The detailed results are in press on Annales Geophysicae

Near Real Time Sea Level Anomaly (SLA) and Sea Surface Temperature (SST) products during 2-years of MFS pilot project: processing, analysis of the variability and of the coupled patterns.

B. Buongiorno Nardelli1, G. Larnicol2, E. D’Acunzo1, R. Santoleri1, S. Marullo3, P.Y. Le Traon2
1
GOS
2CLS
3ENEA

Comparison with Delayed Time data/Accuracy assessment

The Near Real Time (NRT) operational products developed from satellite data (AVHRR, Topex/Poseidon, ERS-2) in the framework of the Mediterranean Forecasting System Pilot Project (MFSPP, autumn 1998-autumn 2000) have been compared to Delayed Time (DT) products over the Mediterranean sea in order to evaluate their accuracy.
Altimeter DT data used were distributed by AVISO (MGC-B, version 2, AVISO, 1996) for the M-GDRS and by the French Processing and Archiving Facility (PAF) CERSAT for the OPRs.
SST DT data were obtained from NASA Pathfinder global dataset (version 4.1, until December 31st 1999) at 9 km resolution, available at the Physical Oceanography Distributed Active Archive Center (PODAAC)
The NRT data were assimilated in MFSPP general circulation model, and consequently it was of great interest first of all to assess the accuracy of the NRT respect to delayed time data, and secondly to characterize the MFSPP years respect to the typical conditions observed in the Mediterranean sea during the previous periods. Actually, NRT data were affected by errors of different nature. In facts, both SLA and SST algorithms for NRT processing are obviously less accurate than delayed time ones. SST operational algorithms process data using constant calibration coefficients, while delayed time products like Pathfinder datasets are obtained from monthly calibrated coefficients. On the other hand, the orbit error contributes significantly to SLA error when estimated in NRT. Moreover, the lower spatial and temporal coverage in the NRT data is also an important source of error (more data are rejected because of the lower quality) and the space-time interpolation scheme (with a shorter time window respect to delayed algorithms) also produces additional errors in the weekly SLA maps. These errors are quantified in tables 1 and 2.

 

 
RMS T0-7
RMS T0-14
annual TPERS
3.76
3.30
TP
3.21
2.49
winter TPERS
4.00
3.43
TP
3.66
2.83
summer TPERS
3.36
3.17
TP
2.53
2.15

Table 1 : RMS of the differences between NRT and delayed time altimeter maps. The column with RMS at T0-7 days corresponds to the actual NRT MFSPP system. The column at T0-14 days stresses the problem of the shift in the observation period.

 
MBE
RMS
annual AVHRR
-0.30
0.93
winter AVHRR
0.38
0.55
summer AVHRR
-0.74
0.74

Table 2 : Mean bias error (MBE) and RMS of the differences between NRT and delayed time SST maps.
 
 

Analysis of the variability

Despite these differences observed between NRT and DT data, the MFSPP dataset described adequately the fundamental aspects of the Mediterranean circulation and allowed to characterize the MFSPP years in a wider temporal context. In fact, both surface height and temperature data coherently describe a surface circulation that is gradually returning to what was known in literature as the ‘classical’ picture for the Mediterranean, after some years characterized by a modified circulation involving the central part of the basin, with an intense anticyclonic circulation in the Ionian sea (1991-1997). Moreover, MFSPP years are clearly characterized by the presence of Ierapetra eddy.


SLA seasonal means deduced from combined maps of TP and ERS. Units are in cm. (a) Winter 1999, (b) Spring 1999, (c) Summer 1999, (d) Fall 1999


SLA seasonal means deduced from combined maps of TP and ERS. Units are in cm. (a) Winter 2000, (b) Spring 2000, (c) Summer 2000, (d) Fall 2000 Coupled Pattern Analysis

Finally, a methodology proposed by Leuliette and Wahr (1999) to investigate the coupling of SSH and SST has been tested on MFSPP data. This multi-variate method consists in the SVD of the covariance between SST and SSH. When this method was applied to 2 years of SST anomalies and SLA data without any additional processing of the maps, the first mode explained almost all the co-variability of the two fields (99% of the covariance), with a low spatial correlation (.20) and a temporal correlation of .57. Despite the temporal coefficients displayed a very clear seasonal signal for both data, it was thus not possible to give a coherent interpretation for these low-correlated spatial patterns. However, once a more refined handling of the SST and SLA maps has been decided, removing the spatial average from each map (‘gradient CPA’), the co-variability of the temperature and sea surface heights fields was distributed in more modes (74% of the covariance for the first mode, 20% for the second, 6% for the third...) containing both interannual (first mode) and seasonal (second mode) signals and characterized by clear and highly-correlated temporal and spatial patterns.
Actually, a longer time period would allow a better identification of varying signals. In this sense the two years of MFSPP satellite data are not the best dataset to perform an advanced analysis of the coupled patterns. However, applying this method to a limited area as the Mediterranean sea represented, on one hand, a test of the capabilities of this methodology on scales smaller than global scale. On the other hand, it represented the first step to the coupled analysis of longer time series of several datasets, which will be performed as soon as coherent datasets will become available (e.g. MFSPP follow on), and that should lead to a more deep analysis of ocean surface dynamics.

 

FIRST MODE

(a)

(b)

(c)

Spatial corr. =.55
Temporal corr.=.78
Explained cov.=.74

(d)

First coupled mode of SLA and SSTA data collected during MFSPP after removing the spatial averages over the basin (‘gradient CPA’). Patterns and associated temporal coefficients. (a) SLA pattern, (b) SLA temporal coefficient, (c) SST pattern, (d) SST temporal coefficient.

(a)

(b)

SLA (a) and SSTA (b) homogeneous correlation maps relative to the first ‘gradient CPA’ mode.

 

SECOND MODE

(a)

(b)

(c)

Spatial corr. =.40
Temporal corr.=.77
Explained cov.=.20

(d)

Second ‘gradient CPA’ mode of SLA and SSTA data collected during. Patterns and associated temporal coefficients. (a) SLA pattern, (b) SLA temporal coefficient, (c) SST pattern, (d) SST temporal coefficient.

(a)

(b)

SLA (a) and SSTA (b) homogeneous correlation maps relative to the second ‘gradient CPA’ mode.