Hood modeling talk

Slide1: 

Recent Advances and Future Challenges in Coupled Physical-Ecosystem Modeling (Implications for Water Quality Modeling and Management) Raleigh R. Hood University of Maryland Center for Environment Science

Slide2: 

Focus: Prognostic marine ecosystem models and coupled physical-biological systems. But this is not a review. It is my view which is strongly influenced by lessons learned in open ocean systems.

Slide3: 

Outline: 1. Some technical and scientific advances 2. Some major challenges and problems A Case Study: Using data assimilation to determine if additional model complexity increases model skill

Slide4: 

Outline: 1. Some technical and scientific advances 2. Some major challenges and problems A Case Study: Using data assimilation to determine if additional model complexity increases model skill

Slide5: 

Numerous Complexities are Now Routinely Considered in Models (not just NPZD anymore) Large and small size classes Carbon N limitation Si limitation P limitation Fe limitation Bacteria DOM Schematic from Leonard et al. 1999

Slide6: 

Representation of Key Functional Groups Microcystis Prorocentrum Coscinodiscus Oceanic Trichodesmium (N2-fixation) Coccolithophorids (Calcite production/ballast) Phaeocystis (DOC production) Estuarine/Coastal Cyanobacteria (freshwater blooms) Diatoms (spring production and export) Flagellates (summer production) E. huxleyi Trichodesmium Phaeocystis pouchetii

Slide7: 

Non-Redfield Elemental Stoichiometry Addressing need for variable ratios And cycling within and among state variables Simplified approaches (e.g., Fennel et al., 2002) -> Complicated approaches (e.g., Moore et al., 2002) Estuarine and water quality models too

Slide8: 

Benthic Biogeochemistry and the Effects of Low Oxygen In shallow, eutrophic systems like Chesapeake Bay biogeochemical processes in the benthos and in hypoxic/anoxic zones can have a profound influence on chemistry, nutrient concentrations and ecosystem dynamics. Proper consideration of low oxygen reduction/oxidation reactions and their effects is crucial and now feasible (but simplified approaches are needed for coupled models). from Codispoti et al., 2001

Slide9: 

Importance of Spatial Resolution In coastal plain estuaries it is crucial to resolve narrow topographic features to properly model the salinity distribution and circulation Increased resolution also allows smaller scales of circulation variability and realism ROMS Grid, courtesy of Ming Li

Slide10: 

Importance of High Resolution Forcing from McCreary et al. 2001 High resolution -> Realistic MLD variability Mixed blooms Complex responses Less predictable Dramatically different solutions compared to climatological forcing

Slide11: 

Data Assimilation Sequential (nudging) Variational (fitting) - initial conditions - boundary conditions - model parameters More exhaustive and objective searches of the parameter space Diagnose structural problems Compare models From Lawson et al 1996

Slide12: 

Outline: 1. Some technical and scientific advances 2. Some major challenges and problems A Case Study: Using data assimilation to determine if additional model complexity increases model skill

Slide13: 

Absence of a Theoretical Foundation No equivalent to Navier-Stokes Dizzying array of ecosystem/biogeochemical models Various structures, formulations and parameterizations No obvious hierarchy among models Where are we going?

Slide14: 

Grazing and Mortality Grazing and mortality terms can profoundly influence model responses and elemental cycles Little attention paid to effects of grazing formulations and mortality Current closure Schemes are ad hoc from Kemp et al. 2001

Slide15: 

Top Down Control Influence of top down control is poorly understood and rarely considered in any realistic way in biogeochemical models Effects of gelatinous species -Rates don’t appear to saturate -Evidence of trophic cascade Effects of fish / seasonal and migratory predators?

Slide16: 

Light Transmission in Estuaries There is a diverse suite of optically active constituents in estuarine waters which include phytoplankton pigments, CDOM, suspended organic matter and sediments. These constituents are highly variable and often have different source functions which are difficult to represent in models. Proper representation of light transmission variability is crucial for modeling primary production and nutrient cycling. In Chesapeake Bay, suspended sediments and particulate matter are very important optical constituents.

Slide17: 

Monospecific Events like HABs Although progess has been made, it is still not possible to model and predict the occurrence of monospecific events like harmful algal blooms. Pfiesteria piscicida From Zhang et al., 2003

Slide18: 

Bacteria and Cycling of DOM Bacteria are responsible for most of the respiration in estuaries and marine systems in general, yet they are usually not represented at all in water quality models. Do we need to model bacteria explicitly? Or are parameterizations of their effects sufficient? The DON, DOP and DOC pools are large in estuaries and their cycling and lability are poorly understood. To what Degree do we need to represent these chemical constituents in water quality models? Proper consideration of bacteria and DOM pools may be crucial. But…

Slide19: 

Grappling with Complex Systems: Parameterizing, understanding and diagnosing very complicated biogeochemical models is a major challenge Will complex models give us the predictive skill we need? (diminishing returns) Will development and use of these models lead to improved understanding? (too many degrees of freedom) Open Source / collaborative approaches may be crucial Data assimilation and objective methods will help We may need to draw from the experiences of the theoretical ecologist who have been grappling with these problems for many years: - Consider “biocomplexity” and emergent properties - Study of general ecosystem characteristics (ascendency, cycling, resilience)

Slide20: 

Outline: 1. Some technical and scientific advances 2. Some major challenges and problems A Case Study: Using data assimilation to determine if additional model complexity increases model skill

Slide21: 

Friedrichs, Hood and Wiggert (in prep.): Comparing models using data assimilation Goal: Quantitatively and objectively compare different biogeochemical model formulations to determine which models perform best over a wide range of conditions. Approach: Force biogeochemical models in exactly the same way using 1-D output from 3-D physical models and observations, then optimize and compare them objectively using data assimilation.

Slide22: 

Differences between three biological models 4-component (P, Z, N, D): (McCreary et al., 2001). 5-component (P, N, D, DON, H): (Hood et al., 2001). 8-component (2P, 2Z, 2D, A, N): (Christian et al., 2002).

Slide23: 

Differences in the physical forcing sets Murtugudde et al. 2001 20-layer 3-D model Weekly output (MLD and w) NCEP forcing McCreary et al. 2001 4.5 layer 3-D model Hourly output (MLD and w) Duirnal cycle FNMOC forcing + mooring Weller et al. 2002 Observed MLD variability Weekly averaged w from Murtugudde et al.

Slide24: 

Post-Assimilation Model Performance All three biological models gave similar results. Effect of different forcing sets greater than effect of different biological models. Increased model complexity leads to decreased predictive skill. Murtugudde et al. forcing-> McCreary et al. forcing-> Weller et al. forcing ->

Slide25: 

Conclusion: Relatively simple biogeochemical models capture the fundamental processes, but they are very sensitive to how physical models represent physical reality. Physical models often fail to properly represent key processes (e.g., mixing variability, diapycnal exchange, etc.) and therefore cause the biogeochemical models to fail. The upside: Biogeochemical models provide a powerful means of discerning subtle problems in physical models. The downside: Do physical simulations have to be “perfect” in order to model biogeochemical cycles and water quality? For simulation, maybe yes.

Slide26: 

Implications for Water Quality Modeling and Management Increases in complexity and resolution are allowing much more realistic simulations, more comprehensive water quality models, and representation of key biological and chemical Processes (e.g., benthic / low oxygen biogeochemical cycling). But where does it end? How much complexity is enough? First order variability is captured with relatively simple model systems. Super complex models are difficult to understand and diagnose, and may have reduced predictive skill. Data assimlation techniques will be crucial for tuning, comparing and diagnosing complex water quality model systems. And also for determining how much complexity is necessary and useful.

Slide27: 

Thank You !

Slide28: 

Priority Activities for Tom Gross and the CCMP: 1. Website development -Basic website in place served from NCBO -Develop model/data serving capability? (or will this be provided via SourceForge ?) 2. Getting existing models and data in place (serving/supporting) Watershed -CBP/HSPF -Other contributions Estuarine -Quoddy -Other contributions 3. Getting new models and data in place (serving/supporting) -ROMS community model (under development) -Other contributions? 4. Proposal development -NOPP -CICEET -CLEANER 5. Workshops and meetings -3rd Annual workshop -Special Sessions?

Slide29: 

Forcing Differences Murtugudde et al. 2001 SWM MLD too deep No eddy effects Too much entrainment gives strong SWM detrainment bloom McCreary et al. 2001 SWM MLD much better But still no eddy effects SWM bloom, but at wrong time Weller et al. 2002 Observed MLD variability Effect of eddy on MLD Lack of upwelling or advection (w from Murtugudde et al.)

Slide30: 

Physical vs. Biological Effects on Model Performance   J1 = Initial cost (no assimilation) J2 = Cost after assimilation Effect of different forcing sets greater than effect of different biological models. Murtugudde et al. forcing set gives best model performance, but it appears to do so for the wrong reasons. Both the Weller et al. and the McCreary et al. forcing fail because neither represent eddy effects properly.