«International Journal of Remote Sensing Archimer March 2012, Volume 33, Issue 6, Pages 1729-1754 ...»
Please note that this is an author-produced PDF of an article accepted for publication following peer review. The definitive publisher-authenticated version is available on the publisher Web site
International Journal of Remote Sensing Archimer
March 2012, Volume 33, Issue 6, Pages 1729-1754
© 2011 Taylor & Francis
The original publication is available at http://www.tandf.co.uk/journals/ Gridded surface wind fields from Metop/ASCAT measurements a, * a Abderrahim Bentamy & Denis Croize Fillon a
The precise knowledge of surface wind vectors over global oceans are one of the main requirements of scientific and operational oceanography and climate projects (see, for instance, http://www.myocean.eu.org/ ). Indeed, they induce heat and momentum transfers between the ocean and the atmosphere and are the main driving forces for surface ocean circulation. Surface winds are routinely derived from radars and radiometers onboard satellites. They are available with various spatial and temporal resolutions associated with the instrument characteristics. Generally speaking, they are retrieved over instrument swaths of several hundred-kilometre widths.
A number of studies have considered the problem of producing synoptic wind maps from scatterometer-derived winds using standard statistical procedures. For example, Woiceshyn et al. (1989) interpolated a 15-day record of SEASAT scatterometer winds to a regular grid and achieved favorable results in wave modeling experiments, as compared to using NWP winds. Tang and Liu (1996) produced 12-hourly wind fields on a 1o latitude-longitude (lat-long) grid, using a successive correction method based on ERS-1 winds and NWP produced winds. Royle et al. (1999) used a Bayesian approach to combine scatterometer winds and NWP winds. Perrie et al (2002) used optimal interpolation (OI) to combine scatterometer and altimeter data with numerical weather prediction (NWP) model wind estimates. Using ERS-1 and ERS-2 scatterometer retrievals, Bentamy et al. (1996; 1998) constructed weekly and monthly wind fields over the global oceans with a spatial resolution of 1° in longitude and latitude. The former were estimated based on the kriging method, which requires the knowledge of the spatial and temporal wind speed, wind stress, and the zonal and meridional component structure functions. Thanks to the QuikSCAT sampling scheme, the objective method has been enhanced for gridded daily, weekly, and monthly global wind field calculations (Bentamy et al, 2002). The related spatial resolution is 0.50° in longitude and latitude. Gridded QuikSCAT data are available from July 1999 through November
2009. The extension of daily, weekly, and monthly averaged wind field availability is highly requested. To meet such requirements, new gridded wind fields are calculated from the Advanced SCATterometer (ASCAT) measurements on board the Metop-A satellite.
The main purpose of this study is to estimate global wind fields from scatterometer wind observations and to assess their accuracy. The paper describes data sets, the temporal and spatial sampling issues, and the methodology used to generate regular in space and time wind speeds and the related components. The validation of the objective method and of the resulting wind fields as well as the investigation of the spatial and temporal characteristics of ASCAT daily winds are provided.
ASCAT scatterometer represents the latest implementation of spaceborne microwave wind-measuring scatterometry. Scientific and technical documentation related to ASCAT physical measurements as well as to ASCAT derived products may be found at the EUMETSAT web site http://www.eumetsat.int/Home/Main/Publications/Technical_and_Scientific_Documenta tion/Technical_Notes/ and at the SAF OSI web site http://www.knmi.nl/scatterometer/. Metop is in circular orbit (near synchronous orbit) for a period of about 101 minutes, at an inclination of 98.59° and at a nominal height of 800 km with a 29-day repeat cycle. ASCAT (Figure 1) has two swaths 550 km wide, located on each side of the satellite track, separated by 700km. It operates at 5.3 GHz 2 (C band). Its fore-beam and aft-beam antennas point at 45 and 135° to each side of the satellite track, respectively. The mid-beam antennas point at 90. The ASCAT beams measure normalized radar cross sections with vertical polarization, 0, which are a dimensionless property of the surface, describing the ratio of the effective echoing area per unit area illuminated. The fore and aft-beams provide backscatter coefficient measurements at incidence angle varying between 34 and 64. The midbeams provide 0 measurements at incidence angles varying between 25 and 53.
Backscatter coefficients are provided with two spatial resolutions of 25km and 12.5km over the global ocean.
All ASCAT products used in this paper correspond to near-realtime data provided by EUMETSAT and by KNMI as the wind component of the Ocean Sea Ice Satellite Application Facility (OSI SAF).
Since 20 March 2007, KNMI processes and makes available surface wind vector derived from ASCAT backscatter coefficients (http://www.knmi.nl/scatterometer/publications/pdf/ASCAT_Product_Manual.pdf). The ASCAT swath datasets used in this study are referenced as ASCAT level 2b (L2b).
ASCAT wind retrievals are provided at WVC of 25 km by 25 km. There are 42 WVC across the two-scatterometer swaths. Data include wind retrievals as well as backscatter coefficients measured over ocean and several associated fields at each valid WVC. From March 2007 through October 2008, ASCAT winds are considered as ‘real winds’ and include the atmospheric stratification impact. Since November 2008, ASCAT retrievals are provided as equivalent neutral surface winds.
To determine accuracy, in-situ data derived from buoys are used. They are provided by Météo-France and U.K. MetOffice (MFUK), the National Data Buoy Center (NDBC), the Tropical Atmosphere Ocean Project (TAO), the Pilot Research Moored Array in the Atlantic project (PIRATA), and by the Research Moored Array for African–Asian– Australian Monsoon Analysis and Prediction project (RAMA). They consist of buoys moored off US coasts (NDBC), off European seas (MFUK), and along the Atlantic (PIRATA), Indian (RAMA), and Pacific (TAO) tropical oceans. For purposes of comparison, all buoy winds are adjusted to a height of 10 meters assuming neutral stability.
The NWP surface winds used in this study are derived from the European Centre for Medium Weather Forecasts (ECMWF) operational analysis. They are routinely provided by Météo-France within the Mersea project (http://www.mersea.eu.org/ ). The time resolution of the ECMWF data is four times daily (00h:00, 06h:00, 12h:00 and 18h:00). They are made available on a regular grid of 0.5° in longitude and latitude.
The ECMWF winds are given at 10 m above sea level in terms of zonal and meridional wind components. Furthermore, the new ERA-Interim reanalysis, which is advanced with respect to ERA-40 (http://www.ecmwf.int/research/era/do/get/era-interim) is also used.
3. ASCAT Sampling SCHEME
As expected, the spatial and temporal resolutions of regular wind fields calculated from ASCAT retrievals are highly related to the scatterometer sampling scheme. Figure 2 shows ASCAT wind speed coverage for January 1st, 2009 (Figure 2a), and the spatial distribution of scatterometer observation sampling length, number of retrievals located within grid cell of 0.25° in longitude and latitude, estimated during 1 – 29 January 2009 (Figure 2b).. Even though ASCAT provides valuable surface wind observations with 3 quite good daily sampling, several oceanic regions are not observed (Figure 2a).
Overall, the grid points located north of 40°N and south of 40°S are observed at least once a day, whereas in the inter-tropical area, the number of ASCAT retrievals falling within each grid point is less than 1 (Figure 2b). Therefore, the calculation of gridded wind fields from ASCAT retrievals, and especially on a daily basis, is a challenging task. Indeed, at several locations, no more than 1 observation a day is expected. The expected impact of such a sampling scheme on the regular wind field calculations is investigated using buoy measurements.
In-situ data are obtained from operational buoy networks involving moorings located off the French and England coasts and maintained by UK Met-Office and/or Météo-France (MFUK hereafter, Figures 3a and 4a) buoys provided by the National Data Buoy Center (NDBC) that are located off and near U.S coasts (Figures 3b 4b), buoys of the TAO array located in the equatorial Pacific (Figures 3c and 4c), and buoys of the PIRATA array located in the equatorial Atlantic (Figures 3d and 4d). Surface winds from buoys are hourly-averaged.
Let wb, u b, vb be the daily-averaged wind speed, zonal, and meridional components calculated from raw and valid buoy data.
wbs, u bs, vbs are daily-averaged wind speed, zonal, and meridional components, respectively, calculated from raw and valid buoy data collocated in space and time with ASCAT WVC retrievals.
The investigation of differences between the two types of averaged buoy data, are used to assess the impact of the ASCAT sampling scheme on daily wind estimations.
Figures 3 and 4 show the mean and standard deviation estimates of differences between wb and wbs calculated during the one-year period. Overall, the biases between the two daily winds are quite low and generally are not significant. The calculation of daily wind speeds from buoy data occurring at ASCAT passes does not yield any systematic bias. The main impact of the ASCAT sampling scheme on daily wind estimation, is clearly found in wb and wbs difference variability (Figure 4). The latter may reach 2ms-1 at some locations of the MFUK and NDBC networks. The highest variability values are found at sites where surface winds are highly variable, such as at the Mediterranean Sea buoys (Figure 4a). However, the lowest values are depicted in the tropical areas, associated with low surface wind variability.
Therefore, the calculation of gridded winds from ASCAT retrievals requires the development and use of a method aiming to provide wind vector estimates at regular space and time grids, and to reduce the impact of the scatterometer sampling scheme.
4. Methodology The main aim is to estimate a regular wind field in space and time using ASCAT remotely sensed data. However, regarding the results of the previous section, to better estimate analyses at each grid point, auxiliary information is also used. The latter are derived from the ECMWF operational 10m-wind analysis. The following linear
relationship between observations and auxiliary data is assumed:
This assumption states that the mean of the difference between observations is independent of space and time separation. Therefore equation 9 leads to
Minimizing functional Var(R(M0)) in weighting space and under unbiased and external
drift constraints leads to the following linear system:
The objective method requires parameterization of the spatial and temporal covariance structure of surface wind speed, zonal and meridional wind components, wind stress magnitude, and wind stress components. The approach used in Bentamy et al. (1996) is adapted for this study. First, the local spatial and temporal stationary is assumed.
Therefore, the covariance does not depend on the precise geographical location and
epoch of data, but only on separations in space and time:
Where δh and δt stand for spatial and temporal separation, respectively.
As the assessment of this assumption is not straightforward, the following assumption
The objective is to determine the covariance matrix involving the main spatial and ~ temporal structure of variable X without any prior gridding or spectral filtering.
Therefore, the investigation of covariance or variogram behavior as a function of space and time separations is performed over several areas of the Atlantic Ocean and the Mediterranean Sea. ASCAT wind retrievals are used to calculate the sample ~ covariance. The observed values of X are then calculated over each satellite WVC and stratified in terms of 1-hourly time windows (WVC time). For each temporal separation (t) varying between 0h and 24h, the covariances of observations spatially separated by a distance (h) varying between 0km and 500km are estimated.
Examples of wind speed, zonal and meridional component variogram behaviors as a function of spatial separation for a lag time less than one hour are shown in Figure 5.
In practice, all ASCAT validated retrievals occurring between 00h:00 and 23h:59mn:59sc UTC of a given day, are collected for daily averaged wind fields calculations, respectively According to ASCAT sampling scheme (Figure 2), in order to reduce the impact of ECMWF wind estimates, extended observation periods are considered for daily ASCAT gridded wind field calculations. Indeed, considering retrievals occurring within a time interval of 12 hours bounding a day of interest, allows almost global coverage.
More specifically, if no ASCAT data are available at a given grid point and during the day of interest, the closest retrievals Xpr (xsat, ysat, tsat) occurring within 50km from the grid point and on the prior day between 12h:00 and 24h:00 UTC, or on the following day between 00h:00 and 12h:00 UTC are selected. The temporally interpolated values 7 of Xpr at 00h:00 or at 24h:00 of the day of interest are used as ASCAT observations for daily calculation.
The temporal interpolation approach is based on the complex empirical orthogonal function (CEOF) method. It aims to determine the advective patterns of wind variables through the determination of the amplitude and phase of the related signal. To achieve this, the twelve ECMWF analyses calculated one day prior, one day after, and during the day of interest are used.
ECMWF wind components um and vm are used to define the following analytic signal: