JOURNAL OF GEOPHYSICAL~EARCH, VOL. 105,NO. C10,PAGES 23,967-23,981, OcroBER 15, 200)

Cyclone surface pressure fields and frontogenesis from NASA scatterometer (NSCAT) winds David F. Zierden, Mark A. Bourassa,and JamesJ. O'Brien Center for Ocean-Atmospheric Prediction Studies, Florida State University, Tallahassee,32306-284{)

Abstract. Two extratropical marine cyclones and their associatedfrontal features are examined by computing surface pressure fields from NASA scatterometer (NSCAT) winds. A variational method solves for a new surface pressure field by blending high- resolution (25 kIn) relative vorticity computed along the satellite track with an initial geostrophic vorticity field. Employing this method with each successivepass of the satellite over the study area allows this surface pressure field to evolve as dictated by the relative vorticity patterns computed from NSCA T winds. The result is a high-resolution surface pressure field that captures features such as fronts and low-pressure centers in more detail than National Centers for Environmental Prediction (NCEP) reanalyses.While using the actual relative vorticity to adjust the geostrophic vorticity ignores the ageostrophy of surface winds, which can be significant in the vicinity of fronts and jet streaks, it is a necessaryapproximation given that the technique uses only surface data. The NSCAT surface pressure fields prove to be nearly as accurate as NCEP reanalyseswhen compared to ship and buoy observations,which is an encouraging result given that NCEP reanalyses incorporate a myriad of data sources and the NSCA T fields rely primarily on one source. In addition, the high-resolution relative vorticity fields computed from NSCAT winds reveal the location of surface fronts in great detail. These fronts are verified using NCEP analyses,in situ data, and satellite imagery.

1. Introduction the assimilation of ERS-1 winds into the European Centre for Medium-Range Weather Prediction (ECMWF) model im- The lack of conventional data over the oceanshas long been pacted the forecasts only marginally [Hoffman, 1993].Andrews a limiting factor in the accuracyof weather forecasting [Atlas et and Ben [1998] demonstrated marked improvements in the ai., 1985]. Often, the only data available are surface observa- United Kingdom Meteorological Office forecasts by assimilat- tions from ships and buoys, which are sparseoutside shipping ing ERS-1 winds, particularly over the Southern Ocean where lanes and the Tropical Ocean-Global Atmosphere experiment conventional data are sparse. More recently, Atlas and Hoff- (TOGA)-Tropical Atmosphere-Ocean (TAD) buoy anay. man [2000] found that the greatest positive impacts of NSCAT Conventional data are now supplemented with satellite data, winds on NWP forecastsresulted from the vertical extensionof and the challenge lies in finding methods to utilize these new surface winds and the modification of surface pressure fields. data sources best. One such source is surface wind vector Some studies have employed scatterometer winds in diag- measurementsform spaceborne scatterometers,which can be nostic studies of midlatitude and tropical cyclones.In many of used to derive surface pressure fields. these studies, scatterometers were only one of many data NASA scatterometer (NSCAT) and other scatterometers sources implemented in improving NWP analysesof the fea- provided wind measurements over the ocean with much ture [Antheset aI., 1983; Tomossiniet aI., 1998;Liu et aI., 1998]. greater resolution and coveragethan were previously available. In contrast, Harlan and O'Brien [1986] assimilated only Sea- Recent research looked to find ways to utilize this high-quality sat-A scatterometer data with National Centers for Environ- data source. A common approach was to form gridded prod- mental Prediction (NCEP, formerly NMC) surface pressure ucts [Liu et al., 1998; Bourassa et al., 1999; Verschellet al., fields to obtain an improved estimate of the central pressurein 1999].These gridded products were used to drive ocean circu- the QE-II storm of 1978. All of these studies showed how lation models, to improve surface fluxes for general circulation scatterometer winds improved estimates of the central surface models, and to study the evolution of regional winds. pressuresand predicted intensities of the systems. The assimilation of scatterometer winds has also had a p0s- Brown and Zeng [1994] developed a method for computing itive impact on numerical weather prediction (NWP). Early surface pressure fields in midlatitude cyclones using ERS-1 impact studies [Baker et aI., 1984, Duffy et ai., 1984] using winds from a single swath and a boundary layer model. Surface Seasat-A winds improved surface analyses significantly, but gradient winds were found using ERS-1 wind data as input to had limited effects on higher levels and forecasts. Duffy and the boundary layer model. Surface pressures were then com- Atlas [1986] first demonstrated improved forecasts with the puted from the gradient winds, and a reference pressure was vertical extension of Seasat-A surface winds, which adjusted located within the field. The computed surface pressure fields mass at higher levels of the model, not just the surface. Later, distinguished fronts and located the centers of cyclones accu- Copyright2(XX) by the AmericanGeophysical Union. rately while giving improved estimates of central pressureover Papernumber 200)J~2 NCEP analyses.Hsu and Wwtele [1997] employed this method 0148-0227/OOf2{XX}J~2$09.00 with Sea.~t-A winds in a similar study. The strength of the 23.967 23.968 ZIERDEN ET AL.: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS

3000 2. Data 2.1. NSCAT 1!11J!11~j~li!li~i!!!i!. !;!i,,!!!!!!!li6m!!:j The primary data used in this study are the NSCAT2level n winds with a resolution of 25 kIn along the satellite's path. 2000 These winds are an updated version processedby the ,JetPro- pulsion Laboratory from measured backscatter using an im- e proved model function. NSCAT operated aboard Japan's .!4 ADEOS for 9 months from late September 1996through June 1997. NSCAT was the first of a new generation of scatterom- 1000 eters; it used many technological advances to improve the quality, coverage, and resolution of near-surface winds. NSCAT's radar operated in Ku band (13.995GHz) rather than the C band as was done by ERS-I. This frequency led to greater accuracy at low wind speeds «4 m S-I), although 0 sensitivity to attenuation by liquid water was increased. Engi- 1000 0 1000 neering advancements in the sensors increased the signal to kID noise ratio of the backscatter measurements,greatly improving Figure 1. Data coverageand resolution along the path of the ambiguity selection. In addition, each wind cell was viewed ADEOS. Dots mark the relative location of each wind sample. from three different angles. The NSCA T radar was dual- polarized from one antenna, providing additional measure- ments to aid in the ambiguity selection. NSCAT was equipped to measure backscatter on both sides of the satellite track, boundary layer approach was twofold: (1) the surface pressure doubling the coverage of ERS-l, which viewed only on one field was derived almost exclusively from scatterometer data, side. and (2) swath data were used directly, without averaging in A digital Doppler filter grouped overlapping backscatter spaceor time. The drawback was that pressurescould only be measurementsfrom the different viewing anglesinto 25 km by computed within the swath of wind data. A discussion of the 25 km cells. The wind speed and direction were computed for each cell using the observed backscatters and a lookup table. accuracy of scatterometer surface pressure fields is given by Calibration/validation of the NSCAT model function was more Zeng and Brown [1998]. These surface pressure fields showed accurate than previous scatterometersbecause of comparisons greatest improvement over NWP analysesover the Southern with high-quality in situ surface observations from research Hemisphere, where the lack of conventional observations can vessels [Bourassa et aI., 1997], National Data Buoy Center cause entire systems to be misplaced or missed all together (NDBC) buoys [Freilich and Dunbar, 1999], and the TOGA- [Brown and Levy, 1986;Levy and Brown, 1991]. TAO array (K. Kelley and S. Dickenson, personal communi- This study makes use of the high-quality NSCAT wind data cation, 1998). In particular, these in situ data included many by deducing surface pressure fields through the use of a vari- observations at low and high wind speeds, enabling accurate ational method. The primary goals are (1) to use NSCAT calI'bration/validation and remoVing the low wind speed biases winds to detemline surface pressure fields, (2) to follow the found in other scatterometers. evolution of surface features descn'bed mostly with NSCAT Attenuation by liquid in the atmosphere, particularly heavy data, (3) to locate and identify surface fronts, and (4) to pro- , is a disadvantage of the Ku band frequency. vide a surface pressure field that could be used to improve Contamination from precipitation droplets can significantly NWP over the oceans. degrade the quality of scatterometer-computed wind vectors. Section 2 descn'bes the data sets, including specifics of Ideally, inclusion of a passive microwave radiometer on the NSCAT and its near-surface wind observations. Section 3 de- satellite platform could have identified contaminated cells and tails the variational method used to detemline surface pres- flagged them appropriately. Unfortunately, mission specifica- sures.The variational method involves the assimilation of rel- tions and funding did not allow for such an instrument to be ative vorticity computed from the NSCA T wind vectors. included with NSCAT, so it is difficult to identify contaminated Surface fronts are located and identified in the relative vortic- cells. Studies are ongoing to determine the effects of precipi- ity field as localized bands of high relative vorticity (section tation on the overall accuracy of the NSCAT winds. 3.1.1). These features are verified as fronts using in situ obser- The ADEOS was a low-altitude, Sun synchronous, near- polar orbiter. In this orbit, NSCAT covered 90% of the ice-free vations and visible GOES 9 imagery. Section 4 uses NSCAT ocean every 2 days. The antenna configuration allowed winds surface pressure fields to follow a case of cyclogenesisand a to be measured in 600 km wide swaths on each side of the case of frontogenesis in the North Pacific. Results show that satellite, with a 400 km gap in the nadir view between the the NSCAT surface pressure fields resolve the structure of swaths (Figure 1). NSCAT2level n wind data covered swaths these features in more detail than the NCEP reanalyses.The on each side of the satellite, with each swath 24 cells wide. NSCAT pressurefields also agree better with NSCAT winds as These rows of 24 cells were perpendicular to the satellite's path far as the location of cyclone centers and the orientation of (Figure 1). horizontal pressure gradients are concerned. Quantitatively, NSCAT proved to be a very reliable instrument, detennining the NSCAT pressure fields compare well with NCEP, espe- near-surface winds (calibrated 10 a height of 10 m) more ac- cially near the reference buoys and where recent satellite data curately and with fewer aliases than previous scatterometers. are available. In the open ocean the chancesof selectin2 an incorrect ambi-

~ ZIERD~ ET AL: SURFACE PRESSURE FIELDS AND FRONI'S FROM NSCAT WINDS 23.969 guity were negligible at wind speedsover 8 m 5-1 (BoIUrlSSGet al.. 1997]. Below that threshold the chances of incorrect am- biguity selection increased with decreasing wind speed. The RMS difference between NSCAT and research vessel winds 'g. was found to be 1.6 m S-1 for wind speed (for wind speem >4 :0... m 5-1) and 13° for direction. They found no statistically sig- oS nificant biases at low or high wind speeds.Freilich and Dunbar [1999] supported these findings in a comparison to quality- controlled NDBC buoy observations. I 165 170 175 180 186 190 195 200 205 210 215 220 225 1.2. NCEP Reanalysesaad NDBC Buoys Lonlilude NCEP reanalysis mean sea level pressures are used to ini- FIgure 2.. Typical daily coverage of NSCAT winds over the tialize the pressure field and to update boundary conditions. study area. Gaps within the swaths indicate mjssing data. Na- The NCEP mean sea level field is different from its surface tional aimate Data Center (NCDC) buoy locations are marked with a square. preaure field, primarily over elevated land surfaces and be. causeof small variations due to the limited spectral resolution of the model. The NCEP mean sea level pressure field is the most accurate representation of surface pressures over the vorticity is computed from the pressure field. A variational ocean. Throughout the remainder of the text, the term "sur- method solvesfor a new geostrophic stream function, minimi2- face pressure" will apply to all preaure values, including the ing the difference between the new geostrophic vorticity and NCEP reanalysis mean sea level pressureproduct. The NCEP the old geostrophic vorticity where satellite data are present mean sea level pressure data are available on a 2.5° global grid and minimimg the difference between the new geostrophic at 6 hour inte~ A third data source is in situ surface pres- vorticity and the old geostrophic vorticity where no satellite sures from NDBC buoys 46X)3, located at latitude 51°51'5"N data are present. The result is an updated surface pressure and longitude 2005'3~, and 51001, located at latitude field that captures the features found in the NSCAT vorticity. 23~'4~ and longitude 1~'1"E. The treatment of NSCAT relative vorticity as gcostrophic ig- nores the ageostrophy of surface winds, which can be signifi- cant in the vicinity of fronts and jet streaks. However, this 3. Methodology approximation is necessaryin the absenceof upper air thermal 3.1. Method and Study Area and mass fields. Repeating the procedure with each new pass A goal of this study is to devise a technique of deriving of the satellite over the domain allows the field to evolve in surface pressure fields from NSCA T winds, which have greater time as dictated by NSCA T data. The steps of this procedure coverageand better resolution than ERS-l winds. Ute Brown are describedin detail in sections3.2--3.6. and Zeng [1994], individual swath data are used, preserving the spatialresolution and small-scalefeatures presentin NSCAT 3.2. Computing ReIatiYeVorticity winds. Unlike Brown and Zeng [1994J, any data within the NSCAT winds are of high spatial density and are located on domain has an influence on the entire pressure field Also, the a regular grid aligned with the satellite path; consequently, surface pressurefield will evolve in time with each satellite pa.u relative vorticity is easily computed using centered finite dif- over the domain. ferences. The speed and azimuthal direction of the winds are The study area is the North Pacific Ocean between 2()0and converted to across-track (u') and along.track (v') compo- 55~ latitude and between 1~ and 225~ longitude. It is nents in a coordinate system aligned with the satellite track. largely free of land and ice (scatterometers only work over The relative vorticity L'..at each interior point in the two swaths water) and large enough to capture synoptic-scale systems. is Midlatitude '-1'cionestrack through the region. Furthermore, ,= (V:+l,j- V:-l,j)/4.I:' - (U:4+1- uv-J/Ay', (1) conventional data are sparse, and nwnerical weather predic- tion analysescan use improvement in this area. The study area where i denotes cell position across the swath,j denotes cell could expect to see three to four passesof the satellite in the position along the swath, and x' and y' are across-trackand ascending node and another three to four passes in the de- along-track locations. 4.1:' and Ay' are twice the cell size and scendingnode each day (Figure 2). All computations and anal- are computed directly from the latitude and longitude of the ysesare performed on a 0$ grid over the domain, preserving corresponding data points instead of being held constant at 50 the small-scalefeatures present in the high-resolution NSCAT kin. They varied between 49 and 51 kIn. If wind data are winds. missing at any of the neighboring cells, the relative vorticity at The technique developed in this study builds on the that point is considered missing. Delunay triangulation and strengths of Brown and Zeng [1994Jand incorporates the vari- interpolation [Renka, 1982] then transfers the satellite relative ational method of HarlDn and O'Brien [1986]. The procedure vorticity onto the 0.250grid. (Plate 1) beginswith an NCEP mean surface pressure field and The RMS difference in NSCAT wind speedsis -1.6 m 5-1 interpolates it onto the 0.250grid over the domain. For each when compared to in situ data [Bounl&la et oJ., 1997;F~ subsequent pass of the satellite over the study area the two and Dunbar, 1999]. This uncertainty propagates through rela- swaths of NSCAT wind data are RuimilAted into the pressure tive vorticity calculations and results in an uncertainty in rela- field. Although surface pressure and winds are physically dif- tive vorticity values of roughly 1 x 10-45-1, similar in mag- ferent data types, they are related through vorticity. Relative nitude to maximum values in strong synoptic-scale systems. vorticity is computed in the swaths from NSCAT winds and However, this RMS uncertainty in NSCAT wind speem in- then interpolated to the 0.250domain grid, while geostrophic cludes both systematic biases and random errors in the 'i:$.m ZIERDEN ET AL.: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS

Update boundary conditions every 12 hours (NCEP)

.~ '3 ~ ~ Assimilate vorticity from ~ each new satellitepass '3 with variationalmethod 1 .3 ~ Compute relative~~~ vorticity from ~

NSCAT winds

-':

~r;~z:~;:g;;;'2~~:~~~::~.; J' .._.'l/ ~"'"- ,...... ~~~~ Updatedpressure field Plate Methodology of computing surface pressure fields from NSCA T winds.

NSCAT winds and the in situ data. Since relative vorticity at higher levels obscures features at lower levels and at the involves the difference in u and v components, most of the surface. Only in well-organized, sharply defined systemscan uncertainty in relative vorticity is due to random error alone in the approximate location of surface fronts be found from such the NSCAT winds. The consistencyof NSCAT relative vortic- passive sensors. Katsaros et at. [l996J used parameters from ity fields with surface features and NCEP geostrophic vorticity active/passivemicrowave sensorsaboard Special SensorMicro- suggeststhat the uncertainty in the NSCA T relative vorticity is wave Imager (SSM/I), Geosat, and ERS-1 satellites to study small «1 X 10-5 S-I) and that random errors in the NSCAT the evolution of marine cyclones. They found that frontal winds are <0.7 m S-I. zones could be identified by large gradients in the SSM/I inte- 3.3. Frontal Detection grated water vapor. Unfortunately, this parameter is a measure A secondaryresult of this study deals with the strong signa- of water vapor over the entire atmospheric column and cannot ture of surface fronts in the relative vorticity fields computed isolate features at the surface. The location of fronts changes from NSCAT winds. The identification and location of fronts with height because of the sloped surface of the interface using satellite remote sensing has long been a topic of great between air masses.Consequently, the integrated water vapor interest. Visible and IR imagery has taught us a great deal can only identify a broad frontal zone representative of many about the structure and evolution of extratropical cyclones levels rather than a sharp line at the surface. Katsaros et at. l~rlsan, 1980; Browning and Roberts, 1994).This type of im- [1996J also used Geosat and ERS-1 altimeter wind speeds to agery, however, ha.~one inherent drawback. Broad cover identify wind speed gradients in the vicinity of fronts. These

~ ZIERDEN ET AL: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS 23,971

Vorticity

10 m/s -10-- 5 =1 1 5 10 10'" s-a Plate 2. NSCAT winds and relative vorticity from two satellite passesaround 1800 UTC, December 20, 1996. Isotherms (degrees Celsius) are from NCDC ship and buoy data (asterisks mark individual observa- tions). A is identified by the band of high relative vorticity (red) in the right swaths.

altimeter data were often obscured by precipitation in the area higher pressure meet. Winds curve cyclonically in response to of interest, especially in the frontal zones. the localized pressure minimum, resulting in high positive rel- Surface fronts can be identified in NSCAT winds by changes ative vorticity values (in the Northern Hemisphere). in wind speed and direction. These changes are often subtle, Plots of relative vorticity in the NSCAT swathsare presented though, making the exact location of a front difficult to deter- showing linear bands of high relative vorticity near-surface mine by visual examination of the wind fields. When relative fronts. The dual swathsof NSCAT winds and relative vorticity vorticity is computed from NSCAT winds, however, even sub- (Plate 2) show a mature cyclone at 1800 UTC December 20, tle changesin wind speed and direction lead to large values of 1996. The cyclone is centered near 3SONand 186°E, as evi- relative vorticity (>1 X 10-4 S-I). Fronts are characterizedby denced by the circulation center and high relative vorticity relatively low pressure at the boundary where air massesof values. Unfortunately, the cyclone center is in the gap between

Plate 3. Goes 9 visible imageryfrom 1800UTC, December20, 1996.Solid white lines mark the study ~ domain. 23,972 ZIERDEN Ef AL.: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS

10 m/s i t) ' -.. -5 ~ 10-& s. Plate 4. NSCAT winds and relative vorticity from two satellite passesaround 1800 UTC, January 5, 1997. Isotherms (degrees Celsius) are from NCDC ship and bouy data (asterisks mark individual observations). A extends form the cyclone center eastward across the northern portion of the domain.

the two passes of the satellite, and its exact location cannot be Warm and stationary fronts have strong NSCAT relative determined. A narrow band of high relative vorticity curves vorticity signaturessimilar to those of cold fronts, though warm southeastward from near the cyclone center, possibly corre- fronts are typically weaker. The NSCAT wind and relative sponding to the change in winds along a cold front. Rough vorticity fields from two passesof the satellite around 1800 contours of surface temperature made from National Climate UTC January 5, 1997, are plotted in Plate 4. A developing Data Center (NCDC) ship and buoy observationsverify a cyclone is roughly centered near 4QONand 185°E, and a band temperature drop behind this feature. Furthermore, GOES 9 of strong relative vorticity extends north and east from the visible imagery (Plate 3) shows a classic coma head at the center along a stationary front. There is a dramatic wind shift low-pressurecenter and a band of cloudiness along the trailing along this feature, with winds from the northeast on the north cold front. These two independent data sources confirm that side and from the southwest on the south side. Surface tem- the high 'vorticity band is indeed a signature of the cold front. perature contours from NCDC surface observations show a

Plate 5. GOES 9 visible imagery from 1800 UTc, January 5, t 997. Solid white lines mark the study domain.

~ ZJERDEN ET AL: SURFACE~ FIELDS AND FRONI'S FROM NSCATWINDS 23.97:

strong tempemture gradient from the high-vorticity line north- ward. GOES 9 visible imagery from the same time (plate 5) R'" (S showsa small cyclone near 300N and 2100w, which is also seen as a concentration of high relative vorticity at the samelocation Outside the swaths in Plate 4. North and west of the small cyclone is a broad band Vi} = L', - L'-- (6) of that extends from 3O"N and 185OWnorth and east across the entire domain. This cloud band corresponds well where R is a reduction factor needed to increase the NSCAT with the band of high relative vorticity in Plate 4, confirming relative vorticity to a geostrophic equivalent. the presenceof a stationary front that was first identified in the The NSCA T relative vorticity is a surface value that needsto NSCAT relative vorticity. The NSCAT vorticity field succeeds be increased to a geostrophic equivalent before it is blended in locating the front with 25-50 kID using the resolution of the with geostrophic vorticity. A simple method for relating NSCA T winds and the width of the high vorticity band, which geostrophic or gradient winds to surface winds usesreduction- is more exact than other satellite data sources. rotation factors: Geostrophic winds are multiplied by a con- stant of 0.6-0.9, depending on boundary layer stability, and 3.4. Computing Geostropbic Vortidty rotated counterclockwise 15°-300[Clarke and Hess,1975]. Har- lan and O'Brien [1986] used a least squares method to find an The variational method requires NSCA T relative vorticity to average reduction constant of 0.83 and a rotation {IM;tor of be blended with the geostrophic vorticity of the initial pressure 27.6° between geostrophic and Seasat-A winds. Brown and field. Geostrophic vorticity is given by Zeng [1994] used their boundary layer model to arrive at a reduction constant of 0.667 and a rotation factor of ISOfor V2p+ I~ "" ',= pi neutral stratification. Herein R is chosen to be 1.5, equivalent to Brown and Zeng's reduction factor for neutral stability. The where p is the surface Pressure.p is taken as a constant 1.225 rotation factor is inconsequential since rotating NSCAT winds kg m-3 (U.S. standard atmosphere),! is the Conolis parame- by a constant angle has no effect on relative vorticity values. ter, /3 = dIldy, and Ug is the zonal component of the geostro- The last term on the right of (3) is a penalty function that phic wind. In centered finite difference form. (2) becomes acts to smooth horizontally the solution field Without this term the only solution is A = 0, andsatellite vorticity is inserted 1 = PI; (pj directly into the field In general, the penalty function involves +PI 2p, '4x. the second derivative of the solution field. often in the form of 1 a Laplacian smoother. In this case, however, the Laplacian of (PIJ- 1 + Pi" -1I',)/4y: p is included in the model, and another penalty function must PI; be used The kinematic geostrophic kinetic energy G(p/j) is minimi7~ [Harlan and O'Brien, 1986]: ~( PIJ+ '24.)' (3) "H +PIJ. G(p,) = 2"1 (u: + v:> - 2"j;2jJ1 Vp . Vp. (7) The initial gu~ forCg is computed from the initial NCEP grid. For each subsequent step the pressure fields and all calcula. The coefficients K, and KE are weights that amtrol the tions are perfonned on the 0.250grid. balance between the amount of smoothing to be done and the data misfits. The cost function must be minimi7.edto arrive at 3.5. the solution field Pij: aF 1 ap; = PI (A'+lJ + A 2A,)/4z

1 + PI (AV+ +-', - 2A,)/Ay F(p" {" Af: pI ~ + Pr (AVo +1 .J/24.y (4) K. -P2f [p -pA,)/4x where the terms on the right-handside are summedover all + (P, pA,)/Ay2j - 0 grid points i andj. The first term on the right-handside is +P, (8) commonly referred to as the model (Cg = (1/pf)V2 p + aF (PIf)ug), the unknown p~ and vorticity fields, multi- -Af + XcVf - 0 (9) plied by a Lagrange multiplier Ai/' This term is known as a a&; "strong constraint" [Saski, 1970]. The second term on the aA;aF= (PI1 V2pf+ 7~ II, -" = 0 right-hand side minimizes the data misfits Vij between the new ) (10) geostrophic vorticity {ij and satellite vorticity (where available) and the misfits between the new and old g~phic vorticities Equation (8) can be written as (outside the swaths). ~ KE In the swaths V2A..+ 7 (~a, = 2Pl V2p, :11) 23.974 ZlERDEN ET AL: SURFACEPRESSURE FlEWS AND FRONTSFROM NSCATWINDS and hasa solutionof the form becausethe feature of interest was located in the center of the study area. In the interior of the domain the solution field is KPc. 1I.ij = 2p:j(Pij - Pr»j), (12) loosely constrained by the boundary values, so their solution field realized the full influence of the assimilated scatterometer where POij is the homogeneous solution to (11) and thus sat- data over the low-pressure system.Also, they assimilated sat- isfies (KE/2pf)V2 POij = O. On the boundaries,11. = 0 and ellite data only once for each NCEP analysis,so their solution POij = pij; therefore POij can be found through successive field was not required to evolve in time. overrelaxation given the boundaxyvalues from the initial pres- Neumann boundary conditions are used for this study. The sure field Combining (9) and (12) and letting K = KE/K, pressure gradient normal to the boundary (as determined from lead$to NCEP reanalysis) is computed at each grid point along the In the swath border. Equations (15) and (16) are then solved holding these normal derivatives constant. This approach allows surface {. = {s, + 2PlK (Pij - P(Wj) pressure values to change with the assimilation of NSCAT (13) vorticity, even at and near the borders. The drawback of using derivative boundary conditions is that the spatial mean pres- Outside the swath sure is not constrained: The mean can drift from the initial K value in a manner other than the true temporal evolution of {II = 'If + 2Pl (PII - PIw/)' (14) the mean. The horizontal gradients and relative highs and lows in the solution field are realistic, but while assimilating one Substitutionof (13) and (14) into (10) yields overpass,the spatial mean would drift between 0 and 6 mbar In the swath from ground truth. Without additional measures this error adds up quickly as the procedure is repeated for new satellite passes. The drift in spatial mean pressure is remedied with refer- ence pressuresfrom within the domain. Ideally, the reference points would be located near the center of the study area and away from sharp horizontal pressure gradients or extreme fea- tures. A constant offset could then be added to or subtracted from the solution pressure field to make the solution and reference pressures equal at the reference points. Unfortu- nately, buoys are only located in the domain by Hawaii and the Aleutians, near the southern and northern borders of the do- 1:1 K. -If (Pi,J+l + Pi,J-J/2dy - 2Pl (Pi; - PfNj) = ".. (16) main (Figure 2). The offset is taken as the average of the differences between buoy and solution pressures at these points. Averaging reduces the influence of the locational errors which are solved using successiveoverrelaxation and constant in sharp gradients near the reference points. nonna! derivative boundary conditions. The derivative boundary conditions still present limitations Lagrange multipliers Aij often have a physical interpretation. on the evolution of features near the borders: Large-scale For example, in (9) the Lagrange multipliers are equal to the features are not able to enter or leave the domain as the data misfits. Results show that their spatial distribution is dom- solution field evolves in time. For eXaDlple,consider a low- inated by small-scale noise, with variations at 1 order of mag- pressure system entering the domain on the western border. nitude less than average vorticity values. No physical struc- To capture correctly this feature as it crossesthe border, the tures, such as the edges of the satellite swaths,are discemable normal derivative should change from positive (increasing to- in their spatial distribution. The Lagrange multipliers corre- ward the interior) to negative. For this reason the boundary spond to grid-scale vorticity differences brought about by the conditions must be updated periodically. On the basis of do- smoothing term in the variational method. main size and the frequency with which new passesof the A stated earlier, K is a coefficient that weights the relative satellite occur, new derivative boundary conditions are com- contnbutions of the two constraints in the cost function. Fur- puted from NCEP reanalyses every 12 hours. The 12 hour thermore, the two constraints are not dimensionally homoge- update cycle is chosen so that new boundary conditions are neous. and the coefficient must account for the difference in implemented at the synoptic times of 0000 UTC and 1200 units in the two terms. A value of K = 1 X 10-13 m-2 UTC. Using additional NCEP data to update the boundary produces a smooth pressure ftelQ while preserving the physical conditions does not lessen the dependencyof the solution field structures present in the NSCAT winds and relative vorticity. on NSCA T vorticity data. It simply provides a framework of Higher values put too much weight on minimizing the geostro- large-scalehorizontal pressuregradients to govern the solution phic kinetic energy, resulting in a pressure field with gradients near the borders. that are too relaxed. 3.7. Viability of the Technique 3.6. Boundary Conditions and ReferencePressures A major goal of this study is to describe the evolution of Solving (15) and (16) for surface pressure requires specifi- cyclones based primarily on NSCA T observations. With the cation of boundary conditions on the borders of the domain. assimilation of data from each new pass of the satellite, less Harlan and O'Brien [1986] held the boundary pressure values infOmlation is retained in the pressure field from the initial constant (Dirichlet), setting them equal to the values from NCEP analysis. In 24 hours, seven to nine passesof the satel- NCEP analyses.This condition was effective for their study lite over the domain cover -75% of the area (Figure 2). In 48 ZlERDEN ET AL.: SURFACE PRESSURE FIE1DS AND FRONTS FROM NSCAT WINDS 23,975 hours the total is over W%, with a majority of the domain The first satellite pass covers only a small comer of the covered at least twice. At this point the geostrophic vorticity domain. The NSCA T pressure field changesvery little from the field is described almost exclusively by NSCAT vorticity. The NCEP initialization, aside from smoothing the discontinuities solution pressure field follows from the geostrophic vorticity caused by NCEP's coarse 2.50 grid (Figure 3a, (KXX)UTC field, constrained only by the pressure field from the previous December 18, 1996). This iteration demonstrateshow the field iteration, the derivative boundary conditions, and the two ref- retains the characteristics of the previous step over areaswhere erence pressures. Continued assimilation of new satellite no new satellite information is available for assimilation. After passes-changesthe geostrophic vorticity field (and correspond- the assimilation of eight passes the NSCAT pressure field ing pressure field) as physical features move about and evolve evolves considerably (1800 UTC, December 18, 1996). The in the domain. Ideally, the process continues throughout the low-pressure system near the northern border weakens,while life cycle of the feature of interest. the low-pressure system near the southern border deepens in The technique does,however, have two limitations. First, the response to the strong vorticity values from the last satellite feature (cyclone, front, etc.) must have a strong signature in pass.The NCEP analysisfrom the same time does not intensify the satellite vorticity field. Results show that high values of this system, yet shows a higher central pressure (1003 mbar relative vorticity are concentrated at frontal zones and cyclone compared to 997 mbar) than the NSCAT field. Also, notice centers. If the feature is weak or diffuse, noise and small-scale how the isob~ tend to "kink" where winds turn sharply in variations in the NSCAT relative vorticity overwhelm the responseto features of high vorticity. Sharp bends in pressure large-scalestructure of the feature. The solution field diverges contours are indicative of a sudden change in horizontal gra- from the true surface pressure field, and this problem is com- dient, often associatedwith frontal zones [DjuriC, 1994]. pounded as more satellite passeswith weak vorticity patterns At 1200 UTC, December 20, 1996, the cyclone reaches its are assimilated. Also, cells contaminated by attenuation from matUre stage (Figure 3c). The NSCAT central pressure is 989 liquid water introduce error into the vorticity field. These er- mbar, close to NCEP's value of 987 mbar. NCEP locates the rors do not appear to affect greatly the solution pressurefields center at 38~ and 186~, 10 east of the circulation center as for strong systemsas the attenuation errors are more local in determined from the NSCAT wind vectors. The NSCAT pres- nature and may not influence the large-scale structure of the sure field places the center at 3~ and 185~ 10 east and pressurefield. It is impossible, however, to determine the exact south of the circulation center. The only other difference in the impact of the attenuation problem on the vorticity and pres- two pressure fields is that NSCAT builds higher pressureat the sure fields without knowing which cells are contaminated. northern edge of the domain, over 1050 mbar. The NCEP The technique also breaks down when the feature moves too analysis only has pressures in this area of a little over 1040 quickly through the study area. Consider a cyclone moving mbar. Elsewhere, the tow fields agree well in the overall struc- west to east through the domain at 10 m S-I. On the first day tUre and the placement of major surface features. The NSCAT the satellite records an area of high relative vorticity corre- ~eld has been able to capture correctly the intensifying system sponding to the center at a longitude of 175~. This feature is with only the assimilation of NSCAT relative vorticity and assimilated into the pressurefield and results in a low-pressure updated boundary conditions. center at that location. The satellite may not pass over the By 1200 UTC, December 21, 1996(Figure 4a), NCEP weak- feature again in 24 hours. Meanwhile, the true center of the ens the system to a central pressure of 994 mbar. The cyclone cyclone would have moved nearly 800 kIn. If the satellite now is now stretched along a major axis running northwest to south- passesover the center at its new position while missing its old east. The center is not well defined and is displaced from the position. the resulting pressure field will erroneously show the circulation center by 50to the north and 30to the west. NSCAT feature as an elongated or two-centered system.The coverage keeps a central pressure of 987 mbar and correctly locates the from NSCAT is insufficient to capture the movement of the center at 38~ and 188°E, coincident with the center of circu- cyclone properly, and the solution field diverges from the true lation of the NSCA T wind vectors. Both fields have tight pres- pressure field. It is foreseen that QuikSCAT, with its wider sure gradients on the northeast side of the cyclone, which coverageand no nadir gap, will alleviate much of this problem, agrees well with the strong southeasterly winds in this area. particularly when multiple scatterometers are in operation. The orientations of the NSCAT pressure contours are more consistent with the wind vectors from the last satellite pass, especially just north of the circulation center. 4. Case Studies Figure 4b depicts the pressure fields at 1800 UTC, Decem- 4.1. Case 1: December 18-24, 1996 ber 22, 1996, after the assimilation of 40 satellite passesover 5 The method is first applied to a case of cyclogenesisthat days. The two pressure fields now agree on the cyclone's cen. occurred December 18-24, 1996. The NCEP surface pressure tral pressure of 993 mbar. NCEP correctly locates the cyclone analysis from OOX>UTC, Derember 18, 1996, initializes the center at the center of circulation, while NSCAT has a double process. The solution pressure field (hereafter call NSCAT low. The tightest gradients and strongest winds are now on the pressure) evolveswith the assimilation of data from 52 satellite western side of the cyclone. Both fields also exhIbit kinking of passesover 7 days. Snapshotsof the NSCAT pressure field are the isobars along the cold front, which extendsto the south and com~ to NCEP reanalyses nearest in time to the latest east of the cyclone center. satellite pass.Also, both the NSCAT and NCEP pressurefields Two days later, at 1200 UTC, December 24, 1996, the cy- are checked for consistency with the NSCAT wind vectors clone reintensifies with a central pressure of 992 mbar. The from the latest satellite pass. This comparison does not con- storm is large, nearly covering the entire domain. NSCAT and stitute validation of the NSCAT pressurefields, as an indepen- NCEP are in good agreement with both the location and in- dent data source is necessaryfor objective results. It is simply tensity of the cyclone center, with NCEP being a little deeper intended to show how the NSCAT pressure fields conform to at 989 mbar. Both fields are consistent with the NSCAT wind features seen in the NSCAT wind fields. vectors. A warm frontal zone now extends eastward from the 23,976 ZIERDEN ET AL.: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS

00 0000 UTC 18 Dec. 1996 0000 UTC 18 Dec. 1996

" ()-5> ;;:1 46 40 j,J ... ~ \\ 30 ~ \1 ~I ~~...,,~';(;-:-- A 202~/ O~.. 165 175 185 195 205 215 226 215 225

~) 1800 UTC 18 Dec. 1996

165 175 185 195 205 ~15 225

(c) 1200 UTC 20 Dec. 1996 " 1200 UTC 20 Dec. 1996

':i'I

~'" .., :~~ ~,~ :' I ~.-,,~ 10_-, 1020 20 r ~ 165 175 185 195 205 215 225 165 175 185 195 205 215 225 Figure 3. Case1: NSCAT surfacepressure field (mbar) after the assimilationof (a) 1, (b) 8, and (c) 20 satellite passes:(left) NSCATwind vectors from the last satellite passand (right) concurrent NCEP reanalysis surface pressures.

cyclone's center. The NSCAT pressure field resolves this fea- NSCAT low-pressure center location exactly matches the cen- ture sharply as seen in the 995 and 1000 mbar contours and ter of circulation. NSCA T also does a better job of showing the strong pressure gradients normal to the front. The NCEP anal- elongated nature of the low, especially on the eastern end ysis has gently cwving contours in this zone, making the exact Notice how the wind vectors parallel the NSCAT pressure location of the front difficult to determine. contours in this area. NCEP does not extend the low far 4.1.. Case 2: January 3-6, 1997 enough east, as the wind vectors cross the contours at unreal- This is a caseof frontogenesis that took place during January istically large angies, Bowing from low to high pressure. 3-6, 1997. The process is initialized at (XX)OUTC, January 3, The low moves little in the next 18 hours (Figure 5c). The 1997,when the only feature of interest is a weak low-pressure center is now located at 4~N and 1920£ in the NSCAT pres- system centered at 4SONand 1~ (Figure 5a). The NSCAT sure field. coincident with the center of circulation. NCEP pressure field picks up another low-pressure lobe entering the continues to place the center too far to the east by ZO.Both domain on the western border at 4fJ'N. This feature is not seen fields have a central pressure of 983 mbar. The NCEP analysis in the NCEP reanalysis. places another distinct low at 42~ and 17TH, where the Figure 5b shows 1800 UTC, January 3, 1997.The low moves NSCAT field has a more continuous trough extendmg east to east and, according to the NSCAT pressure field, is centered at west acrossthe northern portion of the domain. Six hours later, 500N and 1~ with a central pressure of 994 mbar. The at 1800 UTC, January 4,1997, an apparent front has formed NCEP analysis places the center 2° south and east of the and extends across the northern portion of the study area NSCAT location with a central pressureof 987 mbar. The (Figure 6a). The front in the NSCAT pressure field is defined ZIERDEN ET AL.: SURFACEPRmsURE FIElDS AND FRONTSFROM NSCAT WINDS 23.977

1200 UTC 21 Dec. 1996

166 175 185 195 205 J16 225

1800 UTC 22 Dec. 1996

\1>\0 ... Q ~.'(~~:

.- 0 c \~.lf;; 186 175 185 195 205 215 225

(c) 1200 UTC 24 Dec. 1996 1200 UTC 24 Dec. 1996

50 45 40 ~.. ~'" 3530 ~ ~~ 'I, I , ~--~ ...' 25 ~ "'"-"" 20 ~.- ~- 165 175 185 195 205 215 225 165 175 185 195 205 215 225 Figure 4. Sameas Figure3 but for the assimilationof (a) 29, (b) 40, and (c) 52 satellitepasses. by a nearly continuous band of high vorticity and low pressure. now curve gently around the western side of the center rather North of the front, the winds are east-northeast,while south of than shifting sharply as in earlier swaths. Also, beginning at the front, they are from the west-southwest. Also, pressure 1200 UTC, January 5, 1997, a new windshift line fonDSto the gradients have tightened on both sides of the front. The wind southeast of the center, corresponding to the genesis of a shift line coincides well with the line of lowest pressures.The trailing cold front (Figure 6b). This feature is seen in the NCEP analysis persists in separating the feature into two dif- NSCAT pressure field as a sharp trough extending south of the ferent lows. NSCAT wind vectors show no evidence of closed cyclone center. NCEP does not resolve the front, showing only circulation around either of the features to support this analysis. gently curved isobarswith the trough axis placed 100east of the The front is nearly stationary for the next 24 hours and is NSCAT position. sharply delineated by the 1800 UTC, January 5,1997, NSCAT 4.3. Accuracy of the NSCAT and NCEP Pressure Fields pressurefield. The wind shift acrossthe front from northeast to southwest is highly localized along the length of the front. Two generalizations can be made about the NSCAT pres- NCEP finally merges the two lows into one elongated feature, sure field from these case studies. First, the field is more although it is not as linear as was depicted by the NSCAT realistic in the interior of the domain than near the boundaries. pressure field. The NCEP analysis has slightly lower pressure New features moving into the domain are not assimilated into (986 mbar) than NSCAT (989 mbar) on the western portion of the NSCAT pressure field until the satellite passesover the the front. Otherwise, the NSCAT positions of two small low- area, so these features may be totally missed. Also, regions pressure features along the front agree well with the wind near the borders are constrained by the boundary conditions. circulation patterns. Away from the borders, the solution field is free to conform to At 1800 UTC, January 5,1997, the westernmost lobe of low features found in the NSCAT relative vorticity. Second, the pressure is developing into a new cyclone (Figure 6c). Winds NSCAT pressure field is more accurate Where the satellite has 23,978 ZIBRDEN BT Al..: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS

(a) 0000 UTC 3 Jan. 1997 0000 UTC 3 Jan. 1997 ::~~" ~

30 ~? 25 ":/I ./--::;;~::=~~;~1010 :,\&V_~.1110- D'I ~.::=:~~::::::~;----i';;~~--. . 20 ~ . 101~. 185 175 186 196 205 215 225 165 175 185 195 205 21&

(b) 1800 UTC 3 Jan. 1997 1800 UTC 3 Jan. 1997

50 45 /~~ 40 35 1

30 ,- ::==:'1;1)-I::::;'.- ., :;7}'I 'O_': 0 25- - .." 1018/1 20 = -

165 175 185 195 205 215 225 Ie 175 186 195 205 216

165 175 185 195 205 215 225 165 175 185 195 205 215 225 Figure S. Case2: NSCAT surfacepressure field (mbar) after the assimilationof (a) 1, (b) 7, and (c) 12 satellite passes:(left) NSCAT wind vectors from the last satellite passand (right) concurrent NCEP reanalysis surface pressures. paMed more recently. It stands to reason that the portion of buoy observationsare used by NCEP in their reanalyses,so the the domain updated with newer information is more current two are not independent. Second. the observationswere made than an area that has not seen a satellite pass in many hours. at the same synoptic times as the NCEP analyses.The NSCAT Table 1 addresses the accuracy of both the NSCAT and pressurefield. however, may have seen the last satellite passas NCEP pressure fields. Three hourly surface pressure obsexva- far as :3 hours of the synoptic time. Also, the NSCAT pres- tions from ships and buoys (courtesy of the NCDC are com- sure field outside the area coveredby the latest passis basedon pared to values from both the NSCAT and NCEP pressure older satellite information, 12-24 hours from the latest pass. fields for each snapshotin Figures J-6. For each casethe mean Yet another source of error in the NSCA T field is the scarcity and standard deviation of the difference between the pressure of reference pressures. Only two buoys were available, and field values and the in situ observations are computed using all these are both located in the eastern half of the domain. In- available observations for that time. These statistics are com- accuracies in the pressure gradients caused by older satellite puted for all observations in the interior of the domain (at least data lead to increasing errors with distance from the reference 5° from the boundaries). The results show that the NCEP points. The result that NSCAT pressure fields have mean dif- pressurefield is quantitatively more accurate than the NSCAT ferences and standard deviations only slightly higher than field. The NSCA T mean difference rangesfrom 0.1 to 2.5 mbar NCEP's reanalysis product supports the validity of the varia- in magnitude, and the standard deviation is between 2.4 and tional method in deriving surface pressures from NSCAT 6.1 mbar. The NCEP mean difference is small, between 0.1 and winds. 1.2 mbar, and the standard deviation is :s5.8 mbar. NCEP Although the NSCAT pressures may not be quantitatively should be more accurate for severalreasons. First, the ship and more accurate than NCEP over the domain as a whole, the ZIERDEN ET AL.: SURFACE PRESSURE FIElDS AND FRONTS FROM NSCAT WINDS 23.m

(b) 1200 UTC 5 Jan. 1997 ---

45 3511 ~I 4030 ---O.eo 25 ~ 20 165 175 185 195 205 215 225

qualitative advantages are seen in the detailed comparisons vorticity fields gives rise to features in the NSCAT pressure made in sections4.1 and 4.2. Features such as fronts are more fields that are blurred or not seen at all in the NCEP analyses. sharply defined, and pressure gradients are more consistent NSCAT is also better at placing the low-pressure centers with the NSCAT winds. The improved detail in the geostrophic correctly with respect to the center of circulation of the

Table 1. ComparisonWith SurfaceData

Mean Standard Mean Standard Difference, Deviation, Difference, Deviation, Numberof Time and Date mbar mbar mbar mbar ObseIVatiOns

1~ UTC, Dee. 18 -0.2 24 -0.2 1.1 .19 1200 UTC, Dee. 20 1.5 6.1 0.2 4.2 34 1200 UTC, Dee. 21 -0.7 6.1 0.6 3.1- 29 1~ In'c, Dee. 22 -0.1 6.0 -0.1 1.4 37 1200 UTC, Dec. 24 2.0 3.3 0.2 U 37 1~ UTC, Jan. 3 1.0 6.1 1.2 5.8 35 1200 UTC, Jan. 4 -0.2 5.s 0.2 4.7 34 1800 UTC, Jan. 4 -2,5 5.4 0.0 4.3 32 1200 UTC, Jan. 5 -0.8 4.6 -0.3 2.5 33 1800UTC, Jan.5 2.0 4.1 -0.2 3.0 48 23,980 ZIERDEN ET AL.: SURFACE PRESSURE FIELDS AND FRONTS FROM NSCAT WINDS

Table 2. Differencein the Locationsof Low-PressureCenters Compared to Centersof Circulation NSCAT NCEP

1200urc, De<:.20 2.0 280 2.0 170 1800urc, De<:.20 0.0 0 1.0 140 1200urc, De<:.21 1.0 8S 3.0 420 1200urc, Dec. 22 1.0 140 3.0 600 1800urc, De<:.22 0.0 22S 0.0 0 1200urc, De<:.23 2.0 170 2.0 170 1800UI'c, Dec. 23 0.0 0 0.0 1200urc, De<:.24 0.0 0 0.0 0 1800urc, Jan. 3 -1.0 8S 2.0 200 1200urc. Jan. 4 0.0 0 2.0 170 Average 0.7 100 1.5 190

NSCAT winds. An independent data source would lead to a from 0.1 to 2.5 mbar in magnitude with a standard deviation more objective comparison, but the lack of conventional data between 2.4 and 6.1 mbar. NCEP pressures compare slightly over the North Pacific makes the NSCAT wind vectors the best better, with mean differences below 1.2 mbar in magnitude and alternative for detennining the correct location of low-pressure standard deviations below 5.8 mbar. The accuracy of the centers. Table 2 shows the difference in location of low- NSCA T pressure fields increases near the latest satellite pressure centers compared to circulation centers for times swaths and in the interior of the domain. when the circulation center is revealed in the latest satellite Qualitatively, the NSCAT pressure fields resolve the struc- path. In 2 of the 10 cases both NSCAT and NCEP agree ture of cyclones and fronts more realistically and with greater exactlywith the NSCAT winds. In the first case, at 1200 UTC, detail than the NCEP analyses.The NCEP model and other December 1996,the NSCAT low-pressure center is placed 280 spectral models generally represent fronts as broad transition kID from the circulation center. This displacement is causedby zones because of their coarse resolution. The NSCAT winds an older vorticity maximum lying between the two swaths of a and resultant surface pressure fields better represent fronts as new pass,giving a false vorticity signature to the new NSCAT boundaries because of their high spatial resolution. Also, the pressure field (section 4.1). Other than this isolated case, centers of cyclones are placed more accurately in the NSCAT NSCAT low-pressure centers are consistently closer to the pressure fields than in NCEP analyseswhen compared to the circulation centers than NCEP. On average, NSCAT low- centers of circulation from NSCAT winds. The averagediffer- pressure centers are 100 kID from the circulation center, while ence is 100 km for NSCAT pressure fields and 190 km for NCEP averages 1W kID errors. And interesting feature of NCEP analyses. Table 2 is the trend in longitudinal error of the NCEP analyses. Another result of this study is the signature of surface fronts In 7 out of the 10 cases the NCEP center is placed too far in relative vorticity fields computed from NSCAT winds. eastwardwhen compared to circulation centers and in no case Fronts are clearly identified by linear bands of high relative is it displaced to the west. While 10 cases(two different sys- vorticity values. These bands are verified as fronts using sur- tems) are too few to suggest a systematic bias in the NCEP face temperature gradients and satellite imagery. NSCAT vor- analyses,the trend warrants further investigation. ticity fields locate the fronts with an accuracyof 25-50 km and with greater resolution than other satellite data sources. S. Conclusions Although NSCAT winds are a high-quality data source, ef- fective techniques for assimilation into NWP models have A variational method is devised to generate surface pressure proven difficult to develop. Surface pressure fields from fields from NSCAT winds. The method solves for a surface NSCAT winds, however, could provide a more favorable as- pressurefield by smoothly blending relative vorticity computed similation source [Hoffman, 1993; Atlas, 1997]. This study from NSCAT winds with ambient geostrophic vorticity. The brings forth a simple method for determining surfacepressures method ignores the ageostrophy of surface winds as no upper from NSCAT winds and demonstrates its effectiveness. air thefD1alor mass fields are used to make adjustments. The solution pressurefield is updated as new passesof the satellite over the study area provide additional information. Neumann Acknowledgments. Funding for this project is from the NASA JPL NSCAT project. COAPS receives its base funding from ONR's Sec- boundary conditions, updated twice daily with NCEP normal retary of the Navy Grant to James J. O'Bri~n. gradients, allow the surface pressure field to evolve in time. This method is used to study a case of cyclogenesisand a case of frontogenesis in the North Pacific. The NSCA T pres- References sure fields correctly capture these features as they intensify and Andrews, P. L., and R. S. Bell, Optimizing the United Kingdom Me- move through the study area. The NSCAT pressure fields are teorological Office data assimilation of ERS-l scatterometer winds, qualitatively compared to NCEP reanalysis surface pressure Mon. WeatherRev., 126, 736-746, 1998. fields, and both fields are quantitatively compared to NCDC Anthes, R. A., V. H. Kuo, and J. R. Gyakum, Numerical simulations of a case of explosive marine cyclo~nesis, Mon. WeQ!herRev., 111, ship and buoy observations.The domain-averaged differences 1174-1188, 1983. between NSCAT pressures and surface observations range Atlas, R., Atmospheric observations and experiments to assesstheir ZIERDEN ET AL.: SURFACE PRESSURE FIElDS AND FRONTS FROM NSCAT WINDS 23,wt

usefulness in data assimilation, L MeteoroL Sac. Jpn., 75, 111-130, winds: Comparisons with National Data Buoy Center buoys,1. Geo- 1997. phys. Res.. 104, 11,231-11,246,1999. Atlas, R., and R. N. Hoffman, The use of satellite surface wind data to Harlan, J., Jr.. and J. J. O'Brien, Assimilation of scatterometer winds improve weather analysisand forecasting, in Satellites,OceanogTrlphy into surface pressure fields using a variational method, 1. Geophys. and Society,pp. 57-79, Elsevier, New York, 200). Res., 91, 7816-7836, 1986. Atlas, R., E. Kalnay, and M. Halem, Impact of satellite temperature Hoffman, R N., A preliminary study of the impact of the ERS-l soundings and wind data on numerical weather prediction. Opt. C-band scatterometer wind data on the European Centre for Me- Eng., 24, 341-346, 1985. dium-Range Weather Forecasts global data assimilation system,1. Baker, W. E., R. Atlas, E. Kalnay, M. Halem, P. M. Woiceshyn, S. Geophys.Res., 98, 10,233-10,244,1993. Peteherych, and D. Edelmann, large-scale analysis and forecast Hsu, C. S., and M. G. Wurtele, Construction of marine surface pres- experiments with wind data from the Seasat-A scatterometer, L sure fields from scatterometer winds, 1. Geophys.Res., 36, 1249- Geophys.Res., 89, 4927-4936, 1984. 1261, 1997. Bourassa, M. A, D. M. Legler, and J. J. O'Brien, NSCAT winds Katsaros, K. B., J. M. Adams, and N. M. Mognard, Evolution of two validation with research ship observations,Eos Trans.AGU, 78(17), oceanic extratropical cyclones as observed with the special sensor Spring Meet. Suppl., 576, 1997. microwave imager and the Geosat and ERS-l altimeters, 1. Gi!Qphys. Bourassa, M. A, L. Zamudio, and J. J. O'Brien, Noninertial 6ow in Res., 101, 6627-6639, 1996. NSCAT observations of Tehuantepec Winds, L Geophys.Res., 104, Levy, G.. and R. A Brown, Southern Hemisphere synoptic weather 11,311-11,319,1999. from a satellite scatterometer, Mon. WeatherRev., 119, 2803-2813, Brown, R. A, and G. Levy, Ocean surface pressure fields from satel- 1991. lite-sensed winds, Mon. WeatherRev., 114, 2197-2206, 1986. Liu, W. T., W. Q. Tang, and P. S. Polito, NASA scatterometerprovides Brown. R: A, and L. Zeng, Estintating central pressures of oceanic global ocean-surfacewind fields with more structures than numeri- midlatitude cyclones,L AppL Meteorol., 33, 1088-1095, 1994. cal weather prediction, Geophys.Res. Lett.,25, 761-764, 1998. B~, K. A, and N. M. Roberts, Structure of a frontal cyclone, Renka,R, Interpolationof dataon the surface of a sphere,Oak Ridge Q. L R. MeteoroL Soc., 120, 1535-1557, 1994. Nail Lab. Rep. OLNR/CSD-108, Oak Ridge Nat!. Lab., Oak Ridge, Carlson, T. N., AirJIow through midlatitude cyclones and the comma Tenn., 1982. cloud pattern, Mon. WeatherRev., 108, 1498-1509, 1980. Saski, Y., Some basic formalisms in numerical variational analysis, aarte, R. 11.,and G. D. Hess,On the relationbetween surfacewind Mon. WeadaerRev., 108, 875-883, 1970. and pressure gradient, especially in the lower latitudes, L Phys. Tomassini, M., D. LeMeur, and R. W. Sounders,Near-surface satellite Oceanogr.,9, 325-339, 1975. wind obseJVationsof hurricanes and their impact on ECMWF model Djuric. D., WeatherAnalysif, 304 pp., Prentice-Hall, Englewood Oiffs, analysesand forecasts,Mon. WeatherRev., 126, 1274-1286, 1998. N. J., 1994. Verschell, M. A, M. A Bourassa, D. E. Weissman, and J. J. O'Brien, Duffy, D. G., and R. Atlas, The impact of Seasat-Ascatterometer data Model validation of the NASA scatterometer winds, 1. Geophys. on the numeri~ prediction of the Queen Elizabeth n storm, L Res.. 104, 11,359-11,373, 1999. Geophys.Res., 91, 2241-2248, 1986. Zeng, L. X, and R. A Brown, Scatterometer observationsat high wind Duffy, D., R. Atlas, T. Rasmond, E. Barker, and R. Rosenberg, The speeds,1. AppL Meteorol.,37, 1412-1420, 1998. impact of Seasat scatterometer winds on the Navy's operational model, L GeoPhys.Res., 89, 7238-7244, 1984. M A Bourassa, James J. O'Brien, and D. F. Zierden, Center for Freilich. M., and R. S. Dunbar, A prelimiIiary C-band scatterometer. Ocean-Atmospheric Prediction Studies, Florida State University, Tal- model function for the ERS-1 AMI instrument, paper presented at lahassee,FL 32306-2840.([email protected]) First ERS-1 Symposium,European SpaceAgency, Cannes, France, 1992. (Received March 22, 1999; revised February 14, 2(XX); Freilich, M., and R. S. Dunbar, The accuracy of NSCAT 1 vector accepted March 20, 2(XX).)