Impacts of climate change on water resources: A unified approach to simulate hydrologic processes

Transcription

1 Impacts of climate change on water resources: A unified approach to simulate hydrologic processes Martyn Clark, Bart Nijssen, David Rupp, Andy Newman, Dmitri Kavetski, Jessica Lundquist, Nic Wayand, Ethan Gutmann, Pablo Mendoza, David Gochis, Roy Rasmussen, Drew Slater, Chris Landry, Andy Wood, Pedro Restrepo, Levi Brekke, Jeff Arnold, David Tarboton, Vinod Mahat, Danny Marks, Adam Winstral, Pascal Storck, Jim Freer, and Ross Woods

2 Outline Motivation Incomplete characterization of uncertainty in climate impact assessments Improve hydrologic modeling Integrate different approaches to simulate hydrologic processes Preferential selection of modeling alternatives (don t reject entire models, just model components) Directly characterize main sources of uncertainty (understand the interplay between model parameters and process parameterizatrions) Summary & Discussion

3 Methodology to Incorporate Climate Change Information into Water Supply Projections Emissions Scenarios 3+ Scenarios Climate Simulations 16+ GCMs Spatial/Temporal Downscaling 100+ Projections Hydrologic Model Planning Model 100+ Traces

4 Expectations for the future Current approach New approach Accounting for hitherto neglected sources of uncertainty will invariably mean impact assessments will portray increased uncertainty same as the IPCC experience 4

5 Impact of downscaling methodology and hydrologic model on the portrayal of climate change impacts CLM VIC 5

6 Inter-model differences in ET and runoff CLM compared to VIC Comparison of annual water fluxes [mm] 12km CONUS wide VIC produces less ET where > 500mm VIC produces more runoff where <500mm VIC and CLM agree better when they are forced by BCCA than by the other products 3/6/2014 Color code: log10(n)

7 Comparison of extreme runoff Inter-model difference M02 BCCA BCSDd BCSDm AR High flow (Q20yr) [mm/day] CLM VIC Low flow (7Q10) [mm/day] CLM VIC Low flow estimate is more dependent on models

8 Ongoing research: Move from single-model to multi-model approaches Continental-scale application of existing 1-d hydrologic and landsurface models Models applied on either Hydrologic Response Units (HRUs) or grids Routing using the USGS river network topology from the Geospatial Fabric Simulate streamflow at all USGS stream segments Simple time-delay routing models (like used in VIC) Lagrangian kinematic wave routing model (more) Use of default model parameters CLM VIC + PRMS Noah-MP

9 CLM simulations coupled with network-based routing model configured for the USGS geospatial fabric

10 Problems Sub-optimal model fidelity Some models have poor representation of specific processes Ad-hoc characterization of uncertainty Selection of models not constrained to characterize uncertainty in process representation

11 Outline Motivation Incomplete characterization of uncertainty in climate impact assessments Improve hydrologic modeling Integrate different approaches to simulate hydrologic processes Preferential selection of modeling alternatives (don t reject entire models, just model components) Directly characterize main sources of uncertainty (understand the interplay between model parameters and process parameterizatrions) Summary & Discussion

12 Outline Motivation Incomplete characterization of uncertainty in climate impact assessments Improve hydrologic modeling Integrate different approaches to simulate hydrologic processes Preferential selection of modeling alternatives (don t reject entire models, just model components) Directly characterize main sources of uncertainty (understand the interplay between model parameters and process parameterizatrions) Summary & Discussion

13 Understanding areas of model agreement q snow q rain Q lw Q sw T air e air p air u air Q lw Q lw Qsw Qsw Q l Q h canopy wind profile top h can θ ice q unload θ liq q drip Q lw T veg Q hc evap Q lc trans Q lc canopy air space T cas T 1 T 2 T 3 T 4 Q c q thru snow unsaturated soil Q lw θ ice,1 θ ice,2 θ ice,3 θ ice,4 θ liq,1 θ liq,2 θ liq,3 θ liq,4 Q sw q liq Q lg qsurf Q hg bot h can saturated soil Q c q liq q base T n θ ice,n θ liq,n q base bedrock q base

14 a unified approach to hydrologic modeling Mahrt Louis Obukhov Wetted area Capacity limited K-theory L-theory Atmospheric stability Canopy evaporation Linear above threshold Canopy interception Canopy turbulence Liquid drainage Melt drip Topographic drift factors Snow Unloading Linear reservoir Net energy fluxes Blowing snow model Canopy temperature Canopy storage Phase change Spatial variability Canopy radiation Snow temperature Beer s Law 2-stream broadband Solver Snow storage Instant outflow 2-stream vis+nir Supercooled liquid water Liquid water flow Soil temperature Gravity drainage Soil water content Soil water characteristics Aquifer storage Evapotranspiration Soil Stress function Surface runoff Ball-Berry Hydrology Thermodynamics State variables Physical processes XXX Model options Lateral flow Vertical redistribution Infiltration Rooting profile Conceptual aquifer Kinematic Boussinesq Multi-domain Richards Gravity drainage Green-Ampt TOPMODEL VIC Darcy Frozen ground Explicit overland flow

15 Model architecture (spatial variability and hydrologic connectivity) soil soil soil soil soil aquifer aquifer aquifer (e.g., Noah) (e.g., VIC) (e.g., PRMS) (e.g., DHSVM) 15

16 CLM concepts/code used Linkages to CLM development Hierarchal data structures Spatial variability and hydrologic connectivity Canopy radiation Two-stream shortwave canopy radiation parameterization Canopy longwave parameterization Stomatal resistance Ball-Berry Snow Subdivision and merging of snow layers Possible enhancements to CLM Canopy snow interception Hydrologic similarity concepts Lateral flow and hydrologic connectivity (more)

17 Outline Motivation Incomplete characterization of uncertainty in climate impact assessments Improve hydrologic modeling Integrate different approaches to simulate hydrologic processes Preferential selection of modeling alternatives (don t reject entire models, just model components) Directly characterize main sources of uncertainty (understand the interplay between model parameters and process parameterizatrions) Summary & Discussion

18 Preferential selection of modeling alternatives Canopy SW radiation parameterizations Simple application of Beer s Law does not explicitly account for the higher transmission of diffuse radiation at higher zenith angles

19 Preferential selection of modeling alternatives Transpiration Interplay between model parameters and model parameterizations Biophysical representations of transpiration necessary to represent diurnal variability

20 Preferential selection of modeling alternatives spatial variability and hydrologic connectivity 1-D Richards equation somewhat erratic Lumped baseflow parameterization produces ephemeral behavior Distributed (connected) baseflow provides a better representation of runoff

21 Outline Motivation Incomplete characterization of uncertainty in climate impact assessments Improve hydrologic modeling Integrate different approaches to simulate hydrologic processes Preferential selection of modeling alternatives (don t reject entire models, just model components) Directly characterize main sources of uncertainty (understand the interplay between model parameters and process parameterizatrions) Summary & Discussion

22 Characterization of uncertainty Interplay between model parameters and model structure

23 Characterization of uncertainty Absolutely no idea what is correct Different interception formulations Simulations of canopy interception (Umpqua) Again, model fidelity and characterization of uncertainty can be improved through parameter perturbations

24 Characterization of uncertainty The wrong results for the same reasons Different model parameterizations (top plots) do not account for local site characteristics, that is dust-on-snow in Senator Beck Model fidelity and characterization of uncertainty can be improved through parameter perturbations (bottom plots)

25 Outline Motivation Incomplete characterization of uncertainty in climate impact assessments Improve hydrologic modeling Integrate different approaches to simulate hydrologic processes Preferential selection of modeling alternatives (don t reject entire models, just model components) Directly characterize main sources of uncertainty (understand the interplay between model parameters and process parameterizatrions) Summary & Discussion

26 Objectives Better representation of observed processes More precise representation of model uncertainty Summary Approach Detailed evaluation of different modeling approaches Outcomes o Recognizes that different models are based on the same set of governing equations o Defines a master modeling template, to reconstruct existing modeling approaches and derive new modeling methodologies o Provides a controlled approach to model development and evaluation Provided guidance for future model development Improved understanding of the impact of different model development decisions typology of model structural adequacy Improved operational applicability of process-based hydrologic models

27 Component-level model integration can improve hydrologic model simulations Improve model fidelity: Identify preferable modeling approaches (defines the dream model ) Numerical methods Prognostic canopy air space Coupled hydrology and thermodynamics Numerical error control and adaptive sub-stepping Separation of governing equations from their numerical solution Flexible hierarchal data structures Physical representations Variably saturated flow Below-canopy wind profiles Two-stream radiative transfer models Biophysical representation of transpiration Hydrologic similarity concepts to represent sub-grid variability Explicit representation of lateral flow Better characterize model uncertainty Characterize uncertainty in model parameters Represent ambiguity in process parameterizations, where necessary (interception example) Include residual uncertainty, to account for situations where all models are wrong for the same reasons (snow albedo example)