- Linked rainfall recharge models with groundwater models in the National Groundwater Modelling System (NGMS) to allow for scenario runs during a 2012 drought when models were not ready.
- The partial implementation of NGMS helped emulate the manual process used at the time to connect recharge and groundwater flow models.
- Future improvements to NGMS are needed to accommodate new MODFLOW codes, unstructured grids desired by water companies, faster run times, and easier usability.
%+27788225528 love spells in Boston Psychic Readings, Attraction spells,Bring...
DSD-INT 2017 Linking rainfall recharge models with groundwater models in NGMS - Farrell
1. Linked rainfall recharge models
with groundwater models in NGMS
(and looking forward to a future NGMS)
Mark Whiteman1, Rolf Farrell1, Harris Tarnanas1, Marcel Ververs2, Alastair Black3
1 Environment Agency of England
2 Deltares
3 Groundwater Science Ltd.
2. Context
Risk of drought, Spring 2012
Need to undertake scenario runs
Models not ready for it
Hurried updates and over-reliance on rainfall runoff models
Partial NGMS 4R implementation used to emulate manual process undertaken at the time.
Process (following) published in:
Integrated environmental modelling to solve real world problems – Geological Society Special
Publication Vol 408; Splicing recharge and groundwater flow models in the Environment Agency
National Groundwater Modelling System; Rolf Farrell, Marcel Ververs, Paul Davison, Paul Howlett and
Mark Whiteman.
3. 3
Heading
• Level one bullet
• Level two bullet
• Level three bullet
– Level four bullet
» Level five bullet
• Level one bullet
• Level two bullet
• Level three bullet
– Level four bullet
» Level five bullet
WB
WMW
4. Recharge models and groundwater
models
4
Recharge model (4R)
Groundwater model
(MODFLOW96-VKD)
5. 5
Input –
Rainfall, Et
Run
Recharge
Model
Input –
Recharge,
abstractions,
discharges, aquifer
parameters
Run
Modflow
Model
NGMS Output -
Modelled Flow
and level data
Input –
Rainfall, Et
Output -
Modelled
recharge
Input –
Recharge,
abstractions,
discharges, aquifer
parameters
Run
Recharge
Model
NGMS Output -
Modelled Flow
and level data
Input –
Recharge,
abstractions,
discharges, aquifer
parameters
NGMS Output -
Modelled
recharge
NGMS Runs
Modflow
Model
NGMS Runs
Modflow
Model
Recharge
Input –
Rainfall, Et
NGMS Output -
Modelled Flow
and level data
Output -
Modelled
recharge
NGMS Runs
Recharge
Model
Input files Model Process Post Processing
NGMS Processes A Fusion Issue
Stages in the model
Run process
Breaks in the model
run process
4R and
Modflow
4R and NGMS
Modflow
NGMS 4R and
NGMS Modflow
6. 6
Calibrated historic model period
1965 April 2012
4R
1976
Drought
Abs
1995
Drought
4R
G
Abs
4R
Abs
Historic groundwater levels match current event
Historic rainfall matches current scenario event
Historic abstractions closest to predicted current
Drought scenario
1984
GWL
Scenario
drought
Potential groundwater drought event
8. 2017
• Finally fully implement 4R to NGMS
• Needed due to low spring rainfall 2017
• Subsequent rainfall has addressed SW
concerns
• Groundwater drought remains a possibility in
2018; recharge this coming winter still a concern
8
21. Run time…….
• What makes models bigger???
• “Big is best”
• Consultancy steer
• Irregular grids
• Insufficient focus on model efficiency (stability, inactive
cells, grid spacing)
21
22. More usability =
• Control of physical size of model
• Faster Modflow processes
• Faster Delft FEWS processes
• Better (different) hardware
• Avoid unnecessary processes (inactive
layers/cells, running “sub models”, shorter
scenario periods etc)
22
23. Parallel Modflow (PMF) vs “Normal” Modflow
23
9,3
4,8
2,9
0,6 0,4
0,9
0,36
0 0,08
0
1
2
3
4
5
6
7
8
9
10
RUNTIME [Hours] -
on a very fast CPU = 4.4Ghz, 12 core vs GPU = RX480 - 2304 core @
1.2Ghz.
NEAC_Normal NEAC_PMF CBO_Normal
CBO_PMF LON_Normal LON_PMF
EMY_Normal EMY_PMF_NonDry EMY_PMF_NotNonDry
RUNTIME[Hours]
24. A faster running NEAC model
24
0
5
10
15
20
25
30
Regular machine Fast machine ParallelMF
MODFLOW6 / USG can be
notably faster
Parallel MODFLOW further
improvement
Model stability is key
NEAC runtime (hours)
per scenario
25. Reducing run time – an unhelpfully large
model from Southern England……
25
File size (Gb)
MF runtime
(hours)
NGMS Runtime
(hours)
Total runtime
(hours)
As delivered 630 96 96 192
Parallel Modflow 630 24 96 120
Remove inactive layers,
rows and columns 302 19 46 65
Reduce stress periods per
timestep (reduce output
file sizes) 75 13 11 25
Single Scenario 18.8 3.4 2.9 6.3
26. Future NGMS
• New developments in Modflow 6 not
compatible with current NGMS
• Unstructured grids
• Completely new input file format
• New module adapter needed
• New system structure
• Focus on ease of use (actual functionality OK)
• Focus on system reliability/availability
26
27. Questions to the audience
• Benefits of moving to cloud?
• Dealing with physical servers seems a bit old
fashioned
• Handling irregular grids (in 3 dimensions)
• Faster run times in Delft FEWS
27