ECEM’07 The 6th European Conference on Ecological ModellingTrieste, 26-30 November 2007

ASSIMILATION OF WATER QUALITY DATA IN A 3D ECOSYSTEM MODEL OF THE LAGOON OF VENICE

Cossarini G.(1), Solidoro C. (1)

and Lermusiaux P.(2)

(1)Istituto Nazionale di Oceanografia e di Geofisica Sperimentale, Italy (2) Massachusetts Institute of Technology, USA

Aim: “Best estimate” of the nutrient and phyto fields using Data Assimilation for the lagoon of VeniceKnowledge of the state of the system is important for a sustainable management of the lagoon: water quality target, …

• Data assimilation consists of three parts: 1. a model TDM model2. a set of observations MELa1 monitoring3. a DA scheme ESSE - error subspace statistical

estimation

• Several tests on the model error formulation

• Results: insight about ecosystem dynamics and estimation of the nutrient and phyto fields

domain3D coupled model

( )..,,, ITCz

wt si FCKCC

iii +∇∇+

∂∂

=∂∂

Dejak, Pastres, Solidoro et al. (1987-present)

Water temperatureZoo Phyto

Oxy

P NSediment

C

C

Detritus

P N

NutrientsNH4

+

PO43-

NO3-

N2

AIRLight intensityLight intensity

Thermal exchange

Water temperatureZoo Phyto

Oxy

P NSediment

C

Air deposition

C

Detritus

P N

N, P

NutrientsNH4

+

PO43-

NO3-

N2

7000 cells of 300x300x1 m3

TDM - Trophic Diffusive Model

Malamocco Inlet

Lido Inlet

ChioggiaInlet

ChioggiaCuoreMontalbano

Nov.Brenta

Lova

Nav.Brenta

Lusore

CampaltoOselino

SiloneDese

CavallinoMurano

Venezia

ADRIATICSEA

Nutrientloads

DRAINAGEBASIN

Exhange throughlagoon inlets

and boundaries

Main findings:

Model simulations show that:

- nitrogen cycle in the system is mainly driven by external input in the system

- exchanges with Adriatic Sea represents an important ‘purification’ process

- exists a spatial variability of productivity (northern basin more rich than southern one) due to presence of rivers

Mean annual nitrogen budget in the three sub-basins of the lagoon

TDM - Trophic Diffusive Model

M1

M2

C1

C2

C3C4C5

C6

C7

C8

B1 B2 B3

B4B5B6

B7

B8

B9B10 B11 B12

B13

B14B15

B16

B17

B18

B19

B20

MELa1-Monitoring of the Ecosystem of the Lagoon (1st example of monitoring of the whole lagoon from 2001 to 2003)

Network of 30 fixed sampling points:- 2 outside of the inlets- 8 in channel- 20 in shallow water area

12 Monthly sampling campaigns in 2001

Sponsored by Consorzio Venezia NuovaMagistrato alle Acque Venezia

Data for the year 2001 – MELa1

20 water quality parameters:-temp, salinity-DIN (NH4+NO3)-DIP-CHLa…

Main findings of MELa1:

-Strong Nutrient and Chlorophyll a gradients from the inner part towardlagoon inlets. And from northern basintoward southern one

-Trophic index from PrincipalComponent Analysis of water qualityparameters

- the index is fairly well explained by multiple regression with Salinity (a proxy of the fresh water discharge) and Residence Time (a proxy of exchanges through the inlets)

Data for the year 2001 – MELa1

The lagoon is strongly affectedby input and exchanges

kkk v+⋅= CHObs

.. using Error Subspace Statistical Estimation Scheme (Lermusiaux, 1999)

Data Assimilation scheme

( ) dttCFdC ⋅= ,

observation

model

NEW estimation of the state of the system C(+)

Filter scheme: ( ))()()( −⋅−⋅+−=+ CHObsKCC

where the gain matrix is ( ) 1−+⋅⋅⋅⋅= RHPHHPK TT

MERGING SCHEME

estimation of P forecast error covariance?

ESSE (Error Subspace Statistical Estimation) Scheme

Modified from Lermusiaux, 1999

Test on the formulation of:- model error- error of initial condition ICs- error on boundary condition BCs

Test on the formulation of:- model error- error of initial condition ICs- error on boundary condition BCs

iinletxi CC ε+==

iRCriverxi fC υ+⋅Φ==

( ) ii ddttCFdC η+⋅= ,

Testing several stochastic boundary settings:ensemble of BCs is set std of the forecast

BCinlet time evolutionfrom observations

BCriver from monthly estimation ( ΦC) and random function of daily rainfall (fR)

)varN(0,)varN(0,

M

B

∈∈

ηυε ,

var B

=0.3

var B

=0.6

ε, υ time correlatedη not spatial correl.

η, ε, υnot correlated

var M

=0.0

5va

r M=0

.05

ε, υ time correlatedη spatial correlated

Looking for misfit obs-mod(dots) <> model error forecast (maps)

best best choicechoice

Results of ESSE:

1. Eigenvectors of the covariance decomposition (most energetic modes in the system)

2. New estimates of nutrient and phyto fields

3. Evolution of the uncertainty of the estimation

1. Results of ESSE: First Eigenvectors(most energetic dynamics in the system)

1st e

igen

v

march april may juneDIN

~80% of uncertaintyis explainedby 5 first eigenvectors

2ndei

genv

3rdei

genv

external input fromnorthern rivers

other sources

central rivers & internal dynamicsin the northernmost part of lagoon

~99% by 20-25 Eigenvectors

2. Results of ESSE: NEW estimation of the FIELDSa priori field (and observation) a posteriori field (and observation)

DIN april

PHY july

2. Relevance of the results:WATER QUALITY TARGET FOR THE LAGOON (Italian legislation)

0 10 20 30 40 50 60 70 80 90 100

0 10 20 30 40 50 60 70 80 90 100

Percentage of time of the year during which the lagoonis above the limit for DIN WQT=0.35mg/l

Obs ESSE

%%

3. Relevance of the results:Comparison between a priori error and a posteriori error covariance can help in designing the network of sampling stations

Mean of the relative reduction of forecast error

Strong Strong reductionreduction in in northernnorthern partpart butbutsome some marginalmarginal areasareas (and (and southernsouthernbasinbasin) and ) and areasareas closeclose toto inletsinlets stillstillhavehave quitequite high high uncertaintyuncertainty

A priori error

A post error

july

mar

ch

DIN

Conclusions

Lagoon of Venice is a laboratory for testing DA methods (model & data available)

Data Assimilation via ESSE allows to get:- new estimations of nutrient & phytopl. fields- information on dynamics - information on the effectiveness of the monitoring network design

Many Thanks for your attention