26-10-2010
The way forward: Better models, more data, better techniques?
There is either something highly productive in the number of aquatic ecosystem models that exist or…
A case study to demonstrate current limitations
there is something peculiarly unproductive about the effort going into multiple models around the world…
David Hamilton Waikato University, New Zealand Workshop on Lake Ecosystem Modelling Silkeborg, Denmark, 20-22 September 2010
A small subset of the models available: http://stommel.tamu.edu/~baum/ocean_models.html ACADIA ACOM BatTri BOM BRIOS DROG3D CLIO COHERENS DieCAST ECBILT ECOM-si ELCIRC FLAME FMS FRAM FUNDY GMODEL
There is a diversity of models also
Lake models:
Ocean models: GOTM HIM HOPE HYCOM LOAM LSM MICOM MITgcm MOM MOMA NCOM NLOM NUBBLE OCCAM OCCOMM OPA OSMOM
PEQMOD POCM POM POP POSEIDON POSUM QTCM QUODDY ROMS SCRUM SEA SELFE SEOM SPEM TOMS
Mooji et al. (2010): VOLLENWEIDER DYRESM-CAEDYM (1-DV) ELCOM-CAEDYM (3-D) CE-QUAL-W2 DELFT3D-ECO MYLAKE PCLake, SHIRA IPH-TRIM3D-PCLAKE PROTECH SALMO CHARISMA PISCATOR …and DHI DLM DYRESM-WQ BATHTUB MINLAKE
So how are our friends in the climate modelling community doing?
but despite the enormous funding and opportunity there are only c. 13 models
1
26-10-2010
Timelines for water quality model applications years before present 103
102
101
100
10-1
10-2
Present
There is also quality control and evaluation
years into future 10-2
10-1
100
101
102
} } } } } } Hindcasting - past climate -past land-use
Hindcasting recent human disturbance: - eutrophication, - invasive species, - pollutant fate
Hindcasting for detailed temporal understanding: - material fluxes - transport - testing theory
Forecasting: Simulation: Prediction: - response to - lake - future climate weather forecastmanagement -future land use - algal bloom strategies likelihood/fate - extreme events (weather, storms)
- Hambright et al. 2004
- Hipsey et al. (2009) - Arhonditsis and Brett (2004)
- Robson and - Recknagel et al. Hamilton (2003, 2004) (2007) - Burger et al. (2007) - Wallace et al. (2000)
- Elliot et al. (2009a) - De Stasio et al. (1996) - Spillman et al. -Trolle et al. (2010) (2009) - Elliot et al. (2009b) - Hamilton et al. (1999)
…and vigorous scrutiny of these models!
An example of simulations using high frequency: DYRESM temperature simulations of Trout Bog, WI, USA years BP 10-1
}
}
Hindcasting recent human disturbance:
Hindcasting for detailed temporal understanding C Burger et al. (2007)
Outcome: the modeller is happy to have something that reproduces reality but little contribution to model improvement
30
10-2
25
Temperature (ºC)
100
Present
years BP 101
Present
An example of mid-duration simulations: DYRESM temperature simulations of Trout Bog, WI, USA
20
15
T0
T1
T2
T3
T4
T5
M0
M1
M2
M3
M4
M5
O0
O1
O2
O3
O4
O5
10
5
0 2006-08-22 2006-09-01 2006-09-11 2006-09-21 2006-10-01 2006-10-11 2006-10-21 2006-10-31
Outcome: identifying separation of model simulation data from measurements (quantitative values, frequencies) allows opportunities to improve model performance: a focus on process representations
2
26-10-2010
The current situation: data frequency, model time steps and outputs Minutes
Computational time-step
The future situation: data frequency, model time steps and outputs Minutes
Inflow/outflow temperature and conductivity
Hours
Hours
Meteorology
Days
Computational time-step
Inflow nutrients and lake water quality measurements
Inflow/outflow temperature and conductivity
Meteorology
Days Model time-span (3-D)
Weeks
Weeks
Inflow nutrients and lake water quality measurements
Model time-span (1-D)
Months
Model time-span (1-D)
Months
Years
Years
Decades
Decades
Space–time plot: Traditional monitoring and current sensor networks
Model time-span (3-D)
Spatial validation: Biofish measurements v ELCOM-CAEDYM simulation dissolved oxygen
temperature
100 km 27/09/2004
10 km
Spatial 1 km extent (horizontal)
existing sensor networks
Currently available towed/ 100 m autonomous instruments 10 m Fixed point sensors Traditional monthly profiles 1m
11/01/2005
10 cm Annual
Monthly
Weekly
Daily Hourly
Min. Sec.
random selection from ecology (2003)
Temperature (°C) DO (mg L-1) Chlorophyll a (μg L-1)
Frequency of measurement 05/04/2005
RMSE 0.895
R 0.984
n 5250
1.211
0.877
5250
5.031
0.664
5250
Data from Von Westernhagen
3
26-10-2010
Space–time plot: Traditional monitoring and current sensor networks
Landsat-derived chlorophyll a for Rotorua lakes
100 km
10 km
Spatial 1 km extent (horizontal)
Remote sensing
existing sensor networks
100 m 10 m
Allan et al. (2010). Int. J. Remote Sensing
Fixed point sensors Traditional monthly profiles
1m 10 cm Annual
Monthly
Weekly
Daily Hourly
Min. Sec.
random selection from ecology (2003)
Hamilton et al. (2010). Aquat. Sci.
Frequency of measurement Strong vertical and horizontal gradients necessitate the application of highly spatially resolved models
Landsat-derived and ELCOM-derived temperature for
Percentage of current Ohau Channel inflow versus cyanobacteria (as μg chl-a L-1) in Lake Rotoiti
Rotorua lakes and Lake Rotoehu Landsat T
ELCOM T ELCOM U dd 14 Model 0% Ohau
chlorophyll a [ug/L]
12
Model 5% Ohau
10
Model 10% Ohau Model 50% Ohau
8
Model 100% Ohau
6 4 2 0 Jul-01
Geothermal inflows provide a hightemperature end-member to assist remove sensing of temperature
Jan-02
Jul-02
Jan-03
Jul-03
Jan-04
Jul-04
Jan-05
Jul-05
Jan-06
But calibrated parameters used in the 1-D model did not appear to be directly applicable to the 3-D model; the latter model did not capture the effects of 0% Ohau (the inflow diversion) in a spatially realistic way without specific calibration separate to 1-D model Data from Mat Allan
4
26-10-2010
Potential to apply models to major lake ecosystem perturbations: Inflow diversion, Lake Rotoiti
Potential to apply models to major lake ecosystem perturbations: Modified zeolite application to Lake Okaro
Landsat image
Photo: Environment BOP
Thoughts on critical needs • How best to harness the collective expertise and motivation of the lake modelling community? ; • Can we engage ecologists using a modularised system for individual ecological components (benthos, sediment-water, macrophytes etc.)? ; • Should we be scrutinising and applying standards to the models that we use? ; • Investigate scale and time dependence of parameters to ensure they are applicable across field, lab and model applications? ; • Find major perturbations (biomanipulations, flocculents, inflow diversions, etc.) to robustly test model validity.
5