Post on 09-Aug-2020
Gabriel JORDÀ, Loïc HOUPERT, Damià GOMIS, Anthony BOSSE, Pierre TESTOR, Josep LLASSES, Charles TROUPIN
gabriel.jorda@uib.es
INSTITUT MEDITERRANI D’ESTUDIS AVANÇATS
A NEW HYDROGRAPHIC GRIDDED PRODUCT FOR THE MEDITERRANEAN. IMPACT OF XBT CORRECTION ON
HEAT CONTENT ESTIMATES.
The 4th XBT Workshop: XBT Science and the Way Forward 11-13 November 2014, 2014
Beijing, China
The Mediterranean Sea
Marginal Sea
Highly populated coasts
Large Biodiversity (8% of known marine species with
0.8% of surface)
Hot spot for Climate Change (Giorgi 2006)
Historical hydrographic data are sparse in time and space, which is inconvenient for climate studies
Gridded products are required for scientific studies and model validation BUT … A lot of extrapolation is required … Are they accurate enough?
Motivation
Optimal interpolation is usually chosen to generate the maps and to provide an estimate of the formal error, .. but it depends on a-priori assumptions : A more accurate estimate of the associate uncertainties is required
Motivation
Optimal Interpolation
The goal is to produce a new and updated hydrographic gridded product for the Mediterranean
along with a realistic measure of its errors.
Motivation
Data base as collected in Houpert et al (2014) Includes XBT, MBT and CTD data collected by the Medar-MEDATLAS project (till 2000), WDO and National programmes not included in the international databases (from Italy , France and Spain). It also includes ARGO profilers (from Coriolis Data Center) and GLIDER data (EGO project) Quality control procedure applied to all raw data as well as specific corrections for XBT and MBTs (see later). About 140.000 temperature profiles and 75.000 salinity profiles are kept
Methodology -> The in – situ data
Methodology -> The in – situ data
· Relative importance of each type of measurements evolves on time · Before 1965 only MBTs (less accurate) · Few CTDs during last decade (dissemination problem?) · Large contribution of Gliders during the last decade but profiles can be redundant · Few data in the deeper layers
Termosalinometer data is also available but not considered here (it provides only surface data only) Bottles have also not be included
Methodology -> The in – situ data
· Accuracy 0.001 psu (CTD), 0.002 psu (Argo), 0.01 psu (Glider, moorings) · Similar to temperature BUT without XBT and MBT · Calibration of salinity data is very important. An offset O(0.01) psu is possible between historical data · Thermal lag problem specially in GLIDERs and ARGO floats
Termosalinometer data is also available but not considered here (it provides only surface data only) Bottles have also not be included
-Mapping algorithm based on a variational approach (close to Optimal Interpolation). The DIVA tool (Troupin et al., 2012) has been implemented -Correlation length scale and noise-to-signal ratio tuned using data from models -Analysis performed in a grid of 0.2º of spatial resolution and 33 levels from surface to bottom
Methodology -> Mapping procedure
PRODUCT Time span DX Zmax Time Resolution
XBT/MBT Correction
IMEDEA 1950-2013 0.2º Bottom Monthly Cowley 2013
MEDATLAS2 1950-2001 0.2º Bottom Yearly Not
EN3 1950-2011 1º Bottom Monthly Not
EN4 1900-2013 1º Bottom Monthly Not
ISHII 6.7 1945-2006 1º 700m Monthly Ishii 2006
ISHII 6.12 1945-2011 1º 1500m Monthly Ishii 2009
Methodology -> Comparison to other existing products
PRODUCT Time span DX Zmax Time Resolution
XBT/MBT Correction
IMEDEA 1950-2013 0.2º Bottom Monthly Cowley 2013
MEDATLAS2 1950-2001 0.2º Bottom Yearly Not
EN3 1950-2011 1º Bottom Monthly Not
EN4 1900-2013 1º Bottom Monthly Not
ISHII 6.7 1945-2006 1º 700m Monthly Ishii 2006
ISHII 6.12 1945-2011 1º 1500m Monthly Ishii 2009
IMEDEA PRODUCT EN3 PRODUCT
SST January 1950
Methodology -> Comparison to other existing products
PRODUCT Time span DX Zmax Time Resolution
XBT/MBT Correction
IMEDEA 1950-2013 0.2º Bottom Monthly Cowley 2013
MEDATLAS2 1950-2001 0.2º Bottom Yearly Not
EN3 1950-2011 1º Bottom Monthly Not
EN4 1900-2013 1º Bottom Monthly Not
ISHII 6.7 1945-2006 1º 700m Monthly Ishii 2006
ISHII 6.12 1945-2011 1º 1500m Monthly Ishii 2009
7.1 7.15 7.2 7.25 7.3 7.35
x 105
13.4
13.5
13.6
13.7
13.8
13.9
14
14.1MEDITERRANEAN MEAN TEMPERATURE FROM EN3
Volume correctedVolume UNcorrectedSurface Area in EN3=3.8 ·1012
m2 while in IMEDEA is 2.6 ·1012 m2, much closer to the actual values. Also, the averaged depth in EN3 is ~1900m while in reality is 1350 m.
Methodology -> Comparison to other existing products
PRODUCT Time span DX Zmax Time Resolution
XBT/MBT Correction
IMEDEA 1950-2013 0.2º Bottom Monthly Cowley 2013
MEDATLAS2 1950-2001 0.2º Bottom Yearly Not
EN3 1950-2011 1º Bottom Monthly Not
EN4 1900-2013 1º Bottom Monthly Not
ISHII 6.7 1945-2006 1º 700m Monthly Ishii 2006
ISHII 6.12 1945-2011 1º 1500m Monthly Ishii 2009
Methodology -> Comparison to other existing products
PRODUCT Time span DX Zmax Time Resolution
XBT/MBT Correction
IMEDEA 1950-2013 0.2º Bottom Monthly Cowley 2013
MEDATLAS2 1950-2001 0.2º Bottom Yearly Not
EN3 1950-2011 1º Bottom Monthly Not
EN4 1900-2013 1º Bottom Monthly Not
ISHII 6.7 1945-2006 1º 700m Monthly Ishii 2006
ISHII 6.12 1945-2011 1º 1500m Monthly Ishii 2009
Methodology -> Comparison to other existing products
PRODUCT Time span DX Zmax Time Resolution
XBT/MBT Correction
IMEDEA 1950-2013 0.2º Bottom Monthly Cowley 2013
MEDATLAS2 1950-2001 0.2º Bottom Yearly Not
EN3 1950-2011 1º Bottom Monthly Not
EN4 1900-2013 1º Bottom Monthly Not
ISHII 6.7 1945-2006 1º 700m Monthly Ishii 2006
ISHII 6.12 1945-2011 1º 1500m Monthly Ishii 2009
Methodology -> Comparison to other existing products
The error of the product is estimated using an ensemble of surrogate data From a numerical model (NEMOMED8, Beuvier et al. 2010) we extract virtual observations at the same locations and time than the actual observations were. Then, we generate the maps using the same procedure than with real data Finally we compare them with the original model fields “the reality”.
-5 0 5 10 15 20 25 30 3530
35
40
45
NEMO ANOMALY Field for 01-Dec-1999
-2
-1
0
1
2
-5 0 5 10 15 20 25 30 3530
35
40
45
OI ANOMALY Map for 16-Dec-1999 12:00:00
-2
-1
0
1
2
-5 0 5 10 15 20 25 30 3530
35
40
45
OI-NEMO for 16-Dec-1999 12:00:00
-1
0
1
Methodology -> Error estimate
We repeat the procedure N times randomly chosing the year, so for each data distribution we have an estimate of the associated uncertainty
Methodology -> Error estimate
Uncertainties linked to data gaps but also to unresolved scales
and background (first guess) unaccuracies
Estimated Error for surface salinity maps associated to different data coverage
Validation
Mediterranean Averaged SST
Deseasoned and 12-months smoothed time series
PRODUCT Correlation RMSE (ºC)
IMEDEA 0.84 0.13
EN3 0.69 0.21
EN4 0.73 0.19
ISHII v6.7 0.72 0.28
ISHII 6.12 0.81 0.17
Uncertainty of the new product
-5 0 5 10 15 20 25 30 3530
32
34
36
38
40
42
44
46test23 Correlation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1
0
T10m - Correlation
-5 0 5 10 15 20 25 30 3530
32
34
36
38
40
42
44
46test23 RMSE
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 1ºC
0ºC
T10m - RMSE
(From the experiments with model data)
1960 1965 1970 1975 1980 1985 1990 1995 2000
500
1000Number of observations per month TEST23
1960 1965 1970 1975 1980 1985 1990 1995 20000
0.5
1
ºC
RMSE of anomalies TEST23
Number of observations per month
Averaged RMSE per month
Quality evolves in time (seasonally)
Better results in winter (less variability, easier to capture even with less observations)
Uncertainty of the new product
Different from Formal Error!!!
-5 0 5 10 15 20 25 30 3530
32
34
36
38
40
42
44
46test11 RMSE
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0ºC
1ºC
-5 0 5 10 15 20 25 30 3530
32
34
36
38
40
42
44
46test11 Correlation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
1
-5 0 5 10 15 20 25 30 3530
32
34
36
38
40
42
44
46test15 RMSE
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 psu
1psu
-5 0 5 10 15 20 25 30 3530
32
34
36
38
40
42
44
46test15 Correlation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
1
RMSE Correlation
T100m
S 10m
When the characteristic spatial scales of the dominating signal are short the uncertainties increase
(From the experiments with model data)
Uncertainty of the new product
Averaging for different layers and subbasins may increase the reliability of the product
The uncertainty range must be compared to the range of the variability
Uncertainty of the new product
Averaging for different layers and subbasins may increase the reliability of the product
The uncertainty range must be compared to the range of the variability
Uncertainty of the new product
Averaging for different layers and subbasins may increase the reliability of the product
The uncertainty range must be compared to the range of the variability
Uncertainty of the new product
The uncertainty range must be compared to the range of the variability
Averaging for different layers and subbasins may increase the reliability of the product
Uncertainty of the new product
The uncertainty range must be compared to the range of the variability
Averaging for different layers and subbasins may increase the reliability of the product
Uncertainty of the new product
The uncertainty range must be compared to the range of the variability
Averaging for different layers and subbasins may increase the reliability of the product
Uncertainty of the new product
Sources of uncertainty -Lack of data -Background (in particular biases related to decadal variability) -Errors in the raw profiles (those which are beyond instrumental uncertainty; not considered here )
Impact of the XBT-MBT correction
We produce three versions of the product using : -No correction for the XBT/MBT -The Hanawa Correction -Cowley et al., 2013 The MBT data was used until ~1965 The XBT data from then to 1985. After that it is still used although in a relatively small fraction
Impact of the XBT-MBT correction
3D AVERAGE
Impact of the XBT-MBT correction
3D AVERAGE 0-150 m
150 – 600m 600m - bottom
- New hydrographic historical database for the Mediterranean has been generated for the period 1950-2013
- Realistic estimate of uncertainties provided for the maps and averaged quantities
- Non-negligible impact of XBT corrections during the 70’s . However, uncertainty related to sparseness of data is larger …
- Freely available in few weeks upon request
- However, uncertainties are large, specially for salinity and intermediate/Deep layers. Please, use the uncertainty estimates!
Conclusions