Advances in land surface and hydrologic modeling · 1. Most hydrologic modelers share a common...
Transcript of Advances in land surface and hydrologic modeling · 1. Most hydrologic modelers share a common...
Advances in land surface and hydrologic modeling
“Modeling Change in Cold Regions” workshopSaskatoon, Canada, 28 September 201 5
Martyn Clark, Naoki Mizukami, Andy Newman, Ethan Gutmann, Andy Wood, Pablo Mendoza (NCAR)
Bart Nijssen(UW)
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
Motivation: What are the key issues that constrain progress in model development?
• Unsatisfactory process representation▫ Missing processes (e.g., spatial heterogeneity, groundwater)▫ Dated/simplistic representation of some processes
• Limited capabilities to isolate and evaluate competing model hypotheses▫ The failure of MIPs and the need for a controlled approach to model evaluation
• Insufficient recognition of the interplay between different modeling decisions▫ The interplay between model parameters and process parameterizations▫ Interactions among different model components
• Inadequate attention to model implementation▫ Impact of operator-splitting approximations in complex models▫ Bad behavior of conceptual hydrology models
• Ignorance of uncertainty in models and data▫ To what extent does data uncertainty constrain our capabilities to effectively
discriminate among competing modeling approaches?▫ Are we so “over-confident” in some parts of our model that we may reject
modeling advances in another part of the model?
Model constraints?
Model constraints?
Hard coded parametersare the most sensitive ones
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
Improving model physics
• It is difficult to adequately represent model uncertainty using multi-model and multi-physics approaches▫ Wrong results for the same reasons▫ A small collection of model provides poor coverage of the hypothesis space
• It is difficult to understand the importance of individual sources of model uncertainty through the analysis of total (integrated) model errors▫ Model-observation differences emerge through complex compensations
among different sources of model error▫ Therefore…it is very difficult to attribute model-observation differences to
individual model components
• A controlled and systematic approach to model development is needed to understand how individual sources of model uncertainty affect integrated model predictions
Modeling approach
Propositions:1. Most hydrologic modelers share a common understanding
of how the dominant fluxes of water and energy affect the time evolution of thermodynamic and hydrologic states
▫ The collective understanding of the connectivity of state variables and fluxes allows us to formulate general governing model equations in different sub-domains
▫ The governing equations are scale-invariant
2. Differences among models relate toa) the spatial discretization of the model domain;b) the approaches used to parameterize individual
fluxes (including model parameter values); andc) the methods used to solve the governing model
equations.General schematic of the terrestrial water cycle, showing dominant fluxes of water and energy
Given these propositions, it is possible to develop a unifying model frameworkFor example, by defining a single set of governing equations, with the capability to use different spatial discretizations (e.g., multi-scale grids, HRUs; connected or disconnected), different flux parameterizations and model parameters, and different time stepping schemes
Clark et al. (WRR 2011); Clark et al. (2015a; 2015b)
soil soil
aquiferaquifer
soilsoil
aquifer
soil
b) Column organization
a) GRUs and HRUs
The unified approach to hydrologic modeling
Governing equations
Hydrology
Thermodynamics
Physical processes
XXX Model options
Evapo-transpiration
Infiltration
Surface runoff
SolverCanopy storage
Aquifer storage
Snow temperature
Snow Unloading
Canopy interception
Canopy evaporation
Water table (TOPMODEL)
Xinanjiang (VIC)
Rooting profile
Green-AmptDarcy
Frozen ground
Richards’Gravity drainage
Multi-domain
Boussinesq
Conceptual aquifer
Instant outflow
Gravity drainage
Capacity limited
Wetted area
Soil water characteristics
Explicit overland flow
Atmospheric stability
Canopy radiation
Net energy fluxes
Beer’s Law
2-stream vis+nir
2-stream broadband
Kinematic
Liquid drainage
Linear above threshold
Soil Stress function Ball-Berry
Snow drifting
LouisObukhov
Melt drip Linear reservoir
Topographic drift factors
Blowing snowmodel
Snowstorage
Soil water content
Canopy temperature
Soil temperature
Phase change
Horizontal redistribution
Water flow through snow
Canopy turbulence
Supercooledliquid water
K-theoryL-theory
Vertical redistribution
Example simulations
Martyn Clark (left) and Chris Landry (right) at the upper site tower in the Senator Beck basin
The sheltered site at Reynolds Mountain East
Reynolds Creek, Idaho, USA
Senator Beck, Colorado, USA
Reynolds Creek
Senator Beck
Reynolds Creek
Senator Beck
Example application: Simulations of snow in open clearings
• Different model parameterizations do not account for local site characteristics (dust-on-snow in Senator Beck)
• Model fidelity and characterization of uncertainty can be improved through parameter perturbations
Example application: Simulations of snow in open clearings
Example application: Interception of snow on the veg canopy
• Can reproduce observations but rather uncertain about temperature sensitivity
Different interception formulations
Example application: Interception of snow on the veg canopy
• Can reproduce observations but rather uncertain about temperature sensitivity
Different interception formulationsSimulations of canopy interception (Umpqua)
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
17
A- priori (base) parameter estimations + basin-wide calibrations (i.e. basin-wide multipliers)(e.g., SAC/SNOW17 – NWS River Forecast, VIC- USBR- climate change assessment)
Current parameter estimations in large scale modeling
Issues Discontinuity in spatial distributions
of model parameters.
For example, the nationwide VIC simulations used for water security assessments based on a single model with spatially inconsistent parameter estimates
Goal and approach Improve continental parameter
estimates for multiple hydrologic models.
Initial work: Parameter estimation for the Upper Colorado River basin using streamflow from headwater basins
Transferability of calibrated parameters to the other basins?
Not counting for sub-grid variability of soil properties.
18
MPR approach
Samaniego et al. 2010, WRR
Traditional approach
The challenge is regionalization
Continental-scale parameter estimation19
Soil Data (e.g., STATSGO space)
βi
Adjust TF coefficients
Model Layers (e.g., 3 Layers)
��𝑃𝑃𝑖𝑖
Horisontal upscaling
Model Params(e.g., 3 Layers)
𝑃𝑃𝑖𝑖
Vertical upscaling
Model Params(e.g., STATSGO space)
Pi
(Pedo-) transfer function
simulations
Individual basins; donor catchments20
A priori parameter NLDAS Calibrated parameters single basin
Max Soil Moisture Storage in bottom layer
Calibrated parameters – region
Nearest Neighbor
Calibrated multiplier• Cρbulk(basini)• Cd1(basini)• Cd2(basini)• Cztot(basini)i = 1,…15
Individual basins; donor catchments21
A priori parameter NLDAS Calibrated parameters single basin
Max Soil Moisture Storage in bottom layer
Calibrated parameters – region
Nearest Neighbor
Calibrated multiplier• Cρbulk(basini)• Cd1(basini)• Cd2(basini)• Cztot(basini)i = 1,…15
Estimation with default TF coefficients Estimation with calibrated TF coefficients
Max Soil Moisture Storage in bottom layer
Calibrated TF coef. • aρbulk• ad1• ad2• aztot
Transfer function calibration
Initial results are encouraging
• Parameter fields show greater connection with grid-scale variability in terrain or soil properties
• Results at larger and separate flow locations are on par with manually calibrated results
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
Uncertainties in model forcing data
• N-LDAS vs. Maurer▫ Gridded meteorological forcing fields (12-
km grid) across the CONUS, 1979-present
• Opportunities to improve these products▫ Make more extensive use of data from
stations (additional networks) and NWP models (finer spatial resolution) in a formal data fusion framework
▫ Provide quantitative estimates of data uncertainty (ensemble forcing)
CLM simulations over the Upper Colorado River basin for three elevation bands, using two different meteorological forcing datasetsMizukami et al. (JHM, 2014)
Step 1:Estimate probability of precipitation (PoP) via logistic regression, amount and uncertainty at each grid cell (locally weighted regression & residuals)
observations
Example over the Colorado HeadwatersEnsemble Generation
Clark & Slater (2006), Newman et al. (2014, in prep)
Step 2:Synthesize ensemblesfrom PoP, amount & error using spatially correlated random fields
Example over the Colorado Headwaters
observations
Clark & Slater (2006), Newman et al. (2015)
Ensemble Generation
Step 2:Synthesize ensemblesfrom PoP, amount & error using spatially correlated random fields
Example over the Colorado Headwaters
observations
Clark & Slater (2006), Newman et al. (2015)
Ensemble Generation
• 12,000+ stations with serially complete data• Precipitation, temperature or both
CONUS application
Example Output
• Central US Flood of 1993• June 1993 total precipitation
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Continental-scale model benchmarks▫ Data, information, knowledge and wisdom: Can complex process-based
models make adequate use of the data on meteorology, vegetation, soils and topography?
▫ Use of simple models (statistical, bucket) as benchmarks
Routing
CLM simulations coupled with network-based routing model configured for the USGS geospatial fabric
Monthly streamflow over CONUS
cesm4, rcp8.5
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
“Revealing” uncertainties
GCM initial conditions
EmissionsScenario(s)
Global ClimateModel(s)
Downscalingmethod (s)
HydrologicModel
Structure(s)
HydrologicModel
Parameter(s)
scenarios
ens. members
models
Combined uncertainty
projections
35 35
methodsmodels
calibration
GCM initial conditions
EmissionsScenario(s)
Global ClimateModel(s)
Downscalingmethod (s)
HydrologicModel
Structure(s)
HydrologicModel
Parameter(s)methods
scenarios
ens. members
models
models
calibration
Combined uncertainty
projections
36 36
“Revealing” uncertainties
Climate model uncertaintyThe role of internal variability
Change in air temperature
Climate model uncertaintyThe role of internal variability
Change in air temperature
Change in precipitation
Basins of interest for this study
The Colorado Headwaters Region offers a major renewable water supply in the southwestern United States, with approximately 85 % of the streamflow coming from snowmelt. Hence, we conduct this research over three basins located in this area:
- Yampa at Steamboat Springs
- East at Almont
- Animas at Durango
39
Exploring parameter effects at the watershed scaleM
Mendoza et al., JHM 2015
Role of hydrologic model choice and calibration?
Uncalibrated model simulations Calibrated model simulations
Uncalibrated models: Hydrologic change signal in Noah (↑ET and ↑Runoff) differs from the rest of models (↑ET and ↓Runoff).
After calibration, signal direction from Noah-LSM switches to ↑ET and ↓Runoff. Inter-model agreement does not necessarily improve in terms of magnitude and
direction.
Looking at the hydrologic ‘change signals’
40Mendoza et al., JHM 2015
Climate downscaling uncertaintyChoice of statistical downscaling methods
• Widely used and newer methods▫ BCSD-monthly Wood et al (2004)▫ BCSD-daily Thrasher et al (2012)▫ BCCA Hidalgo et al (2008)▫ Asynchronous Regression Stoner et al (2013)
• All simply rescale or shift GCM outputs
Test Downscaling NCEP/NCAR Reanalysis
AR
BCSD
BCCAGutmann et al., WRR 2014
Statistical downscaling: Wet day fraction
Gutmann et al., WRR 2014
Statistical downscaling: Wet day fraction
Gutmann et al., WRR 2014
Impact on Annual water balanceStatistical downscaling methods and hydrologic models
Miz
ukam
i et a
l., J
HM
201
5
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
Are US operational forecasts improving?46
Since 1997, there have been notable advances in capabilities supporting hydrologic prediction. Are we harnessing those advances?
http://www.srh.noaa.gov/abrfc/fcstver/
1997 2012
RM
SE
‘End Member’ Forecast System Approaches
Operational, 1980s => • Provide tailored, but limited
(mostly deterministic) forecasts that are inputs to water management
• Simple conceptual models that can be adjusted manually, in real-time
• Heavy use of model calibration• Reliance on human expertise at the
model/data system level.• Non-reproducible, non-scalable
forecast process• Run on small number of
workstations
Research/experimental 2000s =>• Provide non-tailored, centrally
produced forecasts, typically in form of percentile or frequency analyses
• ‘Physically-based’ high-dimensional models
• Little model calibration• Automated forecast process,
reproducible• Ensemble outputs – require
supercomputing
grid2grid
End Member Forecast System Philosophies
Operational, 1980s => • “The models, data and systems will
always be inadequate, thus human expertise is needed to fix performance on the fly”
• “If the decisions using your model outputs require a certain answer, the models must be simple so that they can be adjusted to provide that answer”
Research/experimental 2000s =>• “The superior physics in new models
and datasets will yield good quality results”
• “Any problems can be fixed with higher resolution and more detailed physical process representation”
• “Decision makers can use risk/hazard levels rather even if flows are poor”
grid2grid
Real-time demonstration and evaluation project
• In water management case study basins, demonstrate and evaluate experimental, automated days-to-seasons flow forecasts using:• various real-time forcing generation approaches• ensemble meteorological forecasts and downscaling techniques• variations in model physics and architecture• automated, objective model calibration• data assimilation• flow forecast post-processing• hindcasting and verification
• Partner with water agencyfield office personnel -- for evaluation-- guide product development-- develop decision calendars
Outline
• Motivation▫ Improve operational applicability of process-based models, while
accounting for model/data uncertainty▫ Improve information content in probabilistic forecasts
• Model development▫ Model architecture and process parameterizations▫ Continental-scale parameter estimation▫ Ensemble forcing▫ Routing
• Applications▫ Climate change▫ Streamflow forecasting▫ Continental-domain benchmarks
Simple models as benchmarks
• The NERD approach (statistical models as benchmarks)• Bucket-style models as a statistical model
Can more complex models extract the same information content from the available data on meteorology, vegetation, soils and topography?
If not, why not?
What work do we need to do in order to ensure that physically realistic models perform better than models with inadequate process representations?
Newman et al., HESS (2015)
CUAHSI-NCAR collaboration
• CUAHSI (Consortium of Universities for the Advancement of Hydrologic Science, Inc.) supports/enables community activities to advance hydrologic science
• New CUAHSI / NSF initiative to improve representation of hydrologic processes in ESMs▫ Accelerate implementation of
state-of-art hydrologic understanding into large-scale land models
▫ Emphasis on model evaluation / benchmarking utilizing catchment-scale observations
▫ Initial focus on Community Land Model / CESM
Winter et al., 1998
Summary
• Hydrologic model development progressing on multiple fronts▫ Model structural inadequacies (spatial architecture, parameterizations)▫ Continental-domain parameter estimation▫ Ensemble forcing▫ Routing
• Applications in climate change and streamflow forecasting▫ Reveal and reduce uncertainties in climate impact assessments▫ Real-time demonstration of advanced streamflow forecasting
technologies
• New projects▫ CUAHSI-NCAR initiative to improve hydrology in ESMs▫ Evaluate uncertainties in process-based hydrologic models▫ New modeling in Hawaii and Alaska
Publications resulting from NCAR efforts• Clark, M. P., B. Nijssen, J. Lundquist, D. Kavetski, D. E. Rupp, E. Gutmann, A. Wood, L. Brekke, J. R.
Arnold, D. Gochis, and R. Rasmussen (2015a), A unified approach to hydrologic modeling: Part 1. Modeling concept, Water Resources Research, 51, doi:10.1002/2015WR017198
• Clark, M. P., B. Nijssen, J. Lundquist, D. Kavetski, D. E. Rupp, E. Gutmann, A. Wood, D. Gochis, R. Rasmussen, D. Tarboton, V. Mahat, G. Flerchinger, and D. Marks (2015b), A unified approach to hydrologic modeling: Part 2. Model implementation and case studies, Water Resources Research, 51, doi:10.1002/2015WR017200
• Elsner, M. M., S. Gangopadhyay, T. Pruitt, L. Brekke, N. Mizukami, and M. Clark (2014), How does the Choice of Distributed Meteorological Data Affect Hydrologic Model Calibration and Streamflow Simulations?, Journal of Hydrometeorology(2014), 1384–1403, doi: 10.1175/JHM-D-13-083.1.
• Gutmann, E., R. M. Rasmussen, C. Liu, K. Ikeda, D. J. Gochis, M. P. Clark, J. Dudhia, and G. Thompson (2012), A comparison of statistical and dynamical downscaling of winter precipitation over complex terrain, Journal of Climate, 25(1), 262-281.
• Gutmann, E., T. Pruitt, M. P. Clark, L. Brekke, J. R. Arnold, D. A. Raff, and R. M. Rasmussen (2014), An intercomparison of statistical downscaling methods used for water resource assessments in the United States, Water Resources Research, 50(9), 7167-7186
• Mendoza, P. A., B. Rajagopalan, M. P. Clark, K. Ikeda, and R. Rasmussen (2014a), Statistical post-processing of High-Resolution Regional Climate Model Output, Monthly Weather Review, in press.
• Mendoza, P. A., B. Rajagopalan, M. P. Clark, G. Cortés, and J. McPhee (2014b), A robust multimodelframework for ensemble seasonal hydroclimatic forecasts, Water Resources Research, 50(7), 6030-6052.
• Mendoza, P. A., M. Clark, M. Barlage, B. Rajagopalan, L. Samaniego, G. Abramowitz, and H. Gupta (2014c), Are we unnecessarily constraining the agility of complex process-based models?, Water Resources Research doi: 10.1029/2014WR015820.
• Mendoza, P. A., M. Clark, N. Mizukami, A. Newman, M. Barlage, E. Gutmann, R. Rasmussen, B. Rajagopalan, L. Brekke, and J. Arnold (2015), Effects of hydrologic model choice and calibration on the portrayal of climate change impacts, Journal of Hydrometeorology, in press
Publications resulting from NCAR efforts (cont.)• Mizukami, N., M. P. Clark, A. G. Slater, L. D. Brekke, M. M. Elsner, J. R. Arnold, and S. Gangopadhyay
(2014), Hydrologic implications of different large-scale meteorological model forcing datasets in mountainous regions, Journal of Hydrometeorology, 15(1), 474-488.
• Mizukami, N., M. Clark, E. Gutmann, P. A. Mendoza, A. Newman, B. Nijssen, B. Livneh, J. R. Arnold, L. Brekke, and L. Hay (2015), Implications of the methodological choices for hydrologic portrayals over the Contiguous United States: statistically downscaled forcing data and hydrologic models, Journal of Hydrometeorology (accepted pending revisions)
• Newman, A., M. Clark, A. Winstral, D. Marks, and M. Seyfried (2014a), The use of similarity concepts to represent sub-grid variability in hydrologic and land-surface models: case study in a snowmelt dominated watershed, Journal of Hydrometeorology, 15, 1717–1738, doi: 10.1175/JHM-D-13-038.1.
• Newman, A., M. Clark, K. Sampson, A. Wood, L. Hay, A. Bock, R. Viger, D. Blodgett, L. Brekke, and J. Arnold (2014b), Development of a large-sample watershed-scale hydrometeorological dataset for the contiguous USA: dataset characteristics and assessment of regional variability in hydrologic model performance, Hydrology and Earth System Sciences Discussions, 11(5), 5599-5631.
• Newman, A., M. P. Clark, J. Craig, B. Nijssen, A. Wood, E. Gutmann, N. Mizukami, L. Brekke, and J. R. Arnold (2015), Gridded ensemble precipitation and temperature estimates for the contiguous United States, Journal of Hydrometeorology (accepted pending revisions)
• Prein, A. F., G. J. Holland, R. M. Rasmussen, J. Done, K. Ikeda, M. P. Clark, and C. H. Liu (2013), Importance of regional climate model grid spacing for the simulation of heavy precipitation in the Colorado headwaters, Journal of Climate, 26(13), 4848-4857.
• Rasmussen, R., K. Ikeda, C. Liu, D. Gochis, M. Clark, A. Dai, E. Gutmann, J. Dudhia, F. Chen, and M. Barlage (2014), Climate change impacts on the water balance of the Colorado Headwaters: High-resolution regional climate model simulations, Journal of Hydrometeorology, 15, 1091–1116, doi: 10.1175/JHM-D-13-0118.1.
• Wood, A., T. Hopson, A. Newman, J. R. Arnold, L. Brekke, and M. Clark (2015), Using a variationalensemble streamflow prediction analysis to understand seasonal hydrologic predictability, Journal of Hydrometeorology (accepted pending revisions)