0.

6
Baseflow Analysis 1

Objectives

discharge (m s )

discharge (m s )
-1

-1
0.4

3
1. Understand the conceptual basis of baseflow analysis.
0.1
2. Estimate watershed-average hydraulic parameters and ground-
water recharge rates. 0.2 straight-line
baseflow recession

Baseflow definition and significance 0 0.01

5/11 5/31 6/20 7/10 7/30 8/19 5/11 5/31 6/20 7/10 7/30 8/19
Portion of (stream) flow that comes from groundwater or other
delayed sources (Tallaksen, 1995. J. Hydrol., 165: 349). Stream discharge gradually Recession hydrographs
decreases after storm events. commonly plot as straight
Understanding of low-flow condition is important for water lines on a semi-long graph.
resource management and environmental protection. Various baseflow “separation”
techniques have been proposed. Q(t) = Q0exp(-at)
→ Why? Q0 : discharge at t = 0
What purpose?
We will review: a : constant (s-1)
(1) Baseflow recession analysis What causes the exponential
(2) Baseflow separation technique Regardless of sophisticated behaviour?
algorithms, they are all arbitrary.
1 2

Reservoir model for recession analysis Physically-based aquifer model piezometric surface, not WT
h = hs
Exponential function is the solution of: Baseflow from a homogeneous,
dS confined aquifer is described by: h0
Q = aS and = −Q (linear reservoir) S
dt
Q ∂  ∂h  ∂h q b
S: volume of water stored (m3)  Kb  = Ss b
∂x  ∂x  ∂t
x=0 x=B
A more general reservoir model is given by: ∂ 2h ∂h
1 T = Sc T = Kb : transmissivity (m2 s-1)
dS p=1 ∂x 2 ∂t Sc = Ssb : storage coefficient or storativity
Q = aSp and = −Q
dt
Q /Q 0

p: dimensionless constant 0.1 p=2 Boundary conditions and initial condition

h(x0) = hs for all t > 0
Non-linear (p > 1) reservoir represents the
effects of complex processes such as the No flow at the divide (x = B) for all t > 0
0.01
transmissivity feedback. 0 1 2 3 4
h = hs + h0 at t = 0 for all 0 ≤ x ≤ B
at
The solution of the non-linear reservoir equation is: Rorabaugh (1964. Int. Assoc. Scientific Hydrol. Pub. 63: 432-441, Eq.1) reported
the Fourier-series solution for flow per shore line, q (m2 s-1):
Q(t) = Q0(1 + at)-p / (p - 1)
q = (2Th0/B)[exp(-at) + exp (-9at) + exp(-25at) + … ]
See Tallaksen (1995) for an excellent review. where a = π2T / (4B2Sc)
3 4
 10 Models for unconfined aquifer
High-order terms in the series are
negligible for tT / (B2Sc) > 0.2. all terms

q / (2Th0/B)
Streams are usually connected to unconfined, not confined,
tc = 0.2 B2Sc / T is called “critical time”. 1 aquifers. The rigorous analysis of unconfined aquifers would
require the solutions of the Richards equation.
∴q ≅ q0 × exp{-at} for t >tc
first term only
where q0 = 2Th0/B The Dupuit-Forchheimer (D-F) approach offers a reasonable
0.1 approximation of complex problems (e.g. Paniconi et al., 2003. Water
Note the similarity between this and the 0 0.2 0.4 0.6
Resour. Res., 39: 1317). The transient flow equation based on the D-F
exponential decay equation of hydrograph t × T/(B2Sc)
in Page 2. approximation is called the Boussinesq equation:
→ What does this mean? ∂  ∂h  ∂h
 Kh  = Sy
∂x  ∂x  ∂t D
h(x,t)
Remember the recession coefficient in this model: a = / π2T (4B2Sc). K: aquifer conductivity (m s-1) Dc
Sy: drainable porosity
The recession coefficient in Page 2 represents the average x=0 x=B
properties of aquifer over the entire watershed.
Exact solution of the non-linear Boussinesq equation is available
only for special cases. Brutsaert (2005, Hydrology – an introduction. Ch.
10, Cambridge Univ. Press) presented a summary of various solutions
for the cross section shown above.
5 6

Early-time drainage Long-term drainage

The first solution considers a D The next solution considers gradual D
riparian aquifer with a thickness D, Dc q drainage of the entire hillslope, Dc q
which is fully saturated at t = 0. The x=0 x=B assuming no flow at x = B. x=0 x=B
depth of the stream (Dc) is assumed
q: flow per shore length (m2 s-1)
much smaller than D. Approximate analytical solution is obtained by “linearizing” the
Boussinesq equation:
Using a technique known as Boltzmann’s transform, early-time
drainage flux from the near-shore zone is (Brutsaert, 2005, Eq.10.64): ∂  ∂h  ∂h ∂ 2h ∂h
 Kh m  = Sy Kh m = Sy
∂x  ∂x  ∂t ∂x 2
∂t
q = 0 . 332 KS y D 3 / t
where hm is the “average” saturated thickness.
Multiplying q by the total channel length L in the watershed Solving the linearized equation, drainage flux per shoreline is:
and 2 (from both sides of the channel), baseflow Q (m3 s-1) q = (2KhmD /B)[exp(-at) + exp (-9at) + exp(-25at) + … ]
measured at the watershed outlet is (Brutsaert, 2005, Eq.10.160):
where a = π2Khm / (4B2Sc)
Q = 0 . 664 L KS y D 3 / t = 0 . 664 KS y D 3 L2 / t Eq. [1]
This is almost identical to the Rorabaugh (1964) equation.
dQ 1 Therefore, for t > 0.2B2Sc / Khm
Note: = − 0 . 664 KS y D 3 L2 / t 3 ∝ − Q 3
dt 2 q = (2KhmD /B) exp{- π2Khmt / (4B2Sc)}
7 8
Brutsaert (2005, p.400) proposed expressing hm as a fraction of D: Watershed-scale flow parameterization
hm = pD where p ≅ 0.35 for Dc << D Eq. [2] Brutsaert & Lopez (1998. Water Resour. Res., 34: 233) used Eqs. [1] and
p ≅ (D + Dc)/(2D) for other cases [3] to estimate average K and Sy for watersheds. This method
Using p, the flux is written as (Brutseart, 2005, Eq.10.116): examines the relation between Q and its time derivative in a form:
dQ
q = (2KpD2 /B) exp{- π2KpDt / (4B2Sc)} = − aQ b a (s-1) and b are constants.
dt
We note that the average distance from the channel It can be shown that:
to drainage divide, B, can be estimated by: A b1 = 3 and a1 = 1.1336/(KSyD3L2) for Eq. [1] (early time)
B
B = A / (2L) b3 = 1 and a3 = π2KpDL2 / (SyB2) for Eq. [3] (late time)
→ Why?
When daily values of Q and -dQ/dt are

-dQ/dt (m3 s-1 d-1)

L
plotted on a log-log graph,
Total baseflow Q (m3 s-1) at the watershed outlet is log[-dQ/dt] = loga + blogQ
(Brutseart, 2005, Eq.10.164):
The cluster of the points is “enveloped”
Q = 2L × (2KpD22L/A) exp{- π2KpDt 4L2/ (4A2Sc)} by lines of b = 3 and b = 1. The envelope
= (8KpD2L2/A) exp{- π2KpDL2t/(A2Sc)} Eq. [3] lines represents the lowest recession
dQ rate (-dQ/dt) for a given Q, which is
Note: ∝ − exp{( − π 2 KpDL 2 t /( A 2 Sc )} ∝ − Q considered the baseflow condition.
dt Q (m3 s-1)
9 10

K or Sy may be estimated from the intercepts (loga) of the envelope Baseflow separation
lines, if other parameters in the equations are known. One can also
Given a hydrograph, “quick” flow and
determine K and Sy simultaneously from a1 and a3 (Brutsaert & Lopez,

discharge
1998, p. 237):
baseflow can be separated by a
number of different methods.
1 . 1336 a3 A 1 . 1336 p 1 - Connecting local minima
K = Sy = π Eq. [4]
π p a 1 ( LD ) 2 a 3 a1 DA
- Variation of local-minima method
In practical computation, daily values of Q and dQ/dt are given by: - Using inflection points time
Q i +1 + Q i dQ Q i +1 − Q i
Q av = = All methods use arbitrary criteria for baseflow, and are time
2 dt ∆t
av consuming for manual operation.

Sample data from the Marmot Creek watershed in Canada will be
used to demonstrate the techniques in the computer exercise.
Automated techniques are at least objective, and are efficient for
processing many data sets.
We will use a digital-filter algorithm of Arnold et al. (1995. Ground
Water, 33: 1010) to demonstrate the usefulness and limitation of
automated baseflow separation.

11 12
Recursive digital filter Baseflow index
The algorithm, originally described by Nathan & McMahon (1990), By applying the digital filter to the entire 2005 summer discharge
calculates the quick flow component qi at time step i from qi-1 at data set (May 1- September 10) for Marmot Creek, it was found
previous time step and total flow Qi and Qi-1: that:
1+ β Total discharge = 2.71 × 106 m3
q i = β q i −1 + (Q i − Q i − 1 )
2 Total baseflow = 1.72 × 106 m3
where β is a filter constant ranging between 0.9 and 0.95.
The ratio of total baseflow to discharge is base flow index (BFI).
Baseflow bi is calculated as: bi = Qi – qi
In this example, BFI = 1.72 / 2.71 = 0.63.
2
In this example from the Marmot raw data Automated baseflow separation offers a convenient tool to
Creek watershed in 2005, the filter 1st pass calculate BFI for multiple watersheds having different size and
2nd pass geology, or for a single watershed in multiple years having

Q (m s )
was successively applied three

-1
3rd pass different meteorological forcing or landuse practice.
times with β = 0.925.

3
1

The first pass appears to have We will use a computer program Baseflow to sample data from
produce “reasonable” results. the Shirakawa watershed in the computer exercise to calculate
0 BFI.
5/30 6/9 6/19 6/29 7/9

13 14

Marmot Creek watershed (14 km2), Alberta, Canada Baseflow Computer Exercise
• Elevation 1700-2700 m, the eastern edge of the Rocky Mountains.
• Mean annual precipitation = 640 mm at the base. Watershed parameter estimation
In this exercise, we will apply the Brutsaert & Lopez (1998) method
to estimate watershed-average values of K and Sy for the Marmot
Creek watershed.

(1) Open Marmot_data.xls file, which contains daily stream

discharge and precipitation values for May 1 - September 10 in
2009, 2010, and 2011.
(2) Open dQ_template.xls file, and paste the discharge and
precipitation data from the raw data file.
(3) Compute Qav = (Qi+1 + Qi) / 2
(4) -dQ/dt is calculated for the days without significant rain, and
with a sufficiently high difference in discharge to minimize the
N

influence of data noises. Columns G and H in the template file

sets the flags (0 or 1) depending on the critical values at the top.
gauging station 1 km Copy the cell formula down and compute -dQ/dt.
15 16
(5) Plot -dQ/dt vs Q for all points, using different symbols for
different years.
(6) Plot straight lines of -dQ/dt = aQb with b = 3 and b = 1 for a
suitable range of Q. Use a as a fitting parameter to adjust the
position of the envelope lines.
(7) Once the values of a are determined for both lines, calculate K
and Sy from Eq. [4]. Assume that A = 14 km2, L = 15 km, D = 2 m
and p = 0.75.
(8) Discuss the magnitude of values.
17
Automated baseflow separation
In this exercise, we will use the Baseflow program to separate
baseflow and compute base flow index (BFI).
(1) Open file.lst file using Notepad or Wordpad, and change the
input and output file names to sk2004.prn and sk2004.out,
respectively. The input file sk2004.prn has already been
prepared in the required format.
(2) Double click on bflow.exe on the file folder to run the program.
(3) Open the output file and covert the date from YYYY, MM, DD to
actual dates by a cell formula “=Date(YYYY,MM,DD)”.
(4) Plot stream flow, pass 1, pass 2, and pass 3 on the same chart.
Observe the quality of separation.
(5) Calculate the total stream flow and total baseflow (for
appropriate pass), and compute the BFI.
(6) Repeat (1)-(5) for 2005 and 2006, and compare the results.
18