Docstoc

Report model for MERSEA

Document Sample
Report model for MERSEA Powered By Docstoc
					Research Project co-funded by the European Commission Research Directorate-General th 6 Framework Programme FP6-2002-Space-1-GMES Ocean and Marine Applications Contract No. AIP3-CT-2003-502885

MERSEA IP
Marine EnviRonment and Security for the European Area - Integrated Project

Deliverable D2.2.10 Implementation of a global SST analysis WP 02 Task 2.2
Ref: MERSEA-WP02-IFR-STR-001-1A.doc October 2007

E. Autret, J. F. Piollé (Ifremer/CERSAT)

Co-ordinator:

Météo-France - France

Report model for MERSEA

Table of Content
1. INTRODUCTION ...................................................................................................3 2. OVERVIEW ...........................................................................................................4 3. L2P INPUT DATA .................................................................................................6 4. PROCESSING OF NATIVE SENSOR (STEP 1) ..................................................10
4.1. PROCESSING SCHEME (STEP 1 AND STEP 2) ........................................................................................ 10 4.2. AVAILABLE FILTERS (IN STEP 1) ........................................................................................................... 11 4.3. L2P OBSERVATIONS REMAPPING (IN STEP 1) ....................................................................................... 13 4.3.1. Available methods ................................................................................................................. 13

5. MERGING OF REMAPPED ORBIT FILES (STEP 2) ...........................................15
5.1. DEFINITION ........................................................................................................................................ 15 5.2. PROCESSING ..................................................................................................................................... 15

6. INTERCALIBRATION OF THE OBSERVATIONS FROM THE SINGLE SENSOR L3 PRODUCTS (STEP 3)............................................................................................17
6.1. DEFINITION ........................................................................................................................................ 17 6.2. PROCESSING ..................................................................................................................................... 17

7. SINGLE SENSOR L3’S OBSERVATIONS MERGING ONTO A MULTISENSOR L3 PRODUCT (STEP 4) ...................................................................................................19
7.1. DEFINITION ........................................................................................................................................ 19 7.2. PROCESSING ..................................................................................................................................... 19

8. ANALYSIS STEP (STEP 5) ..................................................................................22
8.1. ANALYSIS METHOD ............................................................................................................................. 22 8.2. A PRIORI STATISTICS .......................................................................................................................... 22 8.2.1. 8.2.2. 8.2.3. 8.2.4. Background ........................................................................................................................... 22 Structure functions ................................................................................................................. 23 Correlation scales .................................................................................................................. 24 Variances ............................................................................................................................... 25

8.3. ICE AND LAND MASK............................................................................................................................ 25 8.3.1. Land mask ............................................................................................................................. 25 8.3.2. Ice mask ................................................................................................................................ 25

9. REAL TIME CONTROL TOOLS, DATA ACCESS AND FORMAT ......................26 10. REFERENCES ......................................................................................................27

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

1

Report model for MERSEA

Document Change Record

Author E. Autret, JF. Piollé

Modification Creation

Issue

Date 20/10/2007

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

2

Report model for MERSEA

1. INTRODUCTION
One of MERSEA WP2.2 objectives is the implementation of a coherent set of SST analysis at the global scale (global ocean) or at regional scales (Atlantic, Mediterranean sea and Norwegian sea). This reports describes the implemention of the operational analysis system providing daily the global SST analysis at 0.1° resolution. During this implementation, maximum synergy with other analysis systems within MERSEA has been sought, and particularly with the Atlantic fine scale analysis system implemented at Météo-France (Le Borgne at al, 2006). Format specification of input, output and intermediate data are compliant as well as the intercalibration schemes. This report will mostly focus on the description of the processing steps and specific tunings implemented for the ODYSSEA1 v1.0 global scale analysis:      General overview of the processing scheme Input data considered and collected for the global scale analysis Selection and screening of the observation to be used in the analysis Merging of the sensors, including correction and intercalibration of the various observation Analysis method

1

Ocean Data analYsis System for merSEA

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

3

Report model for MERSEA

2. OVERVIEW
The ODYSSEA processing scheme described in this document aims at providing daily gap-free maps of sea surface temperature, referred as L4 product, using satellite data from both infra-red and micro-wave radiometers. The major providers for these input data comply now to GHRSST specification for content and format of swath observations, referred as L2P (Pre-Processed). Only data sources complying to the GHRSST model were considered for the operational processing chain: however it should be raised that these data files are not always fully compliant to this model and that some discrepancies exist between these datasets (ancillary data, flag meanings, confidence estimation,...), which result in a specific tuning for each dataset within the processing chain.

The processing chain described here is fully configurable in terms of:       grid resolution geographical coverage list of inputs remapping techniques input data screening criteria a priori information and analysis method

The processing breakdown is as follow:

Remapping of L2P files onto the final L4 product grid (step1) Input data Description Orbital L2P files The data in original full–resolution swath format (GHRSST L2P) are screened in order to remove all bad measurements (low quality or affected by diurnal cycle) and select the best ones. They are adjusted if necessary to subskin temperature and corrected using the error statistics available within the L2P files. They are then remapped on an isolat-isolon grid, computing a super observation at each grid point. single sensor remapped orbit files

Output data

Merging of the available single sensor remapped orbit files (step 2) Input data Description Single sensor remapped orbit files For each analysis time frame (currently each day), all available single sensor orbit files within a time window centered on the analysis time (currently minus/plus 3 days) are merged together. When several measurements are available at a pixel location, both respective quality and proximity to the

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

4

Report model for MERSEA

analysis time of each measurement are considered in order to select the most relevant. Output data Merged single sensor L3 files (« composite files»)

Intercalibration of the single sensor composite files (step3) Input data Description Single sensor L3 files For each analysis time frame and each sensor, a large scale bias field between the considered sensor and a reference (currently AATSR) is computed (using aggregation of bias information over several days and optimal interpolation to estimate large scale biases). This step allows to correct each sensor using a single reference and to retrieve an adjusted sea surface temperature. Adjusted single sensor L3 files

Output data

Merging of the adjusted single sensor L3 files onto a multisensor L3 file (step 4) Input data
Description

Adjusted sensor L3 files For each analysis time frame, the available adjusted sea surface temperature of all available sensors are merged together into a single multi-sensor file. Respective quality of each measurement and of each sensor is carefully balanced in the merging process in order to select the best observations. Merged L3 multi-sensor file

Output data

Analysis of SST observations to construct the L4 product (step 5) Input data Description Multi-sensor L3 product Analysis method (currently optimal interpolation) constructs gridded SST field from the observations set (currently from multi-sensor L3 product) and a priori statistics. L4 product
Table 1 Main steps of the operational processing chain of ODYSSEA system

Output data

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

5

Report model for MERSEA

3. L2P INPUT DATA
Only SST data complying to GHRSST standard model have been considered, although they comply to this standard at various degree, making the data screening step not as straightforward and generic as it could be. The input data considered are described in table 2:

Origina Satellite/sen l resol. sor 1 km 2 km AATSR AVHRR

product EUR-ATS_NR_2P EUR-L2P-NAR17_SST EUR-L2P-NAR18_SST

Provider MEDSPIRATION/IFREMER ftp://ftp.ifremer.fr/ifremer/medspiration/data MEDSPIRATION/IFREMER ftp://ftp.ifremer.fr/ifremer/medspiration/data

2 km

AVHRR

NAVO-L2P-AVHRR17_L GDAC/NASA NAVO-L2P-AVHRR18_L ftp://podaac.jpl.nasa.gov/pub/GHRSST/data

3 km

GOES/VISSR OSDPD-L2P-GOES11 OSDPD-L2P-GOES12

NOAA/NESDIS ftp://gp16.ssd.nesdis.noaa.gov/pub/goessst/L2P/ MEDSPIRATION/IFREMER ftp://ftp.ifremer.fr/ifremer/medspiration/data GDAC/NASA ftp://podaac.jpl.nasa.gov/pub/GHRSST/data

9 km 10 km

MSG1/SEVIRI AVHRR

EUR-L2P-SEVIRI_SST NAVO-L2PAVHRR17_G NAVO-L2PAVHRR18_G

25 km

AMSRE

USA-RSS-AMSRE-MW- MISST/RSS L2-SST ftp://ftp.misst.org/amsre/swath/nc

25 km

TMI

REMSS-L2P-TMI

MISST/RSS ftp://ftp.misst.org/amsre/swath/nc

Table 2 L2P Input data

Other sources will be considered in the future :   MODIS (1km) METOP (1km)

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

6

Report model for MERSEA

All these datasets have been selected to complement each other in order to get a maximum coverage through:     different coverage areas different sampling time and frequency different resolution different sensitivities (e.g. microwave instruments such AMSRE and TMI are not sensitive to clouds unlike the infra-red radiometers, but have a coarser resolution)

As an example here is an overview of the sampling of each datasets within 3 days around a given date (taking only nighttime data):

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

7

Report model for MERSEA

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

8

Report model for MERSEA

Figure 1 SST (°C) from single sensor composite products on the 15th of October 2007. Observations have been selected within -/+3 days around the current date. These maps illustrate the real time products taking into account the observations realized over about -3/+1 days around the current date

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page

9

Report model for MERSEA

4. PROCESSING OF NATIVE SENSOR (STEP 1)
4.1. Processing scheme (step 1 and step 2)
The computation of the single sensor L3 files is the first step of the global analysis processing scheme. It consists in selecting for each single source (GHRSST L2P dataset), in swath or snapshot pattern, the most relevant data, filtering out erroneous or suspect measurements, and remapping these observations on a grid matching the boundaries of the analysis grid. For high resolution datasets, superobservations (averaging of several single full resolution observations to retrieve coarser resolution data) are computed in order to decrease the volume of data to be managed and processed. The processing of the GHRSST L2P data files to construct the final single sensor L3 file is as follow:

Figure 2Illustration of the two first steps of the processing chain: First, the L2P files are "remapped" on a regular grid. Second, measurements realized within the period taken into account for the analysis step are selected to construct a single file

1. The swath (or snapshot, for geostationary satellites) L2P files are first remapped onto a single gridded file, having the same properties as the final native collated files, containing the observations selected as valid using a set of filters (Table 3). This processing step is performed on the fly as soon as a new L2P file is collected. a) An empty gridded file is created b) Unvalid observations are filtered out from the L2P using a set of selection filters detailed below (table 3) c) The selected observations can possibly be corrected, for instance converting from skin SST to subskin SST and/or using the sensor specific error statistics (SSES

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 10

Report model for MERSEA

bias) provided in the L2P to correct the SST values. For AATSR data, skin to subskin and bias correction are computed at the same time using a statistical anomaly predictive model depending on the available contextual information such as aerosols and wind condition, location, season, data quality and confidence at pixel level and within neighbourhood. This model is adjusted using in situ data. d) The selected observations are remapped onto the grid, using appropriate regridding technique depending on the respective pattern and resolution of the source L2P product and collated file (see details below)  The remapped L2P is merged into the final composite file, for the considered sensor, with the other observations already contained in this file. This merging is performed on the fly as soon as a new remapped L2P is available.

4.2. Available filters (in step 1)
Filter FilterGeo FilterConfidence Description Filter out pixels with unvalid latitude/longitude (occurring in some L2P like AMSRE or TMI) or out of the analysis grid boundaries Filter out pixels out of the specified proximity_confidence range. This allows to select best quality observations and discard data potentially contaminated by clouds. Definition of the proximity confidence may vary depending on the L2P datasets (and not compatible with GDS specifications) such as ASMRE, TMI, GOES products. FilterZenAngle FilterSatZen Filter out pixels out of the specified solar zenital angle range. This filter allows to select only night data for instance. Filter out pixels out of the specified satellite zenital angle. This filter may be of particular interest for geostationary satellites where observations with high incidence angle are usually of poor quality. This angle is almost never provided in L2P (but allowed by GDS specification) and requires an additional ancillary file providing this information for a given geostationary satellite). FilterDWConditio n Filter out pixels potentially affected by diurnal warming. The condition for diurnal warming likelihood are estimated by a maximum wind speed and minimum irradiance, provided as parameters, for which diurnal warming is likely to occur. Many L2P files do not include ancillary data (wind speed and surface solar irradiance) preventing from applying this filter. Besides this screening is very “coarse” and will be improved with respect to recommandations by the GHRSST DW Working Group. FilterAerosol Filter out pixels out of the specified aerosol optical depth range. Many L2P files do not include ancillary data (aerosol optical depth)

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 11

Report model for MERSEA

preventing from applying this filter. Besides this screening is very “coarse” since aerosols fields, when provided, come from forecasting models. FilterIce Filter out pixels out of the specified sea ice concentration range. Many L2P files do not include ancillary data (sea ice concentration) preventing from applying this filter. FilterSSES_bias Filter out pixels out of the specified SSES bias and standard deviation range.

FilterSSES_stdde v Some L2P files do not contain these fields. Besides the corresponding values are not estimated by providers using homogeneous methodologies or using the same in situ reference datasets as input for comparison, which may make this information not reliable or, at least, intercomparable. FilterDTAnalysis Filter out pixels whose confidence flag does not match the provided mask. Some L2P files do not contain this field or the definition of the confidence flag may vary. FilterFlag Filter out pixels out of the specified DT analysis range (distance to a reference SST, climatology or previous analysis). Some L2P files do not contain this field or the used reference may vary a lot (climatology or previous analysis). Will be replaced soon with a control against our own analysis.
Table 3 Selection filters used (if available) in the first step.

USA-RSS-AMSRE-MW-L2-SST

NAVO-L2P-AVHRR17_G NAVO-L2P-AVHRR18_G

NAVO-L2P-AVHRR17_L NAVO-L2P-AVHRR18_L

EUR-L2P-SEVIRI_SST

EUR-L2P-NAR17_SST EUR-L2P-NAR18_SST

OSDPD-L2P-GOES11 OSDPD-L2P-GOES12

EUR-ATS_NR_2P

Correction skin to X subskin Correction SSES with X X X X X X X

FilterConfidence FilterZenAngle

3-4-5 90-180 (night)

4-5 90-180 (night)

3-4-5 90-180 (night)

5 90-180 (night)

3-4-5 90-180 (night)

3-4-5 90-180 (night)

3-4 90-180 (night)

3-4 90-180 (night)

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 12

REMSS-L2P-TMI

Report model for MERSEA

USA-RSS-AMSRE-MW-L2-SST

NAVO-L2P-AVHRR17_G NAVO-L2P-AVHRR18_G

NAVO-L2P-AVHRR17_L NAVO-L2P-AVHRR18_L

EUR-L2P-SEVIRI_SST

EUR-L2P-NAR17_SST EUR-L2P-NAR18_SST

OSDPD-L2P-GOES11 OSDPD-L2P-GOES12

EUR-ATS_NR_2P

FilterSatZen FilterDWConditio n FilterAerosol FilterIce FilterSSES_bias FilterSSES_stdde v FilterDTAnalysis FilterFlag 0-10% 0-0.5 0-10% 0-0.5 0-0.5 0-10%

60

0-0.5 0-10%

0-0.5

Table 4Selection filters configurration for ODYSSEA v1.0

4.3. L2P observations remapping (in step 1)
4.3.1. Available methods
Several methods can be applied to remap the satellite orbit files (swath or snapshot) onto the collated grid, in order to take into account the fact that several satellite pixels may be remapped onto the same single grid cell:  AVERAGING: averaging the pixels falling onto the same grid cell. When averaging the data, it is possible to consider only the pixels having the maximum proximity confidence value (and thus discard pixels with lower proximity confidence value). This method is applied when the satellite pixels have a much higher resolution than the collated grid cells. SAMPLING: selecting, among all the pixels falling into a single grid cell, the closest pixel to the grid cell center. This method is applied when the satellite pixels have a resolution equivalent with the collated grid cells.



Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 13

REMSS-L2P-TMI

Report model for MERSEA

USA-RSS-AMSRE-MW-L2SST 25km 0.25° X X

NAVO-L2P-AVHRR17_G NAVO-L2P-AVHRR18_G

NAVO-L2P-AVHRR17_L NAVO-L2P-AVHRR18_L

EUR-L2P-SEVIRI_SST

EUR-L2P-NAR17_SST EUR-L2P-NAR18_SST

OSDPD-L2P-GOES11 OSDPD-L2P-GOES12

Current software configuratio n EUR-ATS_NR_2P

Pixel resolution Collated resolution Sampling Averaging

1km

2km

2km

3-4km

10km

9km

25km

0.1°

0.1°

0.1°

0.1°

0.1°

0.1°

0.25°

X X X X X X X X

X

Highest X confidence value only

Table 5 Remapping options (step 1) for configuration ODYSSEA v1.0

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

REMSS-L2P-TMI
page 14

Report model for MERSEA

5. MERGING OF REMAPPED ORBIT FILES (STEP 2)
5.1. Definition
The remapped L2P files from a single dataset are merged together in a unique file, referred as “native sensor collated file” only taking into account the observations within a time window (typically +/3 days) centred on the estimation time of the L4 analysis. While the MERSEA L4 analysis is daily, observations from previous days (and next days in delayed mode) are also considered in order to get a better sampling. Of course, data closer to the centre time have higher priority in the merging process (but other criteria such as quality are also considered) and later stronger weight in the objective analysis. The grid definition of the native sensor collated file is identical to the L2P remapped files (same resolution and boundaries).

5.2. Processing
The native sensor collated files are elaborated iteratively as soon as new remapped L2P files are available. The merging of a remapped L2P file into the native sensor collated file is as follow: For each observation of the remapped L2P file :   if there is no observation in the collated file at the same grid point, the observation is added to the collated file if there was already an observation in the collated file at the same grid point, respective representativeness and quality of each one has to be investigated before making a selection. The selection criteria currently implemented are, in hierarchical order : o o o o  best confidence (A) proximity to analysis time (B) minimum risk of diurnal warming effect (not activated) range of satellite zenital angle

The way two criteria A and B are hierarchized is as follow: o if the observation of the remapped L2P is less ranked than the observation already available in the native sensor collated with respect to criterium A, then it is rejected if the observation of the remapped L2P is equally ranked than the observation already available in the native sensor collated with respect to criterium A, then criterium B is investigated however a validity time window (around the centre time of analysis) is associated with each criterium meaning that a criterium is no more effective if the remapped L2P obervation time is out of this time frame (currently 24 hours). For instance, if this observation had a better confidence than the collated observation but much older and out of the associated time frame, then it can not be selected (the more recent observation has priority even if having a lower confidence). The following

o

o

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 15

Report model for MERSEA

criterium in hierarchy is thus investigated. To summarize, a criterium can not be applied (and the next criterium has to be investigated) if :   two observations are equally ranked with respect to a criteria and both in or out the its associated time frame a remapped L2P observation is better ranked but out of the criterium time frame whereas the collated observation is within this time frame

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 16

Report model for MERSEA

6. INTERCALIBRATION OF THE OBSERVATIONS FROM THE SINGLE SENSOR L3 PRODUCTS (STEP 3)
6.1. Definition
The intercalibration step of each dataset aims at removing any large scale bias from the datasets, making all observations consistent and homogeneous before merging into the multisensor collated file. It is performed by computing daily a bias field between the dataset to correct and a reference dataset (currently AATSR). This bias field is then substracted to the sensor observations.

6.2. Processing
The bias field is estimated as follow:  The bias between the dataset to correct and the reference dataset is computed each day for each grid point where observations exist for both datasets, using the data (of the current day only) contained in the respective native sensor collated files. There is one bias observation field for each sensor to correct each day, with the same grid pattern as the corresponding native sensor collated files. The bias observation fields of the last days (typically 10 days) are aggregated, keeping at each grid point the most recent bias observation. This constitutes a kind of collated bias observation file (top panel figure 3) . This bias observation field is optimally estimted over a low resolution grid (typically 5 degrees) and then interpolated (using a bilinear interpolation) at the native sensor collated file resolution. This interpolated bias field is the field to be substracted to the observations of the dataset to correct (middle panel figure 3).





Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 17

Report model for MERSEA

Figure 3 Illustration of the intercalibration step. Top: aggregated local SST biases between the sensor to be calibrated and the reference within +:- 10 days around the processing date (left) and time differences between the processing date and the bias observations (right). Middle: estimated bias. Bottom: histogram of differences between SST observations of the day and the reference before (left) and after (right) correction.

The bias field between the sensor to correct and the reference sensor is re-estimated each day, using a sliding time window of bias observations. The main issue of this approach is the sustained availability of the reference sensor : the processing software is designed to switch easily to a new reference sensor in case the current reference sensor (AATSR) would not be available anymore or not in sufficient sustained manner.

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 18

Report model for MERSEA

7. SINGLE SENSOR L3’S OBSERVATIONS MERGING ONTO A MULTISENSOR L3 PRODUCT (STEP 4)
7.1. Definition
A multisensor collated file (or multisensor L3 file) is the result of the merging of all the native sensor collated files of the day. It aims at providing the most complete field of SST observations for a given day, selecting for each grid point the best and most accurate observation available. It constitutes both an input for the ODYSSEA L4 step and an output product (L3) of our processing chain since it is of high interest for some applications, such as assimilation into ocean circulation models. This product is gridded, with the same grid specification as the final L4 grid. The resolution of this grid may thus differ from the resolution of the native collated files used as input.

7.2. Processing
A global multisensor L3 file is computed each day, as follow:

For each native sensor collated file, the corrected (by the bias correction scheme described above) observations of this file are compared, for each grid pixel, to the observations already available in the multisensor collated file:   If they were not observations yet, they are added If there was already an observation, it is replaced with the new one if and only if the considered sensor has a higher priority at this specific grid cell and both observations are in the same time slice, or if it is in a time slice closest to the analysis time (recent data have priority over older ones). A ranking of all native products was previously established with respect to known respective accuracy and properties of these products. The order of these products may vary depending on the location (some sensors are known to operate better in some regions than others). The time slices are typically 12 to 24 hours, starting from the analysis time.

This selection strategy is being currently upgraded in order to built a more precise decision tree to select the best observation, depending on the SST retrieval conditions (SST range, wind, aerosols, sea ice, cloud proximity, time difference, area) and on a classification of the respective sensor errors under each class of these criteria.

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 19

Report model for MERSEA

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 20

Report model for MERSEA

Figure 4 Multisensor L3 product on 15/10/07. Top : SSTobservations. Middle : time difference. Bottom : sources of SST

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 21

Report model for MERSEA

8. ANALYSIS STEP (STEP 5)
8.1. Analysis method
The analysis system uses estimation theory for mapping a scalar field on a regular grid from sparse and irregular data (see Bretherton et al. (1976), Kaplan et al. (1997) for application to the ocean). The interpolated field, represented by the state vector x, is constructed as the departure from a reference field at the grid points x f . This reference is derived from previous knowledge (climatology or forecast). The analysed field x a is obtained as a linear combination of the innovation ( y o  y f ) and is associated with a covariance matrix P a . The error on the estimation is given by the diagonal of this matrix.

x a  x f  K OI ( y o  y f )
T P a  P  K OI C ao

K OI  C ao (Coo  R ) 1
The K OI matrix is built from the matrices that express the covariances of the field, from grid point to data point ( Cao ), and from data point to data point ( Coo ) and the observation noise covariance matrix R. P is the covariance of the field at grid points.

8.2. A priori statistics
8.2.1. Background
The analysis uses a reference daily climatology. We used the 4km Pathfinder Version 5 (5day) climatology to construct this reference. The main steps of the processing are: a) Using the “Clim_SST_filled” field (ftp://data.nodc.noaa.gov/pub/data.nodc/pathfinder/Version5.0_Climatologies/README.txt): the values inferior to -1.9°C are set to -1.9°C, the pixels beyond an ice limit (defined from ERS PSI products, Gohin et al, 1998) are set to -1.9°C. The remaining gaps are filled by linear interpolation. b) The reference fields are gridded on the global analysis grid (0.1 x 0.1 degree). To avoid introducing small scales artifacts or noise in the interpolated fields due to a noisy background, we choose to prepare a smoothed reference. An illustration of the effect of a noisy background is given on figure 5. c) A linear time interpolation (from the 5-day fields) has been used to create the daily fields.

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 22

Report model for MERSEA

Figure 5 Top left: Extract from a SST mean field with a 0.1 degree resolution calculated (by averaging) from the original 4 km Pahtfinder SST mean. Top right: a set of SST observations. Bottom left: corresponding analysed field presenting artefacts due to the noisy background. Bottom right: We choose to smooth (in space) the original 4 km Pahtfinder SST mean to construct the ODYSSEA analysis background.

8.2.2. Structure functions
We specify the covariances of the field by a structure function modeled by a sum of two Gaussian functions, each associated with specific time and space scales:

C (dx, dy, dt )   i2 exp (
i 1

2

dx 2 dy 2 dt 2   ) 2 L2 2 L2 2 L2 ix iy it

where dx, dy, dt are the space and time separations, Lix , Liy , Lit the corresponding efolding scales. The variances  i control the weight given to each ocean scale. Figure 6 shows an example of the function used (black curve), to calculate the covariance value between two points.
2

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 23

Report model for MERSEA

Figure 6 Example of structure function

8.2.3. Correlation scales
The first scale lengths are taken isotropic and equal to 80 km (to limit the influence radius value set to 5 times the larger length scale in the analysis step). The second scale lengths are isotropic and set to Rossby radius values (bounded by 20 km and by the large scale 80 km). The Rossby radius values have been calculated on a 1x1 degree grid by using the 1998 Levitus climatology. The time length scales are set to 2 days and 1 day. Each analysis takes into account the data measured within the time interval -3/+3 days around the estimation date.

Figure 7 Spatial correlation lengths (km) for the meso-scale

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 24

Report model for MERSEA

8.2.4. Variances
The total variance of the anomaly is to be calculated with a large data set. It will be done with the Pathfinder Version 5 archive. At this time, the total variance is set to a constant value: 3 °C. We assume that this variance is the sum of four terms:
2 2 2 2  tot   12   2   UR   ME 2 2 2 Where  12 and  2 are the variances associated to each scale, the sum  UR   ME is the

2 2 total error variance:  ME corresponds to the instrumental errors and  UR represents small scales unresolved by the analysis and considered as noise. We express the variance associated to each scale as a function of the ocean variance by including normalized weights:

 12  w1 2
2  2  w2 2 2  UR  wUR 2

w1  w2  wUR  1
The relative weights w1 , w 2 , wUR define the distribution of variance over the different scales. In ODYSSEA v1.0, w1 =1/4 (large scale), w 2 =2/4 (meso-scale), wUR =1/4. The error matrix R combines the measurement error and the representativity error, it assumed diagonal.

8.3. Ice and land mask
8.3.1. Land mask
A land mask has been defined on the ODYSSEA v1.0 analysis grid. We extracted the land mask from GMT (Graphics Mapping Tools, Wessel P. and Smith W. H. F,The Generic Mapping Tools, GMT, 2006).

8.3.2. Ice mask
An ice mask is defined for each analysis. In the configuration of the daily system ODYSSEA v1.0, the ice mask is defined daily by using the ice product provided by SAFO&SI (ref.). The variable sea_ice_fraction provided in this product is remapped on our analysis grid and copied in our L4 file. However, for each analysis we extrapolate (with the nearest valid value) the ice data to fill the icy regions (with sea pixels) close to land. The ice mask and the land mask used to affect fillvalues to our L4 product are stored in the variable mask. The variable sea_ice_fraction contain the values provided by the SAFO&SI product.

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 25

Report model for MERSEA

9. REAL TIME CONTROL TOOLS, DATA ACCESS AND FORMAT
The ODYSSEA data are produced and archived at Ifremer. They are available both on ftp and through OpenDAP at the following url :

ftp access ftp://ftp.ifremer.fr/ifremer/medspiration/data/l4hrsstfnd/eurdac/glob/odyssea access is anonymous. OpenDAP access http://www.ifremer.fr/cgi-bin/nph-dods/data/satellite/medspiration/l4hrfnd/eurdac/glob/odyssea The data are in netcdf format, complying to CF-1.0 convention and GHRSST format specification. More details on this format can be found on GHRSST web site or in the following document: http://www.medspiration.org/documents/GHRSST-PP-Product-User-Guide-v1.1.pdf Product visualization pages as well as real time data quality control pages monitoring the overall quality of the ODYSSEA products (with comparisons to other products) and of each single source of SST observations are available on Mersea web site at the following address: http://www.mersea.eu.org/Satellite/sst_validation.html

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 26

Report model for MERSEA

10. REFERENCES
Antonov, J. I., S. Levitus, T. P. Boyer, M. E. Conkright, T. D. O'Brien, and C. Stephens, 1998. World Ocean Atlas 1998 Vol. 1-2-3: Temperature. NOAA Atlas NESDIS 27-28-29, U. S. Government Printing Office, Washington D. C. - Boyer, T. P., S. Levitus, J. I. Antonov, M. E. Conkright, T. D. O'Brien, and C. Stephens, 1998. World Ocean Atlas 1998 Vol. 4-5-6: Salinity. NOAA Atlas NESDIS 30-31-32, U. S. Government Printing Office, Washington D. C. Bretherton F. P., R. E. Davis, and C. B. Fandry, A technique for objective analysis and design of oceanographic experiments applied to MOD-73, Deep Sea Res., 23, 559-582, 1976. Casey K. S., Cornillon P., A comparison of satellite and in situ based sea surface temperature climatologies, Journal of Climate, vol 12, no 6 pp 1848-1863, 1999. Daley, R., Atmospheric Data Analaysis, Cambridge Atmospheric and Space Sciences Series, Cambridge Univ. Press, New York, 1991. Gandin, L. S., Objective Analysis of Meteorological fields, Isr. Program for Sci. Transl., Jerusalem, 1965. Ghrsst-pp International Project Office, The GHRSST-PP Product User Guide, http://www.ghrsstpp.org/documents.htm, 2005. GHRSST-PP Science Team and GDS-TAG working group, The GRHSST-PP Data Processing Specification, GDSv1 Revision 1.7, http://www.ghrsst-pp.org/documents.htm, 2007. Gohin F., Maroni C., ERS scatterometer sea ice polar grids, user manual, IMSI report No. 5, Ifremer/Cersat technical report, C2-MUT-W-03-IF V2, 1998. Kaplan, A., Y. Kushnir, M.A. Cane, and M.B. Blumenthal, Reduced space optimal analysis for historical datasets : 136 years of Atlantic sea surface temperatures. J. Geophys. Res., 102 (C13), 27835–27860, 1997. Le Borgne P., Marsouin A ., Orain F., Roquet H., Coat A., Guichoux Y., Implementation of a fine scale analysis over the Atlantic Ocean, MERSEA deliverable D2.2.8, WP 02 Task 2.2, MERSEA-WP02-MF-STR-002-1A.doc, 2006. Le Borgne P., Marsouin A ., Orain F., Collated files for SST analysis: bias correction, MERSEA WP 02 Task 2.2, MERSEA-WP02-MF-STR-003-1A.doc, 2006. Wessel P., Smith W. H. F, The Generic Mapping Tools, GMT, version 4.1.3, Technical reference Cookbook, General Mapping Tools Graphics, 2006

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 27

Report model for MERSEA

Ref.: 0377eb10-f90c-4f63-ab29-b10c9452e1e1.doc

page 28


				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:10
posted:1/20/2010
language:English
pages:29
Lingjuan Ma Lingjuan Ma
About