TxRR Model Along the Coast

1
TxRR Model Along the Coast

• Victoria Samuels
• May 1, 2001

2
Background of TxRR

• Texas Water Development Board model to evaluate
  needs for instream and freshwater flows to the
  estuarine systems in Texas
• Many of these watersheds are ungaged
• Calibrates rainfall-runoff relationship for a
  TWDB gage watershed
• Relates gaged to ungaged watersheds with
  relationship

SMMAXU (CNG/CNU) SMMAXG
3
Genetic Algorithm Optimization

• Follows the survivalistic behavior of nature
• Nature develops life forms at random
• Weaker life forms are killed off, Successful
  life forms progress
• Successful life modified, tested again
• Pattern continues until MOST successful life form
  found
• Natural or Darwinian Selection

From Introduction to Genetic Algorithms, Nick
Johnson
4
Genetic Algorithm Optimization

• Random sampling of solutions, chromosomes
  undergo natural selection
• Two parent chromosomes are selected from
  remaining population and reproduction occurs,
  form new children
• Crossover
• Mutation

From Introduction to Genetic Algorithms, Nick
Johnson
5
TxRR Model Interface
Courtesy of Venkatesh Merwade
2. Call Fortran Code to run TxRR
1. Windows Based Input Screen
2. Windows Based Output Screen
6
TxRR Input Screen

• RAINFALL/RUNOFF FILES
• .dat files
• currently available only from TWDB

7
TxRR Input Screen

• FIXED WATERSHED CHARACTERISTICS
• MOIST1 assume to be initial soil moisture
  condition
• DRAREA - catchment area in sq. mi., obtained
  from USGS website, http//water.usgs.gov/tx/nwis/s
  w
• abstr1 initial abstraction from direct runoff
  equations, assumed to be 0.2, as in SCS Curve
  Number Method

8
TxRR Input Screen

• GENETIC ALGORITHM SETUP
• Population size number of life forms
  (solutions) to choose from
• Max of Iterations limit on number of
  iterations the optimization routine will run
• Number of children per Chromosome how many
  offspring are formed during reproduction
• Choose Random seed assumed starting point of
  random sampling of solutions

9
TxRR Input Screen

• TxRR PARAMETERS
• GammaA N
• GammaB - k
• QB1 initial baseflow
• A(n) monthly depletion factors, which should
  have a sinusoidal pattern because of its seasonal
  nature
• SMMAX maximum soil moisture
• RECES recession constant for baseflow
• WB baseflow coefficient

10
TxRR Input Screen

• PERIOD OF SIMULATION
• between January 1940 and December 1997
• gages must have data for simulation period, or
  program will not run (and not tell you why)

• FORM OF OPTIMIZATION FUNCTION
• A monthly data
• B daily data
• C volume ratio

11
Run TxRR Code

• Create Input Files
• Go to TxRR DOS code
• Remove return flow?
• 70 zero data warning
• Computer cranks out 200 iterations

Pop Optimized Parameters!
12
TxRR Output Screen
13
Study Area

• Basin Group C
• Along Eastern Coast of TX, near Houston
• Appropriate for TxRR model

14
Study Area-Buffalo Bayou Tidal Watershed
15
Calibrating the Model

• Gage No. 10062/8075500 Sims Bayou at Houston,
  63.0 sq. mi.
• Initial time period of January 1990-December 1992
  ? Results not good
• Shorten time period to January 1991-June 1992 ?
  Improved, but still no good
• Shorten to major storm sequence from December
  1991-June 1992

THE SHORTER THE SIMULATION PERIOD, THE BETTER
16
Calibrating the Model
17
Calibration Parameters

• Using a time period of December 1991 June 1992
• Test Return Flow values
• Test Moist1 values
• Modify Genetic Algorithm information

18
Return Flow Calibration

• Removing return flow decreases the gaged flow by
  the set amount, across the board
• Desired effect to better match baseflow
• Tried 50 cfs, 40 cfs, 45 cfs
• Went with 40 cfs
• Best fit overall
• Minimum value of gaged flow 40 cfs

19
Moist1 Calibration

• Default Value 2.35
• Decreased to 1.75 ? no difference
• Increased to 3.0 ? some peaks were raised, some
  were lowered
• Larger the peak, greater it increased
• This result fit with the gaged flow better
• Tried 4.0, 5.0
• Studied the results by breaking into 4 time
  periods
• Focused approach led to Moist14.0 having the
  best fit

20
Genetic Algorithm Calibration

• Population Size
• Default 100
• Increased to 150, 200
• Results similar if not worse to size of 100
• Increased processing time
• Default retained

21
Genetic Algorithm Calibration

• Number of Children
• Default 1
• Toggled to 2
• Results appeared to be identical
• Default retained
• Random Seed
• Default A, ranged from A F
• Tried each seed option
• Seed D had the best fit

22
(No Transcript)
23
Calibration of Gage 10062 for January - May 1990

• Assumed input parameters used to calibrate the
  first time would apply, with a little tweaking
• Wrong! Peaks far ovestimated, from two to ten
  times greater than the gaged flow
• Changed many of the parameters

24
Calibration of Gage 10062 for January - May 1990

• Removing return flow had no effect
• Decrease Moist1 ? back to default 2.35, 1.5, 0.5
• Decreasing Moist1 reduces peaks above 500 cfs but
  increases peaks below 500 cfs
• Changed Random Seed to A
• Increased range of TxRR parameters

NO REAL EFFECT
25
(No Transcript)
26
My Crazy Idea

• Decrease the drainage area from the USGS reported
  63 sq mi to 50 sq mi
• It worked!
• Trial and Error process to increase the smaller
  peaks and decrease the larger peaks

27
(No Transcript)
28
Calibration of Gage 10061 for January - May 1990

• Removing return flow of 97 cfs
• Try default parameter values USGS Drainage Area
  of 94.9 sq mi
• Baseflow okay, underestimated small peaks,
  overestimated larger peaks
• Same problem as before. Look to the drainage
  area!

29
Use a Drainage area of 30 sq mi vs. USGS 94.9 sq
mi
30
Ungaged Watershed Relationship

• SMMAXU (CNG/CNU) SMMAXG
• Treat one watershed as if it was ungauged and the
  other was, and compare the results

Gauge CN SMMAXG SMMAXU
10061 80.42 0.001 0.0123
10062 78.03 0.0127 0.001
31
Ungaged Watershed Relationship

• Default input values and USGS drainage areas did
  not yield satisfactory results
• Calibrated values led to much better modeled
  flows in both cases
• Upsetting, because the user would not have gaged
  flow to compare to
• Grossly overestimates flow using default,
  uncalibrated factors

32
(No Transcript)
33
(No Transcript)
34
Why?

• Does more rain fall on the gauges than the entire
  watershed, so the contributing area to the gage
  is much less?

Gage Gage Precip. Watershed Precip. Diff. USGS Area Calibrated Area Diff.
10061 47.56 46.09 3.19 94.9 30 68.4
10062 49.43 47.50 4.26 63.0 25 60.3
35
Conclusions

• Calibration is an Art Form
• Only parameter that leads to significant change
  is the drainage area
• Moist1 parameter has slight effect
• Does not serve its purpose well
• Drastically overestimates flow for ungaged
  watersheds (if Im doing this right)
• No real rationale for drainage area calibration