The Pennsylvania State University The Graduate School

EFFECTS OF THE MADDEN-JULIAN OSCILLATION ON THE

CYCLOGENESIS OF HURRICANES EMILY (2005) AND FAUSTO

(2002)

A Thesis in Meteorology by Stephanie E. Zick

c 2008 Stephanie E. Zick

Submitted in Partial Fulfillment of the Requirements for the Degree of

Master of Science

May 2008 The thesis of Stephanie E. Zick was reviewed and approved∗ by the following:

William M. Frank Professor of Meteorology Thesis Advisor

Jerry Y. Harrington Associate Professor of Meteorology

George Young Professor of Meteorology

William H. Brune Professor of Meteorology Head of the Department of Meteorology

∗Signatures are on file in the Graduate School. Abstract

Equatorial waves and the Madden-Julian Oscillation (MJO) play significant roles in the modulation of tropical convection and development of cyclones in all of the major storm basins. Using the Weather Research and Forecasting (WRF) model, case studies are examined for Hurricane Fausto (2002), which formed in the east Pacific during a convectively active phase of the MJO, and Hurricane Emily (2005), which formed in the north Atlantic during a neutral to somewhat convectively inactive phase of the MJO. First, control simulations are run with initial conditions interpolated from the National Centers for Environmental Pre- diction/Nation Center for Atmospheric Research (NCEP/NCAR) reanalysis; the model then is reinitialized and run with the statistically correlated components of the MJO removed. Owing to the newness of the procedure, the case studies are run as ensembles to ensure that the results are statistically significant. Four ensemble members, created by varying the model parameter physics, are performed for each simulation. In response to modified intial conditions, both Hurricane Fausto and Hurricane Emily evolve in very different ways. Mechanisms responsible for these differences and how they are related to the MJO are discussed, including the large scale environment, location of formation, storm track, intensity, and structure. Varying physical parameterizations within the ensemble also has significant effects on the storms and is considered in the results.

iii Table of Contents

List of Figures vi

List of Tables viii

Acknowledgments x

1 Introduction and Background 1 1.1 Previous Studies of Tropical Cyclogenesis ...... 2 1.2 Observations of Equatorial Waves ...... 3 1.3 Characteristics of the MJO ...... 5 1.4 Previous Studies of Wave Interaction with Tropical Cyclones . . . . 7 1.5 Goals ...... 10

2 Observational Data and Model Output 16 2.1 Datasets ...... 16 2.2 Diagnostic analyses ...... 17 2.3 WRF Model ...... 17 2.3.1 Model Initialization ...... 18 2.3.2 Boundary Conditions and Nesting ...... 18 2.4 Modification of Initial Conditions ...... 19 2.5 Ensemble Technique ...... 20

3 Analysis of the MJO’s Influence on Hurricane Fausto - East Pa- cific 24 3.1 Synoptic History ...... 25 3.2 Analysis of MJO wave structure during cyclogenesis and intensification 25 3.3 Model Domain ...... 26 3.4 Comparison of Initial Fields ...... 26 3.5 Evaluation of Genesis Parameters ...... 28 3.5.1 Dynamic Parameters ...... 28

iv 3.5.2 Thermodynamic Parameters ...... 29 3.6 Storm Track ...... 30 3.7 Storm Intensity ...... 34

4 Analysis of the MJO’s Influence on Hurricane Emily - North At- lantic 55 4.1 Synoptic History ...... 55 4.2 Analysis of MJO wave structure during cyclogenesis and intensification 56 4.3 Model Domain ...... 57 4.4 Comparison of Initial Fields ...... 57 4.5 Evaluation of Genesis Parameters ...... 58 4.6 Storm Track ...... 59 4.7 Storm Intensity ...... 60

5 Summary and Conclusions 75 5.1 Comparing Intensities in the Case Studies Using a Student’s t-test . 75 5.2 Conclusions ...... 77 5.3 Future Work ...... 79

A Large-scale Tropical Cyclone Genesis Parameters for Hurricane Fausto 80

B Large-scale Tropical Cyclone Genesis Parameters for Hurricane Emily 83

References 86

v List of Figures

1.1 Spectral Bands for Filtering ...... 12 1.2 Depiction of MJO propagation ...... 13 1.3 Idealized 3D structure of the MJO ...... 14 1.4 Composite MJO-filtered anomalies in the northeast Pacific relative to genesis point and date ...... 15

2.1 Specified and relaxation zones for WRF boundary conditions . . . . 23

3.1 Best track of Hurricane Fausto (from Unisys) ...... 36 3.2 MJO-filtered OLR anomalies averaged over 0-20N ...... 37 3.3 MJO-filtered 850 mb wind and 200 mb wind anomalies averaged over 0-20N ...... 38 3.4 WRF model domain configuration ...... 39 3.5 Unfiltered OLR and 850 mb wind anomalies at 00Z Aug 20 . . . . . 40 3.6 MJO-filtered OLR and 850 mb wind anomalies at 00Z Aug 20 . . . 41 3.7 Initial 850 mb and 200mb wind difference field for 54 km domain . . 42 3.8 Tracks of minimum SLP for 6 km ensemble members ...... 43 3.9 Comparision of Control and MJO-removed 850mb relative vorticity and 850 mb winds at 00Z Aug 21 ...... 44 3.10 Ensemble mean tracks of minimum SLP for 6 km control and MJO- removed simulations ...... 45 3.11 Time series of deep layer mean wind for CEM1 and MEM1 . . . . . 46 3.12 Comparison of 200 mb heights, 200 mb winds, and SLP for CEM1 and CEM2 at 12Z Aug 21 ...... 47 3.13 Comparison of 200 mb heights, 200 mb winds, and SLP for CEM1 and CEM2 at 18Z Aug 21 ...... 48 3.14 CEM2 3 hr convective 3 hr precipitation ending at a) 00Z Aug 24 and b) 06Z Aug 24 ...... 49 3.15 CEM2 3 hr convective 3 hr precipitation ending at a) 00Z Aug 24 and b) 06Z Aug 24 ...... 50

vi 3.16 Timeseries of minimum SLP (mb) for 6 km ensemble members, en- semble mean, and observations ...... 51 3.17 Comparison of 850 mb relative vorticity for CEM1 and MEM1 . . . 52 3.18 Comparison of 850 mb relative vorticity for CEM1 and MEM1 . . . 53 3.19 Comparison of 3 hour convective precipitation for CEM1 and CEM2 54

4.1 Best track of Hurricane Emily (from Unisys) ...... 62 4.2 MJO-filtered OLR anomalies averaged over 0-20N ...... 63 4.3 MJO-filtered 850 mb wind and 200 mb wind anomalies averaged over 0-20N ...... 64 4.4 WRF model domain configuration ...... 65 4.5 Unfiltered OLR and 850 mb wind anomalies at 00Z Jul 10 . . . . . 66 4.6 MJO-filtered OLR and 850 mb wind anomalies (ms −1) at 00Z Jul 10 67 4.7 Initial 850 mb and 200mb wind difference field for 54 km domain . . 68 4.8 Tracks of minimum SLP for 6 km ensemble members ...... 69 4.9 Ensemble mean tracks of minimum SLP for 6 km control and MJO- removed simulations ...... 70 4.10 Time series of deep layer mean wind for CEM1, MEM1, and CEM2 71 4.11 Timeseries of minimum SLP (mb) for 6 km ensemble members, en- semble mean, and observations ...... 72 4.12 Comparison of 3 hour convective precipitation for CEM1 and CEM2 73 4.13 Comparison of cross sections of RH for CEM2 and MEM2 . . . . . 74

vii List of Tables

2.1 Description of Ensemble Members (EMs) ...... 21

3.1 Large-scale Dynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area for Ensemble Means ...... 29 3.2 Large-scale Thermodynamic Tropical Cyclogenesis Parameters Av- eraged Over Genesis Area for Each 6 km Ensemble Mean . . . . . 30

4.1 Large-scale Dynamic Tropical Cyclogenesis Parameters Averaged Over Genesis Area for Each 6 km Ensemble Means ...... 58 4.2 Summary of Appendix Table A.2: Large-scale Thermodynamic Trop- ical Cyclogenesis Parameters Averaged Over Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Mean ...... 59

5.1 Student T-test P-values and Confidence Intervals for Hurricanes Fausto and Emily ...... 76

A.1 Large-scale Dynamic Tropical Cyclogenesis Parameters Averaged Over Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Member 80 A.2 Large-scale Thermodynamical Tropical Cyclogenesis Parameters Av- eraged over Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Member ...... 81 A.3 Large-scale Divergence Averaged Over 4 degree square Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Member ...... 81 A.4 Large-scale VWS Magnitude and Direction Averaged Over 4, 8, and 12 degree Genesis Areas at 00Z Aug 21 for Each 6 km Ensemble Member ...... 82

B.1 Large-scale Dynamic Tropical Cyclogenesis Parameters Averaged Over 8 degree Square Genesis Area at 00Z Jul 13 for Each 6 km Ensemble Member ...... 83

viii B.2 Large-scale Thermodynamical Tropical Cyclogenesis Parameters Av- eraged over Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Member ...... 84 B.3 Large-scale Divergence Averaged Over 4 degree square Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Member ...... 84 B.4 Large-scale VWS Averaged Over 4, 8, and 12 degree Genesis Areas at 00Z Aug 21 for Each 6 km Ensemble Member ...... 85

ix Acknowledgments

I would like to express my sincere gratitude to my adviser, Dr. William M. Frank, for his support and direction in this research and to the other members of my committee, Dr. George Young and Dr. Jerry Harrington. I also thank my office mate Jeff Gall for his guidance in the modelling and for many useful discussions. I thanks my parents, Joy and Bob, my brother, Greg, my boyfriend, Jon, and my best friend Brandy, for their continued encouragement and support and love. This work was supported by the National Science Foundation (NSF), grant ATM-0630364.

x Chapter 1

Introduction and Background

Tropical cyclones form within pre-existing convective regions that are often embedded in larger-scale cyclonic regions such as monsoon troughs, tropical waves, or cyclonic frontal disturbances that move equatorward into the tropics. These cyclonic regions are on a scale much larger than a tropical cyclone core (e.g. Gray 1968; Pasch and Avila 1992; Bracken and Bosart 2000). Frank (1988) presented the idea that there are two distinct stages in tropical cyclone development: genesis, where a mesoscale vortex forms within a poorly organized mesoscale convective system, and intensification. At some threshold the mesoscale convective vortex reaches the intensification stage and can deepen through air-sea interaction processes alone. Large-scale support is no longer necessary, although an unfavorable environment (often strong vertical wind shear or moving over land or cold water) could inhibit intensification or cause the storm to weaken (Frank 1988). Mechanisms for the genesis stage remain somewhat of an enigma despite the availability of advanced global observations and high-resolution numerical models. While the climatological conditions necessary for genesis are well-known, the local conditions present for individual genesis cases are not well understood. One theory is that tropical waves make the local environment more favorable for genesis. Frank and Roundy (2006) showed that the Madden-Julian Oscillation (MJO), equatorial Rossby (ER) waves, and tropical depression type (TD-type) waves (such as African easterly waves) play significant roles in the development of many cyclones every year in all six major basins. Building on this research, the present thesis uses 2 the Weather Research and Forecast Model (WRF) model version 2.2 (WRFV2) to study how the circulation anomalies of the MJO trigger genesis locally within broad climatologically favorable regions. This chapter will lay the background for this research by reviewing previous research on tropical cyclogenesis, tropical waves, the MJO, and the interactions among these phenomena.

1.1 Previous Studies of Tropical Cyclogenesis

The necessary climatological conditions for tropical cyclogenesis are generally agreed upon and can be summarized as sea surface temperatures above 26.5 ◦C overlaying a deep oceanic mixed-layer, deep convection within a region of moist lower and middle levels, positive low-level vorticity anomalies, and weak (or slightly easterly) vertical wind shear (Gray 1968, 1979). While these conditions exist over a large portion of the tropics for extended time periods, only a small percentage of potential disturbances develop into tropical storms. This suggests that there must be additional criteria for individual storms. Many theories about the additional criteria necessary for tropical cyclogenesis have been presented over the past few decades. McBride (1981a,b) and McBride and Zehr (1981) found, by performing a composite of case studies in the north Atlantic and northwest Pacific, that the thermodynamic criteria were generally present throughout the hurricane season while the large-scale dynamics (e.g. vertical and zonal wind shear, upper level divergence, and low-level relative vorticity) varied considerably on daily timescales. Genesis did not occur until the incipient disturbance moved into a region of elevated cyclonic (anitcyclonic) vorticity at lower (upper) levels. Their analysis also suggested that development of a pre-existing disturbance was mainly influenced by large-scale forcings rather than characteristics of the disturbance itself. Several theories for these external forcings have been investigated. Studies have shown that TCs often form when an incipient disturbance interacts with an upper-tropospheric trough (Sadler 1976; McBride and Keenan 1982; Bosart and Bartlo 1991; Montgomery and Farrell 1993; Bracken and Bosart 2000). Briegel and Frank (1997) found that 25 of the 41 northwest Pacific tropical cyclones examined formed when an upper-level trough was present in the 200 hPa wind field. 3

These troughs are believed to force upper level divergence via upper-level vorticity advection, creating a favorable environment for tropical cyclone development. On the other hand, Zehr (1992) found the occurrence of an upper level trough in both developing and non-developing systems. Therefore, a mid-latitude trough to the north-northwest of a cloud cluster is believed to influence genesis, but it has yet to be identified as the triggering mechanism (Briegel and Frank 1997). Love (1985) and Davidson and Hendon (1989) showed that one preferred region for tropical cyclogenesis is within the monsoon trough when anomalous low-level westerly winds develop on its equatorward side, often simultaneously intensifying the subtropical ridge poleward of the monsoon trough, resulting in a large-scale region of positive low-level relative vorticity. Holland (1995) was able to simulate monsoonal flow in the northwest Pacific with a three-dimensional model, along with the subtropical ridge to the north. Holland accounted for the development of the ridge by Rossby wave dispersion and further hypothesized that interactions with the midlatitudes could effectuate enhanced convection. In addition, Holland introduced the idea that easterly waves could grow in the northwest Pacific, through Rossby wave accumulation, where the trade winds meet monsoon westerlies, another preferred region for TC development (Briegel and Frank 1997). Knowledge of the synoptic conditions present at and just prior to genesis have improved greatly, but the specific mechanisms responsible for genesis are still unknown. Weather in the tropics is principally modulated by disturbances that propagate parallel to the equator, or equatorial waves. These disturbances modify the background flow by perturbing the large-scale vertical velocity, low-level and upper-level vorticity, and vertical wind shear fields. The next sections discuss observations of these waves that propagate in the equatorial waveguide.

1.2 Observations of Equatorial Waves

The meridional temperature gradient in the tropics is very small, resulting in little storage of available potential energy. Tropical circulations, therefore, must be driven by other processes such as diabatic heating; indeed, latent heating associated with convective systems is a dominating factor in tropical dynamics. Because there 4 is interplay between air-sea interactions, cumulus convection, mesoscale and large- scale circulations, equatorial waves, and the general circulation, consideration of all temporal and spatial scales of motion is crucial to the study of dynamics in the equatorial region (Holton 1992). Many structural features in the tropics can be studied using equatorial wave theory. Equatorial waves include all eastward and westward propagating disturbances that are trapped near the equator (the disturbance decays away from the equator). While the rotation of the earth greatly influences atmosphere and ocean dynamics, the Coriolis force reduces to zero and changes sign at the equator. In the vicinity of the equator, the Coriolis acceleration terms in the equations of motion are very small. These properties allow the equatorial region to act as a waveguide (Holton 1992). Rather than neglecting Coriolis effects, a linear approximation, the equatorial Beta-plane approximation, generally is used in theoretical applications. The Coriolis parameter is estimated as f ≈ βy where β ≡ 2Ω/a, Ω is the angular velocity of the earth, and a is the radius of the earth (Holton 1992). Using this approximation for the Coriolis parameter in the linearized shallow water model equations, dispersion relations can be obtained for eastward- and westward-moving equatorial waves, e.g. equatorial Rossby waves and Kelvin waves. With the advent of the satellite era and the consequent availability of global datasets, it became possible to study organized convection in the tropics in terms of the dominant wavenumbers and frequencies by spectral analysis of outgoing longwave radiation (OLR), brightness temperature, and precipitable water. Prior to the availability of satellite data, spectral analysis was employed using widely scattered rawinsonde wind data, which were particularly sparse in the tropics. In addition, long time series of rawinsonde data were not available. Despite these limitations, the first observations of stratospheric mixed-Rossby-gravity waves (Yanai and Maruyama 1966) and stratospheric Kelvin waves (Wallace and Kousky 1968) were made in the 1960’s, nearly coincident with the advent of equatorial wave theory (Matsuno 1966; Lindzen 1967). Since these groundbreaking studies, much work has been done to investigate the structure of observed equatorial waves (e.g. Reed et al. 1977), but the role of these waves in tropical weather is still not well understood. Wheeler and Kiladis 5

(1999) identified the dominant temporal and spatial scales of zonally propagating disturbances using wavenumber-frequency analysis of satellite-observed OLR in the equatorial region. After an estimated red background spectrum was removed, the spectral peaks were found to follow the dispersion curves of idealized equatorial wave theory in a shallow water model. In Figure 1.1 (taken from Wheeler and Kiladis, 1999) the prominent convectively-coupled equatorial wave types found were Kelvin, equatorial Rossby (ER), mixed Rossby-gravity (MRG), eastward inertio-gravity (EIG), and westward inertio-gravity (WIG) waves. Also shown in Figure 1.1 are the Madden-Julian Oscillation (MJO) and tropical depression-type waves (TD-type, also known as African easterly waves). Roundy and Frank (2004) conducted a climatology of the five dominant convectively-coupled wave types in the equatorial region: the MJO, ER waves, MRG waves, Kelvin waves, and TD-type disturbances. Whereas Wheeler and Kiladis (1999) applied symmetric and asymmetric filters to their data, Roundy and Frank (2004a) studied the cross-equatorial asymmetry of wavenumber-frequency filtered OLR and precipitable water. Using Fourier filters to separate the time- longitude components for each of the five spectral bands, they found that the MJO and ER waves account for more than 15% of the variance in total OLR within the tropics and more than 50% of local OLR variability in the tropics for time scales of two days or longer. Because OLR is correlated with convection, these waves significantly affect tropical weather through diabatic heating. This thesis focuses on the how tropical disturbances interact with the MJO, which was discovered by Madden and Julian in 1971 and is discussed in detail in the following section.

1.3 Characteristics of the MJO

Using spectral analysis techniques, Madden and Julian (1971) computed spectra and cross spectra from rawinsonde data at Canton Island and Balboa in Panama. Two bands of high coherence were found at Canton Island: the 5-6 day period westward propagating Rossby wave and a 12-100 day period eastward propagating mode which peaked in the 40-50 day range. Under further examination, surface pressure was positively correlated with low-level zonal winds and negatively correlated with upper-level zonal winds. Temperatures in the mid- 6 to upper-levels were also negatively correlated with surface pressure. Madden and Julian (1971) proposed that the oscillation was due to eastward movement of large- scale convection originating in the Indian Ocean. This 40-50 day oscillation is now well-known as the MJO and is the leading mode of intraseasonal variance in large portions of the tropics. Over the past few decades, the MJO has been the subject of a plethora of research endeavors. Origin and propagation mechanisms are still not well understood, although its structure is fairly well resolved. Hendon and Salby (1994) discussed the MJO life cycle. This cycle begins with enhanced convection over the Indian Ocean and reduced convection over the northwest Pacific, generally favoring the summer hemisphere (Roundy and Frank 2004). The region of enhanced convection is related to convergence at 850 mb and divergence at 200 mb. The MJO then propagates eastward at a phase speed of about 5ms −1 when convection is strong and about 10ms −1 when convection is weak (Hendon and Salby 1994). While its signal is most evident in the Indian and Pacific Oceans, the circulation affects the entire tropical troposphere. The MJO is a wavenumber 1-2 oscillation, i.e. there are one or two equatorial regions with enhanced convection and one or two equatorial regions with suppressed convection. Interannually, the MJO varies distinctly, both in the number of cycles and in the strength of the anomalies. Slingo et al. (1999) conjectured that the El Ni˜noSouthern Oscillation (ENSO) was associated with this variability: weak El Ni˜no or neutral years tend to feature a more active MJO. Figure 1.2 (from the Climate Prediction Center’s online MJO summary) depicts the MJO propagating from the Indian Ocean to the western Pacific Ocean. During the beginning of the MJO life cycle, there is enhanced convection over the Indian Ocean and suppressed convection over the Pacific Ocean. An idealized 3-D structure of the MJO is presented in Figure 1.3 (also from the Climate Prediction Center’s online MJO summary). Clear skies, associated with the suppressed convection and due to an enhancement of the mean westerly vertical wind shear, allow incoming shortwave radiation to reach the sea surface and slightly increase SSTs, which can be seen in Figure 1.2. On the eastern edge of the convection, there is convergence in the lowest levels associated with westerly wind anomalies encountering easterly wind anomalies. Friction enhances convergence within the 7 boundary layer. This low-level frictional mass and moisture convergence triggers new convection, and the oscillation advances eastward. In contrast, low-level moisture divergence leads to dissipation of the MJO on its western flank. After passage of the convective region, SST is slightly decreased due to reduced levels of shortwave radiation reaching the ocean (Linacre and Geerts 1997). In the Northern Hemisphere summer, these SST anomalies can be as high as 1−2 ◦C in the strongest northeast Pacific MJO events and tend to lead enhanced convection by about 10 days (Maloney and Kiehl 2001). Nakazawa (1988) studied the structure within the enhanced convective region. This large region of disturbed weather organizes into one or more cloud clusters, or super could clusters. Super cloud clusters travel eastward at 5-10 ms −1 within the MJO as it progresses eastward. Each super cloud cluster consists of smaller cloud clusters that form on its eastern edge and move westward, generally dissipating after one to two days. These smaller cloud clusters have the potential to organize into tropical cyclones and, therefore, are of particular interest in the study of tropical cyclogenesis, especially in basins where nonlinear interactions between wave types or an unstable background state allow disturbances to grow.

1.4 Previous Studies of Wave Interaction with Tropical Cyclones

Tropical cyclogenesis in the north Atlantic and northeast Pacific is often linked to African easterly waves (e.g., Avila and Pasch 1992; Landsea 1993). The strongest easterly waves can propagate across the Atlantic Ocean into the northeast Pacific despite loss of coupled-convection over the cooler central Atlantic (Shapiro 1986). While at least half of the genesis cases in the northeast Pacific and north Atlantic basins can be traced back to an upstream easterly wave, there remains the question of why some systems develop and others do not. Composite and observational case studies have shown relations between the MJO and cyclogenesis in the northwest Pacific and north Indian Oceans (Liebmann et al. 1994), the south Indian Ocean (Bessafi and Wheeler 2006), and the Australian region (Hall et al. 2001). Gray (1979) noted that northeast Pacific hurricanes tend to display 8 active and inactive phases on the timescale of the MJO. Since then, numerous studies have examined the role of the MJO in modulating cyclone activity in that region. Molinari et al. (1997) studied the mean potential vorticity (PV) gradient in the 1991 northeast Pacific hurricane season. This particular season was chosen because of its distinct active and inactive episodes. Three regions displayed a meridional sign reversal in PV which satisfies the Charney-Stern instability condition. Noting that the fluctuations in strength of the low-pass filtered PV sign reversal and low-pass filtered OLR occurred on the timescale of the MJO, they proposed that the MJO entered the northeast Pacific and Caribbean Sea, enhancing convection in these regions. Positive PV anomalies then amplified in the northeast Pacific along the ITCZ and over northern South America, causing the PV sign reversal to also amplify. During this time period, waves were able to grow on the unstable background state and cyclogenesis was more frequent. In a subsequent study, Molinari and Vollaro (2000) looked at the interaction between the ITCZ, the MJO, and easterly waves in cyclogenesis events during the same 1991 season. As in the previous paper, the MJO moved into the northeast Pacific and Caribbean and intensified the sign reversal in the meridional PV gradient. Upstream African easterly waves then were able to grow; the tropical depressions could all be traced back to easterly waves. However, only one storm formed in an easterly wave that preserved its magnitude across central America. It was also noted that each depression appeared to form within the ITCZ. The genesis locations of storms shifted eastward as the MJO progressed eastward. As a result, the MJO was determined to be a major influence in the 1991 season. Maloney and Hartmann (2000a) studied the life cycle of the MJO using 20- 80 day bandpass-filtered 850 mb zonal wind for May-November 1979-1995. The first two principal components of this field were calculated from the leading two empirical orthogonal functions (EOFs), which combined to explain 54% of the variance. An MJO index then was derived from a linear combination of these principal components. Phases were assigned to key events in the dataset, and the events were averaged to create composite events for each phase. The study focused on two phases: one with westerly wind anomalies over the northeast Pacific and the other with easterly wind anomalies. These correspond to positive and 9 negative convective anomalies, respectively. Based on these composites, Maloney and Hartmann (2000a) found that tropical cyclones were four times more likely to form when 850 mb wind anomalies were westerly. Cyclones also became more intense in westerly periods. Dynamical aspects contributing to increased basin activity were diagnosed as cyclonic horizontal shear and low vertical wind shear, both of which are conducive to hurricane formation and intensification. A two part follow-up study by Maloney and Hartmann (2001) and Hartmann and Maloney (2001) examined the role of barotropic dynamics in the northeast Pacific. In part one, Maloney and Hartmann (2001) demonstrated that eddies grow via barotropic eddy kinetic energy (EKE) conversion when 850 mb zonal wind anomalies are westerly. EKE generation during this time also peaked over Central America and the Gulf of Mexico. The Caribbean displayed no peaks in EKE generation though, indicating that upstream wave growth was unlikely. Local dynamics in the northeast Pacific might be sufficient for eddy growth. Analyzing each of the EKE generation terms, zonal variations in the zonal wind were the largest contributor to EKE growth while meridional variations in the zonal wind were less important. This suggested that zonal variations in the zonal wind led to eddy growth and that low-level cyclonic flow might not be crucial for eddy growth in the northeast Pacific. Rather, disturbances may be able to grow in regions where there is convergence in the low-level zonal winds. A similar analysis in the northwest Pacific indicated that all EKE production terms contributed significantly to eddy growth during the convective phase of the MJO. In the second part of this study, Hartmann and Maloney (2001) simulated the MJO using a stochastic barotropic model forced with uniformly spatial vorticity. This model substantiated that barotropic eddy kinetic energy generation is positively correlated with westerly 850 hPa winds. Some work has been done on the influence of the MJO on tropical cyclogenesis in the north Atlantic Ocean, though its influence is largely agreed to be less considerable than in the Pacific and Indian Oceans. However, the MJO often reintensifies as it enters the northeast Pacific. This is asscoiated with enhanced convection, either due to an unstable background state (Molinari et al., 1997) or barotropic growth of eddy kinetic energy (Maloney and Hartmann, 2001). Maloney and Hartmann (2000b) conducted a statistical study similar to their 2000 10 paper for the north Atlantic over the period May-November 1949-1997. Westerly wind anomalies over the northeast Pacific corresponded with cyclonic vorticity anomalies in the Gulf of Mexico. Despite the high mountains in Central America and Mexico, where the MJO is largely believed to weaken significantly due to the topography and the lack of forcing (Madden and Julian 1994), cyclogenesis cases in the north Atlantic during westerly winds anomalies outnumbered cases during easterly wind anomalies by nearly a factor of three. Gulf coast landfalls also significantly increased in westerly episodes. Maloney and Hartmann (2000b) argued that the MJO plays at least as large a role as ENSO in north Atlantic cyclone activity. Bossak (2004) likewise contended that the MJO was a factor in the August and September 2004 surge in north Atlantic hurricanes. Using spectral analysis and filtering for each of the five dominant wave types in the equatorial region, Frank and Roundy (2006) created composite analyses of wave fields with respect to genesis location in each basin. About 60 − 70% of tropical cyclogenesis cases occurred within the negative MJO-filtered OLR anomalies in each of the six basins. In the northeast Pacific (Figure 1.4), cyclogenesis in the composites occurred on the poleward edge of the westerly wind anomaly and on the leading edge of the eastward propagating low-level cyclonic vorticity anomaly. This corresponded with the region of negative MJO-filtered OLR anomalies. Similar features were found in the north Atlantic for MJO-filtered fields; however, these features were less prominent in the Atlantic. The MJO also reduced westerly vertical wind shear at the genesis location prior to cyclogenesis in all basins. Westerly shear is known to inhibit tropical cyclone intensification (Frank and Ritchie, 2001).

1.5 Goals

Understanding how tropical waves influence the formation and structure of hurricanes is crucial to determining the behavior of storms in the present climate and, therefore, how their behavior may change in future climates. A better understanding of tropical cyclogenesis would aid in the prediction of storm counts and intensity. There are also direct benefits of an improved understanding of wave- storm dynamics to the field of forecasting because tropical waves can be forecasted 11 in real-time using statistical models. The formations of Hurricanes Emily (2005) in the north Atlantic and Fausto (2002) in the northeast Pacific are examined using the WRF model. How the MJO produces the dynamical effects observed in the composites of Frank and Roundy (2006) is the main focus of this research.The goal of this research is to improve understanding of how the circulation anomalies of these unique waves trigger genesis locally within broad climatologically favorable regions. This thesis investigates the role of equatorial waves in tropical cyclogenesis using the Weather Research and Forecast (WRF) model version 2.2 (WRFV2.2) and real data. 12

Figure 1.1. Normalized wavenumber-frequency spectrum for a) antisymmetric and b) symmetric OLR power divided by the background power from Wheeler and Kiladis (1999). Contour interval is 0.1, and shading begins at 1.1 for spectral bands. Dispersion curves from equatorial wave theory are superimposed. 13

Figure 1.2. Equatorial vertical cross section depiction of MJO propagation. Red arrows indicate wind direction, and red (blue) SST labels indicate positive (negative) SST anomalies. Figure is taken from the Climate Prediction Center. 14

Figure 1.3. Idealized 3D structure of the MJO. Blue (red) ovals indicate anticyclonic (cyclonic) circulations. Black arrows indicate wind direction and rising (sinking) motion. Figure is taken from the Climate Prediction Center. 15

Figure 1.4. Composite of MJO-filtered OLR anomalies (contours), MJO-filtered wind anomalies, and total OLR (shading) relative to genesis point and date in the northeast Pacific for a) 850 mb wind anomalies and b) 200 mb wind anomalies. Contours for OLR anomalies are every 2W m −2 beginning with ±1W m −2. (Frank and Roundy, 2006) Chapter 2

Observational Data and Model Output

In the following chapter, the datasets used in this study are described (section 1). Procedures for diagnostic analyses are detailed in section 2. Hurricanes Emily and Fausto are chosen using spectral analysis to determine the MJO convective phase. This research takes advantage of the recently released WRFV2.2 (section 3). Control simulations of both case studies are run; the initial conditions then are perturbed by removing an idealized representation of the MJO, and the case studies are rerun. Details of the methodology employed in calculating this MJO- correlated portion are contained in section 4. An ensemble, elaborated on in section 5, of each case study is run for statistical robustness.

2.1 Datasets

For the diagnostic analyses NCEP/NCAR reanalyses of zonal and meridional component winds on various pressure levels are used. The longterm mean for each dataset, calculated using data from years 1968-1996, was also obtained. These data were provided by NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, on their Web site. The NCEP/NCAR reanalysis 1 project uses analyses combined with forecasts to assimilate data to a 2.5 degree latitude x 2.5 degree longitude global grid. The data are available four times daily. For more information on the assimilation technique, see Kalnay et al. (1996). 17

NOAA Interpolated OLR was also obtained from the NOAA/OAR/ESRL PSD Web site. Missing data are filled by temporal and spatial interpolation which is explained in Liebmann and Smith (1996). OLR is interpolated to a 2.5 degree latitude x 2.5 degree longitude global grid and is available once daily. The longterm mean for this dataset is calculated using daily values for 1979-1995. Global final analyses derived from the Global Forecast System (GFS) model were obtained from DOC/NOAA/NWS/NCEP to initialize the WRF model. These data were prepared operationally by NCEP and are available on 1.0 x 1.0 degree grid with 27 vertical level covering the entire globe. Operational gridded, synoptic surface, upper air, and oceanographic analyses are all included in the dataset. In addition, geographical and terrain data were obtained from the WRF users website.

2.2 Diagnostic analyses

The diagnostic analyses involve filtering of OLR and reanalysis wind data for the MJO. The filter used in this study is for periods of 30-100 days, which is broader than the typical 30-60 days often seen in the literature. Periods of 60-100 days are included in an effort to capture the entire MJO. Madden and Julian (1994) showed that the timescale of the MJO covers a wide range: the observed periods of the zonal wind anomalies at Truk Island were from 22 to 79 days. Knutson et al. (1986) found two cases with periods greater than 79 days. Based on these results, we expand the filter to 30-100 day periods. The filtered fields are studied on a longitude-time grid to diagnose MJO convectively active and convectively inhibitive phases. In addition, the filtered OLR anomaly and zonal wind anomalies are plotted on a latitude-longitude grid just prior to and just after tropical cyclone intensification. Synoptic scale features are also considered in relation to cyclone development.

2.3 WRF Model

WRFV2.2 program code was developed collectively by researchers at the National Center for Atmospheric Research (NCAR), the National Oceanic and 18

Atmospheric Administration, the National Centers for Environmental Predic- tion (NCEP) the Forecast Systems Laboratory (FSL), the Air Force Weather Agency (AFWA), the Naval Research Laboratory, the University of Oklahoma, and the Federal Aviation Administration (FAA). Complete documentation on the WRF model is available to the public on the WRF model users website (http://www.mmm.ucar.edu/wrf/users/). The WRF model solves the fully compressible, nonhydrostatic Euler equations using a mass-based terrain-following vertical coordinate. All scalar variables are conservative. A 3rd order Runge-Kutta time-split integration scheme is employed with smaller timesteps for acoustic and gravity-wave modes (Skamarock et al., 2005). WRF is highly modular; various model physics parameters can be specified by the user. Section 5 further addresses the physics options used in this study.

2.3.1 Model Initialization

First, we perform control simulations for each case. The coarse domain in this study is 54 km horizontal grid spacing with nests of 18 km and 6 km. All simulations are run with 29 vertical levels. Initial conditions are created by horizontally interpolating NCEP final analysis data to the 54 km grid and then vertically interpolating to the terrain-following vertical coordinate. Thus, a hydrostatically balanced 3D grid is created as input to the WRF model (Skamarock et al., 2005). A discussion of the specific domains used for simulations of Hurricane Fausto and Hurricane Emily is contained in Chapters 3 and 4, respectively.

2.3.2 Boundary Conditions and Nesting

For real data cases, WRF specifies the horizontal boundary conditions for the coarse grid by a method often referred to as a relaxation or a nudging. There are two components to the lateral boundary (see Figure 2.1): 1) a specified zone, which is derived from the NCEP final analysis and consists of the outermost grid points of the domain and 2) a relaxation zone, which effectively relaxes the boundary conditions toward the WRF model forecast. Boundary conditions at the top of the model are constant pressure with a vertical velocity of zero (Skamarock et al., 2005). 19

Nesting in the simulations is one way, so the only communication with the nested grid is through its boundary conditions, which are interpolated from the parent grid. These lateral boundary conditions are provided at each coarse grid time step. Hence, the nested grid boundary conditions are created in a similar fashion as those for the coarse grid; however, there is no relaxation zone.

2.4 Modification of Initial Conditions

After performing a control run for each case, the filtered wave anomaly field is removed from the initial conditions. The cases then are rerun. By comparing the simulations, the effects of the MJO on the simulated storm can be determined. Initial conditions are modified by running a linear regression for five years of input data versus an EOF time series of the idealized tropical wave. These EOF time series, calculated using spectral analysis to filter OLR anomalies, were created by Dr. Paul E. Roundy at the University of Albany (personal communication). EOF analysis then is applied to those anomalies; the OLR anomalies are projected onto the first 75 EOFs, which account for 90% of the variance. The atmospheric wave is removed by subtracting the correlated part for each input at each gridpoint. This procedure is a new technique and is an idealized approach designed to remove the MJO wave field. A simulation that is initialized with actual observagtions is chosen because it allows verification of the model results. Difference fields of horizontal wind, vertical velocity, and vertical vorticity are calculated by subtracting the wave-subtracted simulation from the control. By comparing the results of these simulations to the control run we will estimate the effects of the MJO upon the real storm. Owing to the newness of the procedure, the case studies are run as ensembles to increase the number of simulations that can be compared and to ensure that the results are statistically significant. The method used to create the ensembles is described in section 5. EOF analysis was introduced by Lorenz (1956). At its basis, a gridpoint areal weighting factor is applied to a spatial field to yield the prominent spatial patterns. Wilks (1999) outlines the theory of EOF analysis, which is also often referred to as principal component analysis (PCA). PCA decomposes a dataset into orthogonal spatial and temporal patterns with the first component explaining 20 the most variance in the dataset. Each subsequent principal component (PC) and corresponding EOF resolves the largest remaining variance. Generally, PCA is 0 applied to anomaly fields x = x−x. The eigen vectors em of the covariance matrix of x0 determine each PC uniquely. From Wilks (1999), the principal components um are determined by

K T 0 X 0 um = e mx = ekmxk, m = 1, ..., M. (2.1) k=1 In Equation 2.1, the m th principal component is calculated from the anomaly

field xk which consists of K variables. The first eigen vector corresponds to the largest eigen value and points in the direction where x0 displays the most variability. Consecutive eigen values decrease in magnitude and explain decreasing amounts of the variance in the anomaly field (Wilks, 1999). The eigen vectors of x0 comprise an orthogonal predictor set: fewer eigen vectors are needed if the data exhibits redundancy. The first 75 PCs are included in this study. A multiple linear regression is run for each model input variable (e.g. geopotential height, temperature, meridional and zonal velocities,...) versus the PCs for the years 2001-2005. In common statistical terminology, the model input variable is the predictand, y; the PCs then correspond to 75 predictor variables, x. Hence, multiple linear regression approximates the MJO-correlated portion of the predictor as

yˆi = b0 + b1x1 + b2x2 + ... + bK xK ,K = 1, ..., 75. (2.2)

This multiple linear regression is performed for all i = 120 model input variables at every gridpoint. Partial correlation coefficients compose b: the correlated portion of each model input with each EOF is contained in this matrix. To remove the

MJO wave anomaly,y ˆi is simply subtracted from each model input yi.

2.5 Ensemble Technique

There are a wide range of methods commonly employed in creating ensembles. Lorenz (1963) found that perfect numerical weather prediction forecasts will never 21

Table 2.1. Description of Ensemble Members (EMs)

Microphysics Scheme Cumulus Parameterization CEM1/MEM1 WRF Single-Moment 6 class scheme Kain-Fritsch CEM2/MEM2 WRF Single-Moment 6 class scheme Betts-Miller-Janjic CEM3/MEM3 Eta microphysics Kain-Fritsch CEM4/MEM4 Eta microphysics Betts-Miller-Janjic CEM5/MEM5 WRF Single-Moment 6 class scheme fully explicit

exist because we will never know the initial conditions well enough, and the introduction of the smallest of errors to the initial conditions can cause the model to diverge rapidly from the truth. Ensemble forecasting was introduced in an attempt to improve individual forecasts by running multiple different numerical forecasts. There are two common methods for creating ensembles: 1) modify initial conditions and 2) use different models and/or model configurations (Leith, 1974). The ensemble technique used in this thesis is akin the latter method, or a multi- model ensemble. Five ensemble members are created by varying the microphysical and cumulus parameterizations. This is outlined in Table 2.1. The C denotes control simulations, and the M denotes MJO-removed simulations. In interpreting the results, the merits of each parameterization are taken into account. Modeling of tropical cyclones is known to be highly sensitive to physical processes (e.g., Puri and Miller, 1990; Braun and Tao, 2000; Davis and Bosart, 2002). This study employs two convective parameterization schemes, the Betts- Miller-Janjic scheme (hereafter BMJ) and the Kain-Fritsch Eta scheme (hereafter KF). Convective parameterizations attempt to represent unresolved vertical fluxes due to convective updrafts and downdrafts. Theoretically, convective parameterizations should only be applied at coarse grid resolutions: in the 5-10 km grid resolution range, the model begins to resolve some convection explicitly. As a result, a fifth ensemble member is initialized with the coarse domain of ensemble member 1 and run with fully explicit convection at 6km. All simulations are run using the Yonsei University (YSU) planetary boundary layer scheme, the Rapid Update Cycle (RUC) model land surface model, the Rapid Radiative Transfer Model (RRTM) longwave radiation scheme, and the Dudhia shortwave radiation scheme. 22

Both the BMJ and KF schemes are commonly used for modeling tropical convection; in fact, the BMJ scheme is an updated version of the Betts-Miller scheme, which was developed from observations of deep tropical convection in the Global Atmospheric Research Program (GARP) Atlantic Tropical Experiment (GATE). The BMJ scheme is a convective adjustment scheme: vertical profiles are adjusted toward reference equilibrium temperature and moisture profiles that closely resemble a moist adiabat. The convective precipitation then is equivalent to the precipitable water difference associated with adjustment to this reference profile. The BMJ scheme parameterizes downdrafts, but updrafts and entrainment are not directly included. In contrast, the KF scheme is a mass conserving, one dimensional plume model, which includes entrainment, updrafts, and downdrafts (Skamarock et al., 2005). Multiple previous studies have observed that convective parameterizations profoundly affect hurricane track and intensity (e.g., Puri and Miller, 1990; Davis and Bosart, 2002; Prater and Evans, 2002) Microphysics parameterizations are also varied within the ensemble system: the WRF Single-Moment 6 class scheme (hereafter WSM6) and the Eta microphysics scheme (also known as the Eta Ferrier scheme). Lord et. al. (1984) found evidence that tropical cyclone simulations of cyclone intensity were sensitive to microphysical processes, specifically that mesoscale features are more prominent when ice processes are included. In contrast, Wang (2002) observed that, while rain only cloud structures were significantly different from cloud structures that included ice species, the intensities were only weakly sensitive to microphysics parameterization. Both the WSM6 and Eta schemes include multiple water phases. The Eta scheme includes four forms: cloud water, rain, cloud ice, and precipitation ice; WSM6 uses 6 water phases, including graupel and the physical processes associated with it. 23

Figure 2.1. Specified and relaxation zones for WRF boundary conditions. Each row and column within these zones is 1 grid cell thick. Chapter 3

Analysis of the MJO’s Influence on Hurricane Fausto - East Pacific

After a notable lull in tropical cyclone activity in the northeast Pacific, Hurricane Fausto formed from an African easterly wave that maintained its structure across the north Atlantic and into the east Pacific. As the sixth named storm of the 2002 northeast Pacific hurricane season, Fausto formed within the monsoon trough in a region of enhanced low-level vorticity. This genesis region lay on the leading edge of a convectively active phase of the MJO. Hurricane Fausto then moved into a favorable large-scale environment for convection with respect to the MJO. This set-up does not match the composites of Frank and Roundy (2006), where the genesis location of the composite storm was poleward of an MJO-filtered westerly wind anomaly and within an MJO-filtered OLR anomaly. As shown later in this chapter, the simulations of Hurricane Fausto differ from the observations of the actual storm, especially when considering the intensity of the hurricane. Although the WRF model simulations of Hurricane Fausto do not verify well with the actual observations, the simulation results can still be compared within the framework of the model. Because of the differences between the simulated storms and the actual storm, the results may not be directly applicable to interpreting the observed behavior. However, the effect of the MJO on the simulated storm can be studied, and the results can be applied to real tropical cyclones in the atmosphere. This chapter examines the effects of the MJO on the simulations of Hurricane Fausto. 25

3.1 Synoptic History

Track and intensity of Hurricane Fausto are displayed in Figure 3.1 (available online at the Unisys). Fausto was designated a tropical depression by NHC at 12Z Aug 21. Forecasters determined that the system developed from an African easterly wave. The tropical depression formed with the monsoon trough in the northeast Pacific basin. After genesis, Hurricane Fausto remained to the south of the Pacific subtropical ridge, following a westerly track as the ridge deepened and moved westward. This ridge continued to deepen, and Fausto turned slightly on a northwesterly track and was steered around the periphery of the ridge. Fausto then quickly developed into an intense hurricane and reached a minimum sea level pressure (SLP) of 936 mb at 18Z Aug 24. Subsequently, the hurricane tracked over cold northeast Pacific waters and weakened to tropical depression strength. Despite continuing to track over cold SSTs, the remnants of Hurricane Fausto maintained tropical characteristics and briefly reintensified into a tropical storm at 00Z Sep 2 after passage under an upper level low. The system was then determined to be extratropical at 00Z Sep 3.

3.2 Analysis of MJO wave structure during cy- clogenesis and intensification

Previous to the development of Hurricane Fausto, more than three weeks had elapsed with no named storms in the northeast Pacific basin. Figure 3.2 displays longitude-time plots, i.e., h¨ovmollers, of MJO-filtered OLR anomalies as well as the locations of development of all named storms in the northeast Pacific for the period May-October 2002. The OLR anomalies are calculated by subtracting the long-term mean of daily OLR from daily OLR data at each location. These anomalies then are filtered for the MJO and are averaged over the tropical Northern Hemisphere between the equator and 20 ◦N. Warm colors indicate that MJO- filtered OLR anomalies are above average (convection is reduced) while cold colors indicate that the anomalies are below average (convection is enhanced). Hurricane Fausto led the way into an active period for the northeast Pacific basin; Tropical Storm Genevieve and Hurricane Hernan also formed at the end of August within 26 a convectively favorable region of MJO-filtered OLR anomalies. H¨ovmollers of MJO-filtered 850 mb and 200 mb wind anomalies in the tropical Northern Hemisphere are shown in Figures 3.3a and 3.3b, respectively. These wind anomalies are calculated via the same method used for the MJO-filtered OLR anomalies in Figure 3.2. Westerly (easterly) wind anomalies at 850 mb (200 mb) are associated with convectively active MJO periods, which correlate with the onset of MJO-filtered negative OLR anomalies in Figure 3.2, although they slightly lag the OLR anomalies. Altogether, it can be determined that Hurricane Fausto formed on the leading edge of a convectively active MJO (negative OLR anomaly).

3.3 Model Domain

Figure 3.4 displays the model domains used for simulations of Hurricane Fausto. Each coarse grid simulation was initialized at 00Z Aug 20. After 24 hours of spin- up, the 18 km and 6 km resolution domains were initialized at 00Z Aug 21. These initialization times were chosen because coarse domain initializations prior to 00Z Aug 20 did not develop a tropical cyclone.

3.4 Comparison of Initial Fields

The unfiltered OLR and unfiltered 850 mb wind anomaly fields are plotted in Figure 3.5. These fields are then filtered to remove the high frequency structures. Specifically, the fields are filtered for the 10-100 day period, which corresponds to the timescale of the MJO. The results of this filtering at 00Z Aug 20 are plotted on a latitude-longitude grid in Figure 3.6. This analysis time coincides with the initialization of the coarse grid. To determine the performance of the linear regression used to modify the initial conditions, the difference fields in initialized 850 mb and 200 mb winds (control minus MJO-removed) are shown in Figures 3.7a and 3.7b, respectively. These wind fields were subtracted from the initial conditions to create the MJO-removed simulations. Ideally, because the linear regression was run versus an EOF time series of the MJO, these fields should be well correlated with the wind anomalies in Figure 3.6. This assumption, however, is limited 27 because only 5 years of the EOF data were used in the linear regression. A longer time series could improve the results. Despite this limitation, examination of the difference in initialized 850 mb zonal winds in Figure 3.7a shows good agreement with MJO-filtered 850 mb wind anomalies in Figure 3.6. Hurricane Fausto formed within an MJO-filtered 850 mb easterly wind anomaly. The cyclone then moved into a region of low-level convergence related to the intersection of westerly and easterly wind anomalies associated with the MJO, which can be inferred from the MJO-filtered negative OLR anomalies in Figure 3.6. From Figure 3.7a, it is also evident that a meridional component of the 850 mb wind was removed from the initial conditionsi in some portions of the domain. The MJO generally is not believed to have a strong meridional component; however, some studies have observed northward propagating cloud zones to 30 ◦N in the Indian monsoon (Madden and Julian, 1994). Most of the north-south flow subtracted from 850 mb initial winds is confined to the portion of the coarse domain that lies in the extratropics. In the extratropics, effects of the MJO are not well understood. Some global models predict a 50 day oscillation in angular momentum. Other studies have revealed 40-50 day variations in angular momentum, geopotential height, and winds, but any connection between these observations and the MJO is at this point tenuous (Madden and Julian, 1994). Considering Figure 3.7b, a 200 mb westerly wind was removed from the initial conditions in the vicinity of the storm. Figure 3.3b showed that Hurricane Fausto formed during a transition from westerly MJO-filtered 200 mb wind anomalies to easterly MJO-filtered 200 mb winds. Upper-level easterly wind anomalies are associated with the convectively active phase of the MJO, whereas upper-level westerly wind anomalies are associated with the convectively inhibitive phase of the MJO. As in the 850 mb wind field, there are strong meridional components of the 200 mb winds that are removed from portions of the initial 200 mb wind field. In fact, to the northwest of Fausto, an upper-level anticyclonic field was removed from the initial conditions. Again, there is little knowledge of the effects of the MJO outside of the tropics, where the MJO may play a role and may induce north-south wind anomalies. 28

3.5 Evaluation of Genesis Parameters

Gray (1968, 1979) first identified the necessary climatological parameters for tropical cyclogenesis, which are listed in Chapter 1. Some studies have attempted to derive a single genesis parameter index from these dynamic and thermodynamic large-scale environmental conditions (e.g. Zehr 1992; DeMaria et al. 2001; Camargo et al. 2007). This section will examine values of ensemble mean dynamic and thermodynamic parameters in the simulations for parameters that are commonly used in these genesis parameter indices. The full tables with values of these indices for each ensemble member can be found in Appendix A.

3.5.1 Dynamic Parameters

The dynamic parameters are contained in Table 3.1 (and Appendix A.1) and include zonal 200-850 mb vertical wind shear (VWSu), meridional 200-850 mb

VWS (VWSv), 850 mb relative vorticity (ζ850), and 850 mb divergence (D850). Each of these values are averaged over an 8 ◦ latitude by 8 ◦ longitude box that is centered on the location of the storm at 00Z Aug 21, when the 6 km domain is initialized. According to observations, Hurricane Fausto reached tropical depression strength at 12Z Aug 21. These genesis parameters thus correspond to measurements 12 hours prior to genesis. The final row in Table 3.1 contains the ensemble mean genesis parameters of the MJO-removed simulations averaged over the area where the respective storm in each control simulation formed. This was done to illustrate the change in the parameters caused by removing the MJO at a specific location. The most striking difference between the genesis parameters in Table 3.1 is the strong increase in vertical wind shear when the MJO is removed. Both meridional and zonal 200-850 mb VWS are larger in the MJO-removed simulations.

Differences in ζ850 and D850 are small in comparision. The storms in the MJO- removed simulations formed within a less favorable environment with respect to shear. However, all of the simulations formed within regions of comparable regions of large-scale ζ850 and D850. Environmental shear magnitudes and directions for each simulation are listed in Table A.4 of Appendix A. Shears are averaged over boxes of 4 ◦, 8 ◦, and 29

Table 3.1. Summary of Appendix Table A.1: Large-scale Dynamic Tropical Cyclogenesis Parameters Averaged Over Genesis Area at 00Z Aug 21 for Each 6 km Ensemble Mean. The parameters include zonal vertical wind shear (VWSu), meridional vertical wind shear (VWSv), 850 mb relative vorticity (ζ850), and 850 mb divergence (D850).

Ensemble Mem- VWSu VWSv ζ850 D850 ber Control -0.93 ms −1 -4.73 ms −1 3.72 × 10−5s −1 −8.10 × 10−6s −1 MJO-removed -2.66 ms −1 -9.50 ms −1 3.77 × 10−5s −1 −5.33 × 10−6s −1 MJO-removed for -7.06 ms −1 -9.81 ms −1 3.82 × 10−5s −1 −6.88 × 10−6s −1 Control Area

12 ◦ latitude and longitude. These environmental shears values are northeasterly in all but two of the calculations: when averaged over the 8 ◦ latitude by 8 ◦ longitude box, the VWS vectors in CEM2 and CEM4 are almost directly from the north. Ritchie and Frank (2007) performed simulations of tropical cyclones with a variable Coriolis environment. In these simulations, a northwest ”beta shear” of 8 ms −1 was observed, which, when added to the environmental shear, resulted in a modification in the shear over the tropical cyclone. Hence, an environmental VWS from the southeast would reduce the negative effects of VWS on the cyclone core and would be the most favorable type of shear for a northern hemisphere cyclone. It is speculated that the increase in northerly shear negatively affected the intensification of Hurricane Fausto in the MJO-removed simulations.

3.5.2 Thermodynamic Parameters

Thermodynamic variables are also important to tropical cyclogenesis. These variables include high relative humidities (RH) at low-levels and high SSTs. An additional parameter, 700 mb θe, is included in this study. Rotunno and Emanuel (1987) presented a thermodynamic argument for tropical cyclogenesis. Initially, a mesoscale vortex cannot intensify due to convective downdrafts in the core which bring low equivalent potential temperature (θ e) air into the boundary layer and suppress development of a warm core. In a follow-up modeling study, Emanuel

(1989) proposed that this effect could be overcome if there is elevated θ e in the 30

Table 3.2. Summary of Appendix Table A.2: Large-scale Thermodynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area at 00Z Aug 20 for Each 6 km Ensemble Mean. The parameters include SST, 1000-700 mb average RH, and 700 mb θe. Ensemble Mem- SST 1000-700 mb RH 700 mb θe ber Control 29.4 ◦C 80.19% 40.6 ◦C MJO-removed 28.8 ◦C 80.85% 40.6 ◦C MEM1-control 28.9 ◦C 83.89% 40.3 ◦C

middle troposphere just above the surface vorticity maximum. In fact, this was determined a necessary condition for tropical cyclogenesis.

Ensemble mean SSTs, 1000-700 mb RHs, and 700 mb θe, all calculated using the same method as was used for the dynamical parameters above, are given in Table 3.2. Thermodynamic parameters for each ensemble member can be found in Table A.2 in Appendix A. These thermodynamic parameters showed little variability within the control and MJO-removed ensembles. Based on these results, the thermodynamical parameters appear to be less important in affecting the location of cyclogenesis. For the case of Hurricane Fausto, these thermodynamic parameters were not significantly modified by removing the MJO.

3.6 Storm Track

Tracks of the location of minimum SLP for each 6 km resolution ensemble member are displayed in Figure 3.8. It is important to note that storm tracks are not of equal lengths because the cyclones in some simulations weakened, resulting in undetectable centers. From this figure, two features are immediately clear. First, the location of minimum SLP for control simulations and MJO-removed simulations at the time of initialization of the 6 km resolution domain are distinctly offset. In addition, the cumulus parameterization had a marked effect on storm track. These differences will be discussed in this section and then compared to ensemble member intensities in section 7. Due to the modification of the initial wind fields, all MJO-removed simulations form the cyclone to the northeast of all control simulations. The modification in 31 the initial wind field transfers the region of low-level relative vorticity anomaly to the north-northeast, as shown in Figure 3.9. This figure also clearly shows the monsoon trough, which Hurricane Fausto formed within. In the MJO-removed simulation, the monsoon trough takes on a more northeasterly tilt, and the cyclone forms to the northeast of the control simulation. This initial location shift results in significant differences in the evolving storm intensities, which are further described in the following section. Figure 3.10 shows control ensemble mean and MJO- removed ensemble mean tracks of the locations of minimum SLP. Ovals represent one standard deviation in latitude and longitude within each ensemble. Differences in initial locations of the control and MJO-removed ensemble mean minimum SLP are quite robust when the standard deviation in location is considered. Spread with regards to track within each ensemble quickly amplifies in time. Comparing the control and MJO-removed ensemble mean locations at each subsequent point in time, the two tracks are greater than one standard deviation apart. Therefore, in the case of Hurricane Fausto, a statistically significant difference in storm track is observed when the MJO is removed from the initial conditions. Further examination of Figure 3.8 reveals that the convective parameterization also has a strong influence on storm track. Control simulations run with the KF scheme and with fully explicit convection most closely follow the observations of a west-northwesterly track. All simulations run with the BMJ scheme move the storm on a more northerly track. While control simulations run with KF scheme and with fully explicit convection closely follow the observed track at the beginning of the storm life cycle, storm movement is too slow. Furthermore, all simulations recurve (shift from a track toward the west to a track toward the east) earlier than the observations of Hurricane Fausto. Altogether, the WRF model has trouble modeling this particular storm. Earlier initialization times failed to produce a storm, suggesting that initial conditions were not well represented in the final analysis at those times. Tropical cyclone motion is influenced by many physical and dynamical pro- cesses, which are summarized in Chan (2005). Intense, symmetric storms are predominantly steered by the environmental flow, which can be approximated by the deep layer mean (DLM) wind. Generally, storm movement is poleward of the DLM wind due to advection of planetary vorticity by the meridional flow 32 associated with the PV anomaly. This effect is commonly referred to as the Beta effect (Chan 2005). Indeed, some insight into the storm tracks can be gained from the deep layer mean wind, calculated by averaging 850-200 mb winds within a 16 ◦ latitude by 16 ◦ longitude box centered on the location of minimum SLP. Shown in Figure 3.11 are the DLM wind vectors for CEM1 (light green), MEM1 (cyan), and CEM2 (red). These time series are different lengths because some of the cyclones became too weak to determine the location of minimum SLP. From Figure 3.11. the environmental flow has an easterly component in all of the simulations. For CEM1 and MEM1, the storms track approximately with or slightly polward of the DLM wind. However, both CEM1 and MEM1 begin to recurve at the end of their lifetimes, diverging from the DLM wind. Altogether, CEM1 and MEM1 tracks agree quite well with Beta drift theory. In comparing the DLM winds to storm track in CEM2, the environmental flow and Beta drift theory are insufficient in explaining the resulting northerly storm track. Other factors must be playing a role in this simulation. Chan (2005) further discusses the present state of knowledge of tropical cyclone motions. Another estimate of tropical cyclone motion is that storms tend to move toward the largest PV tendency. While steering and the beta effect often combine to explain cyclone motion, in a baroclinic environment, inclusion of latent heat release and vertical wind shear can modify the PV distribution. Therefore, an accurate forecast for cyclone motion is contingent on an accurate representation of convection. Because numerical simulations often depend on convective parameterizations, the differences in the distribution of convection that are predicted by these schemes can lead to differences in cyclone motion (Chan, 2005). Davis and Bosart (2002) showed that uncertainties in convective parameterization techniques can lead to larger uncertainties in storm track than those caused by uncertainties in initial conditions. Because the the tropical atmosphere is approximately barotropic, PV gradients are generally largest near the tropopause (Holton, 1992). Wu and Emanuel (1993, 1995a,b) argue that the large scale circulation associated with upper level anticyclones and their associated PV gradients can affect the flow at much lower levels. Furthermore, Shapiro and Franklin (1999) found that large-scale effects may be just as important in track forecasts as anomalies within 1500 km of the 33 storm center. A comparison between CEM1 and CEM2 200 mb heights, 200 mb winds, and SLP is presented in Figures 3.12 and 3.13, valid at 06Z Aug 21 and 18Z Aug 21, respectively. From Figure 3.12, the surface low in both simulations is located at the southern portion of a strong trough in the subtropics. In CEM1, the cyclone is embedded in strong northeasterly flow at 200 mb, especially at the southern periphery of the surface low (Figure 3.12a). At that same time, CEM2 is embedded in a weaker upper-level flow that is predominantly easterly (Figure 3.12b). By 12 hours later, these features have intensified (Figure 3.13). From Figure 3.13a, the core of Hurricane Fausto is under the influence of strong north-northeasterly flow at 200 mb. An upper-level low has also formed, which probably allows the storm to elude the influence of the upper level trough to its north. The surface low in CEM2 simulation in Figure 3.13b, however, remains embedded in the upper-level trough. The simulated storm in CEM2 is probably more strongly influenced by the upper-level trough than the simulated storm in CEM1. These upper-level features continue to influence the storm track as both of these simulations progress in time, and the simulated storm in CEM2 takes a more northerly track. Partly as a result of the diverging storm tracks, the simulations run with the BMJ convective parameterization encounter an intense cyclone which forms at the nest boundary. Anomalous convection at the boundaries is a known issue in WRF, especially when the boundary includes complex terrain (Stauffer 2007, personal communication). Interaction of binary vortices was first described by Fujiwhara (1923, 1931). When two tropical cyclones come within about 900 kilometers of one another, they rotate in a counter-clockwise fashion and tend to converge. If one of the cyclones is stronger, it will dominate and eventually enfold in the weaker vortex. This cyclone interaction is also observed in the subtropics and midlatitudes, where observations have shown cyclone pairs interacting at distances of up to 2000 km (Ziv and Alpert 1995). Figures 3.14a, 3.14b, and 3.15 show 3 hour convective precipitation (mm) and SLP at 00Z, 06Z, and 12Z Aug 24, respectively, for the CEM2 simulation. In Figure 3.14a, an intense cyclone develops over northwestern Mexico at the northern boundary of the 6 km domain. Figures 3.14b and 3.15 then show that the two cyclones begin to rotate counterclockwise around each other. In many of the simulations run with the BMJ convective parameterization, similar 34 vortex interactions occur between Hurricane Fausto and this anomalous convection at the boundaries. The KF scheme produces anomalous convection as well, and binary vortex interactions may explain why the simulated systems recurved.

3.7 Storm Intensity

Due to diverging storm tracks, each simulation of Hurricane Fausto encounters varying synoptic conditions. Minimum SLP versus time for each ensemble member, the ensemble means, and observations are shown in Figure 3.16. It is hypothesized that the WRF model does a poor job of simulating the observed intensity of Hurricane Fausto because of uncertainty in the initial conditions. Observations are sparse over the eastern and central Pacific, which can negatively affect the performance of NWP models and modeling studies. The differences in storm intensity between simulations can be partially explained by differing storm tracks. Both VWS and SST were calculated by averaging values within a radius of 4 ◦ of the location of minimum SLP. The cyclone MEM1 tracks over cooler SSTs into a region of stronger vertical wind shear (not shown). Likewise, for each respective ensemble member, the MEM simulation tracks farther north into a less favorable environment than does the corresponding CEM simulation. In the MEM1 simulation, an increase to nearly 30ms −1 of total VWS is followed by a period of dramatic weakening. Meanwhile, the CEM1 simulation maintained its intensity under about 25ms −1 of total VWS. Owing to such high shear values and because all CEM and MEM simulations encounter higher shear and cooler SST environments as the storms progress, further investigation of storm evolution is necessary to uniquely determine how the MJO affected the genesis and intensification of Hurricane Fausto for each ensemble member. Tropical cyclones often form within the ITCZ as discussed in Chapter 1. All 6 km mesh simulated storms were embedded in the ITCZ at 00Z Aug 21, but some of the simulated storms detached sooner than others as they tracked farther northward. Figures 3.17 and 3.18 depict examples of this detachment with plots of 850 mb relative vorticity and SLP at 00Z Aug 22 and 00Z Aug 23, respectively. The upper panel in each of these figures is CEM1, and the lower panel is MEM1. 35

In Figure 3.17, both hurricanes are embedded in the ITCZ and are entraining positive vorticity from the ITCZ into their cores. By 00Z Aug 23 (Figure 3.18), the hurricane in CEM1 was able to maintain its intensity through entrainment of positive vorticity whereas the hurricane in MEM1 migrated farther northward out of the monsoon trough and rapidly weakened in an unfavorable synoptic environment. Similar results were found for ensemble members 3 and 5, which were run with the KF convective scheme and fully explicit convection, respectively. This argument also can be used to explain the dramatically weaker intensities in the BMJ simulations since the cyclones detach much earlier from the ITCZ in comparison to the KF simulations. However, further examination of the parameterized precipitation is enlightening. Figure 3.19 shows 3 hour convective precipitation ending at 03Z Aug 21 (the first 3 hours of the 6 km simulation) for CEM1 and CEM2. Previous work by Prater and Evans (2002) and Davis and Bosart (2002) found that the KF scheme provides a more realistic representation of precipitation, which is linked to latent heating. Latent heating within the cyclone core and descending motion within the eye combine to produce a warm core, large temperature gradients and therefore high potential energy, which can be converted into kinetic energy (Frank, 1987). By hydrostatic balance, this heating also causes the surface pressure to drop. Heating at large radii reduces the efficiency of this spin-up mechanism. Based on Figure 3.19, in comparison the BMJ scheme, the KF scheme predicts heavier rainfall and rainfall concentrated near the core of the hurricane. Precipitation predicted by the BMJ scheme is more widespread and less intense, which would lead to less concentrated latent heat release and a weaker cyclone. The next chapter investigates a North Atlantic case study, Hurricane Emily. The MJO is usually considered to have less of an influence in the North Atlantic than in the Pacific and Indian Oceans since the MJO tends to diminish in strength as it crosses the mountains of central America. Examination of a more neutral case gives insight into the methodology used in this study. 36

Figure 3.1. Best track of Hurricane Fausto (from Unisys) 37

May02 100

80

Jun02 60

40 Jul02

20

Aug02 0 Time

−20

Sep02 −40

−60 Oct02

−80

−100 0 50 100 150 200 250 300 350 Longitude

Figure 3.2. MJO-filtered OLR anomalies (W m −2) averaged over 0 − 20 ◦N. Circles represent genesis locations for all named storms in the east Pacific basin in 2002. Hurricane Fausto is denoted by the black circle. 38

Figure 3.3. MJO-filtered a) 850 mb wind and b) 200 mb wind anomalies (ms −1) averaged over 0 − 20 ◦N. Circles represent genesis locations for all named storms in the east Pacific basin in 2002. Hurricane Fausto is denoted by the black circle. 39

Figure 3.4. WRF model domain configuration. Domains 1, 2, and 3 are 54 km, 18 km, and 6 km horizontal resolution, respectively. 40

Figure 3.5. Unfiltered OLR (W m −2) and 850 mb wind anomalies (ms −1) at 00Z Aug 20 41

Figure 3.6. MJO-filtered OLR (W m −2) and 850 mb wind anomalies (ms −1) at 00Z Aug 20 42

Figure 3.7. Initial 850 mb and 200 mb wind (ms −1) difference field (control minus MJO-removed) for 54km domain. The red circle indicates the approximate location of Hurricane Fausto at this time. 43

Figure 3.8. Tracks of minimum SLP for 6 km ensemble members. Tracks are marked every 12 hours with circles (asterisks) for control (MJO-removed) simulations beginning at 00Z Aug 21 44

Figure 3.9. Comparision of a) CEM1 and b) MEM1 850mb relative vorticity s −1) and 850 mb winds (ms −1) at 00Z Aug 21 45

Figure 3.10. Ensemble mean tracks of minimum SLP for 6 km control (blue) and MJO-removed (red) simulations. Tracks are marked every 6 hours with circles beginning at 00Z Aug 21. Ovals represent +/- one standard deviation of latitude and longitude. 46

Figure 3.11. Time series of deep layer (850-200 mb) mean wind (ms −1) for CEM1 (green), MEM1 (blue), and CEM2 (red). Actual tracks for each ensemble member are plotted in black. 47

Figure 3.12. Comparison of 200 mb heights (m), 200 mb winds (ms −1), and SLP (mb) for a) CEM1 and b) CEM2 at 06Z Aug 21 48

Figure 3.13. Comparison of 200 mb heights (m), 200 mb winds (ms −1), and SLP (mb) for a) CEM1 and b) CEM2 at 18Z Aug 21 49

Figure 3.14. CEM2 3 hr convective 3 hr precipitation and SLP ending at a) 00Z Aug 24 and b) 06Z Aug 24 50

Figure 3.15. CEM2 3 hr convective 3 hr precipitation and SLP ending at 12Z Aug 24 51

1010

1000

990 CEM1 CEM2 980 CEM3 CEM4 CEM5 Control Mean 970 MEM1 SLP MEM2 MEM3 MEM4 960 MEM5 MJO−removed Mean Observations 950

940

930 08/21 08/22 08/23 08/24 08/25 08/26 Time

Figure 3.16. Timeseries of minimum SLP (mb) for 6 km ensemble members, ensemble mean, and observations. Circles (asterisks) designate control (MJO-removed) simulations. 52

Figure 3.17. Comparison of 850 mb relative vorticity (s −1) for a) CEM1 and b) MEM1 at 00Z Aug 22 53

Figure 3.18. Comparison of 850 mb relative vorticity (s −1) for a) CEM1 and b) MEM1 at 00Z Aug 23 54

Figure 3.19. Comparison of 3 hour convective precipitation (mm) for a) CEM1 and b) CEM2 ending at 03Z Aug 21 Chapter 4

Analysis of the MJO’s Influence on Hurricane Emily - North Atlantic

Hurricane Emily was the fifth named storm of the 2005 Atlantic hurricane season. One of seven major hurricanes in the record-breaking 2005 season, Hurricane Emily intensified to a minimum pressure of 929 mb before making landfall as a category 4 hurricane on the Yucatan peninsula. Hurricane Emily formed over the central Atlantic Ocean from an African easterly wave. Based on MJO-filtered OLR and wind anomalies, the influence of the MJO on the genesis and intensification of Hurricane Emily was weak to nearly neutral. However, some subtle influences may be observed, and it is useful to compare the case study of Hurricane Emily to the case study of Hurricane Fausto.

4.1 Synoptic History

Best track positions from the National Hurricane Center (NHC) are shown in Figure 4.1. An African easterly wave tracked across the eastern and central Atlantic but showed little promise of organizing into a mesoscale system until 00Z Jul 11, when it was designated a tropical depression. Initially, development was inhibited due to lack of moisture caused in part by a ridge of high pressure to the north. The circulation remained broad, yet surface winds increased to tropical 56 storm strength at 00Z Jul 12. Westerly vertical wind shear was observed during this time period (Franklin and Brown, 2006). Despite continued observations of disorganized convection and no obvious synoptic-scale forcing, Emily continued to strengthen. The system interacted with a 200mb anticyclone and subsequently intensified into a hurricane at 07Z Jul 14 (Franklin and Brown, 2006). Hurricane Emily then turned to the west- northwest and increased speed. During this time period Hurricane Emily moved into a more favorable environment with respect to vertical wind shear and rapidly strengthened, reaching category 4 status as it moved into the southeastern Caribbean. After a brief period of weakening to category 2 status, the hurricane intensified again, reaching a peak intensity of 929 mb with sustained winds of 160 miles per hour at 00Z Jul 17. This peak intensity established Hurricane Emily as the first category 5 storm on record to form in the month of July. Hurricane Emily later went on to make two landfalls, first as a category 4 storm near Cozumel, Mexico, and second as a category 3 storm in northeastern Mexico. This study will focus on the earliest stages of the storm’s life cycle.

4.2 Analysis of MJO wave structure during cy- clogenesis and intensification

In Figure 4.2, H¨ovmollers of MJO-filtered OLR anomalies are plotted for the period May-October 2005. Negative OLR anomalies correspond to deep convection and the convectively active phase of the MJO. Positive OLR anomalies correspond to the convectively inhibitive phase of the MJO. Figure 4.3 displays H¨ovmollers of MJO-filtered 850 mb (top panel) and 200 mb (bottom panel) zonal wind anomalies. Warm (cold) colors indicate westerly (easterly) 850mb zonal wind anomalies. Enhanced convection associated with the MJO generally is located on the leading edge of the 850 mb westerly wind anomaly where there is low-level moisture convergence. Based on these H¨ovmollers, Hurricane Emily developed (was classified as a depression) during a weakly inhibitive phase of the MJO. Emily reached hurricane strength at 00Z Jul 14 during a time when the MJO-filtered OLR 57 and 850 mb zonal winds were favorable for convection and, therefore, cyclogenesis. MJO-filtered 200 mb zonal wind anomalies are also presented in Figure 4.3b. Upper-level MJO-filtered zonal wind anomalies were westerly throughout the lifetime of the storm. These upper-level wind anomalies suggest that the MJO was in a convectively inhibitive phase for tropical cyclogenesis. Due to disagreement among MJO-filtered OLR and wind anomalies, it is difficult to determine whether the MJO was convectively active or inhibitive during the latter part of Hurricane Emily’s lifecycle. However, these H¨ovmoller all indicate that the depression formed within a region where the theory of the effects of the MJO would predict reduced convection.

4.3 Model Domain

Model domains for Hurricane Emily are shown in Figure 4.4. The largest domain was 54 km grid spacing and was initialized at 00Z Jul 10. The nested domains of 18 km and 6 km grid spacing were initialized at 00Z Jul 13. A spin-up time of 72 hours was chosen to allow the vortex to organize since Hurricane Emily was loosely organized at the beginning of its life cycle. These simulations were all run through 06Z Jul 16, therefore encompassing the peak intensity of Hurricane Emily’s life cycle.

4.4 Comparison of Initial Fields

Figure 4.5 displays unfiltered OLR anomalies and unfiltered 850 mb wind anomalies at 00Z Jul 10. To determine the influence of the MJO, these fields were filtered for the MJO, and the results are plotted in Figure 4.6. Difference fields (control minus MJO-removed) of initial 850 mb and 200 mb winds are shown in Figures 4.7a and 4.7b, respectively, for comparison. The approximate location of simulated minimum SLP at 00Z Jul 10 is indicated by the red circle. In Figure 4.7a, an 850 mb westerly wind of approximately 2 ms −1 was removed from the region where Hurricane Emily formed. In contrast, Figure 4.6 shows that the overall MJO-filtered wind anomaly near the genesis location is from the northeast. A low- level westerly wind anomaly, such as is seen in Figure 4.7a, is usually associated 58

Table 4.1. Summary of Appendix Table B.1: Large-scale Dynamic Tropical Cyclogenesis Parameters Averaged Over Genesis Area at 00Z Jul 13 for Each 6 km Ensemble Means. The parameters include zonal vertical wind shear (VWSu), meridional vertical wind shear (VWSv), 850 mb relative vorticity (ζ850), and 850 mb divergence (D850).

Ensemble Mem- VWSu VWSv ζ850 D850 ber Control -4.92 ms −1 -0.21 ms −1 4.62 × 10−5s −1 −8.54 × 10−6s −1 MJO-removed 2.43 ms −1 -1.24 ms −1 3.71 × 10−5s −1 −5.44 × 10−6s −1 MJO-removed for 4.39 ms −1 -1.41 ms −1 3.48 × 10−5s −1 −4.46 × 10−6s −1 Control Area

with an active phase of the MJO. However, the 850 mb winds in Figure 4.7a show divergence in the low-level in the region where Fausto formed, which agrees well with MJO-filtered OLR anomalies in Figure 4.6. In the initial 200 mb wind field, a northeasterly component of the circulation was removed. From Figure 4.3b, spectral analysis suggested a westerly anomaly with respect to the MJO at upper-levels. While the 200 mb winds removed from the initial fields contain an easterly component, there is upper-level convergence, and hence subsidence, in the region where the cyclone formed. The circulations removed from the 850 mb and 200 mb initial wind fields would make cyclogeneis more favorable in the MJO-removed simulations. This corresponds with positive MJO-filtered OLR anomalies at the location where Emily formed in Figure 4.6. As with Hurricane Fausto, there are strong correlations with meridional components of the wind in the midlatitudes where the affects of the MJO are not as well understood.

4.5 Evaluation of Genesis Parameters

Tropical cyclogenesis parameters for each simulation are calculated as in Chapter 3.5. The control and MJO-removed ensemble mean dynamic and thermodynamic parameters are listed in Table 4.1 and Table 4.2, respectively. The genesis parameters are also calculated for the MJO-removed simulations but at the locations where the respective control ensemble member simulated storm 59

Table 4.2. Large-scale Thermodynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area for Each 6 km Ensemble Mean. The parameters include SST, 1000-700 mb average RH, and 700 mb θe. Ensemble Mem- SST 1000-700 mb RH 700 mb θe ber Control 28.1 ◦C 80.83% 39.8 ◦C MJO-removed 28.1 ◦C 80.13% 40.1 ◦C MEM1-control 28.0 ◦C 80.38% 40.0 ◦C

was located. Genesis parameters for individual ensemble members are available in Appendix B. Although VWS is higher on average in the control simulations, the shear is easterly, which is known to be more favorable for cyclogenesis than is westerly shear. In the MJO-removed simulations, the VWS is westerly, which would be less favorable for cyclogenesis, but the shear is weak. Comparing the control and

MJO-removed ensemble mean ζ850 and D850, the control simulations show higher values of large-scale vorticity and divergence in the low-levels, which are also more favorable for cyclogenesis. In the third row, VWS is larger and more westerly in the MJO-removed simulations when calculated over the location where the control simulations formed. Examination of the thermodynamics parameters for tropical cyclogenesis in Table 4.2 reveals that the thermodynamic environments are comparable in the control and MJO-removed simulations. These results agree with the results in the case of Hurricane Fausto in that the MJO more significantly affects the dynamical parameters for cyclogenesis. Modification of the VWS by the MJO appears to be the dominant factor in affecting tropical cyclogenesis and the further intensification of the cyclone.

4.6 Storm Track

Minimum SLP tracks for each 6 km resolution ensemble member and NHC observations are shown in Figure 4.8. All tracks start at 00Z Jul 13 and end at 06Z Jul 16. Consistent among all ensemble members, the WRF model predicted the location of Hurricane Emily to be approximately 2 ◦N and 6 ◦E of its actual 60 location at 00Z Jul 13. The simulated storms subsequently track too far north and more slowly than the observations. In examining Figure 4.9, control (red) and MJO-removed (blue) ensemble mean tracks, the minimum SLP in the MJO- removed simulations was slightly farther west and progressed on a more westward track initially in comparison with the control simulations. On 00Z Jul 14, the MJO-removed simulations then tracked more northward and converged toward the control simulated storm tracks. In addition, ensemble spread among MJO-removed simulations is larger with respect to track as evidenced by larger standard deviations in minimum SLP locations. In Figure 4.8, MEM3 and MEM5 simulations track farther south than the remaining ensemble members. Altogether, minimum SLP tracks for control and MJO-removed simulations are in close agreement among all ensemble members in both the control and MJO-removed simulations. DLM winds, calculated over a 16 ◦ latitude by 16 ◦ longitude box centered on the storm, for CEM1 and CEM2 are plotted in Figure 4.10. The simulated storms all track generally with the DLM winds. The Beta gyres may be included in these DLM wind vectors since the simulated storms are well-organized and fairly symmetric, which would account for the storm tracks closely matching the DLM winds. Thus, it is concluded that the simulated storm tracks of Hurricane Emily are explained by synoptic-scale steering and the Beta effect.

4.7 Storm Intensity

Minimum SLP for each ensemble member, control and MJO-removed ensemble means, and observations are plotted versus time in Figure 4.11. As with the case of Hurricane Fausto, the KF convective scheme predicts a more intense hurricane than the BMJ convective scheme. This again can be explained by predicted convective precipitation in CEM1 and CEM2, shown in Figures 4.12a and 4.12b, respectively. The KF scheme predicts intense rainfall near the hurricane core (Figure 4.12a), whereas the BMJ scheme predicts weaker rainfall (Figure 4.12b). Latent heating near the cyclone core due to heavy rainfall allows the hurricane in CEM1 to spin-up more quickly. The effect of the MJO on the intensification of Hurricane Emily varies for each 61 ensemble member. Intensities are comparable in CEM1 and MEM1, but MEM3 reaches a greater intensity than CEM3. In contrast, the hurricanes predicted by the BMJ convective parameterization reach a greater intensity in the control simulations than the hurricanes in the MJO-removed simulations. This appears to be a result of low RH air penetrating into the low-levels near the cyclone core. Figure 4.13 displays vertical cross sections of RH at 12Z Jul 15 for CEM2 (top) and MEM2 (bottom). These cross sections are north-south through the location of minimum SLP in each simulation. In both simulations, low RH air is present in the vicinity of the cyclone core. However, in the MJO-removed simulation (MEM2), the region of low RH is broader. In addition, the area of RH that exceeds 90% covers a smaller area in the lower-levels within the cyclone core. As a result, the thermodynamic environment is less favorable for intensification of the simulated cyclone. Similar results were found in the cross sections of the simulated storms in CEM4 and MEM4. Comparing control and MJO-removed ensemble mean minimum SLP, the control simulations are slightly more intense than the MJO-removed simulations. However, the difference in ensemble mean minimum SLPs is only a few millibars. Based on the predicted tracks and intensities for Hurricane Emily, it appears that the MJO had only a weak influence on the cyclone’s structure. Further discussion of the MJO’s influence on Hurricanes Emily and Fausto is contained in Chapter 5. 62

Figure 4.1. Best track of Hurricane Emily (from Unisys) 63

Figure 4.2. MJO-filtered OLR anomalies (W m −2) averaged over 0 − 20 ◦N. The locations of Hurricane Emily at 00Z Jul 11 and 00Z Jul 14 are denoted by orange and magenta hurricanes, respectively. 64

Figure 4.3. MJO-filtered a) 850 mb wind and b) 200 mb wind anomalies (ms −1) averaged over 0 − 20 ◦N. The locations of Hurricane Emily at 00Z Jul 11 and 00Z Jul 14 are denoted by orange and magenta hurricanes, respectively. 65

Figure 4.4. WRF model domain configuration. Domains 1, 2, and 3 are 54 km, 18 km, and 6 km horizontal resolution, respectively. 66

Figure 4.5. Unfiltered OLR (W m −2) and 850 mb wind anomalies (ms −1) at 00Z Jul 10 67

Figure 4.6. MJO-filtered OLR (W m −2) and 850 mb wind anomalies (ms −1) at 00Z Jul 10 68

Figure 4.7. Initial a) 850 mb and b) 200 mb wind (ms −1) difference field (control minus MJO-removed) for the 54 km domain. The red circle indicates the approximate location of Hurricane Emily at this time. 69

Figure 4.8. Tracks of minimum SLP for 6 km ensemble members. Tracks are marked every 12 hours with circles (asterisks) for control (MJO-removed) simulations beginning at 00Z Aug 21 70

Figure 4.9. Ensemble mean tracks of minimum SLP for 6 km control (blue) and MJO- removed (red) simulations. Tracks are marked every 6 hours with circles beginning at 00Z Aug 21. Ovals represent +/- one standard deviation of latitude and longitude. 71

Figure 4.10. Time series of deep layer (850-200 mb) mean wind (ms −1) for CEM1 (green), MEM1 (light blue). The black vectors indicate storm motion for each simulation. 72

Figure 4.11. Timeseries of minimum SLP (mb) for 6 km ensemble members, ensemble mean, and observations. Circles (asterisks) designate control (MJO-removed) simulations. 73

Figure 4.12. Comparison of 3 hour convective precipitation (mm) for a) CEM1 and b) CEM2 ending at 06Z Jul 13 74

Figure 4.13. Comparison of north-south cross sections of RH (%) through the center of minimum SLP for a) CEM2 and b) MEM2 ending at 12Z Jul 15 Chapter 5

Summary and Conclusions

In Chapters 3 and 4, results from case studies of Hurricane Fausto and Hurricane Emily were presented. Based on these case studies some conclusions can be drawn about ways in which the MJO can influence tropical cyclogenesis and intensification. Section 1 compares differences in intensities in the two case studies using a Student’s t-test. Conclusions and future work are outlined in sections 2 and 3, respectively.

5.1 Comparing Intensities in the Case Studies Using a Student’s t-test

In Chapters 3 and 4, ensemble mean tracks for Hurricane Fausto and Hurricane Emily, respectively, were discussed. To evaluate whether the differences in intensities are statistically significant, a Student’s t-test is used. The t-test in this study is a test for the significance of the difference in the means of two correlated samples. Thus, the null hypothesis is that this difference in the means is zero. The differences, D, in the mean SLPs are calculated by D = XC − XM , where XC and

XM are minimum SLPs averaged over the length of the control and MJO-removed simulations, respectively. Therefore, a negative value of D means that the control simulation was more intense (lower minimum SLP). The Student’s t-test is applied to each ensemble member and to the ensemble mean. P-values and confidence intervals (CIs) for these calculations are shown in 76

Table 5.1. Student T-test P-values and CIs for Hurricanes Fausto and Emily Fausto p-values Fausto CIs Emily p-values Emily CIs CEM1/MEM1 0.0018 -∞:-2.93 0.5708 -9.95:5.54 CEM2/MEM2 0.0008 -∞:-1.14 0.0209 -20.31:-1.74 CEM3/MEM3 0.0000 -∞:-6.95 0.4105 -5.19:12.51 CEM4/MEM4 0.0009 -∞:-0.96 0.0513 -13.96:0.041 CEM5/MEM5 0.0000 -∞:-13.75 0.0243 -11.87:-0.86 Ensemble Means 5.5063 × 10−11 -∞:-6.40 1.63 × 10−8 -5.75:-3.41

Table 5.1. The null hypothesis is often rejected for p-values less than 0.05, i.e. a statistical significance level of 95%. CIs are based on a significance level of 95% and represent the interval of estimates in D for this significance level. If the CI spans zero, then D cannot be deemed significantly nonzero. A one-tailed test was used for the case of Hurricane Fausto because the expectation was that the control mean SLPs would be lower than the MJO-removed mean SLPs. For Hurricane Emily, there was no expectation about SLPs, and a two-tailed Student’s t-test was performed. Statistically significant p-values and CIs are highlighted in red. For Hurricane Fausto, each ensemble member and the ensemble mean D values were significantly nonzero, meaning that the intensities in the control simulations were significantly greater than the intensities in the MJO-removed simulations. The CI of D for the ensemble mean indicates that, on a 95% confidence level, this difference in intensities exceeds 6 mb of central minimum SLP. This result suggests that the MJO significantly affected the genesis and intensification of Hurricane Fausto. In the more neutral MJO case of Hurricane Emily, two of the ensemble members and the ensemble mean had significantly nonzero D values. Neither of the simulations run with the KF cumulus parameterization had significantly different means, while one of the simulations run with the BMJ cumulus parameterization and the simulation run with fully explicit convection were significantly nonzero. It is possible that Hurricane Emily was slightly aided in intensification just after formation by a weak convectively active phase of the MJO. A more favorable large-scale environment then allowed the hurricane to intensify more quickly. If this is true, the timing of genesis of Hurricane Emily may have been influenced by the MJO. Despite some organization, the easterly wave may not have been 77 able to organize initially into a tropical depression within an unfavorable shear environment. However, diasagreement within the ensemble suggests that the model physics were more important in the case of Hurricane Emily. In addtion, if the threshold for statistical significance is increased to 99%, then none of the ensemble members for Hurricane Emily have significantly different means. A larger ensemble or an ensemble created by varying initial conditions would be desirable before coming to any conclusions about the influence of the MJO on Hurricane Emily.

5.2 Conclusions

This thesis presents an analysis of the influences of the MJO in two case studies: Hurricane Fausto (2002) in the East Pacific and Hurricane Emily (2005) in the North Atlantic. A five member ensemble was created by varying model physics in WRF. First, control simulations were run for each case. The initial conditions then were modified by removing the MJO statistically correlated components of the initial fields, calculated via multiple linear regression of each field versus the first 75 PCs of the MJO-filtered wave field. After this procedure, each ensemble member was rerun, thus creating a five member ensemble to compare the influence of the MJO for each hurricane. Based on the research in this study, it appears that the MJO strongly affected the timing and location of genesis for Hurricane Fausto. The subsequent development of Fausto in the MJO-removed simulations then was affected by harsher large-scale environments since the storms in these simulations were located farther to the north in a more baroclinic environment. In simulations run with KF convective parameterization and with fully explicit convection, the hurricanes also detached from the ITCZ sooner and weakened. Another factor that affected hurricane track and intensity was the Fujiwhara effect. Apparently anomalous convection at the boundaries, related to a problem with WRF nesting, caused the storms in the simulations to recurve and rotate counterclockwise around an intense low pressure system over Mexico at the northeast boundary of the domain. In many of the simulations, the remnants of Hurricane Fausto were then wrapped into this very intense system. The anomalous convection may have been amplified by complex terrain at the northern and eastern 78 boundaries. For the case of Hurricane Emily, the influence of the MJO was more subtle. Hurricane Emily intensified into a strong hurricane in all of the simulations due to a favorable large-scale environment and very warm SSTs. Unlike Hurricane Fausto, the simulations all took very similar storm tracks. However, there were obvious differences in intensity for each ensemble member, although there was no agreement as to whether the hurricane in each ensemble member became more or less intense when the MJO was removed. Hence, the effects of the MJO on Hurricane Emily likely were less important than in the case of Hurricane Fausto. Further analysis of differences in intensity revealed that Hurricane Fausto was inhibited from intensifying in the MJO-removed simulations. MJO-removed simulations were at least 6 mb weaker than control simulations on a 95% significance level, and each ensemble member was significantly weaker. As for Hurricane Emily, only two ensemble members had significantly different intensities, both with weaker intensities when the MJO was removed. Furthermore, the ensemble mean minimum SLPs were significantly weaker, on a 95% significance level, by about 3-5 mb. Due to small number of ensemble members, this result is probably a result of the model physics rather than an artifact of the MJO. There are several conclusions that can be drawn from the research presented in this thesis: 1) A 6 km resolution WRF simulation can capture variations in location and orientation of the ITCZ and monsoon trough. For the case study of Hurricane Fausto, the monsoon trough and its associated low-level relative vorticity were displaced to the north when the MJO was removed from the initial conditions. 2) The MJO caused significant variation in VWS. These variations affected the genesis location and intensification of the simulated storms. Therefore, numerical weather prediction models must be able to model circulations such as the MJO to accurately predict the timing and location of cyclogenesis and the subsequent intensification of the cyclone. 3) The WRF model is able to simulate sensitivity of tropical cyclogenesis to to VWS. In both of the case studies of Hurricane Emily and Hurricane Fausto, the simulated storms were highly sensitive to VWS. 4) A multiple linear regression of the first PCs of a wave field versus a time 79 series of GFS final analysis data can be used to remove the correlated components of that wave from the GFS final analysis data used to initialize a model. This was verified by good agreement between MJO-filtered wind anomalies and the calculated MJO-correlated components of the initial analysis field for the case of Hurricane Fausto. 5) Based on the performance of the methodology used to remove the MJO from the initial conditions, the MJO was more prevalent in the initial fields in the east Pacific case study of Hurricane Fausto than in the central Atlantic case study of Hurricane Emily. Whether this is because the MJO is weaker in the central Atlantic cannot be determined since only two cases were studied.

5.3 Future Work

To build upon the results in this study, further research of both case studies may be desirable, especially in the case of Hurricane Emily. An ensemble created by varying initial conditions is particularly appealing. Additional case studies in the East Pacific and North Atlantic could make the results more robust. Case studies in the West Pacific, Indian Ocean, and South Pacific, where the MJO appears to more strongly influence tropical weather, would also be enlightening. Similar research studies with ER waves, MRG waves, and TD-type disturbances may also lead to a better understanding of the roles of equatorial waves in the cyclogenesis and intensification of hurricanes. An analysis of the effect of ER waves on the evolution of Hurricanes Fausto and Emily is currently underway. Appendix A

Large-scale Tropical Cyclone Genesis Parameters for Hurricane Fausto

Table A.1. Large-scale Dynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area for Each Ensemble Member

Ensemble Member VWSu VWSv ζ850 D850 CEM1 -1.81 ms −1 -3.73 ms −1 3.73 × 10−5s −1 −1.65 × 10−5s −1 CEM2 0.32 ms −1 -5.41 ms −1 3.64 × 10−5s −1 −1.35 × 10−5s −1 CEM3 -1.73 ms −1 -4.74 ms −1 3.83 × 10−5s −1 −1.10 × 10−5s −1 CEM4 0.62 ms −1 -5.69 ms −1 3.62 × 10−5s −1 −1.33 × 10−5s −1 CEM5 -2.04 ms −1 -4.08 ms −1 3.79 × 10−5s −1 −1.76 × 10−5s −1 MEM1 -3.61 ms −1 -10.05 ms −1 3.73 × 10−5s −1 −1.22 × 10−5s −1 MEM2 -1.22 ms −1 -8.69 ms −1 3.63 × 10−5s −1 −9.52 × 10−6s −1 MEM3 -3.61 ms −1 -9.74 ms −1 3.98 × 10−5s −1 −7.87 × 10−6s −1 MEM4 -1.18 ms −1 -9.00 ms −1 3.82 × 10−5s −1 −1.03 × 10−5s −1 MEM5 -3.66 ms −1 -10.03 ms −1 3.712 × 10−5s −1 −1.18 × 10−5s −1 MEM1-control -7.63 ms −1 -9.96 ms −1 3.74 × 10−5s −1 −1.95 × 10−5s −1 MEM2-control -5.81 ms −1 -9.45 ms −1 3.89 × 10−5s −1 −1.28 × 10−5s −1 MEM3-control -8.33 ms −1 -9.96 ms −1 3.82 × 10−5s −1 −1.80 × 10−5s −1 MEM4-control -5.84 ms −1 -9.80 ms −1 3.89 × 10−5s −1 −1.89 × 10−5s −1 MEM5-control -7.70 ms −1 -9.90 ms −1 3.75 × 10−5s −1 −1.83 × 10−5s −1 81

Table A.2. Large-scale Thermodynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area for Each Ensemble Member Ensemble Member SST 1000-700 mb RH 700 mb θe CEM1 29.4 ◦C 79.35% 40.7 ◦C CEM2 29.3 ◦C 81.67% 40.3 ◦C CEM3 29.3 ◦C 78.13% 40.9 ◦C CEM4 29.3 ◦C 81.42% 40.3 ◦C CEM5 29.3 ◦C 80.37% 40.6 ◦C MEM1 28.81 ◦C 81.79% 40.6 ◦C MEM2 28.81 ◦C 80.26% 40.5 ◦C MEM3 28.81 ◦C 80.26% 40.8 ◦C MEM4 28.81 ◦C 80.32% 40.6 ◦C MEM5 28.81 ◦C 81.61% 40.4 ◦C MEM1-control 28.87 ◦C 82.63% 40.3 ◦C MEM2-control 28.87 ◦C 82.88% 40.3 ◦C MEM3-control 28.87 ◦C 84.72% 40.4 ◦C MEM4-control 28.87 ◦C 83.05% 40.2 ◦C MEM5-control 28.87 ◦C 86.17% 40.3 ◦C

Table A.3. Large-scale Divergence Averaged over Genesis Area for Each Ensemble Member Ensemble Member D950 D925 D850 CEM1 −5.93 × 10−5s −1 −6.02 × 10−5s −1 −1.65 × 10−5s −1 CEM2 −5.30 × 10−5s −1 −5.32 × 10−5s −1 −1.35 × 10−5s −1 CEM3 −5.87 × 10−5s −1 −5.92 × 10−5s −1 −1.10 × 10−5s −1 CEM4 −5.27 × 10−5s −1 −5.27 × 10−5s −1 −1.33 × 10−5s −1 CEM5 −5.83 × 10−5s −1 −6.00 × 10−5s −1 −1.76 × 10−5s −1 MEM1 −6.83 × 10−5s −1 −6.38 × 10−5s −1 −1.22 × 10−5s −1 MEM2 −5.43 × 10−5s −1 −5.66 × 10−5s −1 −9.52 × 10−6s −1 MEM3 −6.58 × 10−5s −1 −6.03 × 10−5s −1 −7.87 × 10−6s −1 MEM4 −5.55 × 10−5s −1 −5.48 × 10−5s −1 −1.03 × 10−5s −1 MEM5 −6.83 × 10−5s −1 −6.36 × 10−5s −1 −1.18 × 10−5s −1 MEM1-control −5.32 × 10−5s −1 −5.58 × 10−5s −1 −1.95 × 10−5s −1 MEM2-control −4.63 × 10−5s −1 −4.58 × 10−5s −1 −1.28 × 10−5s −1 MEM3-control −5.16 × 10−5s −1 −5.48 × 10−5s −1 −1.80 × 10−5s −1 MEM4-control −4.63 × 10−5s −1 −4.58 × 10−5s −1 −1.89 × 10−5s −1 MEM5-control −5.20 × 10−5s −1 −5.47 × 10−5s −1 −1.83 × 10−5s −1 82

Table A.4. Large-scale VWS Averaged Over 4, 8, and 12 degree Genesis Areas for Each Ensemble Member Ensemble Member VWS4 VWS8 VWS12 CEM1 5.86 ms −1 58 ◦ 4.14 ms −1 26 ◦ 6.42ms −1 11 ◦ CEM2 4.82 ms −1 54 ◦ 5.42 ms −1 357 ◦ 7.09ms −1 17 ◦ CEM3 6.23 ms −1 60 ◦ 4.84 ms −1 30 ◦ 6.53 ms −1 13 ◦ CEM4 4.90 ms −1 50 ◦ 5.72 ms −1 354 ◦ 7.49ms −1 18 ◦ CEM5 7.14 ms −1 55 ◦ 4.56 ms −1 27 ◦ 6.65ms −1 10 ◦ MEM1 9.81 ms −1 37 ◦ 10.67 ms −1 20 ◦ 11.75ms −1 20 ◦ MEM2 7.41 ms −1 41 ◦ 9.39 ms −1 33 ◦ 11.62ms −1 13 ◦ MEM3 10.12 ms −1 33 ◦ 10.39 ms −1 20 ◦ 11.98ms −1 23 ◦ MEM4 6.46 ms −1 53 ◦ 9.08 ms −1 7 ◦ 10.74ms −1 9 ◦ MEM5 9.84 ms −1 38 ◦ 10.68 ms −1 20 ◦ 11.73ms −1 20,◦ MEM1-control 16.95 ms −1 65 ◦ 12.55 ms −1 37 ◦ 12.93ms −1 25 ◦ MEM2-control 13.12 ms −1 62 ◦ 10.87 ms −1 34 ◦ 11.73ms −1 19 ◦ MEM3-control 17.05 ms −1 67 ◦ 12.98 ms −1 40 ◦ 13.26ms −1 30 ◦ MEM4-control 12.69 ms −1 60 ◦ 11.41 ms −1 31 ◦ 11.69ms −1 15 ◦ MEM5-control 16.79 ms −1 65 ◦ 12.54 ms −1 38 ◦ 12.92ms −1 26,◦ Appendix B

Large-scale Tropical Cyclone Genesis Parameters for Hurricane Emily

Table B.1. Large-scale Dynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area for Each Ensemble Member

Ensemble Member VWSu VWSv ζ850 D850 CEM1 0.33 ms −1 -0.51 ms −1 4.51 × 10−5s −1 −8.01 × 10−6s −1 CEM2 0.73 ms −1 -1.86 ms −1 4.19 × 10−5s −1 −1.05 × 10−5s −1 CEM3 -5.27 ms −1 0.33 ms −1 5.01 × 10−5s −1 −8.83 × 10−6s −1 CEM4 0.04 ms −1 -0.60 ms −1 4.30 × 10−5s −1 −7.36 × 10−6s −1 CEM5 -3.73ms −1 1.60 ms −1 5.11 × 10−5s −1 −8.02 × 10−6s −1 MEM1 2.82 ms −1 2.24 ms −1 3.92 × 10−5s −1 −8.29 × 10−6s −1 MEM2 2.89 ms −1 -4.72 ms −1 3.48 × 10−5s −1 −5.80 × 10−6s −1 MEM3 2.22 ms −1 3.76 ms −1 4.03 × 10−5s −1 −1.56 × 10−6s −1 MEM4 2.41 ms −1 -4.26 ms −1 3.28 × 10−5s −1 −6.80 × 10−6s −1 MEM5 1.80 ms −1 1.74 ms −1 3.86 × 10−5s −1 −4.75 × 10−6s −1 MEM1-control 4.93 ms −1 1.39 ms −1 3.93 × 10−5s −1 −6.34 × 10−6s −1 MEM2-control 4.16 ms −1 -5.78 ms −1 3.24 × 10−5s −1 −5.53 × 10−6s −1 MEM3-control 4.47 ms −1 2.85 ms −1 3.88 × 10−5s −1 −5.50 × 10−7s −1 MEM4-control 3.66 ms −1 -5.53 ms −1 2.73 × 10−5s −1 −5.53 × 10−6s −1 MEM5-control 4.75 ms −1 0.00 ms −1 3.60 × 10−5s −1 −4.34 × 10−6s −1 84

Table B.2. Large-scale Thermodynamical Tropical Cyclogenesis Parameters Averaged over Genesis Area for Each Ensemble Member Ensemble Member SST 1000-700 mb RH 700 mb θe CEM1 28.1 ◦C 81.66% 39.2 ◦C CEM2 28.1 ◦C 81.28% 39.2 ◦C CEM3 28.0 ◦C 79.23% 40.6 ◦C CEM4 28.1 ◦C 80.41% 39.2 ◦C CEM5 28.1 ◦C 81.56% 40.7 ◦C MEM1 28.1 ◦C 78.93% 40.7 ◦C MEM2 28.1 ◦C 80.78% 39.2 ◦C MEM3 28.2 ◦C 81.15% 40.7 ◦C MEM4 28.1 ◦C 80.51% 39.1 ◦C MEM5 28.1 ◦C 79.28% 40.7 ◦C MEM1-control 27.9 ◦C 79.59% 40.7 ◦C MEM2-control 28.0 ◦C 80.56% 39.1 ◦C MEM3-control 28.1 ◦C 81.55% 40.6 ◦C MEM4-control 28.0 ◦C 80.20% 39.0 ◦C MEM5-control 28.0 ◦C 80.02% 40.5 ◦C

Table B.3. Large-scale Divergence Averaged over Genesis Area for Each Ensemble Member Ensemble Member D950 D925 D850 CEM1 −5.93 × 10−5s −1 −6.02 × 10−5s −1 −1.65 × 10−5s −1 CEM2 −5.30 × 10−5s −1 −5.32 × 10−5s −1 −1.35 × 10−5s −1 CEM3 −5.91 × 10−5s −1 −5.96 × 10−5s −1 −1.10 × 10−5s −1 CEM4 −5.27 × 10−5s −1 −5.27 × 10−5s −1 −1.33 × 10−5s −1 CEM5 −5.83 × 10−5s −1 −6.00 × 10−5s −1 −1.76 × 10−5s −1 MEM1 −6.83 × 10−5s −1 −6.38 × 10−5s −1 −1.22 × 10−5s −1 MEM2 −5.61 × 10−5s −1 −5.52 × 10−5s −1 9.52 × 10−6s −1 MEM3 −6.58 × 10−5s −1 −6.03 × 10−5s −1 −7.87 × 10−6s −1 MEM4 −5.55 × 10−5s −1 −5.48 × 10−5s −1 −1.03 × 10−5s −1 MEM5 −6.83 × 10−5s −1 −6.36 × 10−5s −1 −1.18 × 10−5s −1 MEM1-control −5.32 × 10−5s −1 −5.58 × 10−5s −1 −1.95 × 10−5s −1 MEM2-control −4.59 × 10−5s −1 −4.41 × 10−5s −1 −1.28 × 10−5s −1 MEM3-control −5.16 × 10−5s −1 −5.48 × 10−5s −1 −1.80 × 10−5s −1 MEM4-control −4.63 × 10−5s −1 −4.58 × 10−5s −1 −1.89 × 10−5s −1 MEM5-control −5.20 × 10−5s −1 −5.47 × 10−5s −1 −1.83 × 10−5s −1 85

Table B.4. Large-scale VWS Averaged Over 4, 8, and 12 degree Genesis Areas for Each Ensemble Member Ensemble Member VWS4 VWS8 VWS12 CEM1 5.86 ms −1 58 ◦ 4.14 ms −1 26 ◦ 6.42ms −1 11 ◦ CEM2 4.82 ms −1 54 ◦ 5.42 ms −1 357 ◦ 7.09ms −1 17 ◦ CEM3 6.23 ms −1 60 ◦ 4.84 ms −1 30 ◦ 6.53 ms −1 13 ◦ CEM4 4.90 ms −1 50 ◦ 5.72 ms −1 354 ◦ 7.49ms −1 18 ◦ CEM5 7.14 ms −1 55 ◦ 4.56 ms −1 27 ◦ 6.65ms −1 10 ◦ MEM1 9.81 ms −1 37 ◦ 10.67 ms −1 20 ◦ 11.75ms −1 20 ◦ MEM2 7.41 ms −1 41 ◦ 9.39 ms −1 33 ◦ 11.62ms −1 13 ◦ MEM3 10.12 ms −1 33 ◦ 10.39 ms −1 20 ◦ 11.98ms −1 23 ◦ MEM4 6.46 ms −1 53 ◦ 9.08 ms −1 7 ◦ 10.74ms −1 9 ◦ MEM5 9.84 ms −1 38 ◦ 10.68 ms −1 20 ◦ 11.73ms −1 20,◦ MEM1-control 16.95 ms −1 65 ◦ 12.55 ms −1 37 ◦ 12.93ms −1 25 ◦ MEM2-control 13.12 ms −1 62 ◦ 10.87 ms −1 34 ◦ 11.73ms −1 19 ◦ MEM3-control 17.05 ms −1 67 ◦ 12.98 ms −1 40 ◦ 13.26ms −1 30 ◦ MEM4-control 12.69 ms −1 60 ◦ 11.41 ms −1 31 ◦ 11.69ms −1 15 ◦ MEM5-control 16.79 ms −1 65 ◦ 12.54 ms −1 38 ◦ 12.92ms −1 26,◦ References

Avila, L. A., and R. J. Pasch, 1992. Atlantic tropical systems of 1991. Mon. Wea. Rev., 120, 2688–2696.

Bessafi, M, and M. C. Wheeler, 2006. Modulation of South Indian Ocean tropical cyclones by the Madden-Julian Oscillation and convectively coupled equatorial waves. Mon. Wea. Rev., 134, 638–656.

Bosart, L. F., and J. A. Bartlo, 1991. Tropical storm formation in a baroclinic environment. Mon. Wea. Rev., 119, 1979–2013.

Bossak, B. H., 2004. ”X” marks the spot: Florida is the 2004 hurricane bull’s eye. Eos Trans. AGU, 85 (50). 541.

Bracken, W. E., and L. F. Bosart, 2000. The role of synoptic-scale flow during tropical cyclogenesis over the North Atlantic Ocean. Mon. Wea. Rev., 128, 353–376.

Braun, S. A., and W.-K. Tao, 2000. Sensitivity of high-resolution simulation of Hurricane Bob (1991) to planetary boundary layer parameterizations. Mon. Wea. Rev., 128, 3941–3961.

Briegel, L. M., and W. M. Frank, 1997. Large-scale influences on tropical cyclogenesis in the western North Pacific. Mon. Wea. Rev., 125, 1397–1413.

Camargo, S. J., A. H. Sobel, A. G. Barnston, and K. A. Emanuel, 2007. Tropical cyclone genesis potential index in climate models. Tellus A, 59, 428–443.

Chan, J., 2005. The physics of tropical cyclone motion. Annu. Rev. Fluid Mech., 37, 99–128. 87

Davidson, N. E., and H. H. Hendon, 1989. Downstream development in the Southern Hemisphere monsoon during FGGE/WMONEX. Mon. Wea. Rev., 117, 1458–1470.

Davis, C. A., and L. F. Bosart, 2002. Numerical simulations of the genesis of Hurricane Diana (1984). Part II: Sensitivity of track and intensity prediction. Mon. Wea. Rev., 130, 1100–1124.

DeMaria, M., J. A. Knaff, and B. H. Connell, 1991. A tropical cyclone genesis parameter for the tropical Atlantic. Wea. Forecasting, 16, 219–233.

Emanuel, K. A., 1989. The finite-amplitude nature of tropical cyclogenesis. Mon. Wea. Rev., 46, 3431–3456.

Frank, W. M., 1988. Tropical Cyclone Formation. A Global View of Tropical Cyclones, R. L. Elsberry, Ed., Office of Naval Research, 53–90.

Frank, W. M., and E. A. Ritchie, 2001. Effects of vertical wind shear on the intensity and structure of numerically simulated hurricanes. Mon. Wea. Rev., 129, 2249–2269.

Frank, W. M., and P. E. Roundy, 2006. The role of tropical waves in tropical cyclogenesis. Mon. Wea. Rev., 134, 2397–2417.

Franklin, J. L., and Brown, D. P., 2006. Tropical cyclon report: Hurricane Emily. (http://www.nhc.noaa.gov/pdf/TCR-AL052005 Emily.pdf)

Fujiwhara, S., 1923. On the growth and decay of vortical system. Quart. J. Roy. Meteor. Soc. 49, 75–104.

Fujiwhara, S., 1931. Short note on the behavior of two vortices. Proc. Phys. Math. Soc. Japan, 13, 106–110.

Gill, A. E., 1982. Atmosphere-Ocean Dynamics, Academic Press, 662 pp.

Gray, W. M., 1968. Global view of the origin of tropical disturbances and storms. Mon. Wea. Rev., 96, 669–700. 88

Gray, W. M., 1979. Hurricanes: Their formation, structure, and likely role in the tropical circulation. Meteorology over the Tropical Oceans, D. B. Shaw, Ed., Royal Meteorology Society, 155–218.

Hall, J. D., A. J. Matthews, and D. J. Karoly, 2001. The modulation of tropical cyclone activity in the Australain region by the Madden-Julian Oscillation. Mon. Wea. Rev., 129, 2970-2982.

Hartmann, D. L., and E. D. Maloney, 2001. The Madden-Julian Oscillation, barotropic dynamics, and North Pacific tropical cyclone formation. Part II: Stochastic barotropic modeling. J. Atmos. Sci., 58, 2259–2570.

Hendon, H. H., and M. L. Salby, 1994. The life cycle of the Madden-Julian Oscillation. J. Atmos. Sci., 51, 2225–2237.

Holland, G. J., 1995. Scale interaction in the western Pacific monsoon. Meteor. Atmos. Phys., 56, 52–79.

Holton, J. R., 1992. An Introduction to Dynamics Meteorology, Academic Press, 511 pp.

Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, W. Ebisuzaki, W. Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, A. Leetmaa, R. Reynolds, R. Jenne, D. Joseph, 1996. The NCEP/NCAR 40-year reanalysis project. Bull. Am. Meteorol. Soc., 77, 437–471.

Knutson, T. R., K. M. Weickmann, and J. E. Kutzbach, 1986. Global-scale intraseasonal oscillations of outgoing longwave radiation and 250 mb zonal wind during northern hemisphere summer. Mon. Wea. Rev., 114, 605–623.

Landsea, C. W., 1993. A climatology of intense (or major) Atlantic hurricanes. Mon. Wea. Rev., 121, 1703–1713.

Leith, C. E., 1974. Theoretical skill of Monte Carlo forecasts. Mon. Wea. Rev., 102, 409–418. 89

Liebmann, B., H. H. Hendon, and J. D. Glick, 1994. The relationship between tropical cyclones of the western Pacific and Indian Oceans and the Madden- Julian Oscillation. J. Meteor. Soc. Japan, 72, 401–412.

Liebmann, B., and C. A. Smith, 1996. Description of a complete (interpolated outgoing longwave radiation dataset. Bull. Am. Meteorol. Soc., 77, 1275– 1277.

Linacre, E., and B. Geerts., 1997. Climate and Weather Explained, An Introductions from a Southern Perspective, Taylor & Francis, Inc., 464 pp.

Lindzen, R. S., 1967. Planetary waves on beta planes. Mon. Wea. Rev., 95, 441–451.

Lord S. J., H. E. Willoughby, and J. M. Piotrowicz, 1984. Role of a parameterizaed ice-phase microphysics in an axisymmetric, nonhydrostatic tropical cyclone model. J. Atmos. Sci., 41, 2386–2848.

Lorenz, E. N., 1956. Empirical orthogonal functions and statistical weather pre- diction, Statistical Forecasting Scientific Rep. 1, Department of Meteorology, Massachusetts Institute of Technology, Cambridge, MA, 57 pp.

Lorenz, E. N., 1963. Deterministic nonperiodic flow. J. Atmos. Sci., 20, 130–141.

Love, G, 1985. Cross-equatorial interactions during tropical cyclogenesis. Mon. Wea. Rev., 113. 1499–1509.

Madden, R. A., and P. R. Julian, 1971. Detection of a 40-50 day oscillation in the zonal wind in the tropical Pacific. J. Atmos. Sci., 28, 702–708.

Madden, R. A., and P. R. Julian, 1994. Observations of the 40-50 day tropical oscillation–A review. Mon. Wea. Rev., 122, 814–837.

Maloney, E. D., and D. L. Hartmann, 2000a. Modulation of hurricane activity in the Golf of Mexico by the Madden-Julian Oscillation. Science., 287, 2002– 2004.

Maloney, E. D., and D. L. Hartmann, 2000b. Modulation of eastern North Pacific hurricanes by the Madden-Julian Oscillation. J. Climate., 13, 1451–1460. 90

Maloney, E. D., and D. L. Hartmann, 2001. The Madden-Julian Oscillation, barotropic dynamics, and North Pacific tropical cyclone formation, Part I: observations. J. Atmos. Sci., 58, 2245–2558.

Maloney, E. D., and J. T. Kiehl, 2001. MJO-related SST variations over the tropical eastern Pacific during Northern Hemisphere summer. J. Climate., 15, 675–689.

Maloney, E. D., and S. K. Esbensen, 2003. The amplification of East Pacific Madden-Julian Oscillation convection and wind anomalies during June- November. J. Climate., 16, 3482-3496.

Matsuno, T., 1966. Quasi-geostrophic motions in the equatorial area. J. Meteor. Soc. Japan, 44, 25–43.

McBride, J. L., 1981a. Observational analysis of tropical cyclone formation. Part I: Basic description of data sets. J. Atmos. Sci., 38, 1117–1131.

McBride, J. L., and R. M. Zehr, 1981. Observational analysis of tropical cyclone formation. Part II: Comparison of nondeveloping vs. developing systems. J. Atmos. Sci., 38, 1132–1151.

McBride, J. L., 1981b. Observational analysis of tropical cyclone formation. Part III: Budget analysis. J. Atmos. Sci., 38, 1152–1166.

McBride, J. L., and T. D. Keenan, 1982. Climatology of tropical cyclone genesis in the Australian region. J. Climatol., 2, 13–33.

Mo, K. C., 2000. The association between intraseasonal oscillations and tropical storms in the Atlanic basin. Mon. Wea. Rev., 128, 4097-4107.

Molinari, J. et al., 1997. Potential vorticity, easterly waves, and Eastern Pacific tropical cyclogenesis. Mon. Wea. Rev., 125, 2699–2708.

Molinari, J., and D. Vollaro, 2000. Planetary- and synoptic-scale influence on Eastern Pacific tropical cyclogenesis. Mon. Wea. Rev., 128, 3296–3307.

Montgomery, M. T., and B. F. Farrell, 1993. Tropical cyclone formation. J. Atmos. Sci., 50, 285–310. 91

Nakazawa, T., 1986. Intraseasonal variations in OLR in the tropics during the FGGE year. J. Meteor. Soc. Japan, 64, 17–34.

Pasch, R. J., and L. A. Avila, 1992. Atlantic tropical systems of 1992. Mon. Wea. Rev., 120, 2688–2696.

Prater, B. E., and J. L. Evans, 2002. Sensitivity of modeled cyclone track and structure of Hurricane Irene (1999) to the convective parameterization scheme. Meteor. Atmos. Phys., 80, 103–115.

Puri, K., and J. Miller, 1990. Sensitivity of ECMWF analyses-forecasts of tropical cyclones to cumulus parameterization. Mon. Wea. Rev., 118, 1709–1742.

Reed, R. J., D. C. Norquist, and E. E. Recker, 1977. The structure and properties of African wave disturbances as observed during Phase III of GATE. Mon. Wea. Rev., 105, 317–333.

Ritchie, E. A., and W. M. Frank, 2007. Interactions between simulated tropical cyclones and an environment with a variable coriolis genesis parameter. Mon. Wea. Rev., 135, 1889–1905.

Roundy, P. E., and W. M. Frank, 2004a. A climatology of waves in the equatorial region. J. Atmos. Sci., 61, 2105–2132.

Rotunno, R., and K. A. Emanuel, 1987. An air-sea interaction theory for tropical cyclones. Part II: Evolutionary study using a nonhydrostatic axisymmetric numerical model. J. Atmos. Sci., 44, 542–561.

Roundy, P. E., and W. M. Frank, 2004b. Effects of low-frequency wave interactions on intraseasonal oscillations. J. Atmos. Sci., 61, 3025–3040.

Sadler, J. C., 1976. A role of the tropical upper troposhere in early season typhoon development. Mon. Wea. Rev., 104, 1266–1278.

Shapiro, L. J., 1986. The three-dimenstional structure of synoptic-scale distur- bances over the tropical Atlantic. Mon. Wea. Rev., 114, 1876–1891.

Shapiro, L. J., and J. L. Franklin, 1999. Potential vorticity asymmetries and tropical cyclone motion. Mon. Wea. Rev., 127, 124–131. 92

Skamarock W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, 2005. A description of the Advanced Research WRF Version 2. NCAR Tech. Rep. TN-468, 100pp.

Slingo, J. M., D. P. Rowell, K. R. Sperber, and F. Nortley, 1999. On the predictability of the interannual behaviour of the Madden-Julian Oscillation and its relationship with El Ni˜no. Quart. J. Roy. Meteor. Soc., 125, 583–609.

Wallace, J. M., and V. E. Kousky, 1968. Observational evidence of Kelvin waves in the tropical stratosphere. J. Atmos. Sci., 25, 900–907.

Wang, Y., 2002. An expicit simulation of tropical cyclones with a triply nested movable mesh primitive equation model: TCM3. Part II: Model refinements and sensitivity to cloud microphysics parameterization. Mon. Wea. Rev., 12 3022–3036.

Wheeler, M., and G. N. Kiladis, 1999. Convectively-coupled equatorial waves: Analysis of clouds and temperature in the wavenumber-frequency domain. J. Atmos. Sci., 56, 374–399.

Wheeler, M., G. N. Kiladis, and P. J. Webster, 2000. Large-scale dynamical fields associated with convectively coupled equatorial waves. J. Atmos. Sci., 57, 613–640.

Wu, C., and K. A. Emanuel, 1993. Interaction of a baroclinic vortex with background shear: application to hurricane movement. J. Atmos. Sci., 50, 62–76.

Wu, C., and K. A. Emanuel, 1995a. Potential vorticity diagnostics of hurricane movement: Part I: A case study of Hurricane Bob (1991). Mon. Wea. Rev., 123, 69–92.

Wu, C., and K. A. Emanuel, 1995a. Potential vorticity diagnostics of hurricane movement: Part II: Tropical storm Ana (1991) and Hurricane Andrew (1992). Mon. Wea. Rev., 123, 93–109.

Wilks, D. S., 1995. Statistical Methods in the Atmispheric Sciences, Academic Press, 467pp. 93

Yanai, M. and T. Maruyama, 1966. Stratospheric wave disturbances propagating over the equatorial Pacific. J. Meteor. Soc. Japan, 44. 291–294.

Zehr, R. M., 1992. Tropical cyclogenesis in the western North Pacfic. NOAA Tech. Rep. NESDIS 61, 181pp. [Available from the U.S. Department of Commerce, NOAA/NESDIS, 5200 Auth. Rd., Washington, DC 20233.]

Ziv, B., and P. Alpert, 2005. Rotation of binary cyclones – A data analysis study. J. Atmos. Sci., 52, 1357–1369.