Impact of CAMEX-4 Datasets for Hurricane Forecasts Using a Global Model

JANUARY 2006 KAMINENI ET AL. 151

RUPA KAMINENI,* T. N. KRISHNAMURTI, AND S. PATTNAIK Department of Meteorology, The Florida State University, Tallahassee, Florida

EDWARD V. BROWELL,SYED ISMAIL, AND RICHARD A. FERRARE NASA Langley Research Center, Hampton, Virginia

(Manuscript received 24 October 2003, in final form 7 February 2005)

ABSTRACT

This study explores the impact on hurricane data assimilation and forecasts from the use of dropsondes and remotely sensed moisture profiles from the airborne Lidar Atmospheric Sensing Experiment (LASE) system. It is shown here that the use of these additional datasets, more than those from the conventional world weather watch, has a positive impact on hurricane predictions. The forecast tracks and intensity from the experiments show a marked improvement compared to the control experiment in which such datasets were excluded. A study of the moisture budget in these hurricanes showed enhanced evaporation and precipitation over the storm area. This resulted in these datasets making a large impact on the estimate of mass convergence and moisture fluxes, which were much smaller in the control runs. Overall this study points to the importance of high vertical resolution humidity datasets for improved model results. It is noted that the forecast impact from the moisture-profiling datasets for some of the storms is even larger than the impact from the use of dropwindsonde-based winds.

1. Introduction of measurements with high temporal and horizontal resolutions. A dominant factor for the improvement of weather Numerical weather prediction requires an accurate forecasts is the quality and space–time distribution of representation of the initial state of the atmosphere. An atmospheric measurements. The spatial distribution of error in the analysis of the initial state can amplify in situ atmospheric measurements is highly inhomoge- through the modeling process and result in drastic er- neous, largely being confined to Northern Hemisphere rors in the predicted state. This is especially true in the continents, shipping routes at the surface, and a limited case of hurricanes and tropical storms. If the initial vor- set of commercial airways confined to a rather narrow tex is not well defined because of lack of data, models range of altitudes and routes between major cities. may fail to predict the track and intensity of the storm. Data-sparse regions include much of the Tropics, the Many field experiments [viz., Stormscale Operational Southern Hemisphere, the Arctic, and much of the oce- and Research Meteorology-Fronts Experimental Sys- anic regions, at least between the surface and commer- tems Test (STORM-FEST), Fronts and Atlantic Storm- cial flight altitudes. The recent inclusion of space- and Track Experiment (FASTEX), Winter Storms Recon- ground-based remote sensing data compensates to naissance (WSR)] have been conducted and forthcom- some degree for the lack of in situ measurements. Sat- ing experiments [viz., The Hemispheric Observing ellite observations now cover the entire globe, and are System Research and Predictability Experiment relatively frequent in time. Space-based remote sensing (THORPEX)] are to be carried out to enhance the has the advantage of yielding relatively good coverage database over oceanic regions. The National Oceanic and Atmospheric Administration’s (NOAA) Hurri- * Deceased. cane Research Division (HRD) and its predecessors recognized that oceanic field experiments were neces- Corresponding author address: Edward V. Browell, NASA sary back in the 1970s, and they have been conducting Langley Research Center, Hampton, VA 23681. annual field experiments in the Atlantic and Pacific E-mail: [email protected] since even before that time. The STORM-FEST field

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 152 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63 experiment was conducted in the central United States Special reconnaissance aircraft operated by NOAA between 1 February and 15 March 1992, and additional and by the U.S. Air Force continue to provide valuable soundings were taken along the west coast on several information for the prediction of hurricanes and severe occasions to assess their effect on forecasts downstream winter storms. Quantitative assessments of hurricane over the STORM-FEST area. This experiment was fol- forecast improvements from including dropwindsonde lowed by FASTEX (Thorpe and Shapiro 1995), which data from reconnaissance aircraft may be found in involved major operational forecast centers in an effort Burpee et al. (1996). Heming (1990) discussed the im- to improve operational numerical weather prediction. pact of surface and radiosonde observations from two Many different platforms were used to collect special Atlantic ships on a storm forecast. Smith and Benjamin datasets for testing various proposed objective and sub- (1993) demonstrated the benefits of wind data assimi- jective targeting strategies and for collecting data nec- lation on short-range forecasts in the eastern half of the essary for evaluating forecasts over the open ocean. United States. Recently Qu and Heming (2002) dis- Later it was noticed that hurricane field experiments cussed the impact of dropsonde data on forecasts of were necessary over other oceanic regions adjacent to Hurricane Debby by the UKMO unified model. the United States; these regions are located off the west The Convection and Moisture Experiment coast, the Gulf of Mexico, and the Caribbean Sea (CAMEX) consists of a series of field experiments (Burpee et al. 1984; Simpson 2003). sponsored by the National Aeronautics and Space Ad- Recent sensitivity experiments on the observing net- ministration (NASA). The first two CAMEX field work in the context of modern data assimilation meth- studies were conducted from Wallops Island, Virginia, ods have indicated that the impact of data strongly de- during 1993 and 1995, and a third field study (CAMEX- pends on the location of the data relative to dynami- 3), which was specifically designed for hurricane stud- cally sensitive parts of the flow. These sensitivity ies, was conducted from Patrick Air Force Base near experiments identify an important component to a suc- Cocoa Beach, Florida, between 3 August and 28 Sep- cessful sampling strategy, that is, the identification of tember 1998. Recently, a fourth field campaign in the areas in which initial condition errors are likely to am- CAMEX series (CAMEX-4) was conducted from 16 plify rapidly (owing to some dynamical instability). Ar- August to 24 September 2001 from a base at the Jack- eas in which the initial condition errors are likely to be sonville Naval Air Station, Florida. This field experi- both large and rapidly growing would be preferred ar- ment was conducted in collaboration with NOAA’s eas to obtain supplementary observations. However, HRD and the U.S. Weather Research Program the recent work by Aberson (2002, 2003) on hurricane (USWRP). CAMEX-4 was focused on the study of targeting revealed that the way the targets are sampled tropical cyclone (hurricane) development, tracking, in- is the most important issue to be addressed. Studies by tensification, and landfalling impacts using NASA- Graham (1990), Heming (1990), Pailleux (1990), Ra- funded aircraft and surface remote sensing instrumen- bier et al. (1996), Pu et al. (1997), and Lord (1996) tation. One of the main objectives of CAMEX is fo- contain good examples of definite positive impacts due cused on studies of atmospheric water vapor and to very few additional vertical profiles. precipitation processes using an array of airborne, Sensitivity tests at the U.K. Met Office (UKMO; space- and land-based remote sensors. Our previous Graham and Anderson 1995) and the National Centers studies showed promising impacts on hurricane fore- for Environmental Prediction (NCEP; Lord 1993) have casts using several CAMEX campaign datasets (Bens- demonstrated that improved wind observations have man 2000; Rizvi et al. 2002). These studies provided the the greatest positive effect on one- to three-day fore- opportunity to address some of the model failures in casts. Also, accurate dropsonde wind measurements in hurricane predictions. the vicinity of western Atlantic and Caribbean hurri- In the present paper we have made an attempt to canes have markedly improved landfall predictions analyze the available CAMEX-4 datasets using a three- (Franklin and DeMaria 1992). Certain specific forecast dimensional variational analysis scheme (3DVAR). problems have been linked to lack of observations in Medium-range hurricane predictions were carried out most of the regions. One example is hurricane track and by integrating the Numerical Weather Prediction intensity forecasting, which is one of the most impor- (NWP) model out to five days with these new initial tant forecast problems. Franklin and DeMaria (1992) conditions. This study was done using the Florida State and Burpee et al. (1996) showed that dropwindsonde University (FSU) global spectral model. This paper is data collected by research aircraft demonstrably led to organized as follows. Section 2 describes the improved track forecasts, which shows the importance CAMEX-4 datasets and the experimental plan using of direct measurements in otherwise data-sparse regions. these new datasets. Details of the analysis scheme and

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 153 forecast model are provided in section 3. The numerical sure vertical profiles of pressure, temperature, humidity results are discussed in section 4. Concluding remarks and wind during descent to the surface. These data are provided in section 5. were measured every 0.5 s (2 Hz) during the descent. The Airborne Vertical Atmospheric Profiling System (AVAPS) was used for monitoring the dropsondes 2. CAMEX-4 datasets from the operational and research aircraft. Additional During CAMEX-3 and CAMEX-4, numerous obser- information on the GPS dropwindsondes can be found vations were collected at various stages of hurricane in Hock and Franklin (1999). During this campaign, up development. During the CAMEX-3 campaign, the ob- to 150 dropsondes were deployed from both the NASA servations were targeted using an adaptive observa- DC-8 and ER-2, and nearly 70 of those were dropped tional strategy based on a technique recently developed on the last three days of the experiment. by Zhang and Krishnamurti (1999). NASA’s ER-2 and Figure 1 shows the sample dropsonde data coverage DC-8, NOAA’s Gulfstream-IV (G-IV) and WP-3D, at various levels from multiple aircraft on 23 September and the U.S. Air Force’s WC-130 weather reconnais- 2001 in conjunction with study of Hurricane Humberto. sance aircraft were available during this campaign. The bottom panel of Fig. 1 shows the coverage at lower levels (850 hPa); the middle panel shows the coverage at middle levels (500 hPa); and the top panel shows the a. Dropsonde upper-level (200 hPa) coverage. On this day, the G-IV One of the NASA aircraft (DC-8) provided a high aircraft covered the large-scale environment of the hur- altitude research platform for a variety of remote sen- ricane and the other aircraft covered the eye and inner sors, dropsondes and microphysical measurements in region. The upper-level data were provided by the DC- the inner core of the storm, while another NASA air- 8, ER-2, and G-IV. The low level cyclonic and the up- craft, the ER-2, provided in situ data over the lower per level anticyclonic patterns can be clearly seen in the stratosphere, and remotely sensed measurements wind data. This intensive data coverage was a highlight through the troposphere. The flight altitude of the of the CAMEX experiments. DC-8 was generally from 30 000 to 40 000 ft and the ER-2 was at nearly 65 000 ft. Two NOAA P-3s were b. Lidar Atmospheric Sensing Experiment used for both hurricane research and reconnaissance. The P-3s penetrated the eyewall repeatedly at altitudes The Lidar Atmospheric Sensing Experiment (LASE) from 1500 to 20 000 ft. The NOAA G-IV was used to instrument was developed at the NASA Langley Re- derive a detailed picture of the upper atmosphere sur- search Center as the first autonomously operating Dif- rounding the hurricanes as well as to provide dropwind- ferential Absorption Lidar (DIAL) system, and it was sonde measurements. It provided measurements of the initially flown on the NASA ER-2 aircraft for profiling upper-level hurricane steering winds that are important water vapor, aerosols and clouds through the entire for determining the track of a hurricane. The G-IV flew troposphere (Browell and Ismail 1995; Browell et al. between 350 and 1500 km from the eye of the hurricane 1997; Moore et al. 1997). The accuracy of the LASE and at altitudes up to 45 000 ft. The U.S. Air Force’s water vapor profile measurements was first demon- WC-130 reported hurricane conditions through a com- strated in the 1995 LASE validation experiment where bination of visual observations and onboard in situ in- comparisons of the LASE water vapor measurements struments. Our data assimilation includes these with in situ water vapor measurements on two addi- datasets at about 3-min intervals. The assignment of tional aircraft (C-130 and Lear Jet), radiosonde pro- error for this analysis assumes that these measurements files, and profiles from a ground-based Raman lidar Ϫ are of the same quality as datasets from commercial showed agreement to be better than 6% or 0.01 g kg 1, aircraft. Further details of these instruments can be whichever is greater throughout the troposphere found on the CAMEX-4 Web site (http://camex.msfc. (Browell et al. 1997). Subsequently LASE was flown on nasa.gov). the ER-2 during the Tropospheric Aerosol Radiative This mission was designed to collect, process, and Forcing Observation Experiment (Ismail et al. 2000; transmit analyses of vertical atmospheric soundings in Ferrare et al. 2000a,b) and in seven other major field the environment of a hurricane. Dropsondes provide experiments onboard the NASA P-3 and DC-8 aircraft. invaluable information about winds in the midlevel hur- Intercomparisons performed during the CAMEX-3 ricane vortex and also the vertical temperature distri- (Ferrare et al. 1999; Browell et al. 2000), Pacific Ex- bution within the eye. The dropsonde system uses glob- ploratory Mission (PEM) Tropics B (Browell et al. al positioning system (GPS) dropwindsondes to mea- 2001) and CAMEX-4 (Kooi et al. 2002) missions pro-

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 154 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

FIG. 1. Sample coverage of dropsonde data from multiple aircraft for CAMEX-4 on 23 Sep 2001, Hurricane Humberto, at (a) upper level (300–200 hPa), (b) middle level (400–600 hPa), and (c) lower level (900–750 hPa).

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 155 duced results consistent with the initial LASE valida- CTRL: Control analysis at T126L14 (triangularly tion experiments. truncated at 126 wave in horizontal and 14 sigma The impact of including LASE data in the hurricane levels in vertical) by using ECMWF analysis at 0.5° analysis compared to the European Centre for Me- resolution ϩ FSU physical initialization. dium-Range Weather Forecasts (ECMWF) analysis CTRL1: CTRL plus all the available wind data dur- without LASE data was recently described by Ka- ing CAMEX-4. It served as background field for mineni et al. (2003). The result of combining the LASE the following experiments. and dropsonde data in hurricane analysis is described in EXP1: CTRL1 plus dropsonde moisture datasets this study. EXP2: CTRL1 plus LASE datasets EXP3: CTRL1 plus dropsonde moisture plus LASE c. Design of experiments datasets. This includes all of the available datasets. A special dataset on 0.5° latitude by 0.5° longitude at 28 vertical levels without inclusion of dropwindsonde 3. Data assimilation data has been provided to us by ECMWF. We have used these data in our analysis. This analysis also in- A typical data assimilation system is composed of a cluded operational global sets from the stream of the forecast model that provides a background, first guess world weather watch, cloud track winds, commercial of the atmosphere by extrapolating information for- aircraft winds reports, surface datasets from marine ward from previous observations, and an analysis that ships of opportunity, oceanic buoy surface reports, combines information from the background and from ocean surface winds from satellite-based scatterom- the entire set of widely scattered observations into a eters, sounding of temperature and humidity from po- model-compatible form for extrapolation to the next lar orbiting satellites. The microwave radiances pro- analysis time. Since data assimilation has observational, vided by Tropical Rainfall Measuring Mission (TRMM) analysis and the modeling components, improvements and Defense Meteorological Satellite Program (DMSP) in one or more of these components are expected to satellites (F13, F14, and F15) provide rain-rate esti- improve forecasts. mates via NASA Goddard algorithms (Olson et al. 1990). The FSU model at a T126 resolution makes use a. Analysis scheme of these data to perform a physical initialization (Krish- Analysis is a process by which various meteorological namurti et al. 1991). The T126 resolution of the FSU observations distributed in space and time from differ- model was noted to be adequate for exploring the ex- ent observing systems are integrated with other infor- tent to which the larger-scale hurricane was resolved by mation, like short-range weather predictions (6 h) from the aircraft datasets. Our definition of a first control an NWP model, previous analysis, climatology, etc., to (CTRL) forecasts includes the ECMWF plus rain-rate form a numerical representation of the initial atmo- estimates in its initial state at the T126 resolution. Sub- spheric state. CAMEX-4 data were analyzed using a sequent to that we generated another control run three-dimensional variational analysis scheme that is (CTRL1), where all the available winds (flight level mainly based on Lorenc’s (1986) concept of minimizing plus dropsonde winds) using CTRL as background a generalized cost function using the Spectral Statistical were assimilated. This is the new control run, which is Interpolation (SSI) technique (Parrish and Derber now being used as background field for all other ex- 1992). The cost function used in this scheme is as fol- periments. Here we have used both the dropsonde and lows: the flight-level data, since our aim is to utilize more and more wind information to define the hurricane struc- 1 T Ϫ1 J͑X͒ ϭ ͕͑X Ϫ X ͒ B ͑X Ϫ X ͒ ture. However, the experiments were designed mainly 2 b b to focus on the impact of water vapor measurements Ϫ ϩ ͓Y Ϫ H͑X͔͒TR 1͓Y Ϫ H͑X͔͖͒, ͑1͒ (either dropsonde or LASE), which is a goal of our obs obs study. The objective of this analysis is to forecast the where X ϭ N-component vector of analysis variables; ϭ broad scale evolution of hurricanes. Xb N-component vector of first guess variables; ϭ ϭ ϫ Dropwindsonde data from the six aircraft (two Yobs M-component vector of observations; B N NOAA P-3s, U.S. Air Force C-130, NASA ER-2 and N background error covariance matrix; R ϭ M ϫ M DC-8, and NOAA G-IV), flight-level in situ data from observational error covariance matrix; H ϭ an operator the DC-8 and P-3 aircraft, and LASE water vapor pro- from analysis to observation space (possibly nonlinear); files collected from the DC-8 aircraft were used in the N ϭ analysis degree of freedom; M ϭ number of ob- following experiments: servations; and ()T ϭ transpose of the matrix.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 156 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

The two terms in J on the right-hand side of Eq. (1) tween the day 1 forecasts and the corresponding veri- deal with the fit of the first guess field and the obser- fying analysis for that date. A 40-day dataset (August– vations with the desired analysis, respectively. In order September 1998) was used to compute these statistics. that the analysis should fit best with both the first guess The amplitudes and scales of the background errors and the observations, J should be minimized with re- have to be tuned to represent the (6, 3 h) forecast errors spect to the analysis variable X. The minimization is in the guess field. This is the reason for the tuning accomplished by using a standard conjugate gradient parameters mentioned above. The statistics that project (CG; Chandra 1978) descent algorithm, which allows R multivariate relations among variable are also derived to be nonlinear. from the NCEP method. Let us define a variable W as b. Forecast model ϭ Ϫ1͑ Ϫ ͒ ͑ ͒ W C X Xb . 2 The global model used in this study is identical in all The matrix C is related to the forecast errors and is respects to that used in Krishnamurti et al. (1991). This given by comprehensive model carries the spectral transform method with the semi-implicit time differencing scheme B ϭ CCT. ͑3͒ for the dynamics. An array of physical parameteriza- tion schemes are used for shallow and deep cumulus Minimization of J leads to the following analysis equa- convection, dry convective adjustment, radiative trans- tion: fer based on a band model for short- and longwave A␦ ϭ f, ͑4͒ radiances, surface fluxes, surface energy balance, definition of clouds for radiative transfer, and the use of the ϭ ϩ T T Ϫ1 ϭ T T Ϫ1 where A [I C L R LC]; and f C L R [Yobs sigma coordinate surface for the treatment of orogra- Ϫ ϩ Ϫ H(Xb CW)] W. phy. The relevant references on these model compo- Using a perturbation technique, the solution of the nents are provided in Krishnamurti et al. (1991). The analysis Eq. (4) is obtained for W. Finally the state current horizontal resolution for the CAMEX studies is vector X, which minimizes J, is determined using (2) T126 and in vertical we use 14 layers. In addition to and (3). The following iterative algorithm is used for these features the model includes a physical initializa- solving the analysis equation: tion component based on Krishnamurti et al. (1991, 2001). The sequence of steps in our modeling starts with Set X ϭ X , W ϭ 0, b a special operational level-three analyzed dataset from Do k ϭ 1, n outer (outer iteration), Ϫ the ECMWF. This dataset is received at FSU at a reso- Evaluate f ϭ CTLTR 1[Y Ϫ H(X ϩ CW)]Ϫ W, obs b lution of 0.5° latitude, 0.5° longitude at 28 vertical lev- Solve A␦ ϭ f, (5) els. These data are interpolated to the FSU transform using an iterative conjugate gradient algorithm, grid at the resolution T126. At this stage precipitation W ϭ W ϩ ␦, estimates based on microwave radiance datasets from X ϭ X ϩ C␦, TRMM Microwave Imager (TMI) and DMSP–Special End do. Sensor Microwave Imager (SSM/I) satellites are used Here one important thing to be noted is that the tan- for the physical initialization that provides a nowcasting gent linear operator is always evaluated with the latest of the initial rain rate for the model. The proposed data X and the nonlinearity in H, which results from the assimilation that incorporates the additional datasets dependence of the sigma coordinates on the surface from the CAMEX-4 research aircraft (dropwindsonde, pressure, is not included in L. However, each time the LASE, and flight level) follows sequentially after the vector f is evaluated, a complete H operator is utilized. completion of physical initialization. The SST analysis Further details about the choice of the analysis vari- for this modeling was obtained from NOAA/National ables, balance between the mass and the wind field, Environmental Satellite, Data, and Information Service vertical coupling, the statistical parameters used, etc., (NESDIS) operational real-time analysis files that were are given in Parrish and Derber (1992), and Rizvi and based on weekly averages (preceding the initial state Parrish (1995). The statistical parameters, like the back- for each storm) and included datasets from surface ground error covariance, empirical orthogonal func- ships, buoys, and satellite-derived estimates. tions (EOF) for vorticity, divergence, temperature, Several techniques have been developed to assess the moisture, and various other statistical parameters de- influence of new datasets for the analysis/forecast is- fining the partitioning between the fast and the slow sues. In this study, the data-denial experiment (DDE) part of the flow, are computed using the difference be- procedure was used to assess the impact of new datasets

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 157 on hurricane analysis and forecasts. In this technique actual forecast/data assimilation cycles are rerun, selec- tively excluding actual observations, and the relative degradation or improvement noted. By examining many such cases, the effect of the exclusion of certain datasets can be assessed.

4. Numerical results and discussion The outputs of the various experiments that were analyzed and the impacts on the hurricane forecasts are discussed in the following subsections. a. Impact on hurricane track and intensity In our sensitivity analysis of the forecasts to the assimilation of CAMEX-4 datasets, we primarily focused on some representative fields such as the mean sea level (MSL) pressure. Comparisons with the verifying ECMWF analysis and officially issued best track from the National Hurricane Center (NHC) were also made. The various forecast tracks for Hurricanes Erin, Hum- berto, and Gabrielle are presented in Figs. 2a,b,c. The legend in these figures distinguishes between the different model runs for each storm. Since the research aircraft flights were limited, generally only one complete set of experiments was possible for each of these storms. Although there were two complete missions flown for Humberto, only one dataset is used here. For Erin and Gabrielle, the forecast track for the control run was generally in the approximate heading of the storms, moving northeastward, but was behind the actual storm position because of the erroneously slow translation speed. On the other hand, forecast tracks with CAMEX-4 datasets were almost parallel to each other and followed the best track closely. However, the initial storm positions in the control and the experiments were very close to the best track position. In the case of Humberto, the forecast track of the control run showed the tendency of recurvature; however, the slow forecast speed increased the distance error substantially reaching 340 km at 72-h forecast. Assimilation of the maximum amount of available data (EXP3) resulted in FIG. 2. (a) Twelve-hour forecast track of Hurricane Erin with the smallest distance errors of 120 and 150 km at 72- initial condition (IC): 1200 UTC 10 Sep 2001, (b) 12-h forecast and 96-h forecasts, respectively. track of Hurricane Humberto with IC 1200 UTC 22 Sep 2001, and Overall we note that, in general, there is a sequential (c) 96-h forecast track of Hurricane Gabrielle with IC 0000 UTC 15 Sep 2001. improvement in track forecasts as more and more datasets were added. The best-track forecasts, in general, were found when the flight-level data, the drop- dropsondes improved the forecast errors well beyond sonde data (wind and humidity), and the LASE mois- what was possible from LASE data alone. This is be- ture profiles were all incorporated within the data as- cause of the much larger coverage of moisture profiles similation. These datasets were added to those of the from the 100 or so dropsondes from six aircraft. LASE control run that only included the ECMWF analysis. provided moisture profiles from one aircraft (DC-8) The addition of moisture data from all the distributed and mainly in cloud-free areas.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 158 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

FIG. 3. A 72-h forecast of sea level pressure (mb) for Humberto with IC 1200 UTC 22 Sep 2001.

These results for Hurricanes Erin and Gabrielle es- lated study, Krishnamurti et al. (1997) noted that if sentially demonstrated that as more and more data several different initial analyses were deployed within were introduced (beyond those of the control experi- the same model, then either a left or a right bias devel- ment), the left bias of the control experiments was gen- oped in the forecasts depending on which initial analy- erally reduced with the addition of each new dataset. sis was used within the same model. This appears to be This left bias is largely related to the positive surface related to differences in the initial vorticity advection vorticity tendency that arises from the larger-scale hori- downwind of the velocity maxima of a hurricane, which zontal advection of vorticity in the lower troposphere. was located at different places for various analysis used This feature is similar for all of the above experiments in that study. This suggests that a left or a right bias and occurs downwind from the wind maximum of the could be reduced somewhat simply from the enhance- hurricane. As dropsonde and LASE datasets were se- ment of the initial datasets or from ensemble forecasts quentially introduced, the hurricane was better re- that use several initial analyses. solved on a smaller scale. The model showed a pro- The MSL pressure of all the three storms was exam- nounced center of rising motion, precipitation, and ined for all the experiments. The forecast intensities of positive vorticity tendency where this storm was now the minimum MSL pressure from all the experiments being resolved. This occurred to the right of the track of were 4–8 hPa lower than those in the control run. Fig- the control run, and the control run’s track appeared to ure 3 shows the MSL pressure on day 3 of forecasts for be mostly dictated by the larger-scale vorticity advec- Humberto for the various experiments. The pressure tion. The local enhancement of the diabatic processes forecasts for the same sequence of experiments (Analy- appears to display a positive tendency for the precipi- sis, CTRL, CTRL1, EXP1, EXP2, and EXP3), as shown tation enhancement from the storm’s newer datasets. It in Fig. 3, are 990, 998, 996, 994, 996, and 990 mb, re- is easy to see how that pushes the positive vorticity spectively. The minimum MSL pressure estimated by maxima to the east of the control experiment. In a re- NHC was 992 mb. Nearly the same differences were

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 159 found for Erin and Gabrielle (not shown). Figure 4 shows track errors for these hurricanes. In the case of Humberto and Erin, the initial errors were small, and these errors were further reduced in the experiments. In case of Gabrielle, the initial large errors were am- plified during model integration; however, these errors were reduced in the experiments. Overall these errors were reduced significantly up to day 3 forecasts. A combination of LASE plus the dropsonde data provided the best result. Furthermore, the results from EXP1 were better than those of EXP2. The possible reason is that the dropsondes from different aircraft provided coverage over many different locations over and around the hurricane. However, all the experiments had improved forecast tracks with stronger intensity and faster move- ment in comparison with the control run. The intensity forecast errors for the maximum storm winds for Erin, Humberto, and Gabrielle are shown in Fig. 5. The wind speed errors are shown in knots at the 850-hPa level. Here we compared the relative error for the two control runs with those of the dropsonde, LASE, and combined observation-based experiments. The best results in all cases came from the use of all FIG. 5. Forecast intensity errors at 850 hPa for Hurricanes (a) humidity datasets (dropsonde plus LASE, i.e., EXP3). Erin, (b) Humberto, and (c) Gabrielle. These show a major improvement over the control results where none of the CAMEX datasets were used. impact on hurricane intensity from the inclusion of im- The major message of Fig. 5 is that there is a positive proved humidity datasets. Given the wind forecasts on a latitude/longitude grid, it is possible to use coordinate transformation to cast datasets in polar coordinates with respect to the storm center, that is, in storm-relative coordinates. This is an objective algorithm that facilitates the examination of a storm’s intensity after an azimuthal averaging is performed. This was done for day 3 of the forecasts derived from the control run (CTRL), CTRL1, EXP3, and for the validated analysis. In Fig. 6 we show the results for the tangential wind. The abscissa is the radial distance from the storm center in km and the ordinate shows the azimuthally averaged tangential velocity at the 850-hPa level. The legend identifies these different datasets. The crosses show the results for the experiment that carries the most datasets. The three panels show results for the different storms: Erin, Gabrielle, and Humberto. Overall we note that the intensity forecasts do show some improvement as we add more and more datasets. What is most interesting here is that there is marked improvement in the intensity forecast as we proceed from CTRL1 to EXP3. This simply means that the addition of humidity data from both the dropsonde and LASE had a significant positive impact on the intensity forecasts over and

FIG. 4. Forecast track errors of Hurricanes (a) Erin, (b) above what was possible from the addition of winds Humberto, and (c) Gabrielle. from dropwindsonde and the flight-level measure-

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 160 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

Ϫ1 FIG. 6. Azimuthally averaged tangential velocity (m s ) vs radial distance (km) from the center of the storm of 72-h forecast for (a) Erin with IC 1200 UTC 10 Sep 2001, (b) Gabrielle with IC 0000 UTC 15 Sep 2001, and (c) Humberto with IC 1200 UTC 22 Sep 2001.

ments. A surprising lack of improvement in the inner LASE data were provided by the DC-8 aircraft flights radii (Ͻ200 km) for Gabrielle may be related to the over a portion of the storm where, in storm-relative lack of the effectiveness of humidity measurements in coordinates, a moist band of air enters the storm. The rain areas. motivation for displaying structure was to see the active quadrants of the storm where moisture enters the hur- b. Impact of extended moisture data coverage on ricane center. The ranger markers are arbitrarily hurricane forecasts marked at every 30° and the radial distance from center to outer radii is 370 km. The flight track along this moist In storm-relative coordinates, the moisture transport band was designated as an “optimal data assimilation vector Vq has an interesting structure. We have verti- flight plan,” and this type of flight plan was used during cally averaged this vector over 200–1000 hPa to exam- the CAMEX-3 (Danielle) and CAMEX-4 (Erin) cam- ine the moisture transports for Erin at 1200 UTC 10 paigns. The goal of this flight plan was to assess the September 2001. This is displayed in Fig. 7 as stream- impacts of high-resolution water vapor and wind mea- lines of Vq. One of the major moist swaths (red) of air surements on the forecasts of hurricane intensity and enters the storm from the south, whereas a dry air tracks. swath (yellow) is seen to spiral in from the north. The Figure 8 illustrates the aircraft flight track of the

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 161

FIG. 7. Moisture transport vector Vq for Hurricane Erin vertically averaged from 200 to 1000 hPa.

DC-8 during a moisture transport mission for studying cific humidity. Figure 9 displays the results based on Hurricane Erin during CAMEX-4. A similar mission Hurricane Erin’s analysis (Figs. 9a–d) and 72-h fore- was also flown for Hurricane Danielle during CAMEX- casts (Figs. 9e–h), respectively. The left panels (Figs. 3. These tracks were specifically designed to provide 9a,b,e,f) show the results for the inner segment of data, moisture datasets to the south of the hurricane (over relatively clear sky areas). The design of this mission was specifically aimed at obtaining a detailed descrip- tion of the moisture transports (Vnq) at two different radii from the storm center. This moisture transport was provided by the dropwindsonde data from the DC-8 and LASE profiles of moisture. Two versions of data assimilation, with and without the LASE data, were used to estimate these humidity transports. We also examine the differences between the LASE-based transports and those of the control experiment that did not include the LASE datasets. For this purpose we determined the vertical cross sections proceeding from west to east following the outer and inner segments of the flight tracks that are shown in Figs. 9 and 10. In this case, dropwindsondes provided the vertical profiles of winds and LASE provided the vertical profiles of spe- FIG. 8. DC-8 flight path around Hurricane Erin on 10 Sep 2001.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 162 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

Ϫ1 Ϫ1 FIG. 9. Moisture transport (m s gkg ) of Hurricane Erin for the (a)–(d) analysis valid on 1200 UTC 10 Sep 2001 and (e)–(h) 2-h forecast with IC 1200 UTC 10 Sep 2001.

and the right panels (Figs. 9c,d,g,h) show the results for port in the inner segment was found to be conspicu- the outer segment. We see here that the moisture trans- ously larger than the transport in the outer segment. ports that included the LASE data were invariably This suggests a divergence of moisture flux between the larger (Figs. 9a,b) than those of the control (Figs. 9c,d) outer and inner segments. The resulting strong evapo- where LASE data were excluded. The moisture trans- ration in the storm’s vicinity may have been a factor in

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 163

Ϫ1 Ϫ1 FIG. 10. Moisture transport (m s gkg ) of Hurricane Danielle, during CAMEX-3, for the (a)–(d) analysis valid on 0000 UTC 31 Aug 1998 and (e)–(h) 72-h forecast with IC 0000 UTC 31 Aug 1998. producing the stronger intensity of Erin, which had analysis (Figs. 10a–d) and 72-h forecasts (Figs. 10e–h), reached a category 3 level. respectively. The left panels (Figs. 10a,b,e,f) show the The corresponding results for Hurricane Danielle in results for the inner segment of data, and the right pan- 1998, where similar flights of the NASA DC-8 were els (Figs. 10c,d,g,h) show them for the outer segment. flown, are shown in Fig. 10. This shows Danielle’s Here we found a similar structure in the analysis (i.e.,

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 164 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

FIG. 11. Days 0 and 3 moisture budget for (a), (b) Hurricane Erin, (c), (d) Hurricane Gabrielle, and (e), (f) Hurricane Humberto for three experiments. day 0) for the moisture transports, although the mag- cay at these observational times. The time series of the nitudes were considerably less. Hurricane Danielle was intensity shows that Erin was weakening slightly a category 1 storm, and did not exhibit as much mass whereas Danielle was in a decaying stage. Thus, large and moisture convergence as was seen for Hurricane impacts of these new datasets were noted in the mag- Erin. The convergences were further reduced when the nitude of the moisture flux convergences that were forecasts for Danielle were examined. The inner and missed in the control run. outer moisture transports were nearly equal. In the case of Danielle, the convection became more tightly c. Moisture budget wrapped around the center and the moisture conver- We shall next discuss the moisture budget of these gence was closer to the center. Furthermore, the selec- storms, where the moisture transports appear to have a tion of the date and the flights flown around these hur- critical role in the overall hurricane behavior. Basically ricanes may have been another factor since the storms the water vapor budget contains elements such as Stor- were going through a life cycle of amplification or de- age ϭ convergence of flux of moisture ϩ evaporation Ϫ

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 165

FIG. 11. (Continued)

Ϫ precipitation. These are integrals over a mass of the maximum wind was around 100 kt (51.44 m s 1). In atmosphere where the following 24 h, the minimum pressure rose to 976 hPa and the maximum winds dropped to 80 kt. 1 p2 r2 ͵ ͑ ͒ dm ϭ ͵ Ͷ ͵ ͑ ͒rdrd␪ dp, ͑6͒ 2) Gabrielle: At 1200 UTC 15 September 2001, the g ␪ m p1 r1 central pressure of Gabrielle (a tropical storm at this stage) had a minimum pressure of 998 hPa and an and these mass elements are enclosed inside circles of intensity of 40 kt. In the following 36 h, the central radii r , and r , around the hurricane from the top of the 1 2 pressure dropped to 991 hPa and the maximum atmosphere p ϭ p to the earth’s surface where the 1 winds reached hurricane strength of 65 kt. pressure is p ϭ p . 2 3) Humberto: At 1200 UTC 22 September 2001, the The three storms, Erin, Gabrielle, and Humberto central pressure of the tropical storm was 1007 hPa were in different stages of their life cycle. and maximum wind of 40 kt. In the next 24 h, the 1) Erin: On 10 September 2001 at 1200 UTC, Hurri- central pressure dropped to 990 hPa and the maxi- cane Erin’s minimum pressure was 969 hPa and mum winds reached hurricane strength of 65 kt.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 166 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

FIG. 11. (Continued)

Given the assimilated datasets and the day 3 forecasts The budget was estimated from various datasets gen- for Hurricanes Erin, Gabrielle, and Humberto it was erated by our model (i) control datasets (CTRL), (ii) possible to evaluate these terms for all of the experi- experiment that includes LASE datasets (EXP2), and ments carried out in this study. Using storm-relative (iii) experiment that includes dropsonde and LASE coordinates, and a local cylindrical frame of reference datasets (EXP3). These datasets are all derived using about the storm center, the moisture budget is ex- the 3DVAR data assimilation. Figures 11a–f show pressed by the following relations: these budgets for Hurricanes Erin, Gabrielle, and Moisture continuity equation: Humberto in storm-relative coordinates for these three datasets. These budget diagrams carry the following in- Ѩq Ѩq␻ t), that is, source orץ/qץ) Ϫ formation: (i) local change ١ ϭϪ ͵ Ѩ dm ͵ · qV dm ͵ Ѩ dm m t m m p sink, (ii) convergence of flux (positive sign), (iii) pre- ϩ Evap Ϫ P, cipitation (P), and (iv) evaporation (Evap). These are all quantities provided by the model at time 0 and at Ѩ Ѩ qV␪ 1 qVrr 72-h forecasts. The following is the summary of mois- qV ϭ ϩ . ͑7͒ · ١ where rѨ␪ r Ѩr ture budget for these individual storms.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 167

FIG. 11. (Continued)

(i) Erin: (Figs. 11a,b). The initial evaporation at a ra- included more and more humidity datasets in their dial distance of 200 to 300 km from storm center initial analysis. With the weakening of Erin we Ϫ increased from 206 to 353 to 503 mm day 1 as we also found spindown of the net convergence of proceed from the CTRL to LASE (EXP2) and to moisture flux. It is possible that an increase in the final experiment with all datasets (EXP3). evaporation rate and storm intensification are con- There was in fact a spin down of evaporation by current indicators, and as such, it may not neces- day 3 of the forecast at these radii to values of 160, sarily imply causality. Ϫ 218, and 298 mm day 1, respectively. This is con- (ii) Gabrielle: (Figs. 11c,d) Gabrielle was an intensify- sistent with the observation that Hurricane Erin ing storm in these cases. Between 200- and 300-km was weakening somewhat. This spindown of radii at the initial time the evaporation increased Ϫ evaporation was also evident at an outer radius sequentially from 222 to 363 to 608 mm day 1 as near 1000 km. Unlike the evaporation, the precipi- we proceed from the control to LASE to the final t), thatץ/qץ) tation amount did not show a significant decrease full data experiment. The local change in intensity with the spindown. In general, larger is, source or sink, convergence of flux, precipita- rainfall rates were found for the experiments that tion, and evaporation, of the moisture budget was

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 168 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

FIG. 11. (Continued)

extremely large at the initial time, perhaps contrib- In general, for day 0 fields we find a gradual increase in uting to initial intensification of Gabrielle from a the convergence of fluxes of moisture as we proceed tropical storm to a hurricane. Therefore, the mois- from the control data–based experiment to LASE da- ture budget reveals an equilibration toward some- ta–based experiment and onto the all-inclusive data– what lower values. based experiments. It is very clear that as the conver- (iii) Humberto: Figures 11e,f displays the moisture bud- gence of moisture flux increases so does evaporation gets for Hurricane Humberto. Here we can see for and precipitation within the radial legs (around the the day 0 an increase in the convergence of fluxes storm). Although the detailed precipitation fields are from 51 to 160 to 245 between 200 and 300 km from not shown here, in the inner radii inside r Ͻ 500 km the the storm center. The forecasts seem to retain some satellite pictures clearly confirm higher rainfall in those of these impacts even at day 3. The differences be- inner belts. tween LASE-based and LASE plus dropsonde- This clearly shows the impact of including the addi- based forecasts are, however, not as large at 72 h tional moisture data from LASE and the dropsondes as as they were at hour 0. However, they do have much well. The control analysis that only includes the mois- larger values compared to the control at 72 h. ture analysis from the world weather watch underesti-

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 169

FIG. 11. (Continued) mates the precipitation compared to the observed esti- datasets of humidity may be important for the modeling mates of rain amounts. of the heavy rain areas. The moisture budgets at 72 h show increased values We have also explored the sensitivity of track fore- for the transports and divergence of flux of moisture as casts to the use of enhanced moisture profiling obser- we sequentially compare results of experiments with vations in the lower versus the upper troposphere. We added datasets (especially moisture datasets). Further- noted that the track forecasts were sensitive to these more, the impacts from the added moisture datasets are variations in the use of moisture datasets. Figure 12a quite robust. The precipitation forecasts in excess of illustrates a comparison of such track forecasts. Here Ϫ 120 mm day 1 along those radial legs from the LASE we compare the control (i.e., where no aircraft obser- and LASE plus dropsonde experiment are in line with vations are used) with those results where the low level the observed estimates. The precipitation rates were (below 850 hPa) moisture data were assimilated and underestimated by the control run. This shows that where the moisture datasets were enhanced only above forecasts that include moisture profile datasets can con- the 850-hPa level. These tracks are here compared to tribute to a positive impact. Since the moist processes the observed best track. We note that, in general, there are most important for a hurricane, such profiling is a stronger positive impact in improving the track

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 170 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

FIG. 12. (a) The 12-h forecast tracks of Hurricane Erin; the vertical profile of moisture from (b) CTRL, (c) dropsonde, and (d) LASE for Hurricane Erin.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 171

Ϫ1 FIG. 13. Observational vertical profiles of specific humidity (g g ) for Hurricane Erin at 33°N at 1200 UTC 10 Sep 2001: (a) CTRL, (b) dropsonde, (c) LASE, and (d) LASE ϩ dropsonde. forecast when the lower tropospheric additional mois- shown in Figs. 13a–d. It is these enhanced moist profiles ture datasets are added. Figure 12a shows the 5-day in our humidity assimilation that seems to impact the forecast tracks for Hurricane Erin at intervals of every forecasts of Hurricane Erin. 12 h. Similar comparisons (not shown) were carried out Figure 14 illustrates hurricane track forecasts from for Hurricanes Humberto and Gabrielle where we several sources for Hurricanes Erin, Humberto, and found a very similar sensitivity. Gabrielle. This includes the forecast from the afore- We have also constructed vertical cross sections of mentioned control run (CTRL, black line), the forecast moisture along the sawtooth flight patterns shown in from the experiment that include the humidity profile Fig. 8 and compared the vertical profiles of moisture datasets (EXP2, blue line), the official forecast of the from LASE and the dropsondes at collocated points. National Hurricane Center (NHC OFC, red line), and a At these locations we found our control analysis to be forecast based on the FSU superensemble (SUPENS, somewhat too dry compared to the LASE and drop- purple line). Also shown are the observed best-track sonde measurements. Figures 12b–d shows vertical pro- positions (green line). We found in all cases that the files of specific humidity at 33°N and 66°W from these experiments that included the humidity profile datasets direct measurements and those interpolated from the perform better than the control and at times were bet- control analysis. In general the dropsondes (Fig. 12c) ter than the official forecasts. The FSU superensemble were slightly more moist compared to the LASE (Fig. has the best performance since it assimilates the results 12d) measurements, but both were wetter than the con- from several models (Williford et al. 2003). trol analysis (Fig. 12a). These differences are more con- The current experiments on Erin, Gabrielle, and spicuous when we look at the respective observational Humberto using LASE and dropwindsonde data were vertical distributions of relative humidity that are not part of the real-time forecasting effort. It is inter-

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 172 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

perhaps provide a useful member for the superensemble. Such a direct test is not possible, since the superensemble requires a training database of as many as 60 forecasts from each member model, and the LASE and dropwindsonde data were not available for such a series of past forecasts.

5. Conclusions This study explores the impact of additional datasets beyond those that are currently available for operational weather prediction. The conventional datasets include surface and radiosondes, ships of opportunity, oceanic buoys, commercial aircraft platforms, cloud and water vapor–tracked winds, and satellite-based soundings. Those were analyzed via the four- dimensional data assimilation by the ECMWF, which was provided for us. The goal of this study was to examine the impact of dropwindsonde (profiles of winds and humidity) and/or LASE (moisture profiles) datasets during CAMEX-4 on hurricane analysis and forecasting. These new observations, especially the winds from the dropsondes and aircraft flight levels and moisture profiles from the dropsondes and LASE, were analyzed sequentially using 3DVAR, and medium- range predictions were then carried out by integrating the FSU global spectral NWP model out to 5 days with these new initial conditions. The forecast of tracks and intensity from the assimilation experiments demonstrate improvement over control experiments. The inclusion of winds only from the dropwinsonde data also shows some improvement in the forecasts over those of control runs that do not include any CAMEX datasets. The additional use of moisture profiling data indicates a clear strong positive impact in several areas. Three-day track forecasts show positive error reduc- tions by as much as 100 km. Even the intensity forecasts for those storms observed during CAMEX-4 show clear improvements compared to the forecasts from a version of model that does not include moisture-profiling data FIG. 14. (a) The 12-h forecast tracks of Hurricane Erin with IC from LASE and the dropwindsondes. We also found 1200 UTC 10 Sep 2001, (b) 12-h forecast track of Hurricane Hum- that the impact of the water vapor–profiling data seems berto with IC 1200 UTC 22 Sep 2001, and (c) 96-h forecast track to be even more encouraging than what we noted from of Hurricane Gabrielle with IC 0000 UTC 15 Sep 2001. the addition of dropsonde winds. A large left bias in the control run was reduced in the track forecasts when the esting to note that only for Erin did the operational newer datasets were included. The inclusion of superensemble perform better than these new experi- CAMEX-4 datasets provided some improvements in ments. For the other two storms, Gabrielle and Hum- the hurricane intensity forecasts as well. A combination berto, our forecasts with the LASE and dropwindsonde of LASE and the dropsonde data also appeared to im- moisture data appear to have done very well in com- prove the intensity forecasts. Our study of the moisture parison to the operational forecasts. This suggests that budget showed enhanced evaporation and precipitation the use of humidity profile data in single forecasts can associated with the use of the additional datasets. A

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC JANUARY 2006 KAMINENI ET AL. 173 large impact of these new datasets was found in the This work is financially supported by NASA CAMEX convergence of mass and moisture fluxes that was Grants NAG8-1848 and NAG8-1537, and NSF Grant missed in the control runs. The experiments with ATM-0108741. Dr. Ramesh Kakar, NASA program CAMEX-4 datasets generated more precipitation near manager for atmospheric dynamic and remote sensing, the storm than the control. The precipitation generated provided the funding for CAMEX-4. in the moisture-data-added experiments was more con- centrated near the storm center than the control run REFERENCES that excluded such datasets, and the overall pattern was closer to the verification. We compared the vertical Aberson, S. D., 2002: Two years of operational hurricane synoptic surveillance. Wea. Forecasting, 17, 1101–1110. structure of specific humidity constructed along the ——, 2003: Targeted observations to improve operational tropical sawtooth flight patterns and noted that the moisture cyclone track forecast guidance. Mon. Wea. Rev., 131, 1613– profiles from the control (interpolated) runs were drier 1628. than those profiles from the runs that included the Bensman, E. L., 2000: The impact of aircraft digital weather data dropsonde and LASE data. in an adaptive observation strategy to improve the ensemble The impact of enhanced observations in heavy rain prediction of hurricanes. Ph.D. dissertation, The Florida State University, 225 pp. disturbances in the Tropics appears to be a very rel- Browell, E. V., and S. Ismail, 1995: First lidar measurements of evant issue. In hurricanes we found that data from en- water vapor and aerosols from a high-altitude aircraft. Proc. hanced observations positively impacts the analysis of 1995 OSA Symp. on Optical Remote Sensing of the Atmo- very moist areas, moisture transports, convergence of sphere, Salt Lake City, UT, Optical Society of America, 212– flux of moisture, and the precipitation forecasts of 214. ——, and Coauthors, 1997: LASE validation experiment. Ad- heavy rains. Improvements in all of these factors have vances in Atmospheric Remote Sensing with Lidar, Springer been seen in the analysis and forecasts of hurricanes Verlag, 289–295. examined here. We believe that this would also be true ——, S. Ismail, and R. Ferrare, 2000: Hurricane water vapor, for other heavy rain producing disturbances over the aerosol, and cloud distributions determined from airborne Tropics and elsewhere. Only one LASE instrument was lidar measurements. Proc. Symp. on Lidar Atmospheric available for measurements during CAMEX, but in the Monitoring, Long Beach, CA, Amer. Meteor. Soc., 65–67. ——, and Coauthors, 2001: Large-scale air mass characteristics future, this capability could be on several aircraft and observed over the remote tropical Pacific Ocean during eventually in space. In the meantime, global humidity March–April 1999: Results from PEM-Tropics B field experi- profiling will be provided by the Aqua satellite; this ment. J. Geophys. Res., 106, 32 481–32 501. points to a natural extension of the present work to Burpee, R. W., D. G. Marks, and R. T. Merrill, 1984: An assess- examine the impact of moisture profiling datasets on ment of omega dropwindsonde data in track forecasts of Hur- Bull. Amer. Meteor. Soc. global weather forecasts. ricane Debby (1982). , 65, 1050– 1058. A major thrust on mesoscale high-resolution model- ——, J. L. Franklin, S. J. Lord, R. E. Tuleya, and S. D. Aberson, ing of hurricanes has emerged in recent years; see, for 1996: The impact of omega dropwindsondes on operational example, Kurihara et al. (1995, 1998), Cocke (1998), hurricane track forecast models. Bull. Amer. Meteor. Soc., 77, Zhang and Altshuler (1999), Pu et al. (2002), Zhang et 925–933. al. (2002), and Wang and Wu (2004). It has been pos- Chandra, R., 1978: Conjugate gradient methods for partial differ- ential equations. Ph.D. thesis, Yale University, 253 pp. sible to observe detailed features of the inner precipi- Cocke, S., 1998: Case study of Erin using the FSU nested regional tating areas (eyewall convection and rainbands) and the model. Mon. Wea. Rev., 126, 1337–1346. matching radar reflectivity (as seen from a computed Ferrare, R., and Coauthors, 1999: LASE measurements of water proxy radar reflectivity using microphysics output and vapor, aerosols, and clouds during CAMEX-3. Proc. Symp. observed radar reflectivity). Furthermore, these models on Optical Remote Sensing of the Atmosphere, Santa Barbara, have also addressed details on the tracks, intensity, and CA, Optical Society of America, 114–116. ——, and Coauthors, 2000a: Comparison of aerosol optical prop- timing of landfall for a number of storms. These recent erties and water vapor among ground and airborne lidars and studies are quite promising for mesoscale thrusts on sun photometers during TARFOX. J. Geophys. Res., 105, data assimilation and forecast experiments on new ob- 9917–9933. serving systems. ——, and Coauthors, 2000b: Comparisons of LASE, aircraft, and satellite measurements of aerosol optical properties and wa- Acknowledgments. CAMEX-4 could not have been ter vapor during TARFOX. J. Geophys. Res., 105, 9935–9947. successful without the cooperation of a large number Franklin, J. L., and K. M. DeMaria, 1992: The impact of omega dropwindsonde observations on barotropic hurricane track of agencies and individuals. We wish to thank all forecasts. Mon. Wea. Rev., 120, 381–391. data producers for their support and cooperation. The Graham, R. J., 1990: The impact of North Atlantic TEMPSHIP ECMWF is acknowledged for the use of their analysis. observations on global model analyses and forecasts during a

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC 174 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 63

case of cyclogenesis. Short-Range Forecasting Research 1990: Recommended algorithms for the retrievals of rainfall Tech. Note 46, Met Office, Reading, England, 32 pp. rates in the tropics using the SSM/I (DMSP-8). University of ——, and S. R. Anderson, 1995: The relative utility of current Wisconsin—Madison Rep., 10 pp. observation systems to global-scale NWP forecasts: Prelimi- Pailleux, J., 1990: The impact of North Atlantic TEMPSHIP nary results. U.K. Met Office, Forecasting Research Division radiosonde data on the ECMWF analysis and forecast. Tech. Rep. 173, 19 pp. ECMWF Research Department Tech. Memo. 174, 36 pp. Heming, J. T., 1990: The impact of surface and radiosonde obser- Parrish, D. F., and J. C. Derber, 1992: The National Meteorologi- vations from two Atlantic ships on a numerical weather pre- cal Center’s Spectral Statistical Interpolation analysis diction model forecast for the storm of 25 January 1990. Me- scheme. Mon. Wea. Rev., 120, 1747–1763. teor. Mag., 119, 249–259. Pu, Z.-X., E. Kalnay, J. Derber, and J. Sela, 1997: Using forecast Hock, T. F., and J. L. Franklin, 1999: The NCAR GPS dropwind- sensitivity patterns to improve future forecast skill. Quart. J. sonde. Bull. Amer. Meteor. Soc., 80, 407–420. Roy. Meteor. Soc., 123, 1035–1054. Ismail, S., E. V. Browell, R. A. Ferrare, S. A. Kooi, M. B. Clayton, ——, W.-K. Tao, S. Braun, J. Simpson, Y. Jia, J. Halverson, A. V. G. Brackett, and P. B. Russell, 2000: LASE measurements Hou, and W. Olson, 2002: The impact of TRMM data on of aerosol and water vapor profiles during TARFOX. J. Geo- mesoscale numerical simulation of Super Typhoon Paka. phys. Res., 105, 9903–9916. Mon. Wea. Rev., 130, 2248–2258. Kamineni, R., T. N. Krishnamurti, R. A. Ferrare, S. Ismail, and Qu, X., and J. Heming, 2002: The impact of dropsonde data on E. V. Browell, 2003: Impact of high resolution water vapor forecasts of hurricane Debby by the Met Office unified cross-sectional data on hurricane forecasting. Geophys. Res. model. Adv. Atmos. Sci., 19, 1029–1044. Lett., 30, 1234, doi:10.1029/2002GL016741. Rabier, F., E. Klinker, P. Courtier, and A. Hollingsworth, 1996: Kooi, S. A., R. A. Ferrare, S. Ismail, E. V. Browell, M. B. Clayton, Sensitivity of forecast errors to initial conditions. Quart. J. V. G. Brackett, and J. Halverson, 2002: Comparison of LASE Roy. Meteor. Soc., 122, 121–150. water vapor measurements with dropwindsonde measure- Rizvi, S. R. H., and D. F. Parrish, 1995: Documentation of the ments during the third and fourth convection and moisture spectral statistical interpolation (SSI) scheme. NCEP Rep., experiments (CAMEX-3 and CAMEX-4). Preprints, Int. La- 38 pp. [Available from NCEP, 5200 Auth. Road, Washing- ser Radar Conf., Quebec City, QC, Canada, ICLAS, 693–696. ton, DC 20235.] Krishnamurti, T. N., J. Xue, H. S. Bedi, K. Ingles, and D. Oost- ——, E. L. Bensman, T. S. V. Vijay Kumar, A. Chakraborty, and erhof, 1991: Physical initialization for numerical weather pre- T. N. Krishnamurti, 2002: Impact of CAMEX-3 data on the diction over the tropics. Tellus, 43A/B, 53–81. analysis and forecasts of Atlantic hurricanes. Meteor. Atmos. ——, R. Correa-Torres, G. Rohaly, and D. Oosterhof, 1997: Phys., 79, 13–32. Physical initialization and hurricane ensemble forecasts. Wea. Simpson, R., 2003: Hurricane! Coping with Disaster. Amer. Geo- Forecasting, 12, 503–514. phys. Union, 360 pp. ——, and Coauthors, 2001: Real-time multianalysis–multimodel Smith, T. L., and S. G. Benjamin, 1993: Impact of network wind superensemble forecasts of precipitation using TRMM and profiler data on a 3-h data assimilation system. Bull. Amer. SSM/I products. Mon. Wea. Rev., 129, 2861–2883. Meteor. Soc., 74, 801–807. Kurihara, Y., M. A. Bender, R. E. Tuleya, and R. J. Ross, 1995: Thorpe, A. J., and M. A. Shapiro, 1995: FASTEX: Fronts and Improvements in the GFDL hurricane prediction system. Atlantic Storm Track Experiment: The Science Plan. 25 pp. Mon. Wea. Rev., 123, 2791–2801. [Available from Dept. of Meteorology, University of Read- ——, R. E. Tuleya, and M. A. Bender, 1998: The GFDL hurricane ing, P.O. Box 239, Reading RG6 2AU, United Kingdom.] prediction system and its performance in the 1995 hurricane Wang, Y., and C.-C. Wu, 2004: Current understanding of tropical season. Mon. Wea. Rev., 126, 1306–1322. cyclone structure and intensity changes—A review. Meteor. Lord, S. J., 1993: Recent developments in tropical cyclone track Atmos. Phys., 87, 257–278. forecasting with the NMC global analysis and forecast sys- Williford, C. E., T. N. Krishnamurti, R. J. Correa-Torres, S. tem. Preprints, 20th Conf. on Hurricanes and Tropical Me- Cocke, Z. Christidis, and T. S. V. Vijaykumar, 2003: Real- teorology, San Antonio, TX, Amer. Meteor. Soc., 290–291. time multimodel superensemble forecasts of Atlantic tropical ——, 1996: The impact on synoptic scale forecasts over the United systems of 1999. Mon. Wea. Rev., 131, 1878–1894. States of dropwindsonde observations taken in the northeast Zhang, D. L., and E. Altshuler, 1999: The effects of dissipative Pacific. Preprints, 11th Conf. on Numerical Weather Predic- heating on hurricane intensity. Mon. Wea. Rev., 127, 3032– tion, Norfolk, VA, Amer. Meteor. Soc., 70–71. 3038. Lorenc, A. C., 1986: Analysis methods for numerical weather pre- ——, Y. Liu, and M. K. Yau, 2002: A multiscale numerical study diction. Quart. J. Roy. Meteor. Soc., 112, 1177–1194. of Hurricane Andrew (1992). Part V: Inner-core thermody- Moore, A. S., Jr., and Coauthors, 1997: Development of the Lidar namics. Mon. Wea. Rev., 130, 2745–2763. Atmospheric Sensing Experiment (LASE)—An advanced Zhang, Z., and T. N. Krishnamurti, 1999: Adaptive observation airborne DIAL instrument. Advances in Atmospheric Re- for hurricane prediction. Preprints, 23d Conf. on Hurricanes mote Sensing with Lidar, Springer Verlag, 281–288. and Tropical Meteorology, Dallas, TX, Amer. Meteor. Soc., Olson, W. S., F. J. LaFontaine, W. L. Smith, and T. H. Achtor, 113–116.

Unauthenticated | Downloaded 10/05/21 06:48 AM UTC