Issues and Strategies for Integrated Model Calibration

1
Issues and Strategies
for
Integrated Model Calibration
IAH Waterloo 2015
Dirk Kassenaar, E.J. Wexler
P.J. Thompson, M.G.S. Takeda
Earthfx Inc.

2
Integrated Modeling
► Integrated modelling can provide
significant insights into the overall
system behavior and response to
complex stresses
► Integrated Modelling: Modelling
with dynamic feedback between
the SW and GW systems
► NOT integrated: Conjunctive or
loosely coupled models that are
manually coupled or explicitly
solved
After USGS

3
Presentation Objectives
► Is integrated modelling different?
 Technical Issues:
► Complex non-linear processes, compensating errors, long run times…
 Non-Technical Issues:
► Knowledge limitations, different conceptual models, biases, terminology…
► Strategies for addressing these issues:
 We present a general strategy and flow chart for model development

4
7dps (2 years in 1.7 minutes)
Click for Animation

5
Integrated Hydrostratigraphic Modelling
► We are already doing integrated modelling…
► Integrated Stratigraphic/Groundwater modelling
 Some GW modellers have only a limited background in geology
► Geology is a “knowledge boundary”
► Desire to independently and completely develop the stratigraphic model
 Re-conceptualization of the stratigraphic model is rarely undertaken once
the GW model calibration process has begun.
► Geologic refinements often addressed with K zones or parameter estimation
(compensating errors)
► Integrated SW/GW modelling
 Similar “knowledge boundaries”, but integration is essential
 Greater chance that “compensating errors” enter into the process

6
Presentation Outline
► Technical Issues and Challenges
 Discussion of issues, with examples of soil zone response and dynamic
GW feedback to illustrate challenges
► Strategies for integrated SW/GW model calibration
 Presentation of an integrated model development “flow chart”
 Other guidelines and recommendations
► Non-technical issues
 Data management, blind spots, “Renaissance Hydrogeology”

7
Customs, Conventions and Blind Spots
► Historic simplifications
 GW: Too many constant heads, assumption of stasis (steady state)
 SW: Lumped parameter models, groundwater as a “decay reservoir”,
separation of hydrology and hydraulics, empirical baseflow separation
► Different in calibration approach and emphasis
 GW: Emphasis on matching heads and spatial patterns (head, xy)
► Less emphasis on regional flux calibration; poor recharge guesstimates
 SW: Emphasis on matching streamflow peaks (flux, time)
► Limited emphasis on spatial and low-flow calibration
► Both SW and GW modellers have “blind spots”
 Conventions and simplifications must be addressed during the integrated
model development process (Watch out for a defensive response!)

8
Integrated Model Development Flowchart
8
1. DATA
COMPILATION
AND SYNTHESIS
2.DEVELOPMENT
OF CONCEPTUAL
MODELS
-STRATIGRAPHY
-HYDROSTRAT
-SOIL ZONE
-WETLANDS
-RIVER SYSTEMS
3. SUB-MODEL
CONSTRUCTION,
PRE-CALIBRATION
AND PRE-PEST
4.INTEGRATION
APPLICATION
AND
INSIGHT
5.
CALIBRATION
PRECIPITATION
RUNOFF
SUB-MODEL
GROUNDWATER
FLOW
SUB-MODEL
ONGOING DATA EVALUATION

9
Model Construction Flowchart
► Model construction includes iterative data evaluation,
conceptualization, sub-model pre-calibration and integration
 Sub-models cannot be independently calibrated, but some components of
the sub-model (e.g. snow pack sub-model) can be more or less developed
► Stages 1-5 are not actually “completed” in a linear manner
 The most important conceptual insights are usually only identified once the
model integration and final calibration is complete.

10
1. DATA
COMPILATION
AND SYNTHESIS
Stage 1: Data Compilation and Synthesis
► Data availability, quality and gaps
 Unevenly distributed, but shallow subsurface data is common
 Watch for an unbalanced calibration: Too much focus on the data
at hand, and not the overall problem
► Integrated relational database
 Reduce barriers to integrated understanding and cross-system
response and calibration assessment
 Integrated data analysis and modelling tools (xyzt)
► Data collection and synthesis end when the report is done!

11
Stage 2: Preliminary Conceptualization
► Multiple conceptual models need to be generally formulated
 Stratigraphy, hydrostratigraphy, soil zone, land cover, vegetation, snow
pack, land use (pervious-impervious), climate
► Pre-identify areas of strong transient SW/GW interaction
 Shallow depth to water – Dunnian rejected recharge
► Enhanced ET in areas with shallow depth to water table
 Dynamic wetlands – storage
 Riparian zones and “contributing areas”
 Reaches with significant river pickup and loss
► Headwaters, springs, intermittent streams, seepage areas
 Identify, but avoid, these areas during initial model construction!

12
GW Feedback Zones
► Dunnian rejected recharge may likely occurs in areas with:
 Depth to water table less than 2 m
 Areas with flowing wells, springs and headwater seeps

13
Stage 3: Sub-Model Development
► Focus on:
 SW and GW model construction and parameter preparation
 Data review, assessment and pattern identification
 Understanding of general sensitivity (Pre-PEST)
► GW: Focus on the deeper GW flow system (steady state)
► SW: Pre-calibrate to a gauged sub-catchment with modest GW/SW
interaction
 Assume parsimony (consistency) when later extrapolating parameters from
the test or “donor” catchment to adjacent catchments.

14
Compensating Errors
► Compensating errors arise when the parameters of one sub-model are
adjusted incorrectly to offset processes and feedback from another
sub-model
► Examples:
 GW feedback (Dunnian rejected recharge) can be incorrectly represented
as low permeability soil
 Wetlands storage (and release) can mask GW discharge
 Cross-watershed GW inflows can be missed as stream gauge underflow
 Excessive hydraulic conductivity variation and PEST can mask structural
and stratigraphic errors.
 Over-parameterization and excessive local sub-catchment calibration
adjustment versus “parsimony”
►

15
GSFLOW: Dunnian Runoff
► Model representation of GW discharge areas, wetlands and shallow aquifers
► Runoff can occur off fully saturated soils (rejected recharge)
 Occurs when the water table is at or near surface
 Not sensitive to surficial material K
► Can create runoff from saturated gravels
 Spatially controlled: Tends to occur in stream valley areas
 Seasonally controlled: Tends to occur in spring when
water table is high
15
Unsaturated
zone
StreamStream
Gravity drainage
Recharge
Ground-water flow

16
Time-varying GW Feedback
► The “contributing area” that
generates true runoff
depends on the time-varying
position of the water table
► Example: Dunnian process
area varies seasonally
between 5 and 25% of the
study area
► Runoff occurs, but it is a
groundwater dependent
process!

17
GW Discharge to the Soil Zone (Daily)
Click for Animation
Daily GW discharge to soil zone

18
Step 4: Integration
► Re-assess the shallow conceptual model
 Review soil zone properties, thickness, storage, drainage, interflow
 Develop compatible groundwater layer 1 geometry and properties
► GW modellers must avoid over-simplifying the shallow system.
 Resist “old habits” previously used to avoid dry GW cells
► MODFLOW NWT – stable representation of shallow complexity
► Watch the terminology
 Infiltration – percolation - recharge

19
SW vs GW Conceptualization
► SW Conceptual Model
 Macropores
 Preferential flow
 Throughflow
 Interflow
 Subsurface stormflow
 Infiltration/percolation/
drainage/recharge
 Event mobilized GW
 Soil/rock contact zone
interface flow
 Seepage faces
► GW Conceptual Model
 1-D or 3-D Richard’s
equation
from Lin, 2010

20
Technical Issues
► The shallow subsurface, where the integration happens, is highly
transient and complex
 Significant fluctuation in system feedback
► GW Feedback is highly variable – wet year/dry year, seasonal
► Empirical baseflow separation is only a first guess
► Strong seasonality means the average conditions never exist
 Steady state calibration can be very limited in the upper system
► In summary, dynamic feedback is reality – get on with it
 Recognizing the dynamic nature is essential to the calibration process

21
Storage and 3D movement of water in the Soil Zone
► Soil zone moisture content
Beach Deposits
Till Upland
- Till uplands drain both vertically and downslope
- Lateral drainage to the beach deposits from the till uplands enhances recharge
- Soil zone storage helps supply rate limited GW recharge to the lower layers
Click for Animation

22
Soil Zone Drainage (GW Recharge)
► When moisture is available (winter months) there is a near constant, but rate
limited, drainage from the soil zone
Click for Animation
Beach DepositsTill Upland

23
Step 5: Integrated Calibration
► Get the models and the team working together
► Continue to re-conceptualize, as necessary
► Write a draft report to formulate your understanding (and impress your
boss/client with your progress)

24
Integrated Model Development Flowchart
24
1. DATA
COMPILATION
AND SYNTHESIS
2.DEVELOPMENT
OF CONCEPTUAL
MODELS
-STRATIGRAPHY
-HYDROSTRAT
-SOIL ZONE
-WETLANDS
-RIVER SYSTEMS
3. SUB-MODEL
CONSTRUCTION,
PRE-CALIBRATION
AND PRE-PEST
4.INTEGRATION
APPLICATION
AND
INSIGHT
5.
CALIBRATION
PRECIPITATION
RUNOFF
SUB-MODEL
GROUNDWATER
FLOW
SUB-MODEL

25
Integration
Time Step
► The timing of the SW
and GW processes is
very different, and
can be a source of
contention
► Daily integration time
step:
 Too fine for GW modelers
 Too coarse for SW
modelers
 No one is happy, so it must
be right!
Click for Animation

26
Iterate: Sub-model Refinement
► After the first fully integrated simulations, it is likely that the uncoupled
models could use another round of refinement
 Update the conceptual model as necessary
 Refine model parameters
 Focus on the timing of the interaction
► GW: Focus on transient shallow system response
 Ensure that surface discharge and groundwater discharge to streams
matches observed wetland patterns and surface stream flows
► SW: Focus on the split between interflow and recharge
 In this final uncoupled refinement phase, the modellers must recognize
that model response will not reflect interaction

27
Step 5x: Final Integrated Calibration
► Lots of re-thinking and even re-conceptualization
 System response timing and lag is sensitive
► Two key benefits of the final integrated calibration process:
 Model Input: Measured total precipitation
 Calibrate to: Measured total streamflow
► Baseflow separation is only good for the preliminary stages
► Focus on matching low flows, and not just the peaks
 Balanced calibration to heads (GW) and flux (streamflow)

2828
Aquifer Head vs. Stream Stage
• GW/SW discharge
reverses during
each storm event
• Baseflow
separation does
not account for
reversals
• GSFLOW Simulated Hydrograph at Oro-Hawkstone stream gauge
Storm Event Reversal:
Stream level higher than aquifer
Dry period:
Aquifer level higher than
stream = GW discharge

29
GW Recharge
► GSFLOW provides
ground water
recharge estimates
on a daily basis
Click for Animation

30
Non-Technical Issues and Strategies
► Expect to do a lot of education: clients and peer reviewers
 Include plenty of simplified details about model integration in your reports
(no one wants to read the manuals!)
► Don’t get too attached to preliminary results
 Integrated conceptual models frequently require change
 Watch for “blind spots”
► Management: Identify someone who knows a little about everything to
oversee integration
 A polymath or renaissance hydrogeologist is needed for mediation, and
“compromise”

31
Conclusions
► Integrated Modelling is different; It requires:
 Integrated calibration strategies
► Don’t become attached to your initial uncoupled calibration estimates!
► Consider re-conceptualization, even late in the integrated process
 Integrated data management
► Data silos and barriers will only hide the relationships and response lag between
the systems
► Integrated modelling and calibration tools
 An integrated and balanced modelling team
► The skill, multi-disciplinary knowledge, and ability of the SW and GW experts to
address their “blind spots” is very important
► Our experience after a number of fully-integrated models:
 It’s hard, but it’s worth it.

1. DATA
COMPILATION
AND SYNTHESIS
2.DEVELOPMENT
OF CONCEPTUAL
MODELS
-STRATIGRAPHY
-HYDROSTRATIGRAPHY
-SOIL ZONE
-WETLANDS
-RIVER SYSTEMS
3. SUB-MODEL
CONSTRUCTION,
PRE-CALIBRATION
4.INTEGRATION
APPLICATION
AND
INSIGHT
5.
CALIBRATION
VALIDATION
PRECIPITATION
RUNOFF
SUB-MODEL
GROUNDWATER
FLOW
SUB-MODEL

Issues and Strategies for Integrated Model Calibration

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Issues and Strategies for Integrated Model Calibration (17)

Recently uploaded (20)

Issues and Strategies for Integrated Model Calibration