Inter-comparison and development of SST analyses over the

Inter-comparison and development of SST analyses over the Mediterranean Sea B. Buongiorno Nardelli, C. Tronconi, R. Santoleri, E. Böhm Istituto di Scienze dell’Atmosfera e del Clima – sezione di Roma Via del Fosso del Cavaliere, 100 – 00133 Roma bruno@gos. ifa. rm. cnr. it

GOS involvement in national and international projects/programmes Mediterranean Forecasting System Adricosm Medspiration GODAE Global High Resolution Mersea SST Pilot Project (GHRSST-PP) PRIMI

MFSTEP (mar. 2003 - feb. 2006) Mediterranean Forecasting System Toward Environmental Predictions RT Observing System satellite SST, SLA, VOS-XBT, moored multiparametric buoys, ARGO and gliders Ecosystem Models Validation/calibration of Coupled physical and biochemical numerical models End-User applications Development of modules for oil spill monitoring, ICZM and fishery management 15 nations involved, 48 institutions Upgrade of present basin scale operational system New model and assimilation Marine forecast downscaling Regional and shelf models nesting Meteorological forecast downscaling 10 km LAMs and 4 km N. H. mesoscale models

MFSTEP sub-regional and shelf systems MFS supports sub-regional (3 km) and shelf models (1 km) nesting: weekly forecasts are produced for ALL the sub-regional models and some shelf models Sub-regional models at 3 km Shelf models at 1. 5 km

The present day MFS (SYS 3) weekly forecasting system ECMWF FC ECMWF AN Wed Thu Fri Sat Sun Mon Tue J-14 J-7 J-5 J-4 J-3 J-2 J-1 Wed J Thu Fri Wed Thu Fri J+1 J+2 J+7 J+8 J+9 SLA SST FORECAST XBT RELEASE ARGO Data are disseminated through a Web/ftp service (www. bo. ingv. it/mfstep)

The MFS SST system MF AVHRR acquisition Atlantic buffer zone + west Med ISAC AVHRR acquisition Entire Mediterranean Night-time SST using MF algorithm Night-time SST using Pathfinder algorithm Cloud detection SST daily composite binning on model grid (1/16 x 1/16) Data merging ISAC Data quality control Optimal Interpolation Data delivery on the GOS-ISAC web-site

CLOUD detection algorithms Cloud detection is an essential step to provide “high quality” SST fields for data assimilation Cloud detection requires a compromise between maximization of coverage and minimization of cloud contaminated pixels Cloud detection in MFSTEP is now performed at various steps: On original images, before the composite is performed: ISAC: building a reference SST and fixing a threshold on the base of the histogram of the differences to this reference (as already described last year by E. Böhm) Before selecting SST data in the optimal interpolation algorithm: comparison to the nearest analysis available (if interpolation error is lower than a fixed value)

SST INTERPOLATION ALGORITHM Basic theory Given n SST observations Φobs at the locations xi (both in space and time) and their associated measurement error εi (assumed to be zero mean and uncorrelated with the signal), the Gauss-Markov theorem states that the optimal least square estimate of the SST at the location x can be obtained as a linear combination of the observations Φobs: where Cxi represents the covariance between the quantity to be estimated and the ith observation: Aij represents the covariance matrix of the observations:

SST INTERPOLATION: practical limitations Any method described as OPTIMAL necessarily becomes strongly SUB-OPTIMAL when implemented to interpolate high resolution satellite data, due to: • Volume of data • Computational limitations • Different scales to be considered in the interpolation Any scheme needs to be built with strong FLEXIBILITY

SST INTERPOLATION: operational implementation Some details about the scheme adopted… • The data used to interpolate at a certain time-space location are selected within a limited sub-domain, close to the interpolation point The most correlated observation is selected first, while all successive data are selected only if they are found along a new direction in the space-time (until n observations are found). • The scheme drives a ‘multi-basin’ analysis to avoid data propagation across land, from one sub-basin to the other.

OI scheme configurable parameters FLAG_INT=1 does not interpolate if a valid observed SST value is present at the interpolation time. FLAG_INT=0 interpolates anyway. LIMIT maximum number of selected data for each interpolation point DIST starting spatial influential radius for data selection. RMAXDIST maximum spatial influential radius for data selection (the influential radius is incremented if data selected within DIST are less then LIMIT). NPIX number of values selected in time for each pixel. THRESH maximum difference admitted between the SST in input and the reference SST (used for residual cloudy pixels removal). The reference SST is either the corresponding day optimal field (for days before the interpolation day J) or the J-1 analysis for the interpolation day (J) and successive days (J+1, J+2, . . . ) RMAXERROPT maximum % error admitted on the reference SST field to activate the residual cloudy pixels removal. IBX dimension of the moving window used for cloud erosion MINSSTVALID minimum SST value considered valid

MF and GOS SST composite images

Daily MF/GOS composite

SST INTERPOLATION

The Global Ocean Data Assimilation Experiment (GODAE) high-resolution sea surface temperature pilot project aims to develop a new generation of global high-resolution (<10 km) SST data products to the operational oceanographic, meteorological, climate and general scientific community, in real time and delayed mode

• L 2 P data products provide satellite SST observations together with a measure of uncertainty for each observation in a common net. CDF format. • L 4 gridded products are generated by combining complementary satellite and in situ observations within Optimal Interpolation systems.

MEDSPIRATION Project • • • European RDAC for the GHRSST-PP Delivery of real-time high quality SST data, matching the GHRSST-PP specification [GDS v 1. 5] Operational and sustained production Generic and scalable system Products : Match-up database (MDB) Gap free high resolution maps (L 4) Direct observations (L 2 P) High-resolution diagnostic datasets

Project organisation ESA/DUE O. Arino Medspiration consortium Project management, Ian S. Robinson, SOC core members system design, implementation, operations support Test users L 2 P product P. Le Borgne, Météo-France Project management/quality, J. Rickards, Vega L 2 P, L 4 products L. Santoleri, B. Buongiorno Nardelli, CNR Archive, dissemination, MDB and L 4 processing J. F. Piollé, IFREMER Expertising (L 4 product) L. Santoleri, B. Buongiorno Nardelli, CNR L 2 P products S. Eastwood, Met. No L 4 product G. Larnicol, CLS HR-DDS D. J. Poulter L 2 P processing software A. Coat, Avelmor

CNR contribution to MERSEA SST activities General Objective: Conduct R&D activities to improve the quality of SST products used by MERSEA modeling and assimilation centers and produce global, Atlantic and Mediterranean Sea analyzed SST fields needed for MERSEA regional and global models Specific CNR work: • Inter-comparison and validation of MEDSPIRATION L 4 and MFSTEP SST products • Tuning of Medspiration L 4_processor • Improvement of MFSTEP analyses

Inter-comparison and validation of MEDSPIRATION L 4 and MFSTEP SST products REMARKS: Medspiration L 4 is at 2 km res. , MFS is at 1/16° Different input -MFS: only AVHRR (CNR+CMS) -Medspiration: L 2 P (AATSR, AVHRR(CMS+Navoceano), SEVIRI…) Different data editing and selection -account for bias between sensors -selection of valid input (confidence values, temporal window…) Different OI algorithm configurations -spatial influential radius (‘bubble’) -strategy for the selection of influential observations within the bubble

Inter-comparison and validation of MEDSPIRATION L 4 and MFSTEP SST products. Methods: Evaluation of processor performance: -qualitative -quantitative MFS XBT data Comparison of SST L 4 against quality controlled in situ XBT data acquired within MFSTEP Test performed: MFS at 1/16° AVHRR by CNR+CMS merging MFS data and SEVIRI/AATSR L 2 P only L 2 P (all infrared) original configuration Medspiration L 4 at 2 km resampled at 1/16° L 2 P original configuration Medspiration L 4 at 1/16° (hereafter Medspiration 1/16 L 4) L 2 P different configurations starting from parameters similar to MFS ones

Inter-comparison and validation of MEDSPIRATION L 4 and MFSTEP SST products MEDSPIRATION L 4 subsampled at 1/16° MFS (only AVHRR)

Inter-comparison and validation of MEDSPIRATION L 4 and MFSTEP SST products MFS and Medspiration L 4 original configuration MFS (only AVHRR) MEDSPIRATION L 4 subsampled at 1/16° MBE=-0. 11 °C MBE=-0. 16 °C Rms=0. 52 °C Rms=0. 55 °C

MERSEA L 4_processors configuration • MFS processor modified to include all infrared sensors (L 2 P) • Medspiration scheme tuned on the base of MFS processor : - similar spatial and temporal influential radius (‘bubble’)… - same correlation function - same grid/resolution • However the two processors have different data editing and selection criteria/strategies -bias between sensors MFS: adjustment to a reference sensor Medspiration: adjustment through OI (error covariance matrix) -selection of valid input (confidence values, clouds) -selection of influential observations within the bubble number of observations data reduction temporal selection

The Mediterranean GOS L 4 SST products flow chart MF AVHRR acquisition Atlantic buffer zone + west Med ISAC AVHRR acquisition Entire Mediterranean Night-time SST using MF algorithm Night-time SST using Pathfinder algorithm Cloud detection SST daily composite binning on model grid (1/16 x 1/16) L 2 P GHRSST Products Data merging ISAC Data quality controll Optimal Interpolation Data delivery on the GOS-ISAC web-site

MFS L 4 (Single-sensor vs multi-sensors) Single sensor Multi sensors

Medspiration L 4 (old vs new configuration) OLD Config. New config.

MERSEA L 4_processors configuration (GOS) Medspiration (new configuration) No Bias correction MBE=-0. 26 °C Rms=0. 52 °C MFS (L 2 P in input) MBE=-0. 11 °C Rms=0. 46 °C

THE SENSOR BIAS ISSUE: background MFS: Reference sensor ”merged files” • Interpolation uses in input ‘merged’ files (1 SST map per day) • Merging procedure selects valid pixels using the sensor sequence below: AATSR NAR 17 AVHRR 17_L SEVIRI NAR 16 AVHRR 16_L • Before adding data to the merged map, the bias between each new image and the pixels that have already been merged is estimated and removed (only if sufficient co-located pixels are found) Medspiration: No preliminar adjustment performed ”collated files” • Data reduction in time (through OAN_KEEP_ALL_MEAS parameter) • Selection of best value for the same sensor: SELMS_LIST > NAR 17_SST AVHRR 17_L NAR 16_SST AVHRR 16_L BIAS adjustment within the OI algorithm: The error covariance is calculated as for points i, j , where b 2, ELW are the variance of the white measurement noise and the bias error coming from a given SSES_Bias_error, respectively.

Example of evaluation of SENSOR BIASES (for existing L 2 P) SEVIRI AVHRR 16 MBE=-0. 07 °C MBE=-0. 56 °C Rms=0. 51 °C Rms=0. 76 °C NAR 17 NAR 16 MBE=-0. 002 °C MBE=-0. 18 °C Rms=0. 49 °C Rms=0. 64 °C

MERSEA L 4_processors configuration (GOS) Main results: MFS (L 2 P in input) MBE=-0. 11 °C Rms=0. 46 °C Medspiration Bias correction – Original signal variance rms similar to MFS, higher bias vs in situ MBE=-0. 21 °C Rms=0. 47 °C MBE=-0. 22 °C Rms=0. 41 °C Medspiration Bias correction (1 to 3 times the estimated MBE for each sensor) Lower signal variance (1 -5 °C) improved rms, bias always there!

Last steps, future work: • Inter-comparison and validation of MEDSPIRATION L 4 and MFSTEP SST products. concluded • Tuning of Medspiration L 4_processor -identified possible evolutions/problems in the L 4_processor code -started tests on 2 -step interpolation stopped due to software licence issues • Improvement of MFSTEP analyses -Run MFS L 4_processor only with L 2 P -Update MFS format to standard GHRSST conventions -install and configure THREDDS, create catalog -include MODIS data in the analyses -Implement new MFS-L 4 production in the operational chain -test on 2 -step interpolation (at 2 km resolution)

Evaluation of SENSOR BIASES (MODIS data) MODIS Terra (11 micron) MODIS Aqua (11 micron) MBE=-0. 21 °C MBE=-0. 34 °C Rms=0. 38 °C Rms=0. 55 °C MODIS Terra (4 micron) MODIS Aqua (4 micron) MBE=-0. 04 °C MBE=-0. 16 °C Rms=0. 33 °C Rms=0. 54 °C

The warm summer 2006 The 2006 SST anomaly was monitored in near real time by the GOS SST Processing system daily SST anomaly respect to the 1985 -2004 climatology Time series of SST mean in the West Med

CNR THREDDS CATALOG STRUCTURE

CNR THREDDS CATALOG STRUCTURE MERSEA According to Thomas. Loubrieu (ifremer) specification, organization SEA SURFACE TEMPERATURE organization we set Near Real Time (NRT) - SST INTERPOLATED MAPS up our Thredds Data Server organization Delayed Time (DT) - SST INTERPOLATED MAPS OCEAN COLOR work in progress…. . organization

CNR THREDDS CATALOG STRUCTURE MERSEA SURFACE TEMPERATURE Near Real Time (NRT) - SST INTERPOLATED MAPS Delayed Time (DT) - SST INTERPOLATED MAPS OCEAN COLOR work in progress…. . organization organization

SST DATA PRODUCT For what concerns the SST product you can find 4 different SST products: • NRT-v 0: Optimal Interpolation SST dataset from AVHRR data only processed in near-real-time mode (http: //www. mersea. eu. org/html/information/catalog/products/GOS-MED-L 4 SST-NRTv 0 -OBS. html) • NRT-v 1: Optimal Interpolation SST dataset from all infrared L 2 P processed in near-real-time mode. (http: //www. mersea. eu. org/html/information/catalog/products/GOS-MED-L 4 SST-NRTv 1 -OBS. html) • DT-v 0: Optimal Interpolation SST dataset from AVHRR data only processed in delayed-time mode. (http: //www. mersea. eu. org/html/information/catalog/products/GOS-MED-L 4 SST-DTv 0 -OBS. html) • DT-v 1: Optimal Interpolation SST dataset from all infrared L 2 P processed in delayed-time mode. (http: //www. mersea. eu. org/html/information/catalog/products/GOS-MED-L 4 SST-DTv 1 -OBS. html)

SST DATA PRODUCT Two different types of time organization: • NOT AGGREGATED • AGGREGATED

SST DATA PRODUCT Two different types of time organization: • NOT AGGREGATED • AGGREGATED

XML THREDDS CATALOG FILES This structure is done with 13 configuration xml catalog files mersea. xml sst_init. xml, sst_NRT. xml nrt_v 0. xml, nrt_v 0_aggr. xml, nrt_v 1_aggr. xml, sst_DT. xml dt_v 0. xml, dt_v 0_aggr. xml dt_v 1. xml, dt_v 1_aggr. xml ocean_init. xml Updated daily for V 1 (NRT, DT) data, and weekly for V 0 (NRT, DT) data

CNR THREDDS CATALOG STRUCTURE MERSEA SURFACE TEMPERATURE Near Real Time (NRT) - SST INTERPOLATED MAPS Single Sensor AVHRR(v 0) From 2004 -05 -19 to 2007 - 04 -16 (last Monday) Not Aggregated: 1062 single downloadable nc files organization product view Aggregated: 1062 single nc files collated in a single nc file weekly updated: every tuesday Multi Sensor (v 1) From 2006 -07 -01 to current data Not Aggregated: 294 single downloadable nc files Aggregated: 294 single nc files collated in a single nc file daily updated view Delayed Time (DT) - SST INTERPOLATED MAPS Single Sensor AVHRR(v 0) From 2005 -12 -01 to 2007 -04 -06 (10 days before last Monday) Not Aggregated: 476 single downloadable nc files Aggregated: 476 single nc files collated in a single nc file weekly updated: every Tuesday Multi Sensor (v 1) From 2007 -07 -01 to 2007 -04 -13 (10 days before this Monday) Not Aggregated: 287 single downloadable nc files Aggregated: 287 single nc files collated in a single nc file daily updated organization product view view

SST DATA ACCESS AND DOWNLOAD You can access these data in two different ways: 1) Directly from our Thredds catalog page: http: //fe 4. sic. rm. cnr. it: 8080/thredds/catalog. html Free access and free download You can download data clicking directly on the files link, for example for NRT-V 0 aggregated data, you have to click on Access: OPENDAP: http: //fe 4. sic. rm. cnr. it: 8080/thredds/dods. C/sst_nrt_v 0_aggr/GOS-MED-L 4 SST-NRTv 0_aggr as ASCII DATA or Binary Data, choosing the variables to download and for each variable the time period you are interested in. 2) From Ifremer link: http: //www. ifremer. fr/thredds 3/ Free access and free download You can search and download data use the Mersea THREDDS browser