Modeling phosphorus in the environment - Chapter 8 pdf

189
8
Modeling Phosphorus
with Hydrologic
Simulation
Program-Fortran
David E. Radcliffe
University of Georgia, Athens, GA
Zhulu Lin
University of Georgia, Athens, GA
CONTENTS
8.1 Brief History of Model Development 189
8.2 Modeling of Hydrology 190
8.3 Modeling of Phosphorus 195
8.4 Modeling of Sediment 199
8.5 Calibration 202
8.6 Case Study: Upper Etowah River Watershed 203
8.7 Comparing HSPF and SWAT 210
8.8 Conclusions 211
Acknowledgments 212
References 212
8.1 BRIEF HISTORY OF MODEL DEVELOPMENT
The Hydrologic Simulation Program-Fortran (HSPF) is a watershed-scale, semi-
distributed model developed from the original Stanford model (Bicknell et al. 2001).
It is one of the two dynamic models intended for modeling watersheds dominated
by nonpoint sources in the U.S. Environmental Protection Agency (EPA) Better
Assessment Science Integrating Point and Nonpoint Sources (BASINS) package
(U.S. Environmental Protection Agency 2004a). The other model is the Soil Water
Assessment Tool (SWAT) described by Arnold et al. (1998) and Neitsch et al. (2002).
The functions and processes included in HSPF were derived primarily from the
following group of predecessor models:

© 2007 by Taylor & Francis Group, LLC
190 Modeling Phosphorus in the Environment
Hydrocomp Simulation Programming (HSP) (Hydrocomp, Inc. 1976, 1977)
Nonpoint Source (NPS) Model (Donigian and Crawford 1976a)
Agricultural Runoff Management (ARM) Model (Donigian and Crawford,
1976b; Donigian et al. 1977)
Sediment and Radionuclides Transport (SERATRA) (Onishi and Wise 1979)
The original development of HSPF was sponsored by the EPA Environmental
Research Laboratory in Athens, Georgia, during the 1970s. HSP was a descendant
of the Stanford Watershed model (Crawford and Linsley 1966). It was first released
in 1980 as Release 5. Later development was sponsored by the U.S. Geological
Survey (USGS) Water Resources Division in Reston, Virginia.
Probably the best-known application of the HSPF model is its use as part of the
Chesapeake Bay Model (Linker et al. 2002). HSPF is one of three linked model
components: a watershed model (HSPF), an airshed model, and a bay model. The
model was used to establish a goal in 1987 of a 40% nutrient — nitrogen (N) and
phosphorus (P) — reduction to the bay by the year 2000. In the current version, the
bay watershed is divided into 94 sub-basins with an average area of 194,000 ha.
A bibliography of articles describing HSPF applications is available at
hspf.com/hspfbib.html (confirmed February 25, 2006). Aside from the Chesapeake
Bay Model, applications modeling phosphorus include Donigian et al. (1996),
Donigian and Love (2002), and Hummel et al. (2003).
8.2 MODELING OF HYDROLOGY
Most of the material in this chapter is taken from the HSPF User’s Manual for
Release 12 (Bicknell et al. 2001), which can be downloaded from the EPA Web site,
www.epa.gov/waterscience/basins/bsnsdocs.html (confirmed February 25, 2006). A
mixture of English and metric units are used in the user’s manual, and this
chapter has aimed for consistency with the manual. HSPF uses elements, zones,
and nodes. One type of element is the land segment, which can be a pervious
land segment (PLS) or an impervious land segment (ILS). Within the PLSs, there

are snow, surface, upper soil, lower soil, and groundwater zones. A segment is
a portion of the land assumed to have uniform (lumped) properties. Another type
of element is a reach element. Within a reach, water moves through a single
zone from an upstream node to a downstream node. A simulation might consist
of a single watershed (completely lumped) or multiple subwatersheds connected
together (partially distributed).
U.S. Environmental Protection Agency (2004b) describes which processes need
to be simulated for PLSs, ILSs, and reaches to simulate hydrology. Basic hydrology
must include PWATER in the PLSs, IWATER in the ILSs, and HYDR in the reaches
elements — there is a process for simulating what happens in the snow layer if that
is applicable.
PWATER is the portion of the model that simulates the water budget for
PLSs. Potential evapotranspiration (PET) is based on U.S. Weather Bureau Class
A pan evaporation times a crop factor, further adjusted by the vegetative cover
percentage in each PLS. Actual ET is calculated by trying to meet the PET from
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 191
five different sources in the following order until PET is satisfied (ET = PET) or
until sources are exhausted (ET < PET): from the groundwater zone as seepage,
from vegetation interception, from the upper soil zone, from the groundwater zone
directly, and from the lower soil zone. The model recognizes that within a land
segment there is variability in ET due to differences in, for example, rooting density.
Rainfall is distributed in PWATER in the following manner. Some of the rainfall
first goes to interception by vegetation (i.e., grass, leaves, stems, and trunks). This
vegetation has a storage capacity that accepts water until it is filled. The intercepted
water is lost through evaporation. The water that remains may infiltrate to upper
zone storage or interflow storage, may enter surface detention storage, or may run off.
None of the conventional methods (e.g., curve number, Green-Ampt) or soil
parameters (e.g., saturated hydraulic conductivity, field capacity) are used to calcu-
late infiltration and interflow. IBAR is the average infiltration rate over the land

segment (in. hr
−1
or in. day
−1
), depending on the time step in use (Figure 8.1). The
model recognizes that within a land segment there is variability in infiltration rates
and that the actual infiltration rate can be less than or greater than IBAR, as this
chapter shows. IBAR is calculated as
(8.1)
where INFILT is a parameter (in. hr
−1
), LZS is the lower zone water storage (in.),
LZSN is the nominal (or average) lower zone water zone storage (in.), and INFEXP
is a parameter (unitless).
FIGURE 8.1 Determination of infiltration and interflow. (Redrawn from B.R. Bicknell, J.C.
Imhoff, J.L. Kittle, Jr., T.H. Jobs, and A.S. Donigian, Jr., Hydrological Simulation Program-
Fortran: HSPF Version 12 User’s Manual, Mountain View, CA, Aqua Terra Consultants,
2001. With permission.)
IBAR INFILT
LZS
LZSN
INFEXP
=






−

potential direct
potential
surface runoff
potential
Interflow
Inches of
water/interval
IMAX
IIMAX
IBAR
IIBAR
line I (infiltration capacity)
line II (interflow +
infiltration capacity)
MSUPY
Inches of
water/interval
IIMIN
IMIN
050100
% of area
© 2007 by Taylor & Francis Group, LLC
192 Modeling Phosphorus in the Environment
This represents infiltration into the lower soil zone; infiltration into the upper
soil zone (a relatively thin layer) is described later. In the case that frozen ground
occurs for a significant period of time, INFFAC accounts for this effect, but it has
been left out of Equation 8.1 for the sake of simplicity. Equation 8.1 shows that as
water content in the lower zone water storage increases during a storm, the infiltration
rate decreases. Once this zone is saturated (LZS = LZSN), the infiltration rate reaches
the minimum asymptotic rate of INFILT. This implies that INFILT is equal to the

field saturated hydraulic conductivity.
The parameter IMAX in Figure 8.1 is the product of INFILD and IBAR. INFILD
is a unitless parameter with a recommended value of 2, so IMAX is twice IBAR. The
water that is available in a time step for infiltration, interflow, or runoff is moisture
supply in inches, MSUPY (Figure 8.1). The total water that infiltrates the lower soil
zone is the area beneath line I and the MSUPY line (clear area). (IBAR is the average
infiltration rate, and this is used in determining the total) As IBAR increases, so does
the amount of water that infiltrates. The water that goes into potential interflow is
the area beneath line II and the MSUPY line and above line I (lightly shaded area
in Figure 8.1).
Interflow is water that moves laterally to a stream due to a restrictive layer in
the unsaturated zone (Fetter 1988). Potential interflow water can become actual
interflow or inflow into the upper soil zone. As IBAR increases, so does the amount
of water that goes to potential interflow. IIMIN and IIMAX are calculated as follows:
(8.2)
where INTFW is a parameter (unitless). It is apparent from Equation 8.2 that as
INTFW increases, line II in Figure 8.1 rises, and the amount of potential interflow
increases. Also, as the water content in the lower soil zone increases, so does the
amount of potential interflow. The water that is available for potential upper soil
zone infiltration, surface detention, or runoff is the area below the MSUPY line and
above line II (darkly shaded area in Figure 8.1).
The remaining water is potential runoff. The fraction of this water that goes into
the upper soil zone (FRAC) is a function of the upper soil zone water content
(UZRAT), which is the upper soil zone water storage in inches (UZS) divided by the
upper soil zone nominal water storage in inches (UZSN). FRAC decreases as the
upper zone water content increases. Note that UZRAT is allowed to be greater than
unity. This is a recognition that UZSN varies from the average (nominal) value over
the pervious land segment.
Once infiltration into the upper soil zone is satisfied, the remaining water goes
into surface storage, runoff, and interflow. Surface storage depends in part on

Manning’s n for roughness (increases with roughness and n), slope length (increases
with slope length), and slope angle (decreases with slope angle). The interflow
component assumes a certain storage capacity for interflow water. The rate at which
IIMIN IMIN INTFW
IIMAX IMAX INTFW
LZS
LZSN
=⋅ ⋅
=⋅ ⋅
2
2
LLZS
LZSN
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 193
water can enter interflow storage depends on the current storage and the rate at
which water is discharging from interflow storage to the stream. The rate of discharge
to the stream is a function of the interflow recession parameter IRC.
Upper soil zone water can percolate into the lower soil zone. Percolation only
occurs when the upper soil zone water content is greater than the lower soil zone
relative water content according to the following empirical relationship (units are
not consistent on both sides of equation):
(8.3)
where PERC is the percolation rate (in. hr
-1
). The fact that PERC is proportional
to INFILT reinforces the idea that INFILT is related to field-saturated hydraulic
conductivity.
Water that percolates into the lower zone from the upper soil zone, plus infil-
trating water, can increase the lower zone soil water storage or pass on through to

groundwater. The fraction that goes to increasing the lower soil zone storage depends
on the lower zone relative water content, LZRAT, which is LZS divided by LZSN.
As the water content increases, less percolation water is retained, and more water
passes through to groundwater.
Infiltrating and percolating water that reaches groundwater storage can discharge
into the stream or can go to deep groundwater storage; this water is essentially lost
from the watershed system. DEEPFR is the parameter (unitless) that determines the
fraction that is lost. Outflow to the stream from the remaining groundwater depends
on the slope of the water table (gradient), the groundwater storage, and two param-
eters, AGWRC and KVARY:
(8.4)
where AGWO is the outflow rate (in. hr
−1
), GWVS is an index to the water table
slope, AGWS is the current groundwater storage (in.), AGWRC is a groundwater
outflow recession parameter (day
−1
), and KVARY is a recession parameter — nonzero
values cause recession to vary as a function of groundwater levels and will produce
seasonal variability in hydrographs.
U.S. Environmental Protection Agency (2004c) provides guidance on how to
choose hydrological parameters. The primary hydrological parameters are as follows.
• FOREST: only used in snow processes.
• LZSN: lower zone nominal soil water storage (in.); recommends an initial
estimate of 1/8 mean annual rainfall + 4 in. (humid regions); min-max
values are 3 to 8 in.; lower values of LZSN will cause more stream flow
(less water lost to ET); default = 4 to 6.5 in. depending on land use.
• INFILT: index to mean soil infiltration rate (in. hr
−1
); provides a range

related to soil hydrologic groups (A: 0.4 to 1.0; B: 0.1 to 0.4; C: 0.05 to
PERC INFILT UZSN
USZ
USZN
LSZ
LSZN
=⋅ ⋅ ⋅ −






01
3
.
AGWO AGWRC KVARY GWVS AGWS=⋅+⋅⋅()1
© 2007 by Taylor & Francis Group, LLC
194 Modeling Phosphorus in the Environment
0.1; D: 0.01 to 0.05); default is 0.16 in. hr
−1
; suggests that Z *INFILT* INTFN
should approximate the long-term infiltration rate, or permeability, in the
soil survey database (untested approach); higher values cause less runoff
and less storm flow in streams.
• LSUR: length of overland flow path (ft) for the pervious land segment; average
length of travel for water to reach a stream; typical values range from 200 ft
for slopes of 15% to 500 ft for slopes of 1%; default = 300 ft; higher values
should cause storm hydrograph to spread out (lower peak value).
• SLSUR: slope of pervious land segment; recommends using digital elevation

data to get this — check change in elevation of pixels in a transect perpen-
dicular to stream, divide by distance between centers of pixels — make
multiple measurements and average; probably has little effect on hydrology
but may affect erosion; default = 0.036 to 0.55, depending on land use.
• KVARY: nonzero values cause seasonal variation in groundwater flow;
increasing the value should cause faster recession during wet months;
default is 0; recommends starting with 0 and adjusting if necessary.
• AGWRC: groundwater recession rate; default = 0.98; recommends finding
this through calibration; higher value causes slower recession; suggests
using higher values for forests.
• PETMAX: used only in snow processes.
• PETMIN: used only in snow processes.
• INFEXP: exponent in the infiltration equation; default = 2.0 and recom-
mends using the default value.
• INFILD: ratio of maximum infiltration rate in a pervious land segment,
IMAX, to average infiltration rate, IBAR; default = 2.0 and recommends
using the default value.
• DEEPFR: fraction of infiltrating water that goes into deep groundwater
storage and is lost from the watershed; default = 0.10; recommends
finding value through calibration; higher value causes less stream flow
overall.
• BASETP: the fraction of a pervious land segment area that has vegetation
able to transpire water directly from groundwater (i.e., riparian or marsh
land vegetation); default = 0.02; recommends calculating this based on
area that is riparian or marsh land vegetation.
• CEPSC: rainfall (in.) intercepted by vegetation; default = 0.10; recom-
mends different values depending on land cover.
• UZSN: upper zone nominal soil water storage (in.); recommends different
values depending on slope, vegetation, and depression storage; overall
rule of thumb is 0.10 LZSN; default = 1.128 in.

• NSUR: n in Manning’s equation for overland flow; larger values of n
indicate a rougher surface and slower flow; default is 0.20; probably has
little effect on water flow but may affect erosion.
• INTFW: interflow parameter; increasing interflow value delays water get-
ting to the stream (otherwise it would become overland flow), so it lowers
the hydrograph peak and spreads the curve out; default is 0.75; recom-
mends using calibration to find value.
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 195
• IRC: interflow recession rate, analogous to groundwater recession rate;
increasing value causes the hydrograph to spread out and decreases peak
value; default = 0.50.
• LZETP: index to lower zone ET related to root distribution; varies between
0 and 1 with 1 representing maximum potential for plant uptake; gives
typical ranges for different types of vegetation; default is monthly varying
values from 0.2 in winter months to 0.4 in summer months.
There are a few parameters associated with the ILSs. Unless the impervious land
area is a large portion of the modeled watershed, these parameters will not have
much effect on model predictions. The most important factor is what percentage of
urban areas is assumed to be impervious; the default is 50%. A few parameters are
also associated with the reaches, which will have little effect on stream flow, although
they may be important for sediment and P transport.
Another source of information on hydrological as well as water-quality param-
eters is HSPF Parameter (HSPFParm) (Donigian et al. 1999). This is a database of
parameter values that have been used by experienced users in 45 HSPF model runs
in 14 states (available at />8.3 MODELING OF PHOSPHORUS
HSPF has a specific routine for modeling P. The module matrix in U.S. Environ-
mental Protection Agency (2004b) shows which modules need to be activated in the
pervious land, impervious land, and reach segments to model P:
• PERLND: activate PWATER, SEDMNT, MSTLAY, and PHOS

• IMPLND: none
• RCHRES: for inorganic P activate HYDR, ADCALC, SEDTRN, OXRX,
and NUTRX for organic P add PLANK
A surface zone, as well as the upper soil, lower soil, and groundwater zones, is
considered. A flow diagram for the pervious land portion of the P routine is shown
in Figure 8.2. Soil P is in organic, soluble, and adsorbed pools. Phosphate is adsorbed
and desorbed using either first-order kinetics (i.e., subroutine FIRORD) or instan-
taneous adsorption using a Freundlich isotherm (i.e., subroutine SV). This chapter
covers only the instantaneous approach. A Freundlich isotherm from the user’s
manual is shown in Figure 8.3. On the y axis, X is the P adsorbed in parts per million
of soil (mg of P per kg of soil), and on the x axis, C is the P in solution in parts
per million of solution (mg of P per L of solution). The y axis intercept of curve 1
and curve 2 is XFIX, the amount of P permanently adsorbed (mg of P per kg of
soil). CMAX is the maximum equilibrium concentration of P in soil solution, and
XMAX is the corresponding maximum adsorbed concentration of P. Adsorbed P is
described by the following equation:
(8.5)XKC XFIX
N
=⋅ +1
1
1
© 2007 by Taylor & Francis Group, LLC
196 Modeling Phosphorus in the Environment
FIGURE 8.2 Flow diagram for P reactions. (Redrawn from B.R. Bicknell, J.C. Imhoff, J.L.
Kittle, Jr., T.H. Jobs, and A.S. Donigian, Jr., Hydrological Simulation Program-Fortran:
HSPF Version 12 User’s Manual, Mountain View, CA, Aqua Terra Consultants, 2001. With
permission.)
FIGURE 8.3 Freundlich adsorption isotherm. (Redrawn from B.R. Bicknell, J.C. Imhoff, J.L.
Kittle, Jr., T.H. Jobs, and A.S. Donigian, Jr., Hydrological Simulation Program-Fortran:
HSPF Version 12 User’s Manual, Mountain View, CA, Aqua Terra Consultants, 2001. With

permission.)
Atmospheric
deposition
Atmospheric
deposition
ORGP
Organic
phosphorus
Organic
phosphorus
mineral-
ization
Adsorp-
tion of
phosphate
P4AD
Phosphorus
adsorbed
Desorp-
tion of
phosphate
P4SU
Phosphate in
solution
PLTP
Plant
phosphorus
Plant
uptake of
phosphorus

Phosphate
immobili-
zation
C, ppm
CMAX
Curve 1
Curve 2
XMAX
XJCT
X, ppm
XDIF
XFIX
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 197
where N1 is the Fruendlich exponent (N1 = 1 is a linear isotherm) and K1 is the
Fruendlich distribution coefficient (units of L per kg when N1 = 1). These parameters
— XFIX, N1, K1, and CMAX — must be supplied for surface, upper, lower, and
groundwater zones.
P mineralization and immobilization are modeled using first-order kinetics with
temperature corrections. The routine requires values for the mineralization rate,
KMP, and immobilization rate, KIMP, at 35°C in units of inverse days or hours,
depending on the time step. These parameters must also be specified for each zone.
The temperature correction equation for P mineralization takes the form
(8.6)
where TMP is the temperature (°C) in the zone, and TH is the correction coefficient
(typically 1.06). There is a similar equation for temperature correction of the immo-
bilization rate. Soil temperature is modeled by HSPF.
Plant uptake is based on a first-order rate or a yield approach. In the first-order
rate approach, for each zone the plant uptake rate parameter (in units of inverse
time) is _KPLP where the underlined space is S, U, L, or K, representing the surface,

upper, lower, and active groundwater zones. After correction for temperature, the
uptake rate takes the form _KPLPK (in units of inverse time). Plant uptake occurs
from the soluble P pool (Figure 8.2). The amount of plant uptake each day is
calculated as the rate times the mass of P in the soluble pool in each zone. The
temperature correction equation takes the same form as Equation 8.6.
The yield approach to plant uptake of P is designed to be less sensitive to soil
nutrient levels and nutrient application rates than the first-order rate option. It allows
crop needs to be satisfied, subject to nutrient and moisture availability, without being
affected by soil nutrient level. In this method, a total annual target is specified by
the user and is then divided into monthly targets during the crop growing season.
The target is further divided into the four soil layers.
Soluble P can percolate down through the soil zones, which requires use of the
MSTLAY module. In the PWATER module, which is used for general hydrology,
some moisture that infiltrates can reach the groundwater in a single time step —
that is, a day or an hour. This has little effect on hydrology, but it is not realistic for
P in many cases. The MSTLAY module takes the fluxes and storages computed in
PWATER and adapts them for runoff, interflow, and percolation through the soil
layers. The revised storages, in inches of water, are also expressed in units of mass
per area units for use in the adsorption and desorption calculations. Percolation
occurs from the surface layer through each of the underlying layers. Percolation of
P from the surface layer to the upper soil zone is described by the following equation:
(8.7)
where SDOWN is the amount of water percolating down (in.), SMST is the amount
of water stored in the surface layer (in.), SLMPF is an arbitrary reduction factor
(< 1), and FSP is the fraction of the soluble P in the surface zone that percolates
(between 0 and 1).
KMPK KMP TH
TMP
=⋅
−35

FSP SLMPF
SDOWN
SMST
=⋅
© 2007 by Taylor & Francis Group, LLC
198 Modeling Phosphorus in the Environment
Percolation of P from the upper zone to the lower soil zone is described by the
following equation:
(8.8)
where ULPF is the factor for retarding percolation (since this variable is in the
denominator, it must be > 1 to cause retardation), UDOWN is the amount of water
percolating down (in.), UMST is the moisture storage (in.), and FUP is the fraction
of the soluble P in the upper zone that percolates (between 0 and 1). There is a
similar equation for percolation from the lower zone to ground water storages.
The surface layer can lose P in surface runoff. Soluble P enters runoff directly
and is adsorbed, and organic P can be removed with sediment. The concentration
of soluble P in runoff is assumed to be the same as the concentration in the surface
layer. Particulate P is removed from the surface layer in proportion to the fraction
of the surface soil layer removed by erosion, although the mass of soil in the surface
layer is a parameter value that does not vary even when material is removed. As
such, an enrichment ratio accounting for the fact that most of the P lost in erosion
is adsorbed to the clay-size fraction (Sharpley 1985) is not employed.
Phosphorus can be added to the system as organic or adsorbed P through
atmospheric deposition or through the special actions block where fertilizer and
manure applications are described. The special actions block is a table of annual or
monthly inputs.
Many processes can be modeled for P in reaches. Most of these occur in the
NUTRX module. They include longitudinal advection of dissolved P, benthal release
of dissolved P, adsorption and desorption of P to suspended sediment in the water
column using a linear adsorption coefficient ADPM(J), which varies for different

sediment size fractions J, and desorption and scour and longitudinal advection of
adsorbed P. In the PLANK module, sources and sinks of P include uptake by
phytoplankton or benthic algae and respiration and inorganic excretion by zooplank-
ton. Atmospheric deposition is also considered.
No guidance document exists — such as the one for selecting values for hydro-
logical parameters — for selecting values for P parameters. The primary parameters
for modeling P are as follows:
• SKPLP, UKPLP: P plant uptake parameter for surface zone, upper zone
• THPLP, THDSP: temperature correction factor for plant uptake, desorption
• KIMP: first-order immobilization rate constant (day
−1
)
• KMP: first-order mineralization rate constant (day
−1
)
• CMAX: maximum equilibrium concentration of P in soil solution (mg L
−1
)
• XFIX: concentration of P permanently adsorbed to soil (mg L
−1
)
• K1: Freundlich distribution coefficient
• N1: Freundlich exponent
• ORGP: initial P storage in each layer for organic P
• P4AD: initial P storage in each layer for adsorbed P
• P4SU: initial P storage in each layer for solution P
FUP
UZS
UZSN ULPF
UDOWN

UMST
=
⋅
⋅
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 199
• PLTP: initial P storage in each layer for plant P
• IP4SU: initial P storage in upper layer for interflow pool
• SLMPF: factor for reducing P percolation from the surface to the upper
soil zone; values < 1 cause a reduction; default = 0.5
• ULPF: factor for reducing P percolation from the upper to the lower soil
zone; values > 1 cause a reduction; default = 2.0
• LLPF: factor for reducing P percolation from the lower soil zone to
groundwater; values > 1 cause a reduction; default = 2.0
• BRCON(I): benthal release rate of P (mg m
−2
per time interval) for aerobic
and anaerobic conditions
• ADPM(J): P adsorption K
d
for suspended sand, silt, and clay fractions
• BNUT: constant concentration of P on bed sediments (mg P mg
−1
sediment)
• CMMP: orthophosphorus Michaelis-Menten constant for P-limited algal
growth (mg P L
−1
); default is 0.015 mg L
−1
8.4 MODELING OF SEDIMENT

In most cases, to model P transport the movement of sediment will also have to be
modeled. Erosion processes that take place on each PLS are described in the SEDMNT
module. Many of the routines are taken from the Agricultural Research Model
(ARM) (Donigian et al. 1977). SLSED represents external lateral input from an
upslope land segment that can be input as a time series by the user. NSVI is any net
external additions or removals of sediment caused by human activities or wind. DET
represents soil that is detached by rainfall and enters detached sediment storage.
AFFIX represents the opposite process: detached sediment that reattaches to soil,
which occurs on days when it does not rain. Once soil is in the detached sediment
storage, it can be washed off in WSSD if the transport capacity is sufficient. Also,
soil can be lost through erosion by scouring (i.e., gully erosion) without being
detached by rainfall splash, which is represented by SCRSD. The total soil lost by
washoff and scouring is SOSED.
The equation for detachment is
(8.9)
where DET is the detached sediment (tons per acre per time interval), DELT60 is
the number of hours per interval (dividing by 60 min), CR is the fraction of land
covered by snow (SNOCOV) and vegetation (COVER), SMPF is the supporting
management factor, KRER is the detachment coefficient dependent on soil properties,
RAIN is the rainfall (in. per time interval), and JRER is the detachment exponent
dependent on soil properties. To simulate reattachment on days when there is no
rainfall, the detached soil sediment storage is decreased by multiplying it by the
factor (1.0 – AFFIX) where AFFIX is a parameter.
DET DELT CRSMPFKRER
RAIN
DELT
=⋅−⋅⋅⋅




60 1 0
60
(. )



JRER
© 2007 by Taylor & Francis Group, LLC
200 Modeling Phosphorus in the Environment
The equation for transport capacity is
(8.10)
where STCAP is the transport capacity (tons acre
−1
time
−1
), KSER is the coefficient
for transport of detached sediment, SURS is the surface detention storage of water
from water modeling routines (in.), and SURO is the surface runoff (in. per time
interval). If the transport capacity is less than the detachment rate, then erosion is
limited by the transport capacity and vice versa.
The equation for scouring is
(8.11)
where SCRSD is the scour of undetached soil (tons acre
−1
time
−1
), KGER is the
coefficient for scour, and JGER is the exponent for scour.
Sediment losses from ILSs are also modeled. These are considered solids that
wash off impervious surfaces, such as dust deposited and then washed off in storms.

The model uses build-up and wash-off curves for ILSs — terminology that comes
from modeling urban areas. Build-up of solids is described by the following equation:
(8.12)
where SLDS is the solids in storage at the end of the day (tons acre
−1
), SLDSS is
the solids in storage at the beginning of the day (tons acre
−1
), ACCSDP is the
accumulation rate of solids storage (tons acre
−1
time
−1
), and REMSDP is the fraction
of solids removed each day, by wind or street sweeping. If no runoff occurs, solids
will build up and approach an asymptote, which is ACCSDP divided by REMSDP.
Wash off of solids is described by the following equation:
(8.13)
where STCAP is the solids transport capacity (tons acre
−1
time
−1
), KEIM is the
coefficient for transport of solids, and JEIM is the exponent for transport of solids.
Sediment transport in the reaches is described by the SEDTRN module. Sus-
pended sediment storage exchanges with the bed load through deposition and scour-
ing. Three options exist for modeling reach sand transport: the Toffaleti method
(Toffaleti 1969), the Colby method (Colby 1964; Colby and Hembree 1955), and the
power function method. These three methods are used to calculate sand transportation
STCAP DELT KSER

SURS SURO
DELT
JSER
=⋅⋅
+






60
60
SCRSD
SURO
SURS SURO
DELT KGER
SURS SURO
DE
=
+
⋅⋅⋅
+
60
LLT
JGER
60







SLDS ACCSDP SLDSS REMSDP=+⋅−(. )10
STCAP DELTKEIM
SURS SURO
DELT
JEIM
=⋅⋅
+






60
60
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 201
capacity in the reach. If the transportation capacity is greater than the sand storage,
then sand is scoured from the stream bed. If the transportation capacity is less than
the storage, then deposition occurs. The transport of cohesive sediment (i.e., silt and
clay) is modeled using deposition and scouring based on shear stress exerted on the
bed surface. The Toffaleti method requires modeling temperature and specifying the
median bed sediment size, D
50
. The Colby method also uses D
50
.

The power function method is the simplest to describe, so some of the details
are shown here. It uses the following equation for sand transportation capacity:
(8.14)
where PSAND is the potential suspended sand load (mg L
−1
), KSAND is the sand-
load suspension coefficient, AVVELE is the average velocity of stream reach (ft sec
−1
),
and EXPSND is the exponent. Sand is scoured or deposited depending on whether
the capacity to transport is less than or greater than the current sand suspended load.
Silt and clay scouring are treated differently because they resist scouring due to
cohesion. The rate of scour of silt and clay sediment is described by the following
equation:
(8.15)
where S is the rate at which sediment is scoured from the bed (mass area
−1
time
−1
),
M is the erodibility coefficient (kg
−2
time
−1
), TAU is shear stress, (lb ft
−2
or kg m
−2
),
and TAUCS is critical shear stress for detachment of bed sediment (lb ft

−2
or kg m
−2
).
There is no scouring as long as TAU < TAUCS. The rate of deposition of suspended
silt and clay is described by the following equation:
(8.16)
where D is the rate at which sediment settles out of suspension (mass area
-1
time
−1
), W
is settling velocity in still water (in. sec
−1
), CONC is the concentration of suspended
silt or clay (mass volume
−1
), and TAUCD is critical shear stress below which sediment
deposits (lb ft
−2
or kg m
−2
).
There is a new technical note selecting values for sediment parameters (U.S.
Environmental Protection Agency 2006). Donigian and Love (2003) can also be
used as a guidance document. The primary parameters for modeling sediment are
as follows:
• SMPF: supporting management practice factor; a table shows values for
alternate practices and slopes in U.S. Environmental Protection Agency
(2006).

• KRER: splash detachment soil coefficient; default = 0.14; ARM manual
(Donnigian et al. 1977) says that this is equal to the product of the USLE
factors K*P where K is soil erodibility and P is the practice factor (usually 1)
PSAND KSAND AVVELE
EXPSND
=⋅
SM
TAU
TAUCS
=⋅ −






10.
DWCONC
TAU
TAUCD
=⋅ ⋅−






1
© 2007 by Taylor & Francis Group, LLC
202 Modeling Phosphorus in the Environment

• JRER: splash detachment exponent; default = 2.0; HSPFParm runs all use 2.0
• AFFIX: fraction by which detached sediment storage decreases each day
due to reattachment; default = 0.03; HSPFParm runs used 0.01 for all
land uses except forest, which had a value of 0.002
• COVER: fraction of land surface shielded from rainfall by vegetation or
mulch; default = 0.88; HSPFParm runs use monthly values that vary by
land use (e.g., forest > urban)
• NVSI: deposition or wind removal of sediment from pervious land; default = 0
• KSER: coefficient in sediment transport; default = 0.10; HSPFParm varies
values from 0.04 to 8.30
• JSER: transport exponent; U.S. Environmental Protection Agency (2006)
recommends 2.5
• KGER: scour or gully erosion coefficient; default = 0.01; HSPFParm runs
use a range of values
• JGER: scouring exponent; default = 1.0; HSPFParm runs all use 1.0
• BEDWID: width (ft) of bed; default = 16 ft
• BEDWRN: depth (ft) of bed sediment; when exceeded produces a warning
message; default = 100 ft
• POR: porosity of bed; default = 0.50; HSPFParm runs varied from 0.6 to 0.8
• DB50: median diameter of bed sediment used with Tofaleti and Colby
methods
• KSAND: coefficient in the sand transport equation; default = 0.10; HSPFP-
arm runs use values 0.001 to 0.1
• EXPSND: exponent in sand transport equation; default = 3.92; HSPFParm
runs use values 1.0 to 6.1; U.S. Environmental Protection Agency (2006)
recommends a starting value of 2.0
• W: fall velocity of sand in still water (in. sec
−1
); default = 0.05; one could
use Stoke’s Law for this

• D: effective diameter of silt–clay particles (in.); default = 0.001; HSPFP-
arm runs used 0.0001 for silt and 0.0004 for clay
• RHO: density of silt–clay particle (gm cm
−3
); default = 3.0; HSPFParm
runs use 2.4 for silt and 2.0 to 2.4 for clay
• TAUCD: critical shear stress for silt–clay deposition (lb ft
−2
); default =
0.10; HSPFParm runs use 0.10 to 0.13 for silt and 0.0002 to 0.12 for clay
• TAUCS: critical shear stress for silt–clay scouring (lb ft
−2
); default = 0.30;
HSPFParm runs use values 0.002 to 2.15
• M: erodibility coefficient for silt–clay (lb ft
−2
day
−1
); default = 0.90;
HSPFParm runs use the same values of 0.0075 to 0.40 for silt and clay
8.5 CALIBRATION
Most studies have used manual calibration of HSPF, often aided by the use of an
expert system tool called HSPEXP, which is described in BASINS Technical Note
#5 (U.S. Environmental Protection Agency 1999). HSPEXP offers advice to the
modeler on parameter changes that can improve the calibration of flow. No guidance
is provided for water-quality calibration. Recent studies such as Doherty and Johnston
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 203
(2003) have performed an autocalibration of HSPF using the Parameter Estimation
(PEST) software developed by Doherty (2004). A limited implementation of auto-

calibration using PEST with HSPF is available in BASINS 3.1.
8.6 CASE STUDY: UPPER ETOWAH RIVER WATERSHED
This chapter illustrates the use of HSPF to predict watershed-scale P transport with
a case study on the Etowah River in the Piedmont region of Georgia. The Etowah
River is the main tributary of Lake Allatoona, which has a Total Maximum Daily
Load (TMDL) limit for P. The objective was to calibrate HSPF using data from the
USGS gauging station (#02392000) at Canton, Georgia (Figure 8.4). The USGS
data contained measurements of daily flow, water temperature, dissolved oxygen,
turbidity, and total P. A study conducted by Kennesaw State University at a sampling
site nearby measured suspended sediment concentration (SSC) and turbidity for several
years (Rose 1999). These data were used to develop a relationship between SSC and
turbidity, and then the USGS turbidity observations were converted into SSC. From
the Canton USGS station, the Upper Etowah River watershed was delineated into
FIGURE 8.4 Upper Etowah River basin with the solid dot indicating the watershed outlet at
Canton, Georgia. The gray lines indicate the stream network, and the dark lines indicate the
watershed and sub-basin boundaries.
Urban
Barren
Transitional
Cropland
Pasture
Forest
Grass Land
Water
10 0 10 20 Kilometers
N
© 2007 by Taylor & Francis Group, LLC
204 Modeling Phosphorus in the Environment
nine sub-basins. The area of watershed was 161,557 ha. The HSPF land-use classes
and area percentages were as follows: forest (88.6%), pasture (7.9%), urban (2.2%),

row crop (0.9%), barren land (0.07%), water (0.04%), and wetland (0.006%). A
threshold for land-use categories was set such that only forest, pasture, and urban
land uses were modeled.
The case study’s approach to finding parameter values was to use soils
information — following the guidance in U.S. Environmental Protection Agency
(2004c) — and HSPFParm values for simulations in the Piedmont region of Virginia
to find initial values and then to use PEST for autocalibration. Even though the values
are adjusted by PEST — within limits set by the user — it is important that the initial
values be good estimates of the final values since PEST uses a local optimization
method.
STATSGO soil data were used, and there were seven mapping units in the study’s
delineated watershed. The soils were primarily hydrologic group B (moderate infil-
tration rate) and C (low infiltration rate). Three parameters were differentiated by
land use: INFILT, CEPSC, and LZETP. All the other parameters were lumped across
land uses. The soils data were used to find area-weighted average values for perme-
ability of the first layer; this was used to get an initial value for INFILT for forest
land use. The assumption was made for the study that INFILT for pasture would be
70% of the forest value and that INFLT for urban would be 50% of the forest value
based on simulations in HSPFParm. An area-weighted average of the available soil
water content in the first layer from the STATSGO database, multiplied by the depth
of the first layer, was used as an initial value for UZSN. A similar approach using
the available soil water content and depths in all the remaining soil layers was used
to calculate an initial value for LZSN.
The final calibrated predicted daily flow is compared with the observed flow at
Canton, Georgia, for the period 1983 through 1991 in Figure 8.5. A log axis was
used for flow to better show low flows. Overall, there was very good agreement
between the model predictions and the observations except during very low flow
periods, such as in Fall 1987.
Donigian and Love (2003) suggested that sediment calibration of watershed
models should consist of two stages: sediment erosion (loads from landscape) cal-

ibration and in-stream sediment transport calibration. Donigian and Love (2002,
Table 2) provide guidelines for choosing HSPF sediment-related parameters in the
SEDMNT and SOLIDS modules. Donigian and Love (2003, Table 1) show typical
ranges of expected erosion rates for different land covers. For forest and pasture,
the typical ranges of erosion rates are 0.05 to 0.4 and 0.3 to 1.5 tons ac
−1
yr
−1
(0.11
to 0.9 and 0.7 to 3.4 mg ha
−1
yr
−1
). A few model parameters in SEDMNT (IMPLND
was ignored) were adjusted accordingly so that the predicted sediment loads from
forest and pasture were within the ranges. Donigian and Love (2003) suggest that
the fractions of sand, silt, and clay entering a model reach should reflect the relative
percentage of the surface material (i.e., sand, silt, clay) available for erosion in the
surrounding watershed but also should include an enrichment factor of silt and clay
to represent the likelihood of these finer particles reaching the channel. The case
study’s analysis based on the STATSGO database showed that the topsoil percentage
of sand, silt, and clay in the Etowah watershed are 48, 30, and 22%, respectively.
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 205
These percentages were used as starting values since the degree of enrichment was
not known.
The final calibrated predicted total suspended solids (TSS) and observed TSS
and turbidity for 1983 through 1991 are shown in Figure 8.6. As is typically the
case, the measured data for contaminant concentrations are relatively sparse, con-
sisting of monthly values in most years, compared to the daily flow data. As a result,

it is difficult to know how well the model performs during storm events. At a
minimum, the model should predict storm concentrations at least as high as the
observed values, which was the case with the present simulations. Performance
during base flow is more readily assessed, and the model tended to underpredict
TSS somewhat under these conditions. According to Donigian and Love (2003, 18),
HSPF undersimulates low concentrations of suspended solids because the model
“employs a relatively gross channel representation, with long reach lengths, that
tends to eliminate localized turbulence and scour conditions.”
The case study assumed that Mehlich III soil test P could be used as an estimate
of the adsorbed P pool in the surface soil zone. Since the concentration of dissolved
P in runoff is the same as the concentration in the surface soil zone, Equation 8.5
describes the relationship between soil test P and dissolved P in runoff. Schroeder
et al. (2004) measured the concentration of dissolved reactive P (DRP) in runoff
from typical Piedmont soils using a rainfall simulator. Similar studies for soils in
other states are listed in Schroeder et al. (2004, Table 1). When DRP (mg L
−1
) was
plotted as a function of Mehlich III P (M3 in mg kg
−1
) in the soil from the 0- to 2-cm
FIGURE 8.5 Simulated and observed daily flow at Canton, Georgia, for 1983 through 1991.
Nash-Sutcliffe coefficient of efficiency is 0.79.
Observed
Simulated
Date
1983
Stream Discharge (cms)
1984 1985 1986 1987 1988 1989 1990 1991
1000
100

10
1
© 2007 by Taylor & Francis Group, LLC
206 Modeling Phosphorus in the Environment
depth, the linear equation DRP = 0.0017M3 + 0.15 (Schroeder et al. 2004, Table 4)
was the best fit to the data. The slope coefficient (0.0017 kg L
−1
) is the ratio of
dissolved P in runoff divided by soil test P. To get an initial estimate of the Freundlich
distribution coefficient, K1, in Equation 8.5, the inverse was used to obtain the value
of 588 L kg
−1
. Since the relationship was linear, the assumption was made that the
Freundlich exponent, N1, was unity and was not calibrated this parameter.
Estimates of the initial store of adsorbed P in the surface soil zone, SP4AD,
were obtained using data from the University of Georgia Agricultural and Environ-
mental Services Laboratories analysis of soil samples submitted by producers in the
Etowah basin counties. These data were not available before 1992, so averages were
used for the period 1992 to 2000. The lab measures Mehlich I soil test P (M1), so
the case study used a relationship between M1 and M3 (M1 = 0.72 × M3 − 1.71)
reported by Shuman et al. (1988) to convert the M1 to M3. For pastures, the average
M3 soil test P in the three counties that dominate the watershed was 126.1 mg kg
−1
(based on 1804 samples). This was converted to pounds per acre (252.1) as the initial
estimate of P4AD in pasture. For forest land use, there were very few samples from
the counties in the basin, so an average was computed for all Piedmont counties
(based on 167 samples). The average M3 concentration was 12 mg kg
−1
or 24.1 lb per
acre as the initial estimate of P4AD in forest. HSPFParm values were used for KIMP

and KMP.
It was assumed that all pasture land received an annual broiler manure appli-
cation of 2 mg ha
−1
yr
−1
in two field applications: late March and late October. It
also was assumed that the manure contained 1.6% P, 90% of which was inorganic
FIGURE 8.6 Simulated daily total suspended solids (TSS) and observed TSS and turbidity
at Canton, Georgia, for 1983 through 1991.
Simulated TSS
Observed TSS
Observed Turbidity
Date
1983
Total Suspended Solids (mg/L), Turbidity (NTU)
1984 1985 1986 1987 1988 1989 1990 1991
1000
800
600
400
200
0
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 207
P based on analysis of broiler manure samples in Georgia conducted by the present
authors. Plant uptake of P was modeled using the yield-based option. The annual
P uptake by forest and pasture was set to 5.6 kg ha
−1
and 28 kg ha

−1
(Cabrera 2005;
Ducnuigeen et al. 1997). Monthly fractions of total annual P uptake and the
fractions of total annual P uptake from soil layers were set according to HSPFParm.
It was assumed that 80% of P uptake was from the upper soil layer. Parameters
were not adjusted in the ILSs, assuming that urban land use would have little
effect on water quality in the study’s watershed. For calibration of P, the total P
concentration from the PLANK module was compared with the observed total P
concentration.
The Discharge Monitoring Reports (DMR) for the period 1998 to 2003 — earlier
dates were not available — were used to determine average point source P loads. There
were two major permitted point sources in the watershed above Canton: the Jasper
Water Pollution Control Plant (average measured effluent discharge of 1668 m
3
day
−1
)
and the Pilgrim’s Pride Poultry Processing Plant (average measured effluent discharge
of 3335 m
3
day
−1
). There were 13 minor permitted point sources in the watershed with
average measured effluent discharge ranging from 9.5 to 758 m
3
day
−1
. Phosphorus
loads were reported in the DMR for the Pilgrim’s Pride Plant and for one of the minor
point sources but not for any of the other point sources. For the point sources

without DMR P loads, the study assumed a P concentration of 1 mg L
−1
because
this is typically the permitted concentration for point sources, although there is
no way of knowing what the actual concentration was. Within each sub-basin with
one or more point sources, a point source was added with the cumulative effluent
discharge and P load.
Before automatic calibration, PEST was used to perform a sensitivity analysis
for flow, sediment, and P. The results are shown in Table 8.1. The most flow-sensitive
parameter by far was the groundwater recession constant, which controls the shape
of the stream hydrograph. Other flow-sensitive parameters were related to upper and
lower zone soil water storage, the interflow recession constant, and evapotranspira-
tion. The groundwater recession constant was also the most sensitive parameter by
far for predicting sediment. The next most sensitive parameters were a group of
sediment parameters related to stream transport. The most sensitive parameter for
predicting total P was the y-intercept value for the Freundlich P asdorption isotherm
in topsoil, XFIX. Other parameters related to P adsorption were also sensitive: N1
and K1. The second, fourth, and fifth most sensitive parameters for predicting P
were parameters related to flow that affect runoff, INFILT_F and INFILT_P, and
evapotranspiration, LZETP_P. Initial adsorbed P in pasture was a sensitive variable,
and so was the variable controlling benthic release of P in streams. Oddly, P
prediction was not sensitive to any of the sediment parameters. This may be due to
the fact that there was little erosion from pasture and forest land uses and that most
of the P lost is soluble P.
The sensitivity analysis was used to decide which parameters to use in auto
calibration with PEST. The study first calibrated for flow and then for sediment and
P concentrations in the Etowah River at Canton, Georgia. The final calibrated
predicted and observed total P 1983 through 1991 are shown in Figure 8.7. As was
the case with TSS, the observed P data set was sparse. Model predictions matched
© 2007 by Taylor & Francis Group, LLC

208 Modeling Phosphorus in the Environment
TABLE 8.1
Top 15 Parameters in Terms of Sensitivity to Flow, Sediment, and P
Parameter
Sensitivity
to Flow Parameter
Sensitivity
to Sediment Parameter
Sensitivity
to P
AGWRC 8.647√ AGWRC 1.4283 XFIX 0.2260√
UZSN 0.0625√ EXPSND 0.2347√ INFILT_F 0.0879
LZETP_F 0.0589√ TAUCSS 0.1996√ N1 0.0767
IRC 0.0575√ TAUCSC 0.1996 INFILT_P 0.0752
INFILT_F 0.0494√ KSAND 0.0645 LZETP_P 0.0605
LZSN 0.0377√ TAUCDS 0.0344√ K1 0.0605√
LZETP_P 0.0298√ M 0.0197√ IRC 0.0538
CEPSC_F 0.0182 LZETP_F 0.0180 INTFW 0.0367
INTFW 0.0165 LZSN 0.0156 SPFAD_P 0.0303√
INFILT_P 0.0161 UZSN 0.0138 AGWRC 0.0276
DEEPFR 0.0079 IRC 0.0118 BRPO4 0.0238√
LZETP_U 0.0052 INFILT_F 0.0094 UZSN 0.0191
LSUR 0.0040 LZETP_P 0.0085 SLMPF 0.0173√
NSUR 0.0040 INFILT_P 0.0068 KMP 0.0170√
INFILTU 0.0038 INFILT_U 0.0053 LZETP_F 0.0159
Notes: A check mark indicates that variables were used in autocalibration.
FIGURE 8.7 Simulated daily total P and observed total P at Canton, Georgia, for 1983
through 1991.
Simulated
Observed

Date
1983
Total Phosphorus (P mg/L)
1984 1985 1986 1987 1988 1989 1990 1991
2
1.6
1.2
0.8
0.4
0
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 209
or exceeded observed P concentrations during storm events, as one would hope.
Predicted storm flow peak concentrations were usually in the range of 0.3 to 1.5 mg
L
−1
. Base-flow concentrations of P were simulated accurately as well, usually less
than 0.1 mg L
−1
.
The values of the most important variables and initial values, bounds for
calibration, and final calibrated values of all parameters are shown in Table 8.2.
For the flow-related variables (PWATER module), autocalibration resulted in
changes that caused more runoff than the initial parameter values (decrease in the
infiltration parameter and increases in the upper and lower soil water zone storage
parameters). Most of the changes in the sediment parameters caused greater
suspended sediment during base flow (increase in sand transport exponent and
decreases in critical shear stress threshold values). Changes in the P-related param-
eters during autocalibration were a mixture of variables that increased runoff P
(i.e., lower initial adsorbed P in pasture land use) and decreased runoff P (i.e.,

decrease in the Freundlich partition coefficient).
The predicted average P loads from point and nonpoint sources for the period
of calibration are shown in Table 8.3. Point sources accounted for only 3% of
TABLE 8.2
Initial Values, Bounds, and Final Values of HSPF Parameters
Calibrated in the Upper Etowah River Basin Case Study
Parameter
a
Module Initial Value Bounds Calibrated Value
INFILT_F PWATER 0.29 0.05–0.4 0.1734
LZETP_F 0.7 0.6–0.8 0.8
LZETP_P 0.5 0.4–0.6 0.57
LZETP_U 0.3 0.1–0.4 0.34
LZSN 5.29 2–15 9.23
UZSN 0.79 0.05–2 1.38
AGWRC 0.98 0.85–0.999 0.986
IRC 0.5 0.3–0.85 0.566
EXPSND SEDTRN 3.92 1–6.1 4.51
TAUCD_S 0.1 0.1–0.13 0.1
TAUCD_C 0.2 0.0002–0.12 0.1
TAUCS_S 0.4 0.002–2.15 0.316
TAUCS_C 0.4 0.002–2.15 0.316
M 0.2 0.05–0.4 0.05
SLMPF MSTLAY 0.9 0.001–1 0.925
KMP PHOS 0.04 0.001–1 0.047
XFIX 10 1–100 1
K1 588 1–1000 525
P4AD_P 252 0.5–400 231
BRPO4 NUTRX 0.01 0.001–1 0.0093
a

Extensions _F, _P, and _U indicate forest, pasture, and urban land use, respec-
tively. Extensions _S and _C indicate silt and clay, respectively.
© 2007 by Taylor & Francis Group, LLC
210 Modeling Phosphorus in the Environment
the average annual P load at Canton, Georgia. Pasture land use accounted for
most (51%) of the average annual load despite the fact that the number of acres
in pasture was small compared to forest land use. The pasture load was high due
to the high specific yield of P per hectare of land (2.75 kg ha
−1
yr
−1
) compared
to forest land use (0.20 kg ha
−1
yr
−1
). The urban load was relatively small due to
low total land area.
8.7 COMPARING HSPF AND SWAT
As noted in the introduction, HSPF is one of two dynamic watershed-scale models
that are part of the BASINS software. The other model is SWAT. Since the present
authors have some experience in running both models, this section briefly compares
their approaches to modeling flow, sediment, and P. Both models are calibrated, so
at least some of the parameter values are adjusted. However, estimates of parameter
values are needed to minimize the number of calibrated parameters if manual
calibration is used and to make it more likely that a global optimum is found if
autocalibration is used. In general, the SWAT parameters for flow tend to be con-
ventional soil parameters (e.g., available water content, saturated hydraulic conduc-
tivity, soil drainage class), whereas HSPF parameters for flow are unique to the
model (e.g., INFILT, INFEXP, INTFW, UZSN). As a result, the SWAT soil parameters

can be obtained from soil survey data. In fact, the BASINS version of SWAT can
read some of these parameters directly from a soil survey database. The only way
to obtain initial values for many HSPF parameters is to go to the HSPFParm database.
HSPF models infiltration using a unique equation (Equation 8.1) and includes
interflow (Equation 8.2). SWAT uses the curve number approach for runoff, which
can only be used on a daily time step. There is an option for using the Green-Ampt
equations for infiltration and an hourly time step in SWAT, but most users have not
exercised this option.
SWAT uses a Modified Universal Soil Loss Equation approach for estimating
erosion from uplands. HSPF uses a series of equations that estimates detachment
and transport capacities to model erosion. Again, SWAT uses more conventional
soil parameters, whereas the HSPF parameters are relatively unique and best
TABLE 8.3
Simulated Average P Loads from Different Land Uses
at Canton, Georgia, for the Period 1983 through 1991
Source
Land Use Specific Yield
Average Annual Load
Area (ha) (kg ha
−1
yr
−1
) (kg yr
−1
) (%)
Urban nonpoint 1,206 0.80 9,620 12
Forest nonpoint 143,276 0.20 28,000 34
Pasture
nonpoint
15,202 2.75 41,828 51

Point NA NA 2,532 3
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 211
determined from HSPFParm. Stream processes for sediment transport are described
in more detail in HSPF than in SWAT. SWAT does not include a sediment-settling
velocity, threshold shear-stress values, or different particles sizes of sediment. On
the surface, the soil P routines in both models appear similar, but there are some
important differences. SWAT uses three pools of inorganic P: labile, active, and
stable pools. HSPF uses two pools of inorganic P: phosphate in solution and
adsorbed P (Figure 8.2). SWAT implements the P routine developed in EPIC and
described in Jones et al. (1984) and Sharpley et al. (1984). In SWAT, it is clear that
labile P refers to anion exchange resin extractable P, but in HSPF it is not clear
whether phosphate in solution refers to dissolved P or some other form. In SWAT,
labile P is converted to soluble P in runoff using a phosphorus soil partitioning
coefficient, but in HSPF the concentration in runoff is the same as the concentration
of solution P in the surface layer. However, the relationship between P in solution
and sorbed P in the surface layer (Equation 8.5) used in HSPF can be interpreted
as a relationship between runoff P concentration and labile P in the topsoil, in which
case the Freundlich partitioning coefficient is similar to the phosphorus soil parti-
tioning coefficient in SWAT. An enrichment ratio is used to predict particulate P in
runoff in SWAT but not in HSPF. In-stream P processes are modeled in greater detail
in HSPF than in SWAT. SWAT models algal growth in streams and the movement
of P among soluble, organic, and algal P pools. Organic P can settle out, and
inorganic P can be released from the stream bed. HSPF includes these processes
and adds to this the interaction of P with suspended sediment and scouring of bed
sediment that contains P. HSPF also models phytoplankton uptake and release of P
and zooplankton release of P.
Saleh and Du (2004) compared SWAT and HSPF predictions of flow, sediment,
and nutrients (i.e., N, P) in the Upper North Bosque River watershed in Texas. They
found that HSPF did a better job of predicting daily flow than SWAT — based on

the Nash-Sutcliffe model efficiency — and attributed this to use of the curve number
approach in SWAT, which does not account for rainfall intensity. HSPF also did a
better job of predicting sediment, but SWAT was a better predictor of nutrient
loading. They attributed this to the ease of modeling agricultural practices such as
fertilizer and manure applications in SWAT.
Both models allow the user to simulate farming practices that occur on specific
dates such as tillage and fertilizer and manure applications, but it is considerably
easier to do in SWAT than in HSPF.
8.8 CONCLUSIONS
HSPF is a state-of-the-art, dynamic, watershed-scale model that is being used in
many projects modeling P dynamics and transport, most notably the Chesapeake
Bay Model. It is a semidistributed model that divides a watershed into sub-basins.
Within each sub-basin it models water, sediment, and P movement in land segments
representing the dominant land uses and in a stream-reach segment. The equations
describing infiltration and interflow in HSPF are unique to the model, and as a result
the parameters for these processes are not readily measured. U.S. Environmental
Protection Agency (2004c) is a good guide on selecting parameter values for
© 2007 by Taylor & Francis Group, LLC
212 Modeling Phosphorus in the Environment
modeling water movement in HSPF. Donigian and Love (2003) and U.S. Environ-
mental Protection Agency (2006) are helpful in finding parameters for modeling
sediment. HSPFParm is a database of parameter values that has been used by
experienced users in 45 HSPF model runs. A technical note is needed to guide users
in finding parameters for modeling P with HSPF.
As a case study, HSPF was used to model flow, sediment, and P in the Upper
Etowah River Basin; the study found good agreement between predictions and
observations after calibration for the period 1983 through 1991. The model showed
that nonpoint sources dominated the P load in the watershed and that the largest
source of P was pasture land use associated with poultry operations. The most
sensitive parameters for predicting P were those related to soil adsorption and initial

stores of soil adsorbed P, infiltration, and benthic release of P.
ACKNOWLEDGMENTS
This work was supported by a grant from the USDA-CSREES Water Quality grant
GEO-2003-04944 entitled “A Framework for Trading Phosphorus Credits in the
Lake Allatoona Watershed.”
REFERENCES
Arnold, J.G., R. Srinivasan, R.S. Muttiah, and J.R. Williams. 1998. Large-area hydrologic
modeling and assessment. Part I: model development. J. Am. Water Resour. Assoc.
34:73–89.
Bicknell, B.R., J.C. Imhoff, J.L. Kittle, Jr., T.H. Jobs, and A.S. Donigian, Jr. 2001. Hydro-
logical Simulation Program-Fortran: HSPF Version 12 User’s Manual. Mountain
View, CA: Aqua Terra Consultants.
Cabrera, M.L. 2005. Personal communication. University of Georgia, Athens.
Colby, B.R.1964. Practical computations of bed material discharge. J. Hydrol. Div. ASCE
90(HY2) 217–246.
Colby, B.R. and C.H. Hembree. 1955. Computation of total sediment discharge, Niobrara
River near Cody, Nebraska. U.S. Geological Survey Water Supply, Paper 1357.
Crawford, N.H. and R.K. Linsley. 1966. Digital simulation in hydrology: Standford water-
shed model IV. Technical Report 39, Department of Civil Engineering, Stanford
University, CA.
Doherty, J. 2004. PEST — Model-Independent Parameter Estimation User’s Manual, 5th ed.
Brisbane, Australia: Watermark Numerical Computing.
Doherty, J. and J.M. Johnston. 2003. Methodologies for calibration and predictive analysis
of a watershed model. J. Am. Water Resour. Assoc. 29:251–265.
Donigian, Jr., A.S., D.C. Beyerlein, H.H. Davis, Jr. and N.H. Crawford. 1977. Agricultural
Runoff Management (ARM) Model version II: refinement and testing. U.S. Environ-
mental Protection Agency, EPA 600/3-77-098, Environmental Research Laboratory,
Athens, GA.
Donigian, Jr., A.S., R.V. Chinnaswamy, A.S. Patwardhan, and R.M. Jacobson. 1996. Watershed
modeling of pollutant contributions and water quality in the LeSueur Basin of

Southern Minnesota. Pp. 109–111 in WATERSHED '96 — Moving Ahead Together.
Conference Proceedings, June 8–12, Baltimore, MD.
© 2007 by Taylor & Francis Group, LLC
Modeling Phosphorus with Hydrologic Simulation Program-Fortran 213
Donigian, Jr., A.S., and N.H. Crawford. 1976a. Modeling pesticides and nutrients on agricultural
lands. U.S. Environmental Protection Agency, EPA 600/2-7-76-043, Environmental
Research Laboratory, Athens, GA.
Donigian, Jr., A.S., and N.H. Crawford. 1976b. Modeling nonpoint pollution from the land
surface. U.S. Environmental Protection Agency, EPA 600/3-76-083, Environmental
Research Laboratory, Athens, GA.
Donigian, Jr., A.S., J.C. Imhoff, and J.L. Kittle, Jr. 1999. HSPFParm: an interactive database
of HSPF model parameters, version 1.0. U.S. Environmental Protection Agency, EPA-
823-r-99-004, Washington, D.C.
Donigian, Jr., A.S., and J.T. Love. 2002. The Connecticut watershed model — a tool for BMP
impact assessment in Connecticut. WEF-Watershed 2002, February 23-27, 2002. Ft.
Lauderdale, FL, WEF-2002 Specialty Conference Proceedings on CD-ROM.
Donigian, Jr., A.S. and J.T. Love. 2003. Sediment calibration procedures and guidelines for
watershed modeling. WEF TMDL 2003, November 16–19, 2003, Chicago, IL. WEF
Specialty Conference Proceedings on CD-ROM.
Ducnuigeen, J., K. Williard, and R.C. Steiner. 1997. Relative nutrient requirements of plants suitable
for riparian vegetated buffer strips. Interstate Commission on the Potomac River Basin,
Report Number 97-4, Virginia Department of Environmental Quality, Richmond, VA.
Fetter, C.W. 1988. Applied Hydrology, 2nd. ed. Columbus, OH: Merrill Publishing Company.
Hummel, P.R., J.L. Kittle, P.B. Duda, A. Patwardhan. 2003. Calibration of a watershed model
for metropolitan Atlanta. WEF TMDL 2003, November 16–19, 2003. Chicago, IL.
WEF Specialty Conference Proceedings on CD-ROM.
Hydrocomp, Inc. 1976. Hydrocomp Simulation Programming: Operations Manual, 2nd ed.
Palo Alto, CA: Hydrocomp, Inc.
Hydrocomp, Inc. 1977. Hydrocomp Water Quality Operations Manual. Palo Alto, CA:
Hydrocomp, Inc.

Jones, C.A., C.V. Cole, A.N. Sharpley, and J.R. Williams. 1984. A simplified soil and plant
phosphorus model. I: documentation. Soil Sci. Soc. Am. J. 48:800–805.
Linker, L.C., G.W. Shenk, P. Wang, K.J. Hopkins, and S. Pokharel. 2002. A short history of
Chesapeake Bay modeling and the next generation of airshed, watershed, and estuarine
models. Presented at the Watershed Conference, February 23–27, Fort Lauderdale, FL.
Neitsch, S.L., J.G. Arnold, J.R. Kiniry, J.R. Williams, and K.W. King. 2002. Soil and Water
Assessment Tool theoretical documentation. U.S. Department of Agriculture Agricul-
tural Research Service, Report 02-01, Grassland, Soil, and Water Research Laboratory,
Temple, TX.
Onishi, Y. and S.E. Wise. 1979. Mathematical Model, SERATRA, for Sediment-Contaminant
Transport in Rivers and its Application to Pesticide Transport in Four Mile and Wolf
Creeks in Iowa. Richland, WA: Battelle, Pacific Northwest Laboratories.
Rose, P. 1999. Lake Allatoona Phase I Diagnostic-Feasibility Study for 1992–1997. Kennesaw,
GA: A.L. Burruss Institute of Public Service. Kennesaw State University.
Saleh, A. and B. Du. 2004. Evaluation of SWAT and HSPF within BASINS program for the
Upper North Bosque River watershed in central Texas. Trans. ASAE 47(4):1039–1049.
Schroeder, P.D., D.E. Radcliffe, M.L. Cabrera, and C.D. Belew. 2004. Relationship between
soil test phosphorus and phosphorus in runoff: effects of soil series variability. J.
Environ. Qual. 33:1452–1463.
Sharpley, A.N. 1985. The selective erosion of plant nutrients in runoff. Soil Sci. Soc. Am. J.
49:1527–1534.
Sharpley, A.N., C.A. Jones, C. Gray, and C.V. Cole. 1984. A simplified soil and plant
phosphorus model. II: prediction of labile, organic, and sorbed phosphorus. Soil Sci.
Soc. Am. J. 48:805–809.
© 2007 by Taylor & Francis Group, LLC