Follow this and additional works at:

Transcription

1 University of Miami Scholarly Repository Open Access Theses Electronic Theses and Dissertations Development and Analysis of the Systematically Merged Atlantic Regional Temperature and Salinity (SMARTS) Climatology for Satellite-Derived Ocean Thermal Structure Patrick C. Meyers University of Miami, Follow this and additional works at: Recommended Citation Meyers, Patrick C., "Development and Analysis of the Systematically Merged Atlantic Regional Temperature and Salinity (SMARTS) Climatology for Satellite-Derived Ocean Thermal Structure" (2011). Open Access Theses This Embargoed is brought to you for free and open access by the Electronic Theses and Dissertations at Scholarly Repository. It has been accepted for inclusion in Open Access Theses by an authorized administrator of Scholarly Repository. For more information, please contact

2 UNIVERSITY OF MIAMI DEVELOPMENT AND ANALYSIS OF THE SYSTEMATICALLY MERGED ATLANTIC REGIONAL TEMPERATURE AND SALINITY (SMARTS) CLIMATOLOGY FOR SATELLITE-DERIVED OCEAN THERMAL STRUCTURE By Patrick C. Meyers A THESIS Submitted to the Faculty of the University of Miami in partial fulfillment of the requirements for the degree of Master of Science Coral Gables, Florida August 2011

3 2011 Patrick C. Meyers All Rights Reserved

4 UNIVERSITY OF MIAMI A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science DEVELOPMENT AND ANALYSIS OF THE SYSTEMATICALLY MERGED ATLANTIC REGIONAL TEMPERATURE AND SALINITY (SMARTS) CLIMATOLOGY FOR SATELLITE-DERIVED OCEAN THERMAL STRUCTURE Patrick C. Meyers Approved: Lynn K. Shay, Ph.D. Professor of Meteorology and Physical Oceanography Terri A. Scandura, Ph.D. Dean of the Graduate School Arthur J. Mariano, Ph.D. Professor of Meteorology and Physical Oceanography Eric W. Uhlhorn, Ph.D. Research Scientist NOAA AOML Hurricane Research Division Miami, Florida

5 MEYERS, PATRICK C. (M.S., Meteorology and Physical Oceanography) Development and Analysis of The Systematically (August 2011) Merged Atlantic Regional Temperature and Salinity (SMARTS) Climatology for Satellite-Derived Ocean Thermal Structure Abstract of a thesis at the University of Miami. Thesis supervised by Professor Lynn Shay. No. of pages in text. (99) A new oceanic climatology to calculate ocean heat content (OHC) was developed for application year-round in the Atlantic Ocean basin. The Systematically Merged Atlantic Regional Temperature and Salinity (SMARTS) Climatology blends temperature and salinity fields from the World Ocean Atlas 2001 (WOA) and Generalized Digital Environmental Model v.3.0 (GDEM) at 1/4 resolution. This higher resolution climatology better resolves features in the Gulf of Mexico (GOM), including the Loop Current and eddy structures, than the previous coarser 1/2 products. Daily mean isotherm depths of the 20 C (D20) and 26 C (D26) (and their mean ratio), reduced gravity (e.g., 2-layer model), mixed layer depth (MLD), and OHC were estimated from the blended climatology. Using SMARTS with satellite-derived surface height anomaly and SST fields, daily values of D20, D26, MLD, and OHC were calculated from 1998 to 2010 using a twolayer model approach. Airborne and ship-deployed expendable BathyThermographs (XBT), long-term moorings, and Argo profiling floats provided the in-situ data to blend and assess the SMARTS Climatology. A clear, direct relationship emerged from the detailed analysis between satellite-derived and in-situ measurements of isotherm depths

6 and OHC. This new climatological approach created a more accurate estimation of isotherm depths and OHC from satellite radar altimetry measurements, which can be used in hurricane intensity forecasts from the Statistical Hurricane Intensity Prediction Scheme (SHIPS). The Mainelli (2000) technique of calculating OHC was reexamined to most accurately project sea surface height anomalies (SSHA) into changes in depths of D20, D26, and MLD. SSHA surface features were tracked to determine realistic drift velocities ingested into the objective analysis algorithm. The former OHC algorithm assumed a climatological MLD, however observations show large temporal variability of MLD. Using a SSHA-dependent MLD for the OHC estimation improves the two-layer model by 5%. Upper ocean thermal structure estimations improved by 25% using the SMARTS Climatology as compared to that of Mainelli (2000).

7 ACKNOWLEDGMENTS The research team gratefully acknowledge support from NOAA NESDIS, the University of Miami Fellowship program, NSF (ATM ), NASA Hurricane Science Program (NASA Award NNX09AC47G), NOAA Joint Hurricane Testbed (NOAA Grant: NA17RJ1226) program, and the Department of the Interior's Minerals Management Service Dynamics of the Loop Current Study (MMS Contract M08PC20052). The project continues to be grateful to the NOAA Aircraft Operation Center (Dr. Jim McFadden) who make it possible to acquire high quality data during hurricanes through the Hurricane Forecast Improvement Project (HFIP) and the collaborative ties with NOAA's Hurricane Research Division directed by Dr. Frank Marks at AOML and NOAA's Environmental Modeling Center directed by Dr. Hendrick Tolman at NCEP. This research would not be possible without the support of committee chairman Dr. Nick Shay, who provided insight and guidance to the scientific process. Drs. Arthur Mariano and Eric Uhlhorn offered valuable knowledge and edits as members of the thesis committee. I would like to acknowledge the members of the University of Miami s Upper Ocean Dynamics Laboratory, particularly Jodi Brewster and Dr. Benjamin Jaimes for assistance with data processing and acquisition. iii

8 TABLE OF CONTENTS Page LIST OF FIGURES... LIST OF TABLES... LIST OF ABBREVIATIONS... vi viii ix CHAPTER 1 INTRODUCTION AND BACKGROUND Background Approach Data World Ocean Atlas Generalized Digital Environmental Model v Altimetry Data Satellite Sea Surface Temperatures In-Situ Profiles Expendable Bathythermographs Argo Profilers PIRATA Moorings Airborne Expendables METHODOLOGY In-Situ Profiles Objective Analysis Daily Climatology Two-Layer Model Statistics of Interest Regional Analysis Weighting Scheme CREATING THE SMARTS CLIMATOLOGY Weighting Maps Realism of the SMARTS Climatology Mainelli (2000) vs. SMARTS Climatology Normalized RMSD Justification of MLD Adjustment Realistic Characteristics of Satellite Field iv

9 3.5.1 Spatial Characteristics XBT Transects Temporal Characteristics Argo Profilers SENSITIVITY OF THE TWO-LAYER MODEL TO SST FORCING Sensitivity of OHC to SST NOAA/NESDIS Operational OHC Limits of Predictability of OHC CASE STUDY: OCEAN THERMAL STRUCTURE BELOW HURRICANES GUSTAV AND IKE Updated Satellite Technique Recalculating Reduced Gravity Available Altimetry Data Synoptic History Gustav Ike In-Situ to Satellite Comparisons: Gustav In-Situ Observations Satellite Estimations In-Situ to Satellite Comparisons: Ike In-Situ Observations Satellite Estimations Discussion Satellite Differences Analysis Near-Inertial Wake Daily Variability Thermistor Drifters Seasonal Thermostad Analysis of Updated Methodology SMARTS Impact on SHIPS Changes to SHIPS OHC Field between Mainelli and SMARTS Discussion Summary and Future Work Summary Future Work REFERENCES v

10 LIST OF FIGURES CHAPTER Page 1 INTRODUCTION AND BACKGROUND 1.1 OHC field using pre-katrina using Mainelli (2000) and SMARTS Differences in Mainelli and SMARTS OHC field pre-katrina Tracks of satellite altimetry data Locations of XBT launches Locations of Argo profiles PIRATA mooring array Drop-points of airborne expendable ocean profilers METHODOLOGY 2.1 Hovmoller of SSHAs in the Loop Current Methodology of estimating a daily climatology Schematic of OHC estimation Comparison of GDEM satellite estimates to in-situ observations CREATING THE SMARTS CLIMATOLOGY 3.1 SMARTS weighting maps for D SMARTS weighting maps for D SMARTS weighting maps for MLD Justification of combination method for the SMARTS Climatology Comparison of SMARTS satellite estimates to in-situ observations Normalized RMSD of oceanographic parameters Standard deviation of oceanographic parameters MLD adjustment justification Location of XBT transect SMARTS estimates comparison to XBT transect SMARTS estimates comparison to Argo profiler SENSITIVITY OF THE TWO-LAYER MODEL TO SST FORCING 4.1 Observation errors of AMSR-E and Reynolds SSTs Observation errors of GOES-POES operational SST product Sensitivity of OHC to SST and two-layer methodology Comparison of AXBT profile to climatology and satellite estimation vi

11 5 CASE STUDY: OCEAN THERMAL STRUCTURE BELOW HURRICANES GUSTAV AND IKE 5.1 Locations of AXBTs deployed in 2008 field campaign Comparison of SST and OHC fields prior to Hurricane Gustav Characterization of GOM watermasses Available altimetry for Hurricane Gustav analyses Available altimetry for Hurricane Ike analyses Changes to thermal structure from two collocated AXBTs Comparison of observations to satellite estimations of D20 for Gustav Comparison of observations to satellite estimations of D26 for Gustav Comparison of observations to satellite estimations of MLD for Gustav Comparison of observations to satellite estimations of SST for Gustav Comparison of observations to satellite estimations of OHC for Gustav Profiles in LC and WCR with subsurface thermostad Comparison of observations to satellite estimations of D20 for Ike Comparison of observations to satellite estimations of D26 for Ike Comparison of observations to satellite estimations of MLD for Ike Comparison of observations to satellite estimations of SST for Ike Comparison of observations to satellite estimations of OHC for Ike H*Wind fields of Hurricanes Gustav and Ike over the LC Changes in observed D20 after Hurricane Ike Track of drifter launched during Hurricane Gustav Time series of drifter observations compared to satellite estimates Collocated AXBT profiles in the LC before and after Hurricane Ike SMARTS Impact on SHIPS 6.1 SHIPS intensity correction for OHC Comparison of Mainelli (2000) and SMARTS intensity correction Forecasted intensity differences for Gustav and Ike by using SMARTS.. 90 vii

12 LIST OF TABLES CHAPTER Page 1 INTRODUCTION AND BACKGROUND 1.1 Available altimetry data METHODOLOGY 2.1 Parameters used in Objective Analysis CREATING THE SMARTS CLIMATOLOGY 3.1 RMSD of ocean estimates using Mainelli (2000) and SMARTS SENSITIVITY OF THE TWO-LAYER MODEL TO SST FORCING 4.1 RMSD of OHC estimation using several SST products RMSD of OHC for the operational GOES/POES SST product Sensitivity of OHC to SST and two-layer methodology CASE STUDY: OCEAN THERMAL STRUCTURE BELOW HURRICANES GUSTAV AND IKE 5.1 AXBTs deployed in 2008 field campaign RMSD of satellite estimates during hurricane Gustav and Ike viii

13 LIST OF ABBREVIATIONS AMSR-E Advanced Microwave Scanning Radiometer for EOS AVHRR Advanced Very High Resolution Radiometer AXBT Airborne Expendable Bathythermograph AXCP Airborne Expendable Current Profiler AXCTD Airborne Expendable Conductivity, Temperature, and Depth CCE Cold Core Eddy D20 Depth of the 20 C isotherm D26 Depth of the 26 C isotherm GCW Gulf Common Water GDEM Generalized Digital Environmental Model GOES Geostationary Operational Environmental Satellite GOM Gulf of Mexico LC Loop Current MLD Mixed Layer Depth NOAA National Oceanic and Atmospheric Administration NRMSD Normalized Root Mean Squared Difference OHC Ocean Heat Content POES Polar Operational Environmental Satellite RMSD Root Mean Squared Difference SHIPS Statistical Hurricane Intensity Prediction Scheme SMARTS Systematically Merged Atlantic Temperature and Salinity Climatology SSHA Sea Surface Height Anomaly ix

14 SST Sea Surface Temperature TC Tropical Cyclone WCR Warm Core Ring WOA World Ocean Atlas XBT Expendable Bathythermograph x

15 Chapter 1: Introduction and Background A new oceanic climatology to calculate upper ocean thermal structure was developed for application year-round in the North Atlantic Ocean basin. The Systematically Merged Atlantic Regional Temperature and Salinity (SMARTS) Climatology blended temperature and salinity fields from the World Ocean Atlas 2001 (WOA) and Generalized Digital Environmental Model v.3.0 (GDEM) at 1/4º resolution for use in a two-layer model to project sea surface height anomalies (SSHA) into the vertical. This higher resolution climatology better resolves features in the Gulf of Mexico (GOM), including the Loop Current (LC) and eddy shedding field, than the coarser 1/2º products of previous studies (Mainelli 2000). SMARTS was calculated from the monthly GDEM and WOA climatologies with an applied 15-day running average to eliminate discontinuities when transitioning between months. Daily mean isotherm depths of the 20ºC (D20) and 26ºC (D26) (and their mean ratio), reduced gravity, and mixed layer depth (MLD) were estimated from the climatology. Using SMARTS with satellite-derived surface height anomaly and SST fields, daily values of D20, D26, MLD, and Ocean Heat Content (OHC) were calculated from 1998 to 2010 using a two-layer model approach. OHC is an important scalar when determining the ocean s impact on tropical cyclone intensification, because it is a predictor of sea surface temperature (SST) cooling during hurricane passage and controls the magnitude of air-sea enthalpy fluxes. Airborne and ship-deployed expendable BathyThermographs (XBT), long-term moorings, and Argo profiling floats provided the in-situ data to assess the SMARTS Climatology by comparing isotherm depths and OHC. Based on over 45,000 profiles of temperatures from the Argo quasi-lagrangian floats, 1

16 2 XBTs, and long-term moorings from 1998 to 2010, a clear, direct relationship emerged from the detailed analysis between satellite-derived and in-situ measurements of isotherm depths and OHC. 1.1 Background Tropical cyclones (TC) are some of the strongest forcing events to the ocean. Strong winds create very strong currents in the mixed layer, force mixing events and entrainment of colder thermocline waters at the bottom of the mixed layer, and cause upwelling of thermocline water by Ekman pumping. The magnitude of the currents and sea surface temperature (SST) cooling due to mixing and entrainment is partly tied to the thermal structure of the ocean lying beneath the TC. These ocean parameters affect the air-sea interaction and the enthalpy fluxes of heat and moisture at the surface. The ocean and atmosphere are not independent in the TC environment, so it is necessary to understand their interactions in order to improve forecasting of both systems. Hurricane effects on the upper ocean can be observed as a cold wake behind tropical cyclones, which is caused by entrainment of cooler thermocline waters. Atmospheric forcing causes turbulence and mixing in the ocean by three mechanisms: i) wind driven shear of the near-surface waters driven by the wind stress; ii) shear at the base of the mixed layer due to energetic near-inertial motions; and iii) convective motions due to surface heat and salt fluxes (D Asaro 1985). The mixed layer is homogeneous, with minimal density stratification with depth. Any weak shear can cause vertical mixing of any density perturbations in the upper layer. Stronger stratification occurs in the thermocline, making it much more difficult to entrain water into the mixed layer via Richardson Number instabilities. When strong enough, currents in the mixed

17 3 layer and, more importantly, shear across the OML base, can cause entrainment of the thermocline waters, essentially eroding away the top of the thermocline. This acts to deepen the mixed layer with two important consequences: i) the wind driven transport reduces velocity because it is spread over a deeper layer, and ii) SSTs decrease because of the cool water entrainment (Sanford et al. 1987). Reducing SSTs possibly acts to inhibit fluxes of latent and sensible heat across the air-sea interface into the hurricane boundary layer. The translational speed of the storm has a significant impact on the effects of the mixed layer. Chang and Anthes (1978) found the following differences between fast and slow moving storms based on modeling: i) The speed of the ocean currents is independent of the translational speed because the balance between wind stress, vertical mixing, and pressure gradient forces respond quickly compared to the residence time of the storm; ii) The deepening of the mixed layer is greater for slow-moving storms because mixing occurs for a longer amount of time; iii) The longer duration of mixing also causes sea surface temperatures to decrease more in slow storms; iv) The pattern of the near-inertial pumping is elongated for storms moving faster than the first baroclinic mode phase speed; and v) The enhancement of mixing on the right side of the storm is greater for faster moving storms due to relatively higher wind speeds on the right side of the storm track. For slow moving storms, the ocean thermal structure is extremely important to diagnose the extent of ocean cooling. Knowing the OHC value is a good proxy for thermal structure, as it represents the amount of thermal energy in the upper ocean available to the storm, although most of this available energy is not utilized.

18 4 Thus, two major factors exist to determine the amount of SST cooling after the passage of a hurricane: the amount of entrainment that occurs at the base of the mixed layer and the original depth of the mixed layer. If cold water is near the surface, i.e. the mixed layer is shallow, then any mixing will create a greater amount of cooling in the mixed layer. If inertial currents are occurring during the entrainment process, then the mixing is enhanced by increased shear instabilities (Price 1981). Also, the upper ocean is significantly affected by the advection of heat by inertial currents. Under some circumstances, the rate of thermal energy advected into a region can be as great as the energy lost to entrainment, especially in frontal regimes (Jacob and Shay 2003). The most notable ocean features in the Atlantic basin that can affect hurricane intensity are the warm core eddies that spin off of the LC in the Gulf of Mexico. The thermal structure of the ocean cannot be determined solely by the SSTs. Significant areas of deep warm water may affect a storm s maximum intensity. The reduction of the negative intensity feedback mechanism in areas of extremely deep and warm water can provide significant amounts of thermal energy to the storm via heat and moisture fluxes at the surface. Between a minimal decrease of SST and the advection of warm waters in the atmospherically forced region, a nearly constant source of energy can be provided to the storm (Shay et al. 2000). This forcing potentially creates conditions which cause intensity and structural changes to a storm, so the initial state of the ocean must be known to make the most accurate hurricane forecasts (Halliwell et al. 2008; Shay and Uhlhorn 2008). The oceanic energy source for TCs has long been known to be related to SST (Palmen 1948; Fisher 1958). Leipper (1967) introduced the concept of OHC as a quantity

19 5 of thermal energy in the ocean above the minimum threshold for TC formation, assumed to be 26 C from Palmen (1948). OHC = Sfc D26 c P ρ(t 26 )dz (1.1) Where c p is the specific heat of seawater, ρ is the density of water, and T is temperature. OHC quantifies the amount of thermal energy potentially available to storms in the upper ocean. OHC is directly related to MLD, D26, and SST, such that thick layers of warm water (deep MLD and D26 and warm SST) are the most favorable for TC. Several case studies have shown the importance of OHC and ocean thermal structure in the rapid intensification of TCs. In 1995, Hurricane Opal rapidly intensified as it traveled over a warm core ring (WCR) in the Gulf of Mexico (Shay et al. 2000) under favorable atmospheric conditions with Opal s juxtaposition to an upper level trough (Bosart et al. 2000). Over a 14-hr period, Opal s unforecasted intensification from a Category 1 to Category 4 occurred directly over a warm feature which enhanced air-sea feedback during hurricane passage. Coupled modeling studies showed that Opal s central pressure was 10 mb higher in the absence of the WCR (Hong et al. 2000). In 2005, Hurricanes Katrina and Rita interacted with a WCR in the Gulf of Mexico. WCRs characteristically have higher values of OHC and deeper values of MLD, D20 and D26. For both storms, sea surface pressure decreases were directly correlated with large values of D26 and OHC (Shay 2009). Mainelli et al. (2008) found using OHC as a predictor of intensity in the Statistical Hurricane Intensity Prediction Scheme (SHIPS) improved forecasting of intensity of Category 5 storms by on average by 5 to 6%. The WCR that Katrina and Rita crossed acted to inhibit mixing of cooler thermocline waters, enhancing air-sea enthalpy fluxes at the surface (Jaimes and Shay 2010). This

20 6 WCR was measured with in-situ temperature profile measurements, as well as a satellite estimate using Sea Surface Height Anomalies (SSHA). An accurate method of determining ocean thermal structure from SSHAs is needed to replace costly in-situ measurements of a limited domain in hurricanes. 1.2 Approach To diagnose the spatial features of ocean thermal structure, Shay et al. (2000) proposed an empirical approach to calculate OHC using satellite altimetry data. Briefly, SSHAs measured by satellite radar altimeters correspond with changes of isotherm depths relative to a climatological value. The changing D26 isotherm is connected to SSHA based on thermal expansion of the water column. To estimate OHC and isotherm depths over the entire basin, a two-layer climatology of isotherm depths, MLDs, and reduced gravity is required. The previous climatology for the Atlantic basin, developed by Mainelli (2000), blended older versions of GDEM and WOA at 1/2 resolution in an undocumented and subjective manner. The development of the SMARTS Climatology aims to systematically determine a blending strategy of the GDEM and WOA climatologies dependent upon the individual climatology s success in calculating oceanographic parameters using the two-layer model compared to over 45,000 in-situ profiles. The SMARTS Climatology has 1/4 resolution which better resolves features such as the LC and Gulf Stream (Figure 1.1). The higher resolution increased the predictability of the location and extent of coherent structures using satellite altimetry data. Similar studies inferred upper ocean temperature profiles by a regression of SSHA and magnitudes of empirical orthogonal functions of ocean profiles (Carnes et al.

21 7 1990; Fox et al. 2002). While this methodology created realistic temperature profiles, the technique was strictly statistical and did not include any explicit dynamics. The two-layer technique assumes a direct correlation between SSHA and D20 dependent upon reduced gravity. In addition to point-bypoint evaluation of the calculation of ocean thermal structure, the spatial and temporal characteristics of the Figure 1.1 OHC field pre-katrina using the Mainelli 0.5 climatology (top) and the GDEM 0.25 climatology (bottom) using the same satellite altimetry data. The Loop Current is more defined in the GDEM climatology, as well as eddy features in the Gulf of Mexico. satellite-derived data were compared to in-situ observations. Using long-term moorings with thermistor chains, variability could be measured on daily to annual time scales. Additionally, expendable profilers deployed by aircraft encompassed synoptic scale features, such that observed slopes of isotherm surfaces can be compared to satellitederived fields. Thus, the overarching goals of the development and analysis of the SMARTS Climatology are (i) to design a methodology to systematically blend two preexisting climatologies dependent on their accuracy in estimating observational data using

22 8 satellite altimetry data and (ii) to examine the strengths and weaknesses of the satellitederived ocean thermal structure spatially and temporally. 1.3 Data To create the SMARTS Climatology, two oceanographic climatologies were analyzed to determine the individual strengths of each climatology in calculating D20, D26, and OHC using a two-layer model. The Generalized Digital Environmental Model v3.0 (GDEM) (Carnes 2009) and the World Ocean Atlas 2001 (WOA) (Conkright et al. 2002) were examined here. Profiles of temperature and salinity were calculated globally at 1/4 resolution. While these two climatologies were similar on the large scale, differences of thermodynamic structure by over 30% in regions of interest necessitated a thorough performance analysis of the individual climatologies (Figure 1.2). Climatological values of D20, D26, MLD, and reduced gravity were extracted from these datasets for use in the two-layer model. Both of these climatologies were at 1/4 resolution, which was an improvement to the previous 1/2 climatologies used in Mainelli (2000). The increased resolution better resolved longlived frontal features near the LC, eddy shedding region, and Gulf Stream. GDEM and WOA did not assimilate the in-situ data used in this study, allowing for unbiased evaluation of the Figure 1.2 Difference in satellite-derived OHC between GDEM and WOA. two climatologies.

23 World Ocean Atlas 2001 WOA is a product of NOAA s National Oceanographic Data Center. The monthly climatology was computed globally from the World Ocean Database, composed from over 7 million profiles primarily from Conductivity, Temperature and Depth (CTD) data (1772-low resolution profiles; 1967-high resolution), mechanical bathythermographs (1941), expendable bathythermographs (1966), and moored buoy data (1990) (Conkright 2002). The profile data were interpolated to levels at 24 standard depths from the surface to 5500 m. The data were grouped into 1 x 1 boxes, from which the local arithmetic mean, standard deviation, and standard error of the mean are computed for all months. The data were then objectively analyzed to 1/4 resolution from 60 S to 60 N at all longitudes following the technique of Barnes (1964). The climatology was smoothed with a median smoother using data from 5 grid boxes in all directions from the data point Generalized Digital Environmental Model v3.0 GDEM is a monthly, quarter-degree climatology of temperature and salinity developed by the Naval Oceanographic Office. GDEM was constructed from a set of four-dimensional steady-state models. The profile data source was the Master Oceanographic Observation Data Set, containing over 5.5 million profiles dating back to 1920 (Carnes 2009). Profiles were calculated globally at 78 depths down to 6600 m in all ocean regions with depths deeper than 100 m. Observational data were gridded onto each depth surface over the entire domain and was then objectively analyzed to a standard 1/4 grid. Values of temperature, salinity, and their respective variances were calculated at all grid points. The GDEM domain extends from 40 S to 60 N globally.

24 Altimetry Data Satellite radar altimeters measure sea surface height along repeated tracks. These polar-orbiting satellites actively transmit a radar pulse and determine the height of the sea surface based on the time it takes for the pulse to reflect off the surface and return to the satellite and make a correction based on gravity effects on the orbit. These satellites repeat their exact paths on a 10, 17, or 35-day basis depending on the mission (Figure 1.3). All available satellite altimetry data were acquired for this study. At least two active satellites were required for accurate mesoscale applications of altimetry data (Rosmorduc 2003). Typically, two or more altimeters were operational at any given time (Table 1.1). Altimetry data were acquired from Naval Oceanographic Office and SSHAs were determined using a background mean SSH from the Collecte Localisation Satellites Combined Mean Dynamic Topography (Rio and Hernandez 2004). Figure 1.3-Tracks of all available altimetry data from 2003 to Altimeters repeat their exact tracks every 10 (blue), 17 (red), or 35 (green) days.

25 11 Altimeter (period) '98 '99 '00 '01 '02 '03 '04 '05 '06 '07 '08 '09 '10 TOPEX (10d) Oct ERS (35d) Oct GFO (17d) Jan Jun Jason-1 (10d) Oct Envisat (35d) Oct Jason-2 (10d) Jun Table 1.1- Available satellite altimetry from 1998 to At least two altimeters are always in service, providing global coverage every ten days Satellite Sea Surface Temperature Global SSTs were acquired from Remote Sensing Systems Optimally Interpolated (OI) data, which blends data from satellite radiometers on the Tropical Rainfall Measuring Mission (TRMM) equatorial satellite and NASA s Aqua polar orbiting satellite. The TRMM Microwave Imager (TMI) radiometer covers a region from 40 S to 40 N, while Aqua s Advanced Microwave Scanning Radiometer for EOS (AMSR-E) produces global SSTs. The OI data were gridded daily at 1/4 resolution. SST retrieval was inhibited in regions of rain, sun-glitter, and close to land, and data gaps were filled in by interpolation In-Situ Profiles Over 45,000 profiles provided the in-situ oceanographic data to evaluate the empirical calculation of D20, D26, and OHC. Profiles covering the Atlantic basin were collected from drifting Argo profilers, ship-borne expendable BathyThermographs (XBTs), Airborne expendable BathyThermographs (AXBTs), Conductivity, Temperature, and Depth (AXCTDs), and Current Profilers (AXCPs), as well as longterm PIRATA moorings.

26 12 Expendable Bathythermographs (XBTs) XBTs are deployed by vessels of opportunity during cross-basin cruises, creating a snapshot of the spatial variability of temperature in the water column (Figure 1.4). The XBT payload contains a thermistor, whose data is transmitted by a fine copper Figure Location of over 8,000 XBTs from wire back to the ship. Thermistors measure the resistance of a conductive material, which is directly related to the environmental temperature surrounding the thermistor. Older XBTs are accurate to approximately ±0.5 C, while newest models are accurate to ±0.1 C. The probe is precision weighted and spin-stabilized which allows for a predictable rate of decent leading to depths accurate to ±2% in the upper 200 m (Sippican; Singer 1990). Quality controlled XBT data used in this study were acquired from NOAA s Atlantic Oceanographic and Meteorological Laboratories (AOML). Profiles are quality controlled at AOML by (1) testing for duplicate profiles, (2) comparing profiles to climatological means to identify potential outliers, (3) comparing profiles to other observations from the same cruise to test for spatial consistency between drop points, (4) testing vertical temperature sections between profiles from the same cruise to test for internal consistency, and (5) mapping monthly averages of SST, temperature at 150 m, and average temperature of the upper 400 m to find outliers (Bailey et al. 1994). Data flagged as inconsistent, doubtful, or bad were excluded from this study. Data collected from XBTs are valuable to diagnose spatial characteristics of ocean thermal structure. XBT transects provide high resolution data across features of

27 13 interest, such as the Gulf Stream. These data will help to address the skill of the satellite algorithm in determining horizontal variability of temperature in regions large structural gradients, such as the Gulf Stream. Temporal changes of these features can be diagnosed by examining repeated ship routes. Argo Profilers Basin-wide profiles of temperature and salinity are available from the Argo buoy array (Figure 1.5). At any given time, approximately 3000 Argo floats are deployed over the world s oceans. The typical sampling mode of a float is to descend to a parking Figure 1.5- Over 13,000 Argo profiles with SST>26 C in the northern Atlantic. depth of 1000 m, where it remains for approximately 10 days. At this point, a hydraulically-controlled bladder empties and the profiler descends to 2000 m. The bladder then begins to inflate, causing the profiler to rise, at which point temperature and salinity data begin to be recorded. On its path to the surface, data are stored at approximately 200 depths. Once the profiler reaches the surface, satellites determine the position of the float and profile data are transmitted from the float to the satellite. Salinity measurements, via conductivity ratios, are accurate to ±0.01 psu. Temperature measurements are reportedly accurate to ±0.005 C, with depths accurate to ±5m (Carval 2010). Such accuracy is required because of the long-term mission of Argo to assess climate variability of deep circulation, temperature, salinity, and upper ocean heat content. The real-time data product is used because temperature and depth measurements

28 14 are sufficiently accurate, allowing for the past year s data to be used. Nineteen automated quality control checks of the real time data are performed to ensure a realistic profile. Argo data are valuable in assessing the SMARTS climatology because data are collected in all regions of the open Atlantic. Where XBTs are typically restricted to repeated ship tracks, Argo profiles cover the data gaps. However, observations are sparse in the Gulf of Mexico because of the short residence time of floats in the LC and geography restricting the profile from completing a full 2000 m profile. PIRATA Moorings The Pilot Research Moored Array in the Tropical Atlantic (PIRATA) off the African coast provides temperature data along in the upper ocean (Servain 1998) (Figure 1.6). Data from the surface to 500 m are measured by a thermistor chain at a vertical resolution of 20 m in the top 140 m, and at depths of 180, 300, and 500 m. The Figure Location of 9 PIRATA mooring locations providing daily thermistor chain data as early as ratio of mooring length to water depth is kept above 0.985, so there is minimal change of depths of the thermistors based on buoy movement. Daily temperature data accurate to ±0.01 C at the surface and ±0.1 C subsurface are sampled for analysis. Daily quality control of data is performed to identify errant data. Temperature changes of over 5 C from the previous day and unrealistic vertical temperature gradients are flagged for inspection. Temperature values falling over 3 standard deviations from a

29 15 90-day climatological mean are also flagged. Visual inspection of the data by PIRATA scientists removes errant data from the dataset. The mission of PIRATA is to monitor air-sea interactions in the tropical Atlantic that affect regional climate variability on seasonal, annual, and inter-annual timescales. Daily in-situ measurements of D20, D26, and OHC can be compared to satellite-derived fields of the same parameters. While Argo data are quasi-lagrangian in nature, PIRATA data are in the Eulerian framework, showing how oceanic thermal structure varies with time at set location. Power spectrum analyses of in-situ and satellite data show the dominant time scales of variability to ensure the satellite-derived data are temporally realistic. Airborne Expendables Ocean observations on synoptic scales and mesoscale can be collected using expendable probes launched from an airborne platform. Airborne expendable bathythermographs (AXBTs), conductivity, temperature, and depth profilers (AXCTDs), and current profilers (AXCPs) are launched by NOAA WP-3D aircraft during Figure Location of over 2000 drop points of airborne expendable profilers. These profilers fill in a large data-gap in the Gulf of Mexico. reconnaissance flights (Figure 1.7). When the profiler is launched by the aircraft, a parachute is deployed to minimize surface impact. Once on the ocean surface, a saltwater activated release mechanism disengages and the payload drops. A thermistor transmits data to the surface float via wire, and the data are transmitted by RF signal to a recorder

30 16 on the aircraft, and then converted into a temperature signal. Depth is estimated using a probe fall rate determined by the hydrodynamics and buoyancy of the probe. A majority of the aircraft data were collected during NOAA Hurricane Hunter reconnaissance missions prior to, during, or after hurricane passage. The remainder of the airborne data was collected in response to the Deepwater Horizon (DWH) oil spill of Airborne expendable data used in this study were processed and quality controlled by University of Miami s Upper Ocean Dynamics Laboratory. The raw profile was passed through a 9-element median filter twice to remove spikes in the data. The profile was then visually inspected to remove any remaining noise or questionable data. To check for potential gross bias of the data, profiles were compared to surrounding profiles. Any questionable or excessively noisy profiles were removed from the dataset. AXBTs are accurate to ±0.5 C, while AXCTDs and AXCPs are accurate to ±0.1 C (Bane and Sessions 1984). Comparisons of AXCTD data from the EPIC field program to simultaneous shipbased CTD measurements confirmed that AXCTDs are a viable method of collecting accurate hydrographic data (Shay et al. 2002). Airborne oceanography permits sampling of mesoscale and synoptic features at time scales such that the feature does not significantly change over the sampling period. The airborne expendable data fill in a major data gap in the Loop Current and the Gulf of Mexico. The flights typically targeted regions of high eddy variability and locations of anomalously high or low sea surface heights measured by satellites. These locations should correspond to points where D20, D26, and OHC are significantly different than the climatological values. In 2008, arrays of AXBTs were deployed prior to and after the passage of hurricanes Gustav and Ike.

31 17 Comparing the two resulting datasets quantifies the upper ocean cooling and deepening of the mixed layer. These data can be compared to satellite-derived data to determine if the satellite data were able to resolve large changes in ocean thermal structure after significant wind forcing events. Collectively, the in-situ profiles provided the necessary data to best create and analyze the SMARTS Climatology. Chapter 2 explains the two-layer model and methodology for blending the GDEM and WOA climatologies. Improvements to Mainelli (2000) and a detailed analysis of the performance of the SMARTS Climatology are in Chapter 3. Chapter 4 explores the satellite OHC estimations sensitivity to SST. A case study of changes to upper ocean thermal structure and satellite estimations during passage of Hurricanes Gustav and Ike is discussed in Chapter 5. Chapter 6 describes expected changes to the SHIPS model due to use of the SMARTS Climatology, and conclusions and future work are presented in Chapter 7.

32 Chapter 2: Methodology 2.1 In-Situ Profiles Each temperature profile was interpolated to a 2-meter increment using a cubic interpolation scheme to maintain a realistic profile at depths where data were less dense, such as deep in the PIRATA array and at depth for XBTs. The profiles were interpolated from 2 m, to avoid artificially high ocean skin temperatures, down to the deepest observation. The depths of the 20 and 26 C isotherm were extracted from the profile by linearly interpolating between the two depths surrounding the isotherm of interest of the interpolated profile. The mixed layer depth was determined by finding the deepest point of the interpolated profile which is within 0.5 C of the temperature at 2 m. This threshold temperature was chosen following Monterey and Levitus (1997) and because it was outside the precision of the instrumentation. Strictly using a temperature difference from a near surface value was preferred to using a temperature gradient threshold due to the large spatial variability of the structure of the thermocline (de Boyer Montegut et al. 2004). Temperature inversions were not of concern because these typically only occur in polar regions, outside of the domain of this study. OHC was calculated using a trapezoidal integration from the surface to D26 of OHC Obs = Sfc D26 c P ρ T,S (T 26 )dz (2.1) Density was calculated at every depth using a 4 th order polynomial (Miller and Poisson 1981) dependent on temperature and an assumed salinity of 35 psu. The specific heat of water, c p, was taken as 4200 J kg -1 C

33 Objective Analysis SSHA data were unevenly spaced over the basin, which necessitates an Objective Analysis (OA) scheme to interpolate data to a regular grid. The parameter matrix algorithm of Mariano and Brown (1992) grids non-stationary fields using time-dependent correlation functions. SSHA data from 5 days before and 5 days after the date of interest were input, which assures basin-wide coverage by at least the 10-day altimeters. Parameters for spatial and temporal correlation scales followed Mainelli (2000) and are presented in Table 2.1. The SSHA field was assumed to have uniform variance and measurement noise was uncorrelated. The OA was performed locally point-by-point, using 20 influential data points determined by the correlation model from Mariano and Brown (1992) which accounts for Operational Mode Dynamic Height Roughness Parameter 1.1 Tension 0.9 Spline Bicubic Influential points 20 Zero crossing scale 1.8 e-folding scale 1.2 Temporal decay scale 20 days Confidence level 0.95 Table 2.1 Key parameters used in the OA following Mariano and Brown (1992) phase speeds, zero-crossing scales, and spatial and temporal decay scales. The previous OA assumed a uniform drift velocity over the entire Atlantic basin to the west-southwest (0.03 /day westward, 0.01 /day southward). However, drift velocities vary spatially with latitude and around bathymetric features (Chelton and Schlax 1996). To assess the spatial variability of feature drift velocity, SSHAs from were OAed without any drift velocity correction. Three years of OA SSHA fields were plotted in Hovmöller diagrams with a north-south and east-west cross-section for all 10 x10 grid boxes (Figure 2.1). If the SSHA showed clear trends, the drift

34 20 velocity for that grid was determined by taking the average speed of three subjectively chosen trend lines. The entire SSHA dataset was reanalyzed to make adjustments using these estimated drift velocities. 2.3 Daily Climatology GDEM and WOA climatologies provided the 4- dimensional data needed to construct the climatology necessary for the implementation of the empirical twolayer model for calculation of ocean thermal structure. Extracted profiles of temperature and salinity were interpolated to a 1 m resolution using a cubic interpolation scheme. An (cm) Figure 2.1 Hovmöller of SSHA at 15 N in Caribbean Sea for two years. Drift velocities are on the order of 0.1 /day. The prior OA technique assumed basin-wide velocity of 0.3 /day. iterative process identified the depth of the 20 and 26 C isotherms and applied a linear interpolation between the two points surrounding the isotherm depth to approximate the appropriate depth. MLD was defined as the depth where the temperature deviates from the SST by a magnitude of 0.5 C. The two-layer model depended on a density difference between upper (ρ 1 ) and lower (ρ 2 ) layers. The interface between layers was assumed to be the 20 C isotherm

35 21 because it was typically within the thermocline. Density was calculated at each level of the climatology and the density of each layer (ρ 1 and ρ 2 ) was estimated as the arithmetic mean of the layer. From the values of density, the reduced gravity, g, was calculated by g = g ρ 2 ρ 1 ρ 2 (2.2) with an acceleration due to gravity, g, of 9.81 m s -2. If g was less than 0.2 m s -2, g was set to 0.2 m s -2 to avoid unrealistic values when applied in the two-layer model. A 15-day running mean applied to the monthly climatologies ensured smoother monthly transitions (Figure 2.2). This blending technique created a daily climatology which can be used as a background for altimetry calculations. Figure A 15-day running mean was used to create the 'daily' climatology. For example, to create the October 2nd climatology, the September climatology was weighted 6/15 and the October climatology is weighted 9/ Two-Layer Model A reduced-gravity model adjusted the climatological values of D20, D26, and MLD in relation to SSHA. In a two-layer model, changes in SSH reflect changes in the depth of the interface between layers. If the SSHA was positive (negative), then the upper (lower) layer was thicker (thinner) and the interface depth was deeper (shallower). The inferred depth of the 20 C isotherm (D20 ) is a function of the reduced gravity, the climatological depth of the 20 C isotherm (D20 ) and the SSHA, η.

36 22 D20 = D20 + ρ 2 ρ 2 ρ 1 η (2.3) A uniform stretching or shrinking of the upper layer was assumed in order to calculate the depth of the 26 C isotherm (D26 ) and MLD. Therefore, the displacement of D26 and MLD was determined by the climatological ratio between the variable of interest and D20. D26 = D26 D20 D20 MLD = MLD D20 D20 (2.4) (2.5) Calculations of D20, D26, and MLD were performed at every grid point in the ocean domain with depths greater than 100 m. If the depth of the calculated isotherm was deeper than the bathymetry, then the isotherm depth was set to be the depth of the ocean bottom. OHC was then calculated by adjusting equation (1.1) assuming a temperature profile with a homogeneous mixed layer with temperature from TMI-AMSR-E SSTs and a constant temperature gradient from the base of the mixed layer to D26. OHC = 1 2 ρ 1c P (D26 + MLD )(SST 26 ) (2.6) The estimation of OHC is presented graphically in Figure 2.3. Previous studies assumed a climatological mixed layer (Mainelli 2000, Shay and Brewster 2010). This study compared MLD to in-situ MLD values to determine if equation (2.5) is a valid technique for estimating MLD for OHC estimations. Figure Schematic of OHC calculation. The red shading shows the true OHC by integrating the black temperature profile. The dashed blue line shows the approximated temperature profile of the upper ocean.

37 23 The in-situ data (D20, D26, OHC, MLD) were then compared to the satellitederived data at the nearest corresponding grid point. To account for errors in the OA, only profiles with corresponding OA errors of less than 0.5 were compared. This empirical method of calculating ocean thermal structure was confirmed to be practical using the GDEM climatology (Figure 2.4). 2.5 Statistics of Interest A set of statistical parameters was necessary to adequately assess GDEM and WOA in calculating temperature structure. The standard deviations of the in-situ and satellite data determined if the satellite-derived data accurately captured the natural variability of ocean parameters. Root mean squared deviations (RMSD) measured the differences between the observations and satellite calculations, which measures the accuracy of the estimator. Count Count Count Figure 2.4 Scatterplots compare in-situ observations from nearly 7000 XBTs to satellite derived values of D20 (left), D26 (center), and OHC (right) using the GDEM climatology. Histograms of error (bottom) show errors are evenly distributed around 0.

38 24 RMSD = n i=1 (x i x i ) 2 n (2.7) The prime denotes estimated value of n observations. A normalized value, NRMSD was used to avoid gross bias errors and account for inherent variability of the datasets NRMSD = RMSD x i (2.8) where the overbar denotes an arithmetic mean. Basic correlation coefficients explained how much error is inherent in the satellite calculations. Regression calculations of in-situ versus satellite data for lines of form y=mx demonstrated if the two-layer method produced data falling near the 1:1 line, such that regression slopes should be near m=1. Similarly, a robust regression analysis of the form y=mx+b ignored outliers and determined a regression slope and bias. 2.6 Regional Analysis It was expected that the performance of each climatology in calculating ocean thermal structure would vary spatially and temporally due to its handling of sub-grid scale variability. Attempts to divide the Atlantic basin into broad analysis regions resulted in minimal differences between skill of the GDEM and WOA climatologies. Instead, the Atlantic basin was divided into 5 x5 boxes in which blending weights were to be determined for each box. In border areas between two regions, there was a 2 linear transitional region to smooth the weighting map and avoid abrupt borders. The 2 buffer zone was larger than the Rossby radius of deformation (~1 in the GOM) to avoid gross distortion of features as they translated through transitional zones. Weights were determined for the hurricane season (June 1 November 30) and off-season (December 1 May 31). Seasonality of GDEM and WOA necessitated weighting schemes for different times of the year.

39 Weighting Scheme Local RMSD of each climatology determined the blending weights for the SMARTS climatology. This parameter was selected because it best quantified the predictive strength of the individual climatologies. For each 5 box, the weights of GDEM and WOA were determined based on the relative magnitudes of RMSD following SMARTS = GDEM RMSD 2 WOA 2 +WOA RMSD GDEM RMSD2 WOA +RMSD GDEM 2 (2.9) In regions with fewer than 50 observations, WOA and GDEM were weighted evenly due to insufficient sample size. Additionally, an absolute weighting map was produced using a weight of 0 or 1 depending on which climatology performed better in a specific grid box. The resulting SMARTS Climatology was used to recalculate isotherm depths and OHC from the satellite altimetry data. Comparisons of upper ocean data derived from SMARTS and Mainelli (2000) climatologies highlighted expected changes to SHIPS parameterizations for hurricane intensity forecasting.

40 Chapter 3: Creating the SMARTS Climatology Using the previously described methodology, the SMARTS Climatology was created by blending the GDEM and WOA climatologies. The use of SMARTS for ocean thermal structure estimation improved the accuracy of OHC calculation basin-wide, especially in the critical GOM and LC regions. A large portion of the improvements to the Mainelli (2000) climatology can be attributed to more accurate GDEM and WOA climatologies which have significantly improved over the past decade. As part of this research effort, regional coherence of the weighting parameters showed both local strengths and weaknesses in GDEM and WOA. 3.1 Weighting Maps The weighting maps resulting from RMSD analysis depicted the locations where either GDEM or WOA were better suited for ocean thermal structure calculation. The D20 weighting map was created first, considering the new D20 calculations were needed for estimations of D26, MLD, and OHC. The D20 weighting map was also used for the blended climatology of densities for the upper and lower layer in the two-layer model. For the hurricane season, the two-layer model calculated D20 better using the GDEM climatology in the GOM and most of the LC regime (Figure 3.1). In fact, GDEM also outperformed WOA along the U.S. coastline, being slightly better in the colder waters north of where the Gulf Stream separates from the coastline near Cape Hatteras, North Carolina. Generally, the subtropical waveguide was dominated by GDEM, particularly along the African coast. The southern extent Caribbean Sea and the center of the subtropical gyre tended to be better predicted using WOA. In this regard, a blending technique of these two climatologies was warranted for SMARTS. 26

41 27 Figure 3.1 Weighting maps for D20 used to create the SMARTS climatology for the hurricane season (top row) and outside of the hurricane season (bottom row). Weighting maps were created using relative RMSD values (left) and absolute values (right). Figure 3.2 Same as Figure 3.1 but for D26.

42 28 Outside of the hurricane season, observations were more sparse and the ventilation of D26 at the surface moves southward during the winter and spring months. Again, GDEM outperformed WOA in the equatorial wave guide. WOA predicted D20 better near the Caribbean islands and in the western GOM whereas GDEM only slightly improved calculations along the mid-atlantic and western Florida coasts. Once the weighting maps of D20 were calculated, D20 was recalculated for the entire basin. Daily values of D26 and MLD could then be estimated based on the ratio between GDEM and WOA climatological values of these parameters and the SMARTS climatological value of D20. Following this outline, analysis showed GDEM typically performed better in the Caribbean Sea and the western GOM, although neither climatology showed marked improvements in the GOM eddy shedding region (Figure 3.2). Elsewhere in the Atlantic Ocean basin, WOA typically outperformed GDEM except in isolated locations. Generally, WOA was better for D26 calculations outside of the hurricane season. The weighting maps for MLD were more erratic, although there was some regional coherence (Figure 3.3). In the GOM during hurricane season, neither climatology stood out in particular, although GDEM was slightly more accurate in the Caribbean Sea and the LC. Elsewhere, WOA was typically more accurate other than a few locations where GDEM held a slight advantage. Outside of the hurricane season, WOA was better for MLD calculations in the GOM and the western equatorial Atlantic Ocean. GDEM marginally improved satellite MLD calculations off the northern South American and African coasts, as well as in the central Atlantic Ocean basin.

43 29 Figure 3.3 Same as Figure 3.1 but for MLD. Using these weighting maps with climatological values for D20, D26, and MLD, the SMARTS Climatology was created. Daily values of D20, D26, MLD, and OHC were recalculated using satellite altimetry and the new climatology, such that it could be compared to results from analyses utilizing the Mainelli climatology and determine if the relative or absolute weighting method was preferable. 3.2 Realism of the SMARTS Climatology To justify the use of the SMARTS Climatology, the resulting blended fields must depict realistic fields of these spatially varying variables. When using the absolute blending technique, there was also a risk of creating an unrealistic field due to sharp transitions between the GDEM and WOA climatologies. Visual inspection of a sample of climatological fields based on the absolute blending scheme showed no evidence of the transitional areas (Figure 3.4). The absolute and relative blending schemes show realistic fields for all parameters needed for the two-layer model.

44 Mainelli (2000) vs. SMARTS Climatology The Mainelli climatology was created based on limited in-situ observations during one hurricane season. The SMARTS Climatology was developed using year round observations over a much longer time period with updated and Figure 3.4 SMARTS Climatology on September 15 for D20 using the absolute weighting scheme.. improved GDEM and WOA climatologies. The SMARTS Climatology vastly improved calculations of ocean thermal structure based on the crude two-layer model by reducing errors in the various files as listed in Table 3.1. Mainelli (2000) typically overestimated D20 throughout the entire basin by over 30 m. This total bias was essentially negligible (<1 m) with the SMARTS climatology where RMSD was reduced by 37%. In the GOM and LC, Mainelli s bias reduced to Basin RMSD (bias) D20 (m) D26 (m) MLD (m) OHC (kj/cm²) Mainelli 48.9 (34.2) 24.1 (7.7) 32.2 (-21.4) 18.3 (-0.6) SMARTS Relative 31.0 (-0.5) 18.2 (-8.4) 17.2 (-3.6) 15.0 (-3.0) SMARTS Absolute 30.2 (-0.7) 17.9 (-7.3) 16.9 (-3.5) 14.8 (-2.6) GOM RMSD (bias) D20 (m) D26 (m) MLD (m) OHC (kj/cm²) Mainelli 47.6 (17.9) 32.5 (11.4) 22.1 (13.3) 40.1 (21.7) SMARTS Relative 41.9 (0.6) 24.5 (-11.6) 17.7 (-5.7) 26.1 (-0.5) SMARTS Absolute 40.7 (-0.8) 24.5 (-12.0) 17.5 (-6.2) 25.5 (-1.1) LC RMSD (bias) D20 (m) D26 (m) MLD (m) OHC (kj/cm²) Mainelli 49.3 (20.5) 30.9 (11.1) 25.4 (14.0) 36.9 (17.3) SMARTS Relative 41.3 (-0.3) 23.1 (-10.4) 19.8 (-5.0) 24.7 (-1.4) SMARTS Absolute 39.8 (-1.5) 22.7 (-10.2) 19.5 (-5.3) 24.0 (-1.6) Table 3.1 RMSD and bias (in parenthesis) of satellite calculation of upper ocean thermal structure using Mainelli (2000) and the relative and absolute weighted versions of the SMARTS Climatology. Values were calculated using all available in-situ observations basin-wide (top), and in the GOM (middle) and LC (bottom).

45 31 about 20 m and the SMARTS bias was still nearly negligible. RMSD was larger in the GOM and LC due to larger depth variability of observations. Implementing SMARTS reduced the RMSD of D20 to about 40 m in the GOM and LC or a reduction of nearly 20%. Across the entire basin, RMSD values decreased by using SMARTS to estimate D26 by 25%, however it should be noted that biases were also of the same magnitude. RMSD decreased from 24 m to 18 m using SMARTS, but where Mainelli (2000) overestimated D26 by about 8 m, SMARTS underestimated D26 by the same magnitude. Similarly in the GOM and LC where substantial spatial variability exists, RMSD was reduced by 25% to about 24 m using SMARTS. In this complex regime, Mainelli overestimated D26 by 11 m in the GOM and LC, while SMARTS underestimated D26 by 11 m. There were large MLD biases in the Mainelli climatology, underestimating MLD basin-wide by 21 m and overestimating in the LC and GOM by nearly 14 m. Such large differences led to a basin-wide RMSD of 32 m and 23 m in the GOM and LC, respectively. Mainelli s estimation of upper ocean thermal structure assumed a climatological MLD, whereas the updated two-layer model with the SMARTS Climatology uses an adjusted MLD. Using SMARTS reduced basin-wide RMSD by almost 50% and a small negative bias of 4 m. In the GOM and LC, SMARTS improvements were more modest with a negative bias of 6 m and a 20% improvement in RMSD. Using all in-situ data from the entire Atlantic Ocean basin, the Mainelli climatology performed well when calculating OHC, with a bias of less than 1 kj cm -2 and

46 32 RMSD of 18 kj cm -2. SMARTS decreases RMSD to 15 kj cm -2 with a small 3 kj cm -2 underestimate. In the region of greatest concern where hurricanes tend to reach severe status in the GOM, Mainelli overestimated OHC by an average of 20 kj cm -2 in the LC complex whereas the bias from SMARTS was nearly negligible. SMARTS decreased RMSD in the GOM and LC by 35%, from about 40 to 25 kj cm -2. Such large changes in OHC calculations could have significant impacts on hurricane intensity forecasting in the SHIPS model (DeMaria et al. 2005). Scatterplots comparing in-situ values to SMARTS-derived data showed strong correlations, with values typically falling near the perfect fit curve of slope one (Figure 3.5). Histograms of absolute differences between satellite and in-situ data showed errors of SMARTS were evenly distributed around the mean bias where 69% of satellite calculations of D20 were within 20 m of the observed value and 89% were within 40 m. The negative bias of D26 was visually noticeable in the histogram as only 65% of D26 SMARTS calculations were within 15 m of the observed value, while 90% were within 30 m. For OHC, 85% of satellite calculations were within 20 kj cm -2 of observed values Normalized RMSD To more accurately identify the regions of uncertainty of the two-layer model, a normalized value of RMSD (hereafter NRMSD) was examined. NRMSD is the RMSD as calculated by equation 2.8 normalized by the mean of the local in-situ observations. Maximum values of NRMSD were typically highest in the GOM and the northern Atlantic Ocean for most variables of interest (Figure 3.6). D20 had a large NRMSD in the north Atlantic Ocean basin because D20 often ventilates in this region. Even if satellite SSTs were just above 20 C, the calculation of D20 would not be adjusted, leading to

47 33 Figure 3.5 Scatterplots (left) of D20 (top), D26 (middle), and OHC (bottom) comparing in-situ measurements to values derived by the two-layer model using the SMARTS Climatology. Data points with differences greater than 4 standard deviations from the mean were removed (<0.5% of points). Histograms (right) of errors have bin widths of 10 m, 5 m, and 5 kj cm -2 for D20, D26, and OHC, respectively. large errors and a manifestation of the drawbacks of the crude two-layer model. Additionally, this was an area of high variability due to the Gulf Stream eddy field, as measured by the standard deviations of observations (Figure 3.7).

48 34 Figure 3.6 Normalized RMSD for D20, D26, MLD, and OHC during the hurricane season. The contours represent the differences relative to the variables average magnitudes. NRMSD is greatest in the GOM and northern Atlantic. Figure 3.7 Standard deviations of observed values during hurricane season for D20, D26, MLD, and OHC.

49 35 The other region of relatively high NRMSD was the GOM. From a broader perspective, the GOM region had high variability of upper ocean thermal structure, which partially explained the inflated errors. Attribution of the differences was difficult, however the most likely candidate was probably problems inherent to the two-layer model in a dynamically evolving oceanic regime. Generally, areas of high sub-grid scale variability coincided with high NRMSD, such that the two-layer model could not solely account for all upper ocean variability as expected. 3.4 Justification of MLD Adjustment The two-layer model used by Mainelli (2000) confined MLD to a climatological value. Observations show regions of higher variability of MLD, particularly in the LC, GOM and Gulf Stream, which are all characterized by a strong eddy field where MLDs are maintained by a variety of forcing mechanisms (Figure 3.7). At the ocean surface, momentum fluxes act to mix the upper ocean while latent and sensible heat fluxes force buoyancy driven mixing, while shear instabilities entrain cooler thermocline waters at the base of the mixed layer. While the climatological approach is stable, using climatology did not capture many occurrences of deep MLD structure (Figure 3.8). Adjusting the MLD following the same logic used to calculate D26 led to a more realistic scatter of satellite values. The adjusted methodology allowed for shallow mixed layers that often mix quickly when subjected to strong surface winds via shear instability which were typically not present in the climatology. Additionally, adjusting MLD resulted in balancing the low bias for deep MLD in-situ observations. The slope of the regression line between in-situ and calculated MLD increased from 0.57 to 0.68, which was an improvement but still well below the 1:1 line. Overall bias of calculations

50 36 Figure 3.8 Scatterplots of MLD (left) and OHC (right) comparing in-situ observations to satellite estimations using climatological MLD (top) and the adjusted MLD (bottom). The red dashed line represents a 1:1 relationship and the linear regression line is solid blue. decreased by a magnitude of about 2 m, however this was accompanied by an increase of RMSD on the same scale. Bias actually increased slightly when considering only data below the in-situ median MLD. Overall MLD bias decreased by adjusting MLD because using only the climatology as an MLD estimate could not capture occurrences of particularly deep MLDs. RMSD increased primarily in areas of shallow observed MLD where the adjustment using the two-layer model was disproportionate to the magnitude of the true MLD.

51 37 Similar results were observed when considering OHC calculated with the climatological and adjusted value. The slope of the regression line improved from 0.83 to 0.9, yet RMSD increased due to the increased scatter of calculated MLDs. The regression analysis suggested a potential link between MLD and SSHA, yet further analysis could not confirm a direct link. The climatological MLD approach changes of thermocline stratification by only adjusting D26, although observations did not suggest a strong correlation between SSHA and thermocline lapse rates. Adjusting MLD moderately improved the estimation of the depth of the thermocline. 3.5 Realistic Characteristics of Satellite Field Spatial Characteristics XBT Transects Repeated transects across the Atlantic provided snapshots of vertical temperature structure, which allowed for comparison to SMARTS-derived values. Three annual September transects between 2004 and 2006 across a region east of Florida through the Gulf Stream provided over 120 XBT profiles for analysis (Figure 3.9). Along this transect, the satellite algorithm using the SMARTS Climatology captured the horizontal variability of OHC kj cm -2 within the error of observations (Figure 3.10). Most of the variability of OHC could be attributed to east/west changes in SST, Figure 3.9 Satellite estimated OHC on September 15, 2005 during one of three annual XBT transects along the red line.

52 38 Figure 3.10 Average temperature observed by XBTs along the transect in Figure 3.9 with observed (solid) and SMARTS-derived (dashed) depths of D26 (black) and D20 (white). Observed OHC (red line, top) is compared to satellite derived values with error bars equivalent to 2σ. considering there was little variability of D26. Satellite calculations of D20 had a shallow bias, particularly near local minima along the 20 C isotherm. The satellite methodology accurately calculated the locations of relative maxima and minima of D20, only the magnitudes were not accurate. The two-layer model often exaggerated the slope of the D20 surface, although the locations of maxima and minima were accurate Temporal Characteristics Argo Profilers Argo profilers provided valuable data for analysis of the two-layer model with the SMARTS Climatology. The drifters completed one full profile of the water column every 10 days, the same as the time window of SSHA data, which ensured a corresponding time series of independently calculated satellite fields. Figure 3.11 depicts data from 2008 from an Argo profiler west of the Caribbean Islands with characteristics representative of the full dataset.

53 39 Figure 3.11 Comparisons of D20, D26, SST, and OHC between Argo profiler (solid black) and SMARTS-derived (dashed red) from D20 typically did not change greatly between subsequent profiles, although satellite calculations showed higher temporal variability in D20. The two-layer model s calculation of D20 was sensitive to SSHA changes, often overestimating observed in-situ changes but still capturing change on a monthly timescale. Satellite-derived values of D26 also showed such overestimated fluctuations, which was expected considering its dependence on D20. In-situ values of D26 fluctuated enough such that climatology alone could not accurately depict the ocean subsurface structures. Generally, the two-layer model accurately determined the change of direction of the 26 C isotherm, although the magnitude of change was often overestimated. Differences in OHC were a result of disparities between in-situ and satellite values of D26.

54 Chapter 4: Sensitivity of the Two-Layer Model to Sea Surface Temperature Forcing An accurate representation of SSTs was critical for a realistic calculation of OHC by satellite. Based on previous rough calculations of skill among different SST products, estimates of OHC were calculated using SSTs from Remote Sensing Systems TMI/AMSR-E for the University of Miami s Upper Ocean Dynamics Laboratory OHC product. A more complete analysis was necessary to justify the use of AMSR-E for OHC calculation. AMSR-E was compared to two daily Reynolds SST products optimally interpolated to a 0.25 grid. The first used Advanced Very High Resolution Radiometer (AVHRR) infrared SST data. The second combined the AVHRR data with AMSR microwave data (AVHRR-AMSR). Both Reynolds products incorporated in-situ data from vessels and buoys. It should be noted that the in-situ temperatures used in this comparison were at a depth of two meters. Using the 2 m depths could cause some errors, as this was not the skin temperature measured by the radiometers. RMS differences and bias between each satellite product and in-situ SSTs were of primary importance for application to OHC calculations. The AVHRR products were compared with AMSR-E only where observations were available from both sources and in-situ SST was greater than 26 C. The AVHRR-only data had a low bias of 0.14 C and RMSD of 0.50 C. AMSR-E outperformed AVHRR, with a smaller low bias of 0.03 C and RMSD of 0.40 C. Not surprisingly, adding the AMSR data to AVHRR improved the satellite observations. RMSD values were similar between AMSR-E (0.39 C) and AVHRR-AMSR (0.39 C). AMSR-E had a slight low bias of C and AVHRR- AMSR had a low bias of 0.06 C. The post-processed AMSR-E data had errors that were 40

55 41 smoothly distributed, whereas the AVHRR data were unevenly distributed around zero (Figure 4.1). Figure 4.1 Distribution of differences between in-situ near surface temperatures and satellite SSTs from RSS AMSR-E and Reynolds-AVHRR-AMSR. Bins are centered around 0 at 0.1 C increments. The curves with outlined markers show the expected distribution if the distributions were Gaussian. 4.1 Sensitivity of OHC to SST RMSD analysis of SST suggested AVHRR-AMSR was the slightly better product, however AMSR-E had a smaller bias. To identify which version of SST was optimal for OHC calculations, OHC was calculated using satellite estimated depths and the individual SST products. The RMSD of OHC was minimized using the Reynolds AVHRR-AMSR SSTs, including the GOM and LC regions (Table 4.1). This analysis demonstrated that AVHRR-AMSR is slightly better for OHC calculation in the two-layer model framework, however this does not settle the debate of which SST product was superior on the whole.

56 42 a.) RMSD (kj/cm 2 ) Basin-wide Gulf of Mexico Loop Current AMSR-E AVHRR-AMSR b.) Bias ( kj/cm 2 ) Basin-wide Gulf of Mexico Loop Current AMSR-E AVHRR-AMSR Table (a.) RMSD calculation for OHC using different SST products. AVHRR-AMSR minimizes RMSD basinwide and (b.) reduces OHC bias in the GOM and LC NOAA/NESDIS Operational OHC Starting in the summer of 2011, NOAA s National Environmental Satellite, Data, and Information Service (NESDIS) is implementing the SMARTS Climatology and twolayer OHC calculation technique in near-real time. In this application, NESDIS is using a blended daily SST product observed by Geostationary Operational Environmental Satellite (GOES) and Polar Operational Environmental Satellite (POES) platforms. These satellites use AVHRR for SST measurements. The available data date back to 2009, so there are a limited number of in-situ observations in the SMARTS database (Figure 4.2, Table 4.2). The GOES/POES data were compared to the Reynolds and AMSR-E products. GOES/POES had a slightly larger RMSD and bias than Reynolds and AMSR-E, however, it was sufficient for OHC calculations. RMSD (kj/cm²) ALL (6,244) GOM (1,224) LC (1,001) AMSR-E AVHRR-AMSR GOES/POES Table 4.2 RMSD for OHC calculation using three SST products. The number of observations for each region are in parentheses. Data was from 2009 and 2010, corresponding to over 6,200 observations basin-wide

57 Limits of Predictability of OHC Figure 4.2 SST errors from the AMSR-E (blue), daily Reynolds AVHRR-AMSR (red), and GOES/POES (green) in 0.1 C bins. When first introduced, Mainelli s (2000) two-layer model approach for OHC calculations was verified with a small dataset of airborne expendables deployed in the GOM during hurricane aircraft reconnaissance flights. Now with two orders of magnitude more observations, the approach could be evaluated basin-wide. Even if a perfect climatology existed such that it results in exact depth calculations by satellite altimetry, there would still be inaccuracies in the trapezoidal OHC calculation in the two-layer model. To validate the calculation of OHC by trapezoidal approximation, all in-situ upper ocean profiles were simplified to a homogeneous mixed layer with a temperature equal to the SST and a thermocline with constant stratification (dt/dz) from the base of the mixed layer to D26 (Figure 4.3). The

58 44 resulting OHC from simplified profiles was then compared to the in-situ profiles, effectively setting the limit of accuracy of the fully satellite-derived OHC. It was expected that the trapezoidal calculation overestimated OHC because of the typical concavity of temperature structure; profiles were typically most strongly stratified at the top of the thermocline and stratification weakened at greater depth. Basin-wide, the trapezoidal approximation overestimated OHC by 8.5 kj cm -2. In the GOM and LC, most observations occurred during hurricane season such that surface temperatures, and therefore OHC, were artificially high. The trapezoidal calculation led to a high bias of 22.2 kj cm -2 in the GOM and 25.3 kj cm -2 in the LC. This bias led to large RMSD values basin-wide (13.5 kj cm -2 ), in the GOM (29.9 kj cm -2 ) and LC (33.73 kj cm -2 ). In Figure 4.3 Trapezoidal (black dashed) and satellite estimations are compared to an in-situ profile (black solid) in the Gulf of Mexico prior to the passage of Hurricane Gustav is The trapezoidal approach uses in-situ values of SST, MLD, and D26 to give a rough upper ocean temperature profile. The derived profiles show sources for errors due to satellite SST (green dashed) and estimated depths (blue solid), and their combined error (red dashed).

59 45 essence, this sets the bar for what can be expected for OHC calculations using the previously described two-layer model. Even with perfect predictability of MLD and D26 and accurate satellite SST, significant differences still exist between observations and satellite estimation. Further insights to the source of errors could be found by replacing satellite observed parameters with in-situ values (Figure 4.3, Table 4.3). Basin-wide, satellite depth estimates were more responsible than satellite SST for inaccuracies of OHC. Conversely, in the LC and GOM regions, satellite measurements of SST cause more of the error. In the LC and GOM, D26 was generally deeper such that the satellite-derived D26 adjustment was relatively smaller than the climatological D26. Compare this to other regions where the D26 adjustment is on the order of the climatological value. As a result, a.) ALL Observed Satellite Satellite All (kj/cm²) Observed Trapezoid Depth SST Satellite RMSD Bias Std Dev a.) GOM Observed Satellite Satellite All (kj/cm²) Observed Trapezoid Depth SST Satellite RMSD Bias Std Dev a.) LC Observed Satellite Satellite All (kj/cm²) Observed Trapezoid Depth SST Satellite RMSD Bias Std Dev Table 4.3 Errors arising from the trapezoidal approach to calculating OHC from satellite derived depths and SST. Satellite depths are used with in-situ SST, and satellite SST is used with in-situ D26 and MLD. The last column uses both satellite depths and temperatures.

60 46 the influence of SST was the principal factor causing errors of OHC in the GOM and LC. Errors arose due to deviations from the idealized temperature profiles. The trapezoidal approach does not account for seasonal subsurface thermostads. In late winter when MLD is greatest, surface waters are gradually heated due to sensible heating at the surface. This warmer surface water does not mix with the subsurface cooler waters from the winter mixed layer until a strong wind forcing event or internal forcing occurs. The WOA and GDEM climatologies do not resolve these subsurface thermostads in the monthly climatology (Figure 4.4). Prior to Hurricane Gustav, thermostads of approximately 29 C and 27 C existed below the surface from 30~60 m and 80~110 m, respectively. It is impossible to resolve these features with the two-layer model without incorporating atmospheric forcing. Figure 4.4 A worse case in-situ profile in a WCE prior to passage of Hurricane Gustav (black) with the climatological profile from GDEM (blue) and the satellite-derived profile (red).

61 Chapter 5: Case Study Ocean Thermal Structure below Hurricanes Gustav and Ike In 2008, hurricanes Gustav and Ike crossed through the GOM, interacting with the LC and eddy field. In cooperation with NOAA s Hurricane Research Division (HRD) and Aircraft Operation Center (AOC), 222 GPS sondes and 407 AXBTs (see Table 5.1) were deployed from WP-3D aircraft as part of a field campaign to simultaneously observe the hurricane atmospheric boundary layer and upper ocean temperature field over the life cycles of Gustav and Ike (Figure 5.1). Prior to the passage of both hurricanes, an array of AXBTs was deployed along the forecasted path of the storm. During in-storm flights, AXBTs and GPS sondes were launched during transects through the hurricane in a Figure 5.4 pattern. After landfall, an array of AXBTs was deployed in cross sections along the storm track. With observations before and after the storm, the impacts of the storm-induced cooling and deepening of the oceanic mixed layer could be examined spatially. Gustav Ike Date Flight AXBT Date Flight AXBT 28 Aug RF43 49(2) 08 Sep RF43 47(2) 29 Aug RF42 16(0) 09 Sep RF42 6(0) 30 Aug RF43 19(2) 10 Sep RF42 10(2) 31 Aug RF42 16(1) 10 Sep RF43 20(7) 31 Aug RF43 19(1) 11 Sep RF42 10(1) 01 Sep RF43 19(0) 11 Sep RF43 22(3) 03 Sep RF43 54(4) 12 Sep RF42 10(4) 12 Sep RF43 20(4) 15 Sep RF43 61(5) Total 7 191(10) 9 216(28) Table 5.1 Deployment of AXBTs for hurricanes Gustav and Ike. A pre- and post-storm array was deployed along the hurricane track and in-storm AXBTs were launched in hurricane conditions. Total number of probes for each flight is listed, with failures in parentheses. 47

62 Figure Locations of AXBTs launched during the 2008 hurricane season for Hurricanes Gustav (top) and Ike (bottom). Drop locations are layered over satellitederived OHC. 48

63 49 The GOM is of particular interest for hurricane intensity forecasting, due to its complex oceanographic dynamics and economical and societal sensitivity along its coast. During the hurricane season of 2008, SSTs were typically higher than 30 C well above the Figure 5.2 SSTs (top) and OHC (bottom) observed by AXBTs prior to Hurricane Gustav. Note the small variation of SSTs relative to large spatial differences in OHC. threshold necessary for tropical cyclone formation (Cione and Uhlhorn, 2003). While SSTs are fairly uniform over the GOM, the subsurface temperature structure fluctuates spatially (Figure 5.2). As the LC protrudes into the GOM, warm core rings (WCR) shed off the main current and drift into the GOM. During a WCR shedding event, an oppositely rotating cold core eddy (CCE) is also formed. The WCR/LC, CCE, and Gulf Common Water (GCW) have unique temperature structures as observed in the Gustav and Ike datasets (Figure 5.3). The LC and WCR are characterized by deep warm waters with weak stratification in the upper 200 m of the water column. The CCE and GCW generally have a shallow mixed layer above a very strong thermocline. Due to these structural differences, WCR/LC regions typically have much greater values of OHC than in CCE/GCW regimes and are therefore areas where rapid intensification could be anticipated. During the life span of hurricanes Gustav and Ike, an eddy shedding event was occurring. Over the course of the few weeks surrounding the two storms, a large

64 50 intrusion of the LC began to break off to create the foundation of a WCR. Both Gustav and Ike traveled over the WCR after crossing Cuba and entering the Gulf. These storms did not undergo rapid intensification above the WCR due to synoptic scale atmospheric conditions, however the effects of atmospheric forcing on the ocean can be examined. 5.1 Updated Satellite Technique Recalculating Reduced Gravity Initial satellite estimations Figure 5.3 SSHA field prior to Hurricane Ike (top) with the locations of profiles of GCW (black), WCR (red) and CCE (blue). SSTs are similar between profiles, however OHC varies greatly. demonstrated that the reduced gravity term for estimation of the D20 isotherm was too small, resulting in exaggerated displacements of the thermocline. Upon further review, it was determined that the climatological density calculation following Mainelli (2000) and Shay and Brewster (2008) using the Kraus-Businger estimation does not accurately estimate the density in the deep ocean because it lacks depth dependence. Updating the density calculation was necessary to estimate a lower boundary depth for the bottom layer of the two-layer model. Using SSHA estimations with observed and climatological values of D20, the reduced gravity term, g'*, can be estimated by g = ρ 2 ρ 1 ρ 2 g = ηg (D20 Obs D20) (5.1)

65 51 Observations suggested a minimum cutoff value of g'* near 0.03 m s -2 in the GOM and a lower boundary for the lower layer was estimated to be 400 m, which is within the scale of the thermocline. Additionally, satellite estimations of the D26 isotherm were errant due to the climatological ratio of D26/D20. In the eddy shedding region of the GOM, the D26/D20 in GDEM and WOA was below 0.3, however in-situ measurements showed an observed ratio above 0.4. This discrepancy necessitated a minimum D26/D20 ratio of 0.4 in the GOM and LC. These adjustments to the two-layer framework improved thermocline estimations and can be implemented basin-wide Available Altimetry Data A change in the methodology of estimating ocean thermal structure was necessary to eliminate satellite SSHA observations outside of the relevant time frame. The standard 10-day window of SSHA observations centered on the day of interest was shifted such that there was no interference of SSHA data observed after (before) storm passage in the pre-storm (post-storm) dataset. Maintaining a 10-day window of SSHA data ensured full basin coverage by the 10-day altimeter, and satellite SST estimates from the day of AXBT deployment provided the surface boundary condition. Altimetry data were not available directly over the core of the LC prior to Hurricane Gustav, although the western and northern fronts of the LC were sampled by altimeters (Figure 5.4). The CCE to the west of the LC/WCR complex was not directly observed, but the region of shallow thermocline depths directly to the north was thoroughly covered. The western WCR was crossed by an altimeter, however this feature was not measured by AXBTs in the post-storm data array. In the days after Gustav s

66 52 passage, there were several altimeter tracks passing over the LC region. To the northwest in the CCE, the SSHA data were observed over a week after hurricane passage. Due to the quick succession of the two storms, the same altimetry data were used to estimate the post-gustav and pre-ike fields (Figure 5.5). The satellite estimates during the two deployments were not identical because of updated SSTs and correcting SSHAs for Figure 5.4 Available altimeter tracks in a 10-day window before (top) and after (bottom) passage of Hurricane Gustav overlaid on AXBT observed OHC. The color of the track represents the time of SSHA observation relative to the date of airborne reconnaissance. estimated drift. The pre-ike AXBT array was well covered by altimeters, although the core of the CCE west of the LC was not directly sampled. SSHA data were more sparse for the post-ike array, with a lack of altimetry data in the CCE west of the LC and the northern frontal region of the LC/WCR complex. The western WCR and the northern CCE regions were well covered by altimetry in Figure 5.5 Same as Figure 5.15, but for Hurricane Ike altimetry both pre- and post-storm flights.

67 Synoptic History Gustav Hurricane Gustav developed from a wave originating during a period of enhanced wave generation off near Cape Verde. Gustav intensified to a Category 1 hurricane and then weakened just prior to landfall in Haiti on August 26 th. Gustav meandered westward through the Caribbean as a tropical storm due to geographical effects from Hispaniola and a ridge over Florida. Deep warm LC waters and favorable atmospheric conditions sparked rapid deepening, intensifying Gustav from a tropical storm to a major hurricane in 24 hours. As a Category 4 hurricane, Gustav made landfall in western Cuba on August 30th. The storm emerged to the south of Cuba on August 31 st, maintaining its Category 4 status. An upper-level ridge, as well as dry air entrainment caused Gustav to weaken over the GOM over the next two days, despite quickly traversing a WCR. Although maximum wind speed weakened, Gustav s wind field expanded laterally over the GOM, with tropical storm force winds extending over 200 nm from the storm's center. Gustav eventually made landfall near Cocodrie, LA on September 1 st as a Category 2 hurricane and rapidly weakened as the storm's remnants traveled north to the Great Lakes (Beven and Kimberlain 2009) Ike The atmospheric wave that eventually became Hurricane Ike propagated off the African coast two weeks after the wave that became Gustav. Ike reached major hurricane status on September 4 th when the storm was still well east of the Caribbean. Eastern Cuba received the brunt of the hurricane force winds when Ike made landfall as a Category 4 hurricane on September 8 th. The storm weakened as it traveled westward over the Island

68 54 until its eye emerged just to the south of Cuba. Although Ike was centered over the Caribbean Sea, its interaction with the Cuban coastline disturbed the inner core dynamics, limiting organization and intensification. Ike then made another Cuban landfall and crossed into the GOM on September 10th, where the wind field expanded. Once in the GOM, the storm passed south of the WCR as it was shedding from the LC, and Ike intensified from a Category 1 to Category 2 storm. However, an eyewall replacement cycle limited the ability for the storm to rapidly intensify due to angular momentum considerations. Weak inner core convection and a massive wind field also limited Ike's intensification. Ike made landfall as a Category 2 hurricane on September 12 th just north of Galveston, TX, bringing with it a large storm surge that inundated many coastal communities (Berg 2009). Hurricane Ike crossed directly through the pre-storm AXBT array on a NNW trajectory. After emerging from Cuba as a slightly disorganized Category 1 hurricane, Ike redeveloped its structure and strengthened to a Category 2 storm over the WCR. While maximum wind speeds suggests Ike was less powerful than Gustav, Ike had a massive wind field of TS force winds. While over the LC/WCR, Gustav had an integrated kinetic energy of TS force winds of 63.1 TJ (1 TJ = J) (1330 UTC 31 Aug), whereas Ike was 10 kt weaker but had an integrated KE of 149 TJ (1930 UTC 10 Sep) (Powell et al. 1998). Such a large wind field had a dramatic effect on the ocean thermal structure and SSTs.

69 In-Situ to Satellite Comparisons: Gustav In-Situ Observations Prior to Hurricane Gustav s emergence over the GOM, 49 AXBTs were deployed on August 28 th on in a lawn mower pattern transecting the LC and frontal region three times and then sampling an area where satellite altimetry suggested the location of a mature WCR and CCE (Figure 5.3). At this time, the LC protruded NNW with the frontal region extending to 26 N with a slight westward hook due to a CCE located to the west (CCE W ) and north (CCE N ). AXBT observations did not suggest that a WCR had been shed from the LC, with maximum observed D20 values of 290 m along both transects through the LC (Figure 5.7a). The D26 field had similar spatial characteristics as the D20 field and also did not suggest eddy separation, with maximum depths of 160 m in the LC and only 35 m in CCE W (Figure 5.8a). The MLDs were largest in the core of the LC at 50 m, and were as shallow as 15 m in CCE W and CCE N (Figure 5.9a). Elsewhere in the GOM, D26 values were approximately 40 to 50 m and MLD values were near 20 m. Outside of the LC/WCR, temperature profiles indicated shallow mixed layers above a strongly stratified thermocline which weakened with depth (d²t/dz² > 0). Despite the variability of subsurface temperature, SSTs had a small range of variations from 29.2 to 30.5 C (Figure 5.10a). Hurricane Gustav crossed directly over the LC as a Category 3 storm. Over the course of the storm s passage, post-storm observations from September 3 rd demonstrated that the eddy had begun to detach from the LC. A local maximum of D20 existed near - 86 W, 25.0 N, suggesting a closed eddy circulation independent of the LC assuming geostrophy (Figure 5.7d). This closed feature did not show up in the D26 observations

70 56 (Figure 5.8d). To the south of the eddy, observations showed westward propagation of the eddy/lc, as the maximum depths shift 50 km westward. MLD reached a maximum of over 70 m approximately 100 km to the east of the LC core, directly underneath Gustav s path. The shallowest MLD was 30 m in CCE N (Figure 5.9d). Subsequent to storm passage, SSTs showed much more variability, ranging from 26.7 to 29.6 C (Figure 5.10b). In areas with AXBTs collocated in the pre and post storm flights, the impacts of Gustav s forcing on the ocean could be examined. Above the LC, maximum winds peaked at 100kts in Gustav s NW quadrant on August 31 st according to HRD H*Wind analysis (Powell et al. 1998). While over the LC, maximum wind speed was steady; however the hurricane force wind field expanded 20 km radially, extending 40 km west and 120 km east of the center of circulation. Gustav passed to the east of the eddy separation area, such that storm induced mixing would theoretically have the largest effect on the eastern half of the WCR/LC, particularly to the east of Gustav s track. The changes of D20 and D26 in the LC showed the westward migration of the WCR as it separates from the LC, with shoaling D20 and D26 isotherms to the east and deepening to the west (Figures 5.7g & 5.8g). The area near (88 W, 24 N) which suggested extensive deepening of the thermocline must be ignored due to a lack of collocated AXBT deployments. Beneath the storm center, some shoaling of deeper waters was expected due to Ekman pumping (Gill 1984). During this shedding event, it is difficult to attribute changes in thermal structure below the thermocline to either the formation of the WCR or wind forcing from Gustav. Wind-forced mixing induced a deepening of the LC mixed layer underneath the storm track of up to 40 m (Figure 5.6).

71 57 In this region, the upper ocean is weakly stratified, such that wind or buoyancy forcing can cause relatively large deepening of the surface mixed layer. In the LC/WCR complex, SSTs decrease by only 0.5 to 1.0 C (Figure 5.10g). Despite the SST cooling, OHC values to the west of the WCR increased due to its westward propagation (Figure 5.11g). Due to the pre and post storm flight patterns, changes occurring to the CCEs and in the GCW were difficult to analyze. The post-storm flight revealed a portion of CCE N advecting Figure 5.6 Profiles taken at the same location in the LC before (blue) and after (red) passage of Hurricane Gustav. The mixed layer deepened and cooled due to passage, and the deeper waters were upwelled. around the northwestern extent of the WCR. This feature was not clearly resolved during the pre-storm flight. Scale analysis estimated isotherm displacement forced by Ekman pumping (L=τ/ρfU) on the order of 15 m, which was consistent with ~20 m upwelling of D20 and D26 observed after Gustav s passage over the WCR. The largest SST decrease occurred to the north of the observed array along the east side of the storm track. Assuming SSTs were relatively uniform over the entire GOM prior to Gustav, SSTs decreased by as much as 2.5 C in this northern region Satellite Estimations The pre-storm estimations of D20 were accurate and comparable to in-situ observations (Figure 5.7bc). The maximum D20 isotherm depth in the core of the LC/WCR complex was observed and estimated to be near 280 m, despite poor coverage

72 58 by altimeters (Figure 5.4). Satellite estimations accurately identified the location of the western LC frontal region along AXBT transects at 24.0 and 25.5 N, however a large region of negative SSHAs in the southwest of the array caused a misrepresentation of the eastern extent of the LC. The northern extent of the LC/WCR complex was accurately located and a SSHA minimum to the north suggested a sharper horizontal gradient relative to in-situ observations. To the west in the CCE, the D20 isotherm was underestimated by 20 m, although this may be attributed to the OA of in-situ data in an area where an AXBT was not deployed. Despite the accuracy of D20, satellite estimations of the D26 isotherm surface in the LC/WCR complex were underestimated due to the D26/D20 ratio (Figure 5.8bc). In the core of the LC, the observed D26/D20 ratio was near 0.55, much larger than the 0.4 in the climatology. Comparisons to data from other hurricane seasons showed that the observed ratio in 2008 was much larger than other years and the 0.4 climatological ratio was acceptable. Outside of the LC, estimations of D26 were accurate within 10%. AXBT profiles in the LC contained a weakly stratified layer below the ocean mixed layer with a temperature just above 26 C, causing large satellite underestimations of the D26 isotherm while not contributing greatly to the in-situ value of OHC (Figure 5.12), which is discussed further in MLDs were accurate throughout most of the LC other than the region of low SSHAs in the southeast of the array, and MLDs were too shallow in the CCE area by 10 to 15 m. The AMSR-E SSTs were accurate for the pre-storm flight grid, with absolute errors of at most 0.5 C (Figure 5.10bc) which did not show spatial coherence with the locations of the oceanographic features. Despite the underestimation of the D26 isotherm

73 59 in the LC, the maximum magnitude of OHC was accurately diagnosed to be near 150 kj cm -2 (Figure 5.11bc). Again, the southeast portion of the LC was underestimated because of the SSHA minimum. OHC in the region of the CCE was very accurate, while the western WCR was underestimated by satellite altimetry due to the center position of the eddy being too far to the southeast. After Gustav s passage, the D20 isotherm field was accurate in areas with collocated AXBTs and altimetry data. The location and maximum depth of the LC/WCR complex was accurate with nearly negligible differences (Figure 5.7ef). The region of the CCE was not well covered by altimetry (Figure 5.4) such that satellite estimates differed by 80 m. The in-situ and estimated changes of the D20 isotherm followed a similar pattern, with a deepening of the ocean mixed layer of approximately 25 m. A large area extending N/S near 88 W suggested a region of shoaling by over 75 m, which was not realistic and was in an area with low SSHA sampling density. Elsewhere in GCW, estimated shoaling of the thermocline by 20 m was accurately determined. The post-storm D26 isotherm field closely followed the D20 isotherm field spatially, again underestimating D26 isotherm depths in the LC because of the in-situ sub-mld feature (Figure 5.8ef). Near the core of the LC, the two-layer model estimated a deepening of the thermocline by 30 m where the D20 isotherm was observed to shoal by 20 m, an inaccuracy attributed to a lack of satellite altimetry data. Satellites suggested a strong CCE near (87 W, 27 N) along Gustav s track, however this feature was also in an area without direct altimeter coverage. To the north, satellite estimations were accurate to <10%.

74 60 More notable differences existed between in-situ measurements and satellite observations after Gustav s passage, where the satellite measurements underestimated the cooling in the CCR by as much as 1.0 C (Figure 5.10ef). The apparent overestimated cooling in the northern LC/WCR frontal region was an artifact of the OA of in-situ profiles, although evidence showed that the cooling was overestimated in this region by about 0.4 C. Elsewhere, SST differences were not coherent on large scales and within the error range of the satellite measurements. OHC differences in the post-storm field were directly connected to the underestimations of D26 and MLD west of the LC, yet elsewhere OHC estimates were typically accurate to 15% (Figure 5.11ef). The satellite and in-situ data from hurricane Gustav demonstrated that the magnitude and location of maximum values of D20 and OHC can be accurately estimated in the LC. The locations of frontal regions were also accurate in areas where satellite altimetry was available. The climatological ratio of D26/D20 can be errant due to unique characteristics of the in-situ field, causing large underestimations of D26 while not largely affecting the accuracy of OHC.

75 Figure 5.7- D20 from observed in-situ profiles (left column), satellite-derived values (center column), and the satellite bias (right column) before passage of Hurricane Gustav (top row), after (center row), and the change of D20 (bottom row). Unfilled contours show regions of uncertainty of the OA of in-situ data. Scatterplot compares in-situ AXBT measurements and collocated satellite estimates. Locations of AXBTs launches are identified form pre-storm (dots) and post-storm (plusses) flights. 61

76 Figure 5.8 Same as Figure 5.7, but for D26. 62

77 Figure 5.9 Same as Figure 5.7, but for MLD. 63

78 Figure 5.10 Same as Figure 5.7, but for SST. 64

79 Figure 5.11 Same as Figure 5.7, but for OHC. 65

80 In-Situ to Satellite Comparisons: Ike In-Situ Observations A similar analysis was conducted from the passage of Hurricane Ike where an array of AXBTs was deployed across the forecasted storm track. The flight pattern passed over the LC/WCR and CCE to the west on subsequent N/S legs. Afterward, a mature WCR was sampled further to the west before sampling the GCW to the north along an E/W transect. These data confirmed the existence of the closed circulation of the WCR (Figure 5.13a). Maximum depths of D20 and D26 isotherms in the WCR were approximately 300 m and 210 m, respectively (Figure 5.14a). The MLD was still about 60 m deep due to recent passage of Gustav which did not allow for a new surface mixed layer to form (Figure 5.15a). OHC values throughout the LC/WCR are nearly uniform near 120 kj cm -2 (Figure 5.17a). Satellite altimetry shows the CCE was moving slowly southward between the two WCRs. In this eddy, the depths of D20 and D26 isotherms were much more shallow, reaching only 70 m and 30 m, respectively, and the MLD was only 30 m (Figure 5.15a). The OHC values in the CCE were only 40 kj cm -2 which was much lower than the LC due to its shallower thermal structure. Profiles from the western WCR had similar features to its eastern counterpart. The local maximum D20 was about 225 m (Figure 5.13a), and D26 was 130 m (Figure 5.14a). MLDs were 50 m deep, which was roughly the same as the LC/WCR prior to hurricane Ike (Figure 5.15a). This WCR did not experience much direct atmospheric forcing from Gustav, such that the mixed layer did not undergo deepening on the scale observed in the LC/WCR complex.

81 67 Along the northern extent of the observed array, temperature profiles consistently showed shallow MLD of approximately 30 m above a strongly stratified thermocline containing the 26 C isotherm at 50 m and 20 C isotherm at 100 m. SSTs were highest in the western half of the array and the northern extent of the separating WCR at 29.5 C (Figure 5.16a). The eastern portion of the array was slightly cooler at 28.5 C. AXBT transects along 86 W before and after Ike s passage showed structural changes that occurred in the LC/WCR. That is, the D20 and D26 isotherms shoaled by approximately 30 m in the center of the WCR, an area where Ike intensified to a Category 2 storm (Figures 5.13dg and 5.14dg), consistent with the scale analysis suggesting isotherm displacement due to Ekman pumping of approximately 20 m. The WCR broadened eastward during Ike s passage, causing deepening and presumably warming of the thermocline. In the WCR eastern region, Ike caused massive deepening of the mixed layer on the right of the storm track. MLD deepened nearly 100 m to over 140 m (Figure 5.15dg). In the center of the WCR, MLD deepened by 40 m to almost 100 m, while SSTs decreased by 1.5 C. When coupled with the shoaling of D26, OHC decreased by 40 kj cm -2 (Figure 5.17dg). Directly westward in the CCE, observations along 88 W suggested deepening of both the D20 and D26 isotherms by 50 and 30 m, respectively. Wind forced mixing deepened the MLD by 25 m along this transect and the corresponding SSTs decreased by only 0.5 to 1.0 C (Figure 5.16dg), suggesting that the SST minimum in the post-storm OA is artificial. Because of the deepening of the thermal structure and small SST decrease, OHC actually increased by 20 kj cm -2 in this area.

82 68 The pre-storm flight grid did not capture the core of the CCE, such that full analysis of Ike s effects on the CCE is not nearly as comprehensive as Gustav. The poststorm flight shows the core of the CCE is NW of the LC/WCR. Based on the OA of the observational data in the core of the CCE, which happened to be directly under the storm track, both the D20 and D26 isotherms shoaled by 40 m with minimal changes to MLD. Notice that SSTs plummeted in this region to the east of the storm track, to as low as 24 C, representing a drop of over 4.5 C (Figure 5.16dg). This drastic drop eliminated all 40 kj cm -2 of OHC near the CCE core. Over this region, TS force winds extended 150 km radially. Finally, the WCR to the west was southwest of the storm track and was only skirted by TS force winds. D20 and D26 isotherms each shoaled by less than 20 m and MLD deepened by 15 m. The corresponding SSTs decreased by roughly 0.75 C resulting in minimal changes to OHC Satellite Estimations Altimetry measurements were scarce over the LC/WCR complex prior to Hurricane Ike s passage, such that the center of the LC was not accurately located and the maximum D20 isotherm was overestimated by 25 m (Figure 5.13bc). Similar to the Gustav case, a region in the southeast of the array with artificial negative SSHAs caused 100 m underestimations of the D20 isotherm. The north and west frontal regions of the LC/WCR complex and the center of the western WCR were very accurately located, although D20 depths were overestimated in the WCR by 20 m. The CCE to the west was estimated to be too shallow because of the OA, where the horizontal gradients along the front were well measured by altimetry, but not over the core (Figure 5.5).

83 69 The D26 isotherm was generally underestimated throughout the LC/WCR complex because of the weakly stratified subsurface watermass slightly above 26 C (Figure 5.12) and the inaccuracy of the estimated center of the D20 maximum. The depths of D26 in the CCE were relatively accurate, although the location was inconsistent between the OA of in-situ and satellite data (Figure 5.14bc). The D26 isotherm in the western WCR was underestimated by over 20 m because of a similar subsurface layer as in the LC (Figure 5.12). Figure 5.12 (Top) Location of AXBT deployments in the LC (blue square) and WCR (red circle) overlaid on SSHAs. (Bottom) Profiles comparing characteristics in the LC (blue) and WCR (red) prior to passage of Hurricane Ike. AMSR-E accurately depicted SSTs prior to Ike s passage, with maximum errors on the order of 0.5 C and in the margin of error of the satellite and AXBTs (Figure 5.16bc). Maximum OHC in the LC was overestimated by 15 kj cm -2 and the southeast extent of the array was greatly underestimated by 50 kj cm -2 because of inaccurate SSHA OA data (Figure 5.17bc). The northwest extent of the LC/WCR complex was poorly located by satellite, leading to large differences between in-situ observations and satellite estimates in that small area. Pre-storm OHC in the western WCR was overestimated due to of deep MLD estimates (Figure 5.15b).

84 70 After hurricane passage, the estimated location of the deepest D20 isotherm was again too far to the north, however the magnitude of D20 was comparable between in-situ and satellite measurements (Figure 5.13ef). The westward extent of the LC/WCR complex was consistent with in-situ measurements, such that the accuracy of the satellite estimation was near 10%. Altimetry suggested that the CCE retracted to the north, however this region was not well covered by altimeters in both the pre- and post-storm timeframe. The shape of the region of GCW was accurate, although the D20 isotherm was underestimated by 40 m. Satellite estimates were consistent with in-situ observations in the southern portion of the LC, suggesting shoaling of 20 m, but to the north, deepening of the D20 isotherm by 30 m was observed where the satellite product showed shoaling of over 40 m. In the CCE, the overestimated deepening resulted from errors in the pre-storm OA, and modest shoaling of D20 in the WCR of 25 to 40 m was comparable to in-situ changes. Estimated values of the D26 isotherm were again significantly underestimated because of the subsurface temperature structure. The estimated changes between pre- and post-storm D26 isotherms mirrored the changes of D20 (Figure 5.14ef). The satellite technique did not depict the 75 m of deepening of the MLD caused by Ike s passage (Figure 5.15e), leading to a satellite overestimation of OHC decrease by 50 kj cm -2 (Figure 5.17ef). The two-layer framework lacks atmospheric forcing, such that the MLD deepening could not be resolved. There was a very pronounced cold wake behind Hurricane Ike, particularly outside of the LC (Figure 5.16h). Satellite SSTs showed the southern extent of this cooling reaching further south than observed leading to underestimations of OHC in the

85 71 area. Due to the large region of satellite SSTs below 26 C there was a large region of no heat content in the area of the CCE after Ike s passage, which was consistent with in-situ measurements. In the LC, the great underestimation of MLD deepening led to an overestimation of the reduction of OHC by over 40 kj cm -2 (Figure 5.17h). The satellite data correctly identified a region where OHC increased, however the amount of heating was overestimated by 40 kj cm -2.

86 72 Figure 5.13 D20 from observed in-situ profiles (left column), satellite-derived values (center column), and the satellite bias (right column) before passage of Hurricane Ike (top row), after (center row), and the change of D20 (bottom row). Unfilled contours show regions of uncertainty of the OA of in-situ data. Scatterplot compares in-situ AXBT measurements and collocated satellite estimates. Locations of AXBTs launches are identified form pre-storm (dots) and post-storm (plusses) flights.

87 Figure 5.14 Same as Figure 5.13, but for D26. 73

88 Figure 5.15 Same as Figure 5.13, but for MLD. 74

89 Figure 5.16 Same as Figure 5.13, but for SST. 75

90 Figure 5.17 Same as Figure 5.13, but for OHC. 76

91 Discussion The field experiment associated with Gustav and Ike provided a unique opportunity to analyze hurricane impacts on upper ocean thermal structure in two storms (e.g., a second following the first with similar tracks). Hurricane Gustav was a relatively compact storm, with a small wind field during passage through the GOM, while Hurricane Ike was much more expansive and had larger impacts on the GOM structure (Figure 5.18). At the time of passage over the LC, the integrated kinetic energy of hurricane force winds for Gustav and Ike were 16 TJ and 65 TJ, respectively (Powell et al. 1998). Additionally, Gustav passed over the GOM much faster, with storm motion averaging about 16 kts, compared to about 8 kts for Ike. Considering the differences of wind field size and translational speed, it was expected that ocean thermal structure was more affected by Ike than Gustav. In the LC, Figure H*Wind field produced by HRD for Gustav at 1030 UTC on August 31 (left) and Ike at 1930 UTC on September 10 (right). At these times, Gustav s maximum wind speed was 99kts (Category 3) and Ike s was 89kts (Category 2).

92 78 MLDs deepened by 30 m in Gustav and over 60 m in Ike. Likewise, MLDs in the CCE deepened twice as much after Ike s passage compared to Gustav. Deeping of the ocean mixed layer is coupled to SST cooling. Near-inertial currents in the upper ocean forced by air-sea momentum fluxes erode the top of the thermocline, entraining cooler thermocline water. For both storms, the strongest cooling occurred to the right of the storm track in the northern portion of the array. MLDs were relatively shallow and the thermocline was strongly stratified in this area of GCW and a CCE. Currents in the ocean mixed layer are stronger in CCEs than in the LC /WCR complex, which enhances shear-induced mixing at the bottom of the mixed layer (Price 1983, Jaimes and Shay 2009). Cold thermocline water mixes with the thin mixed layer, leading to pronounced SST cooling. SSTs in GCW and CCEs cooled by up to 3 and 4.5 C in hurricanes Gustav and Ike, respectively. In LC and WCR regimes, however, SST cooling is inhibited because of the deeper depths that tend to be resistive to shearinduced mixing (Shay and Uhlhorn 2008). Mean mixed layer currents are weaker which limits shear-induced mixing. The thermocline is more weakly stratified in WCRs and the LC, such that any water that is entrained is not much colder than mixed layer temperatures. Also, the thermocline water is entrained throughout a deeper mixed layer, diminishing the amount of cooling of surface waters. Such physical processes likely occurred during passage of Gustav and Ike, considering the enhanced cooling in CCE and GCW regions Satellite Differences Analysis The satellite estimation of ocean structure using the SMARTS climatology was successful in identifying the primary oceanographic features during the passage of

93 79 hurricanes Gustav and Ike. Mean errors and biases from pre-, in-, and post-storm flights provided insights to the abilities of the two-layer model (Table 5.2). Satellite estimations of the D26 isotherms for the 2008 hurricane season were 10 to 20% more erroneous than observed in other research campaigns due to the weakly stratified layer beneath the OML just above 26 C. Despite the absolute difference between in-situ measurements and satellite estimations of the D26 isotherm, OHC was still relatively accurate, particularly for Hurricane Gustav (RMSD=23.2 kj cm -2 ). However in Hurricane Ike, OHC errors were greater due to the significant underestimations of mixed layer deepening. AMSR-E typically captured the pre- and post-storm SST field. Pre-storm fields were fairly uniform, which was observed by AXBTs and satellite altimetry. The poststorm satellite field was generally accurate in the magnitude of SST cooling, but the spatial extent of the cold wake was inaccurate for Hurricane Ike. During in-storm flights, SSTs were overestimated by AMSR-E by nearly 1 C between both storms. Post-storm MLD errors highlighted the shortcomings of the altimetry based MLD adjustment technique. The satellite technique did not capture the extensive deepening of Gustav D20 (m) D26 (m) MLD (m) SST ( C) OHC (kj/cm²) Pre-Storm (36) 37.6 (1.8) 24.4 (-11.4) 12.4 (-5.7) 0.47 (0.20) 21.1 (-1.6) Post-Storm (38) 38.9 (-6.5) 26.9 (-17.0) 14.1 (-10.7) 0.43 (0.01) 25.6 (-11.1) All (118) 29.9 (-5.1) 30.4 (-18.0) 18.1 (-12.5) 0.64 (0.31) 23.2 (-5.1) Ike D20 (m) D26 (m) MLD (m) SST ( C) OHC (kj/cm²) Pre-Storm (33) 33.8 (-12.0) 35.3 (-25.2) 15.2 (-7.3) 0.31 (0.12) 19.7 (-5.7) Post-Storm (31) 43.4 (-31.1) 39.5 (-28.5) 37.8 (-28.0) 0.55 (-0.33) 31.8 (-24.3) All (129) 46.6 (-22.0) 41.6 (-30.7) 33.0 (-24.1) 0.88 (0.21) 30.4 (-14.7) Table 5.2 RMSD values for D20, D26, MLD, SST, and OHC from the pre-storm and post-storm AXBT flight grids. Satellite bias is in parentheses. Only AXBTs recorded and processed by University of Miami s Upper Ocean Dynamics Laboratory were used for this analysis.

94 80 the mixed layer, particularly for Hurricane Ike with its larger wind field. In the two-layer model framework, vertical mixing of upper ocean waters above the D20 isotherm would not significantly affect SSHAs. In Gustav, where there was more modest mixed layer deepening, MLD calculations worsened by 7 m between pre- and post-storm surveys. Errors were much larger after Ike s passage, with an average MLD underestimation of 28 m using updated SSHAs. As anticipated, the MLD calculation technique did not perform well after a strong wind-forcing event. Based on the errors in diagnosing MLD deepening, inferences of heat exchange between the atmosphere and ocean must be avoided when using only satellite data, particularly when two storms cross over the GOM in a small timeframe. In the area of the WCR/LC complex during Hurricane Ike, satellite estimations suggested heat loss in the upper ocean by 50 kj cm -2, where in-situ data determined only 20 kj cm -2 of heat loss occurred. A large portion of this heat loss was due to a mass flux out of the upper ocean. In most post-storm environments, it should be expected that MLD, and therefore OHC, will be underestimated by satellite data and the associated heat loss will be exaggerated Near-Inertial Wake In-situ evidence of a potential storm wake existed, likely caused by near-inertial currents. The near-inertial oceanic response to hurricanes has been well documented in models and observations (Chang and Anthes 1978; Price et al. 1981; Gill 1984; Shay and Elsberry 1987; Shay et al. 1998; Jaimes and Shay 2009). In the wake of Hurricane Ike, a coherent pattern of positive and negative D20 anomalies was observed along the storm track (Figure 5.19). The amplitude of the wave on the 20 C isotherm surface was approximately 40 m, with a full wavelength of 400km. At this latitude, the inertial period

95 81 (IP) is 28.7 hours, which put the post-storm flight approximately 3-4 IPs after storm passage, so near-inertial currents would still be expected to exist (Shay and Elsberry 1987). Ike took approximately 30 hours to travel Figure 5.19 Changes in observed D20 between pre- and post-storm flights showing evidence of an inertial wake. between the two D20 anomaly minima, which corresponded with the inertial period and provided support that the anomalies were due to the near-inertial wake Daily Variability Thermistor Drifters An array of drifters was deployed in front of Hurricanes Gustav and Ike during the 2008 hurricane season. Each drifter was equipped with a series of thermistors to measure hourly changes in upper ocean temperature structure. Comparing the hourly data from the drifters to the daily data derived using SMARTS revealed the variability of OHC on timescales shorter than 24 hours which cannot be resolved by the daily product. The inertial wake of the storms could be seen in the trajectories of the drifters (Figure 5.20). Inertial currents forced Ekman pumping, which caused D26 isotherms to Figure 5.20 Track of drifter launched ahead of Hurricane Gustav during the 2008 hurricane season overlaid on SSHA measured by altimetry data. The inset shows the track of the drifter with red circles at 0000Z each day.

96 82 fluctuate by 10 m around a mean of about 42 m. Outside of periods when SSTs dropped due to mixing and surface fluxes, SSTs remained relatively stable with diurnal cycling of about 0.5 C (Figure 5.21). Depending on the phase differences between the near-inertial currents and the diurnal temperature cycle, daily OHC variation was either dampened or enhanced. Hourly data from the drifter dataset provided only a small portion (<1%) of data for the SMARTS analysis. Using airborne expendable profilers in the post-storm environment with strong near-inertial surface currents, values of OHC could vary by more than 20 kj cm -2 depending on the timing of AXBT launches. Such sizeable Figure 5.21 D26 (top), SST (middle), and OHC (bottom) from drifter (solid black) showing diurnal cycles due to near-inertial currents forced by Hurricane Gustav. Satellite data using the SMARTS Climatology (dashed red) cannot capture this daily variability.

97 83 oscillations contributed to the differences between nearly instantaneous observations by the profilers and 10-day altimetry data used in the SMARTS calculations Seasonal Thermostad The satellite-derived thermal structure after hurricane passage did not capture the immense deepening of the mixed layer due to wind forced mixing. In the LC after Ike s passage, MLDs in some locations were observed to be deeper than 130 m in an area where climatological MLDs are less than 30 m. Such extensive deepening occurred in part due to subsurface thermostad water, a phenomenon unresolvable in the trapezoidal upper ocean of the two-layer model. Over the course of winter, sensible heat loss from the ocean to the atmosphere causes SSTs to cool and buoyancy-driven mixing and deepening of the mixed layer. Once atmospheric temperatures start to increase, the surface waters are warmed. This buoyant surface layer is primarily mixed downward by wind forced mixing, such that in the absence of a very strong atmospheric forcing event, a homogeneous mixed layer with a temperature of approximately the SST lies on top of a nearly homogeneous layer with a temperature near the winter mixed layer temperature. Such a phenomenon was observed in the GOM during the 2008 hurricane season. In the LC/WCR complex, a layer of water existed below the surface mixed layer from approximately 75 to 200 m. This layer was weakly stratified with temperatures ranging from 26.0 to 27.5 C. Consulting the AMSR-E SSTs from the previous winter, minimum winter SSTs in the LC fell within that temperature range. In this subsurface layer, the temperature gradient was approximately 1.5 C/100 m, whereas directly below, the temperature gradient was 5.0 C/100 m in the permanent thermocline. This subsurface

98 84 feature was also observed in the WCR to the west, which shed from the LC in late May The subsurface thermostad waters are particularly prevalent in the LC rather than GCW because of the climatological structure of the upper ocean. In GCW, the thermocline is much shallower and more strongly stratified than in the LC (Figure 5.3). Therefore, under similar season atmospheric buoyancy and shear forcing, the winter mixed layer is much deeper in the LC than in the GCW. Once atmospheric temperatures begin to rise, a new surface mixed layer is formed. In the shallow mixed layer of GCW, atmospheric forcing easily mixes the warmer surface waters down to the base of the winter season mixed layer, such that there is no significant evidence of mode water. In the LC, mechanical atmospheric forcing does not mix surface waters to the base of the much deeper winter mixed layer, resulting in the formation of a subsurface layer of mode water with the temperature of the winter mixed layer. In 2008, the winter MLDs in GCW and the LC were approximately 75 m and 200 m, respectively. The passage of hurricanes Gustav and Ike directly over the LC provided the necessary atmospheric forcing to mix surface water deep into the water column (Figure 5.22). Prior to Hurricane Ike, the surface mixed layer was about 60 m, sitting above a layer of weakly stratified water down to 175 m. After passage, the MLD deepened to about 125 m in the same location, and the water below the mixed layer is more strongly stratified.

99 85 Figure 5.22 Temperature profiles in the LC at 23.5N, 85.0W before (blue) and after (red) Hurricane Ike. Climatological profiles from GDEM (solid) and WOA (dashed) are also shown. This example shows the limitations of narrowing down the upper ocean characterization to a single parameter. At this location, D26 moderately decreased from 147 to 143 m, and similarly, OHC decreased from 121 to 111 kj cm -2. Despite the small changes in these values, the upper ocean underwent drastic structural changes which would affect the dynamical properties. For instance, when surface currents are created by surface wind stress, the current is typically roughly uniform through the mixed layer (Shay et al. 2011). In areas with shallow mixed layers, the mean mixed layer currents are relatively stronger because the momentum is distributed over a thinner layer, which enhances mixing by shear instabilities at the base of the mixed layer. Therefore, differences in MLD affect SST cooling, which has direct implications in fluxes at the airsea interface. Although subsurface mode water formation can occur on an annual basis, the GDEM and WOA climatologies did not portray this annual cycle. Observed MLDs near the core of LC were typically more than 20 m deeper than the climatology. Also, there is

100 86 not a strongly stratified thermocline in the climatologies as seen in most of the 2008 observations, but rather a layer with nearly constant stratification Analysis of Updated Methodology The new estimation of climatological reduced gravity greatly improved the twolayer framework for estimating upper ocean thermal structure. The updated g'* value was typically larger such that the magnitude of the estimated D20 isotherm anomaly was suppressed. Previous estimations accurately located the oceanographic features while overestimating the D20 anomalies, which led to overestimations of OHC in the LC and GOM. Improvements were primarily found in D20 estimates, reducing RMSD for Hurricanes Gustav and Ike by 30% and 10%, respectively. Moderate reductions of RMSD for D26 isotherms and MLDs improved OHC calculations by about 15%.

101 Chapter 6: SMARTS Impact on SHIPS SHIPS is an operational forecasting tool used by the National Hurricane Center to help forecast changes in tropical cyclone (TC) intensity. The statistical-dynamic model was developed using multiple linear regression of atmospheric and oceanographic parameters found to affect TC intensity (DeMaria et al. 2005). Regressions were individually calculated for each forecasting period from 6 to 120 hours in 6 hour increments. For a predictor to be included in SHIPS, it must be significant at the 1% level for at least one forecast period. Operationally, the model utilizes near real-time atmospheric and oceanic observations and the latest forecasted fields from the GFS forecasting model every six hours. Currently, OHC is a switch in the SHIPS model, which turns on when OHC is above 60kJ cm -2. Without this threshold, the impact of OHC in SHIPS was not significant (DeMaria et al. 2005). This threshold OHC was typically met in the GOM, LC, and Caribbean during hurricane season. Previous studies (DeMaria et al. 2005; Mainelli et al. 2008) calculated the impact of OHC on the SHIPS model by using OHC as a correction to the SHIPS output due to sample size restrictions. More recently, OHC was included with all other predictors in the multiple regression calculation, resulting in reducing the impact of OHC on intensity forecasts in the 2 to 4 day forecasting period (DeMaria, personal communication), although there are no physical principles explaining this apparent contradiction. The 2010 coefficients for intensity change based on OHC provided by DeMaria showed moderate (<5kts) intensification of TCs in regions of OHC greater than 150kJ cm -2 for the first 60 hours of the forecast period (Figure 6.1). From 60 to 96 hours, the SHIPS coefficients actually suggested higher values of OHC would act to 87

102 88 weaken storms. For forecasts beyond 96 hours, OHC enhanced intensification. It should be noted that TCs rarely experience OHC greater than 120 kj cm -2 for longer than 4 days, other than for particularly slow moving storms or storms moving longitudinally along the LC. Figure 6.1 Intensity correction for all forecast periods from all TC data in the SHIPS database. Track-averaged values of OHC determine the magnitude of intensity change based on a linear regression. 6.1 Changes to SHIPS OHC Field between Mainelli and SMARTS As shown in the previous chapter, Mainelli (2000) typically overestimated OHC in the GOM and LC on average by 20 kj cm -2. This thereby reduced the apparent effect of OHC on TC intensification in SHIPS. Under the most recent SHIPS calculations, the wind speed correction due to OHC would be underestimated in the GOM and LC. As suggested by Figure 6.2, the OHC differences were estimated for Gustav and Ike with the Mainelli and SMARTS climatologies and the resulting fields of SHIPS wind speed adjustment for the 120h forecast using one month of satellite data surrounding hurricanes

103 89 Figure 6.2 OHC (left) and SHIPS OHC wind speed adjustment (right) estimated using Mainelli (top) and SMARTS Climatology (middle) with differences (bottom) of SMARTS minus Mainelli. Gustav and Ike. The intensity adjustment was largest at this forecast length and showed the largest impact on TC intensity change in SHIPS. Using the SMARTS Climatology instead of Mainelli decreased the satellitederived value of OHC in the LC and Caribbean by more than 30 kj cm -2 and decreased the satellite value of the LC in the GOM by more than 20 kj cm -2. In the Atlantic Ocean Basin, the SMARTS Climatology caused the OHC estimation to increase by upwards of 30 kj cm -2, however OHC remained below the 60 kj cm -2 threshold and therefore did not

104 90 contribute a correction to the intensity calculation in this area. Comparisons of estimated wind speed correction using the 2010 SHIPS OHC coefficients showed SMARTS reduced the intensity adjustment nearly everywhere. The adjustment decreased between 5 and 10 kts in most of the LC and the peak reduction was 15 kts near the Yucatan Strait. Predictors in SHIPS are track-averaged, so a TC forecasted to travel over the LC and Yucatan Strait and into the GOM would be forecasted to be about 10 kts weaker using SMARTS instead of Mainelli. Such large differences suggested that implementing the SMARTS Climatology into the SHIPS calculation could significantly impact the SHIPS coefficients and threshold value. 6.2 Discussion Estimating upper ocean thermal structure with the SMARTS Climatology will require recalculation of SHIPS coefficients because of a bias correction from the Mainelli climatology. The large differences in OHC resulted in sizeable changes to the SHIPS intensity adjustment, particularly over the LC and GOM along common TC tracks (Figure 6.3). By correcting for the large bias, the threshold OHC necessary to be a factor in SHIPS calculations must be reduced. Previous studies (Leipper and Volgenau, 1972; Shay et al. 2000; Hong et al. 2000; Uhlhorn and Shay, 2004) demonstrated that less than 20 Figure 6.3 Average difference of SHIPS OHC 120h wind speed adjustment using SMARTS and Mainelli (2000) for 30 days surrounding passage of Gustav and Ike in 2008.