of the experiments that assimilated IR (Figs. 12b2–d2), we find that both the GTS+TDR+IR and GTS+IR+TDR analyses at this time have a stronger secondary circulation than the GTS+IR analysis (Figs. 12b2,c2). As a result of assimilating TDR observations, the upper-level outflow strengthened and lifted vertically relative to the GTS+IR analysis whereas the lower-level inflow strengthened within the inner core. This effect is more pronounced when assimilating TDR before IR (Fig. 12d2). Based on Fig. 12d2, we conclude that the GTS+TDR+IR analysis has a stronger upper-level outflow than in the GTS+IR+TDR analysis. Additionally, the low-level inflow within 50-km radius of the TC center is stronger in the GTS+TDR+IR analysis, yet outside of 50-km radius it is stronger in the GTS+IR+TDR analysis. To summarize, simultaneously assimilating TDR and IR observations weakens the TC vortex in the middle to upper troposphere, tightens the TC vortex in the lower troposphere, and strengthens the secondary circulation relative to the GTS+IR analysis, with the net impact being more pronounced if TDR is assimilated before IR.
It is important to note that Dorian was still a sheared tropical storm at the time of this first TDR cycle. Consequently, azimuthal averages of the secondary circulation might not tell the full story of the impacts of TDR assimilation when the average is taken over the entire vortex. Therefore, we investigated differences in the secondary circulations using azimuthal averages for each shear-relative quadrant (Fig. 13). Inspection of these shear-relative differences (note the color bar is different from Fig. 12) reveals that the left of shear quadrants contain the dominant contributions to the vortex-wide azimuthal averages in Fig. 12.
Experiment differences at 2200 UTC 27 Aug between the EnKF analysis mean azimuthally averaged radial velocities for those experiments that had the same background mean at the start of the first TDR cycle for the (a1)–(d1) downshear right (DR); (a2)–(d2) downshear left (DL); (a3)–(d3) upshear left (UL); and (a4)–(d4) upshear right (UR) quadrants.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
The cumulative effects of simultaneously assimilating TDR and IR observations on the low-level vortex are illustrated as Hovmöller diagrams in Fig. 14. Inspection of this figure reveals that the experiments assimilating IR observations produced high (e.g., 16 m s−1) azimuthally averaged 10-m wind speeds sooner than experiments that did not. Although the TDR experiments were slower to develop high 10-m wind speeds, they ended up producing wind speeds that exceeded the GTS+IR experiment during the second TDR phase of cycling (2100 UTC 28 August–0000 UTC 29 August). Further inspection of Fig. 14 reveals that the region of 10-m wind speeds exceeding 16 m s−1 extended noticeably farther from the TC center in the GTS+IR experiment than in the experiments that assimilated both IR and TDR observations. Consequently, these Hovmöller diagrams are consistent with the finding that simultaneously assimilating TDR and IR observations tightened the low-level vortex relative to the GTS+IR experiment.
Temporal evolution of 10-m wind speeds in the EnKF analysis mean for the (a) GTS Only, (b) GTS+IR, (c) GTS+TDR, (d) GTS+TDR+IR, and (e) GTS+IR+TDR experiments.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
As before, we further compare the analyzed and forecasted inner-core wind speeds to the SFMR-retrieved surface wind speeds (Fig. 15) and 3-km observed wind speeds (Fig. 16). Inspection of Fig. 15 reveals that assimilating TDR radial velocities improves the analyzed and forecasted inner-core surface wind structure. At most times, the overly broad vortex of the GTS+IR is narrowed when TDR observations are assimilated. Furthermore, the two experiments that assimilated both IR and TDR observations better capture the intensity and location of the eyewall maxima. These conclusions apply to the 3-km wind speed as well (Fig. 16). Ultimately, assimilating TDR radial velocities improves the analyzed and forecasted inner-core wind speeds, with the greatest benefits coming from experiments that assimilate it simultaneously with IR observations.
Observed surface wind speeds together with simulated surface wind speeds along line segment AB within each inset for the EnKF analysis mean at (a1) 0000 UTC 28 Aug and (b1) 0000 UTC 29 Aug, as well as the (a2),(b2) 48-h and (a3),(b3) 72-h forecasts initialized from them. Observed surface wind speeds were retrieved from the SFMR instrument aboard NOAA aircraft.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
As in Fig. 15, but for 3-km wind speeds. Observed 3-km wind speeds were measured by flight-level probes aboard NOAA aircraft.
Citation: Monthly Weather Review 149, 7; 10.1175/MWR-D-20-0338.1
4. Discussion and conclusions
In this study we have shown the value of simultaneously assimilating TDR radial velocities and GOES-16 all-sky IR BTs on both the analyses and forecasts of Hurricane Dorian. Results of deterministic forecasts show that forecasts initialized from analyses that assimilated TDR radial velocities after conventional GTS observations had 25% lower average track errors at a lead time of 48 h and greater compared to any other experiment. In terms of intensity, forecasts that assimilated all-sky IR BTs had average intensity errors lower than forecasts that did not assimilate all-sky IR BTs at lead times of 48 h and greater. The improvements brought to forecasts when assimilating IR observations were shown to be a result of better analyzed cloud and moisture fields, as well as more intense initial primary and secondary circulations. Simultaneously assimilating TDR radial velocities and all-sky IR BTs further reduced the average intensity errors relative to the GTS+IR experiment by 37% at lead times of 48–72 h. These improvements were shown to be a result of a more realistic inner-core wind structure when TDR and IR observations are assimilated together. Ultimately, simultaneously assimilating TDR radial velocities and all-sky IR BTs has the potential to achieve more accurate track and intensity forecasts than only assimilating all-sky IR BTs.
We have shown that the order in which these observations are assimilated makes a difference in the results. At lead times of 48–72 h, the intensity errors of forecasts that assimilated all-sky IR BTs before TDR radial velocities are, on average, 27% lower than those of forecasts that assimilated all-sky IR BTs after TDR radial velocities. This dependence of the analysis on the order of assimilation when using a serial filter has been demonstrated in other studies as well (Nerger 2015; Kotsuki et al. 2017). The assimilation order can influence the results of sequential methods like EnSRF as a result of localization of the increments coming from each observation [i.e., Kalman gain localization (Kloc)]. Furthermore, Kotsuki et al. (2017) demonstrated that the analysis can be significantly improved in the EnSRF by changing the assimilation order. Determining the optimal order of assimilation when simultaneously assimilating TDR radial velocities and all-sky IR BTs with a serial EnKF is something that deserves more attention in the future, especially considering that serial EnKF methods that employ Kloc are used in operational practice and are common in ensemble DA research. For instance, Environment Canada has been running a Kloc EnKF system (Houtekamer and Mitchell 2001) globally since 2005 (Houtekamer and Mitchell 2005; Jacques et al. 2017). Also, the NCAR experimental real-time convection-allowing ensemble prediction system (Schwartz et al. 2015) is based on a Kloc EnKF method (Anderson 2001). There is also a plethora of ensemble DA research that employs Kloc EnKF (e.g., Romine et al. 2013; Thompson et al. 2015; Zhang et al. 2018, Y. Zhang et al. 2019).
The improvements brought to deterministic forecasts by all-sky IR BT assimilation in this study, as well as other studies, is encouraging. Additionally, assimilating TDR observations, even with their limited spatial (~1/6 that of the IR observations) and temporal (only 9 out of the 37 cycles when IR was assimilated) coverage, brought added improvements to those obtained from all-sky IR BT assimilation. These results suggest that more frequent TDR flights through burgeoning storms will substantially improve hurricane forecasts, even with the availability of dense and frequent IR observations.
Given the expense of TDR flights, it is necessary to test the setup of this experiment on other storms to confirm that the case of Hurricane Dorian is not a statistical anomaly. As such a test, we conducted additional experiments on Hurricane Laura (2020), the only other storm since GOES-16 became operational that was poorly forecasted and for which there were TDR measurements prior to rapid intensification. In the Hurricane Laura experiments, we used the same experimental design as for Hurricane Dorian, but without the GTS Only and GTS+TDR experiments. We initiated the ensemble at 1800 UTC 22 August and spun up until 0600 UTC 23 August. We initialized the GTS+IR experiment from the ensemble at 0600 UTC 23 August and cycled hourly until 1400 UTC 23 August. We initialized the GTS+IR+TDR and GTS+TDR+IR experiments from the background of the GTS+IR experiment at 1000 UTC 23 August and cycled hourly until 1400 UTC 23 August. We initiated three 5-day deterministic forecasts from the EnKF analysis means at 1200, 1300, and 1400 UTC 23 August.
Although the track forecasts had substantial errors, the intensity forecasts of the GTS+TDR+IR and GTS+IR+TDR experiments more accurately predicted the rapid intensification of Laura than the GTS+IR experiment. Furthermore, the GTS+IR+TDR had a mean absolute error of maximum surface wind speed that was slightly smaller than that of the GTS+TDR+IR experiment. These results are consistent with the results of the Hurricane Dorian experiments and provide support for additional TDR measurement campaigns focused on future storms. Future studies should also investigate more deeply the interplay between the adjustment of the TC vortex by the assimilation of all-sky IR BTs and TDR radial velocities in order to reap the maximum benefits from each.
Acknowledgments
This work was supported by ONR Grant N00014-18-1-2517, NGGPS and HFIP through Subcontract 3004628721 with the University of Michigan, and NOAA Grant NA18NWS4680054. Computing was conducted at the Texas Advanced Computing Center (TACC). All conventional GTS observations were obtained from NCAR RDA (datasets 351.0 and 461.0). TDR data were obtained from
REFERENCES
Aberson, S. D., A. Aksoy, K. J. Sellwood, T. Vukicevic, and X. Zhang, 2015: Assimilation of high-resolution tropical cyclone observations with an ensemble Kalman filter using HEDAS: Evaluation of 2008–11 HWRF forecasts. Mon. Wea. Rev., 143, 511–523, https://doi.org/10.1175/MWR-D-14-00138.1.
Anderson, J. L., 2001: An ensemble adjustment Kalman filter for data assimilation. Mon. Wea. Rev., 129, 2884–2903, https://doi.org/10.1175/1520-0493(2001)129<2884:AEAKFF>2.0.CO;2.
Andreas, E. L, L. Mahrt, and D. Vickers, 2015: An improved bulk air–sea surface flux algorithm, including spray-mediated transfer. Quart. J. Roy. Meteor. Soc., 141, 642–654, https://doi.org/10.1002/qj.2424.
Avila, L. A., 2019: Hurricane Dorian Advisory Number 33. National Hurricane Center, accessed 15 September 2020, nhc.noaa.gov.
Avila, L. A., S. R. Stewart, R. Berg, and A. B. Hagen, 2020: Tropical cyclone report: Hurricane Dorian (24 August–7 September 2019). NHC Tech. Rep. AL052019, 74 pp., https://www.nhc.noaa.gov/data/tcr/AL052019_Dorian.pdf.
Bao, J.-W., 2016: Physical processes in tropical cyclone models. Advanced Numerical Modeling and Data Assimilation Techniques for Tropical Cyclone Prediction, U. C. Mohanty and S. G. Gopalakrishnan, Eds., Springer, 107–144.
Barker, D. M., W. Huang, Y.-R. Guo, A. J. Bourgeois, and Q. N. Xiao, 2004: A three-dimensional variational data assimilation system for MM5: Implementation and initial results. Mon. Wea. Rev., 132, 897–914, https://doi.org/10.1175/1520-0493(2004)132<0897:ATVDAS>2.0.CO;2.
Bauer, P., G. Ohring, C. Kummerow, and T. Auligne, 2011: Assimilating satellite observations of clouds and precipitation into NWP models. Bull. Amer. Meteor. Soc., 92, ES25–ES28, https://doi.org/10.1175/2011BAMS3182.1.
Cangialosi, J. P., 2020: National Hurricane Center Forecast Verification Report: 2019 Hurricane Season. NHC Tech. Rep., 75 pp., https://www.nhc.noaa.gov/verification/pdfs/Verification_2019.pdf.
Chan, M.-Y., J. L. Anderson, and X. Chen, 2020a: An efficient bi-Gaussian ensemble Kalman filter for satellite infrared radiance data assimilation. Mon. Wea. Rev., 148, 5087–5104, https://doi.org/10.1175/MWR-D-20-0142.1.
Chan, M.-Y., F. Zhang, X. Chen, and L. R. Leung, 2020b: Potential impacts of assimilating all-sky satellite infrared radiances on convection-permitting analysis and prediction of tropical convection. Mon. Wea. Rev., 148, 3203–3224, https://doi.org/10.1175/MWR-D-19-0343.1.
Chen, X., and F. Zhang, 2019: Development of a convection-permitting air-sea-coupled ensemble data assimilation system for tropical cyclone prediction. J. Adv. Model. Earth Syst., 11, 3474–3496, https://doi.org/10.1029/2019MS001795.
Chen, X., R. G. Nystrom, C. A. Davis, and C. M. Zarzycki, 2021: Dynamical structures of cross-domain forecast error covariance of a simulated tropical cyclone in a convection-permitting coupled atmosphere–ocean model. Mon. Wea. Rev., 149, 41–63, https://doi.org/10.1175/MWR-D-20-0116.1.
Christophersen, H., A. Aksoy, J. Dunion, and K. Sellwood, 2017: The impact of NASA Global Hawk unmanned aircraft dropwindsonde observations on tropical cyclone track, intensity, and structure: Case studies. Mon. Wea. Rev., 145, 1817–1830, https://doi.org/10.1175/MWR-D-16-0332.1.
Emanuel, K., and F. Zhang, 2016: On the predictability and error sources of tropical cyclone intensity forecasts. J. Atmos. Sci., 73, 3739–3747, https://doi.org/10.1175/JAS-D-16-0100.1.
Emanuel, K., and F. Zhang, 2017: The role of inner-core moisture in tropical cyclone predictability and practical forecast skill. J. Atmos. Sci., 74, 2315–2324, https://doi.org/10.1175/JAS-D-17-0008.1.
Eyre, J. R., 2016: Observation bias correction schemes in data assimilation systems: A theoretical study of some of their properties. Quart. J. Roy. Meteor. Soc., 142, 2284–2291, https://doi.org/10.1002/qj.2819.
Gaspari, G., and S. E. Cohn, 1999: Construction of correlation functions in two and three dimensions. Quart. J. Roy. Meteor. Soc., 125, 723–757, https://doi.org/10.1002/qj.49712555417.
Geer, A. J., and P. Bauer, 2011: Observation errors in all-sky data assimilation. Quart. J. Roy. Meteor. Soc., 137, 2024–2037, https://doi.org/10.1002/qj.830.
Geer, A. J., S. Migliorini, and M. Matricardi, 2019: All-sky assimilation of infrared radiances sensitive to mid- and upper-tropospheric moisture and cloud. Atmos. Meas. Tech., 12, 4903–4929, https://doi.org/10.5194/amt-12-4903-2019.
Gosset, W. S., 1908: The probable error of a mean. Biometrika, 6, 1–25, https://doi.org/10.2307/2331554.
Green, B. W., and F. Zhang, 2013: Impacts of air–sea flux parameterizations on the intensity and structure of tropical cyclones. Mon. Wea. Rev., 141, 2308–2324, https://doi.org/10.1175/MWR-D-12-00274.1.
Green, B. W., and F. Zhang, 2014: Sensitivity of tropical cyclone simulations to parametric uncertainties in air–sea fluxes and implications for parameter estimation. Mon. Wea. Rev., 142, 2290–2308, https://doi.org/10.1175/MWR-D-13-00208.1.
Han, Y., P. van Delst, Q. Liu, F. Weng, B. Yan, R. Treadon, and J. Derber, 2006: JCSDA Community Radiative Transfer Model (CRTM)—Version 1. NOAA Tech. Rep. NESDIS 122, 33 pp., https://repository.library.noaa.gov/view/noaa/1157.
Han, Y., F. Weng, Q. Liu, and P. van Delst, 2007: A fast radiative transfer model for SSMIS upper atmosphere sounding channels. J. Geophys. Res., 112, D11121, https://doi.org/10.1029/2006JD008208.
Hendricks, E. A., M. T. Montgomery, and C. A. Davis, 2004: The role of “vortical” hot towers in the formation of Tropical Cyclone Diana (1984). J. Atmos. Sci., 61, 1209–1232, https://doi.org/10.1175/1520-0469(2004)061<1209:TROVHT>2.0.CO;2.
Honda, T., and Coauthors, 2018: Assimilating all-sky Himawari-8 satellite infrared radiances: A case of Typhoon Soudelor (2015). Mon. Wea. Rev., 146, 213–229, https://doi.org/10.1175/MWR-D-16-0357.1.
Hong, S.-Y., and J.-O. J. Lim, 2006: The WRF single-moment 6-class microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129–151.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, https://doi.org/10.1175/MWR3199.1.
Houtekamer, P. L., and H. L. Mitchell, 2001: A sequential ensemble Kalman filter for atmospheric data assimilation. Mon. Wea. Rev., 129, 123–137, https://doi.org/10.1175/1520-0493(2001)129<0123:ASEKFF>2.0.CO;2.
Houtekamer, P. L., and H. L. Mitchell, 2005: Ensemble Kalman filtering. Quart. J. Roy. Meteor. Soc., 131, 3269–3289, https://doi.org/10.1256/qj.05.135.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, https://doi.org/10.1029/2008JD009944.
Jacques, D., W. Chang, S.-J. Baek, T. Milewski, L. Fillion, K.-S. Chung, and H. Ritchie, 2017: Developing a convective-scale EnKF data assimilation system for the Canadian MEOPAR project. Mon. Wea. Rev., 145, 1473–1494, https://doi.org/10.1175/MWR-D-16-0135.1.
Kotsuki, S., S. J. Greybush, and T. Miyoshi, 2017: Can we optimize the assimilation order in the serial ensemble Kalman filter? A study with the Lorenz-96 model. Mon. Wea. Rev., 145, 4977–4995, https://doi.org/10.1175/MWR-D-17-0094.1.
Krishnamurti, T. N., S. Pattnaik, L. Stefanova, T. S. V. V. Kumar, B. P. Mackey, A. J. O’Shay, and R. J. Pasch, 2005: The hurricane intensity issue. Mon. Wea. Rev., 133, 1886–1912, https://doi.org/10.1175/MWR2954.1.
Meng, Z., and F. Zhang, 2008: Tests of an ensemble Kalman filter for mesoscale and regional-scale data assimilation. Part III: Comparison with 3DVAR in a real-data case study. Mon. Wea. Rev., 136, 522–540, https://doi.org/10.1175/2007MWR2106.1.
Minamide, M., and F. Zhang, 2017: Adaptive observation error inflation for assimilating all-sky satellite radiance. Mon. Wea. Rev., 145, 1063–1081, https://doi.org/10.1175/MWR-D-16-0257.1.
Minamide, M., and F. Zhang, 2018: Assimilation of all-sky infrared radiances from Himawari-8 and impacts of moisture and hydrometer initialization on convection-permitting tropical cyclone prediction. Mon. Wea. Rev., 146, 3241–3258, https://doi.org/10.1175/MWR-D-17-0367.1.
Minamide, M., and F. Zhang, 2019: An adaptive background error inflation method for assimilating all-sky radiances. Quart. J. Roy. Meteor. Soc., 145, 805–823, https://doi.org/10.1002/qj.3466.
Montgomery, M. T., M. E. Nicholls, T. A. Cram, and A. B. Saunders, 2006: A vortical hot tower route to tropical cyclogenesis. J. Atmos. Sci., 63, 355–386, https://doi.org/10.1175/JAS3604.1.
Nerger, L., 2015: On serial observation processing in localized ensemble Kalman filters. Mon. Wea. Rev., 143, 1554–1567, https://doi.org/10.1175/MWR-D-14-00182.1.
Nystrom, R. G., and F. Zhang, 2019: Practical uncertainties in the limited predictability of the record-breaking intensification of Hurricane Patricia (2015). Mon. Wea. Rev., 147, 3535–3556, https://doi.org/10.1175/MWR-D-18-0450.1.
Nystrom, R. G., R. Rotunno, C. A. Davis, and F. Zhang, 2020: Consistent impacts of surface enthalpy and drag coefficient uncertainty between an analytical model and simulated tropical cyclone maximum intensity and storm structure. J. Atmos. Sci., 77, 3059–3080, https://doi.org/10.1175/JAS-D-19-0357.1.
Osuri, K. K., U. C. Mohanty, A. Routray, and D. Niyogi, 2015: Improved prediction of Bay of Bengal tropical cyclones through assimilation of Doppler weather radar observations. Mon. Wea. Rev., 143, 4533–4560, https://doi.org/10.1175/MWR-D-13-00381.1.
Otkin, J. A., R. Potthast, and A. S. Lawless, 2018: Nonlinear bias correction for satellite data assimilation using Taylor series polynomials. Mon. Wea. Rev., 146, 263–285, https://doi.org/10.1175/MWR-D-17-0171.1.
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., 73, 4911–4925, https://doi.org/10.1175/JAS-D-16-0017.1.
Rajeswari, J. R., C. V. Srinivas, P. R. Mohan, and B. Venkatraman, 2020: Impact of boundary layer physics on tropical cyclone simulations in the Bay of Bengal using the WRF model. Pure Appl. Geophys., 177, 5523–5550, https://doi.org/10.1007/s00024-020-02572-3.
Romine, G. S., C. S. Schwartz, C. Snyder, J. L. Anderson, and M. L. Weisman, 2013: Model bias in a continuously cycled assimilation system and its influence on convection-permitting forecasts. Mon. Wea. Rev., 141, 1263–1284, https://doi.org/10.1175/MWR-D-12-00112.1.
Rotunno, R., Y. Chen, W. Wang, C. Davis, J. Dudhia, and G. J. Holland, 2009: Large-eddy simulation of an idealized tropical cyclone. Bull. Amer. Meteor. Soc., 90, 1783–1788, https://doi.org/10.1175/2009BAMS2884.1.
Schmit, T. J., M. M. Gunshor, W. P. Menzel, J. J. Gurka, J. Li, and A. S. Bachmeier, 2005: Introducing the next-generation Advanced Baseline Imager on GOES-R. Bull. Amer. Meteor. Soc., 86, 1079–1096, https://doi.org/10.1175/BAMS-86-8-1079.
Schwartz, C. S., G. S. Romine, R. A. Sobash, K. R. Fossell, and M. L. Weisman, 2015: NCAR’s experimental real-time convection-allowing ensemble prediction system. Wea. Forecasting, 30, 1645–1654, https://doi.org/10.1175/WAF-D-15-0103.1.
Shen, F., J. Min, and D. Xu, 2016: Assimilation of radar radial velocity data with the WRF hybrid ETKF–3DVAR system for the prediction of Hurricane Ike (2008). Atmos. Res., 169, 127–138, https://doi.org/10.1016/j.atmosres.2015.09.019.
Shen, F., D. Xu, J. Min, Z. Chu, and X. Li, 2020: Assimilation of radar radial velocity data with the WRF hybrid 4DEnVar system for the prediction of Hurricane Ike (2008). Atmos. Res., 234, 104771, https://doi.org/10.1016/j.atmosres.2019.104771.
Smith, A. B., 2020: U.S. billion-dollar weather and climate disasters, 1980–present (NCEI Accession 0209268). NOAA/National Centers for Environmental Information, accessed 15 September 2020, https://doi.org/10.25921/stkw-7w73.
Thompson, T. E., L. J. Wicker, X. Wang, and C. Potvin, 2015: A comparison between the local ensemble transform Kalman filter and the ensemble square root filter for the assimilation of radar data in convective-scale models. Quart. J. Roy. Meteor. Soc., 141, 1163–1176, https://doi.org/10.1002/qj.2423.
Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117, 1779–1800, https://doi.org/10.1175/1520-0493(1989)117<1779:ACMFSF>2.0.CO;2.
Tong, M., and Coauthors, 2018: Impact of assimilating aircraft reconnaissance observations on tropical cyclone initialization and prediction using operational HWRF and GSI ensemble–variational hybrid data assimilation. Mon. Wea. Rev., 146, 4155–4177, https://doi.org/10.1175/MWR-D-17-0380.1.
Uhlhorn, E. W., and P. G. Black, 2003: Verification of remotely sensed sea surface winds in hurricanes. J. Atmos. Oceanic Technol., 20, 99–116, https://doi.org/10.1175/1520-0426(2003)020<0099:VORSSS>2.0.CO;2.
Weng, F., 2007: Advances in radiative transfer modeling in support of satellite data assimilation. J. Atmos. Sci., 64, 3799–3807, https://doi.org/10.1175/2007JAS2112.1.
Weng, Y., and F. Zhang, 2012: Assimilating airborne Doppler radar observations with an ensemble Kalman filter for convection-permitting hurricane initialization and prediction: Katrina (2005). Mon. Wea. Rev., 140, 841–859, https://doi.org/10.1175/2011MWR3602.1.
Weng, Y., and F. Zhang, 2016: Advances in convection-permitting tropical cyclone analysis and prediction through EnKF assimilation of reconnaissance aircraft observations. J. Meteor. Soc. Japan, 94, 345–358, https://doi.org/10.2151/jmsj.2016-018.
Whitaker, J. S., and T. M. Hamill, 2002: Ensemble data assimilation without perturbed observations. Mon. Wea. Rev., 130, 1913–1924, https://doi.org/10.1175/1520-0493(2002)130<1913:EDAWPO>2.0.CO;2.
Zhang, F., and Y. Weng, 2015: Predicting hurricane intensity and associated hazards: A five-year real-time forecast experiment with assimilation of airborne Doppler radar observations. Bull. Amer. Meteor. Soc., 96, 25–33, https://doi.org/10.1175/BAMS-D-13-00231.1.
Zhang, F., C. Snyder, and J. Sun, 2004: Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter. Mon. Wea. Rev., 132, 1238–1253, https://doi.org/10.1175/1520-0493(2004)132<1238:IOIEAO>2.0.CO;2.
Zhang, F., Y. Weng, J. A. Sippel, Z. Meng, and C. H. Bishop, 2009: Cloud-resolving hurricane initialization and prediction through assimilation of Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 137, 2105–2125, https://doi.org/10.1175/2009MWR2645.1.
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, L15810, https://doi.org/10.1029/2011GL048469.
Zhang, F., M. Minamide, and E. E. Clothiaux, 2016: Potential impacts of assimilating all-sky infrared satellite radiances from GOES-R on convection-permitting analysis and prediction of tropical cyclones. Geophys. Res. Lett., 43, 2954–2963, https://doi.org/10.1002/2016GL068468.
Zhang, F., M. Minamide, R. G. Nystrom, X. Chen, S.-J. Lin, and L. M. Harris, 2019: Improving Harvey forecasts with next-generation weather satellites: Advanced hurricane analysis and prediction with assimilation of GOES-R all-sky radiances. Bull. Amer. Meteor. Soc., 100, 1217–1222, https://doi.org/10.1175/BAMS-D-18-0149.1.
Zhang, J. A., 2018: Evaluating the impact of model physics on HWRF forecasts of tropical cyclone rapid intensification. Developmental Testbed Center, 16 pp., https://dtcenter.org/sites/default/files/visitor-projects/DTC_project_final_report_JunZhang-final.pdf.
Zhang, J. A., R. F. Rogers, and V. Tallapragada, 2017: Impact of parameterized boundary layer structure on tropical cyclone rapid intensification forecasts in HWRF. Mon. Wea. Rev., 145, 1413–1426, https://doi.org/10.1175/MWR-D-16-0129.1.
Zhang, Y., F. Zhang, and D. J. Stensrud, 2018: Assimilating all-sky infrared radiances from GOES-16 ABI using an ensemble Kalman filter for convection-allowing severe thunderstorms prediction. Mon. Wea. Rev., 146, 3363–3381, https://doi.org/10.1175/MWR-D-18-0062.1.
Zhang, Y., D. J. Stensrud, and F. Zhang, 2019: Simultaneous assimilation of radar and all-sky satellite infrared radiance observations for convection-allowing ensemble analysis and prediction of severe thunderstorms. Mon. Wea. Rev., 147, 4389–4409, https://doi.org/10.1175/MWR-D-19-0163.1.
Zhu, L., and Coauthors, 2015: Prediction and predictability of high-impact western Pacific landfalling Tropical Cyclone Vicente (2012) through convection-permitting ensemble assimilation of Doppler radar velocity. Mon. Wea. Rev., 144, 21–43, https://doi.org/10.1175/MWR-D-14-00403.1.
