A HYDROGRAPHIC AND BIO-CHEMICAL CLIMATOLOGY OF THE MEDITERRANEAN AND THE BLACK SEA: SOME STATISTICAL PITFALLS (modb.oce.ulg.ac.be/medar) Michel Rixen 1, Jean-Marie Beckers 2 and Catherine Maillard 3 The Color of Ocean Data Brussels, Belgium, November SOC, Southampton, UK 2.GHER, University of Liège, Belgium, 3. SISMER, Ifremer, Centre de Brest, BP70, Plouzane, France
Task I, II, III, V At 15:20 Recent advances in oceanographic data management of the Mediterranean and Black Seas: The MEDAR/MEDATLAS 2002 data base By C. Maillard and E. Balopoulos (France, Greece)
Objective analysis –Optimal interpolation (OI) (+ sub-optimal schemes) –Successive corrections (SC) (+ sub-optimal schemes) –Variational inverse model (VIM) (stat. Equiv. to OI) –…. Task IV: climatology
The Variational Inverse Model Dimensional analysis+Bessel K1 correlation function
Finite element mesh
VIM:no bias OI: Information crosses boundaries
Computational cost Field Error field OI VIM OI VIM
Climatology: some details 25 standard vertical levels (Obsolete: automatic QC: –data rejected if outside 3*std locally) Sandwell bathymetry at 2’ –Used for contours and FEM Reference field=climatic field –(semi-normed analysis) T,S,Alkalinity,DOX,NH4,NO2,NO3,PO4, SiO4,H2S,pH,Chl Climatologic, seasonal, monthly, inter-annual and decadal temporal windows when relevant 20km x 20 km, 8 km x8 km or 5 km x 5 km resolution Analyzed and error fields
A good example: enough data
Another good example: enough data
Even more good examples…
Alboran, 200m, many data Levantine basin, 200m, few data VIM and OI: statistical hypotheses - gaussian frequency distributions - statistics are homogeneous and isotrope - uncorrelated noise
Nitrite Salinity PhosphateSilicate Ph Temperature
Vertical distribution of temperature
Yearly distribution of salinity
Monthly distribution (salinity)
Ionian 1980 Months Months Months Months Months Months 1 Levantine 1984 Months Months Months Months Months 1 2 Temp Possible bias ? Temp
2D analysis appropriate?
PO4 at 30m: coastal (<18km) and/or shallow sounding (<50m)
With coastal data Without coastal data Difference Phosphate (mmole/m3)
Ionian (36-37 ºN, 19-20ºE)
Some potential problems… Statistical hypothesis Correlation length= km, so at least data homogeneously distributed needed! Few data at deeper levels: –extrapolate from upper levels? Coastal data bias beyond the physical diffusion/ advection through correlation length: in several areas the only existing data Last but not least: obvious errors in the raw data (e.g. instrument calibration)
Selection of robust fields Annual, seasonal and monthly climatology –Temperature, Salinity Annual and seasonal –Oxygen, Silicate, Phosphate –Hydrogen sulphide (H 2 S) in the Black Sea Annual only –Nitrate, Nitrite, PH, Ammonium, Alkalinity, Chlorophyll So far: (almost) the best we can do…
Future More data! Other parameters (e.g. ADCP, DIC, POM, DOC,…) Multivariate analysis and QC (bio-chemical data!) 3D Variational analysis? More high level final products Your feedback MEDAR Climatology: “modb.oce.ulg.ac.be/medar” Free access to fields figures animations