Modeling phosphorus in the environment - Chapter 12 pps

The GWLF approach conceptualizes the watershed as a system of different land areas — hydrologic response units HRUs — that produce surface runoff and erosion and a single groundwater res

with the Generalized

Watershed Loading

Functions (GWLF) Model

Elliot M Schneiderman

New York City Department of Environmental Protection, Kingston, NY

CONTENTS

12.1 History of Model Development 277

12.2 Spatial and Temporal Resolution 279

12.3 Predicting Infiltration and Runoff 279

12.4 Predicting Phosphorus in Runoff 284

12.5 Predicting Phosphorus Leaching 285

12.6 Simulating Management and BMPs 285

12.7 Simulating In-Stream Processes 286

12.8 Example Simulations 287

12.8.1 Use of GWLF to Evaluate BMPs 287

12.8.2 Accuracy of GWLF: Comparison of Simulated to Measured Loads 289

12.8.3 Simulation of Runoff Volumes and Source Areas 292

12.9 Sensitivity Analysis 295

12.10 Availability of Model 295

References 296

12.1 HISTORY OF MODEL DEVELOPMENT

The Generalized Watershed Loading Functions (GWLF) model was originally developed at Cornell University by Douglas Haith and associates (Haith and Shoemaker 1987; Haith et al 1992) as “an engineering compromise between the empiricism of export coefficients and the complexity of chemical simulation models” (Haith et al., 1992, p.1) The GWLF approach conceptualizes the watershed as a system of different land areas — hydrologic response units (HRUs) — that produce surface runoff and erosion and a single groundwater reservoir that supplies base flow

Dissolved and suspended substances (i.e., nutrients, sediment)in stream flow areestimated at the watershed outlet by loading functions that empirically relatesubstance concentrations in runoff, sediment, and base flow to watershed- andHRU-specific characteristics The strength of this approach lies in its fairly robusthydrologic formulation of a daily water balance and in the ability to adjust loadingfunctions through calibration and for specific watershed conditions to an everincreasing body of knowledge and data on the factors that influence the export ofsubstances in stream flow from a watershed.

In addition to the original model there are currently several versions of GWLF inuse ArcView GWLF (AVGWLF) (Evans et al 2002) was developed by PennsylvaniaState Institutes of the Environment for Pennsylvania watersheds AVGWLF provides ageographic information systems (GIS) interface to GWLF, has a modified sedimentalgorithm for channel erosion, and incorporates best management practices (BMP)reduction factors Variable Source Loading Function (VSLF) Model (Schneiderman et

al 2002, 2006) was developed by the New York City Department of EnvironmentalProtection (NYC DEP) for the New York City water supply VSLF has a modifiedrunoff algorithm to account for saturation-excess runoff; adds optional snowmelt, evapo-transpiration (ET), and groundwater algorithms for tuning hydrologic simulation tovaried physiographic settings; modifies the sediment algorithm; adds BMP reductionfactors; utilizes Vensim visual modeling software (http://www.vensim.com) for trans-parent viewing of model structure and for viewing tables, graphs, and statistics for allmodel variables at daily, weekly, monthly, annual, and event time steps; and has built-

in model calibration and testing tools BasinSim (Dai et al 2000) is a Windows-basedversion of the original GWLF model developed at Virginia Institute for Marine Science.For research at the Choptank River Basin tributary of Chesapeake Bay, Lee et al (2000,2001) and Fisher et al (2006) converted GWLF to Visual Basic with an ArcView andArcMap GIS interface and added error analysis and adjustments for nonlinear agricul-tural land-use effects and hydric soils These versions of GWLF have incorporatedvarious different modifications from the original BASIC program, but all, including thevarious teaching tool versions, adhere to the basic water balance formulation and loadingfunction philosophy

The various versions of GWLF are commonly used to predict how stream flowand nutrient loads from a watershed are affected by land-use, watershed-manage-ment, and climatic conditions The U.S Environmental Protection Agency (EPA)has classified GWLF as a model of mid-range complexity that can be used fordeveloping Total Maximum Daily Load (TMDL) limits for impaired water bodies(U.S Environmental Protection Agency 1999) GWLF has been applied to theChoptank River Basin tributary of Chesapeake Bay (Lee et al 2000, 2006),including an application to historical land cover changes (Fisher et al 2006); theNew York City water supply watersheds (New York City Department of Environ-mental Protection 2005; Schneiderman et al 2002, 2006); and throughout Pennsylvania(http://www.avgwlf.psu.edu) A Web search at the time of this writing shows that

a version of GWLF is being used in at least 12 U.S states — Arizona, Georgia,Illinois, Iowa, Kansas, Michigan, Mississippi, North Carolina, Pennsylvania,New York, Utah, and Virginia — to meet EPA requirements for development ofTMDLs

12.2 SPATIAL AND TEMPORAL RESOLUTION

GWLF has flexibility in the spatial and temporal resolution of model output Thebasic time step for the hydrologic water balance calculations is daily, which can beaggregated up to larger time steps The original GWLF model aggregates hydrologicand water-quality output to a monthly representation, mainly because it does notaccount for drainage-area-based delays on flood peaks following storms Dailyoutput from GWLF will generally give too fast a response to rain events becausethe model computes a daily water balance but does not include routing VSLFincorporates a time delay for runoff at the watershed outlet using an exponentialdecay function that can be calibrated VSLF provides daily, weekly, monthly, annual,and event time step outputs, as desired

Spatial resolution depends on how a GWLF model application is set up InGWLF the watershed area is divided into HRUs, which are land areas that share asimilar hydrologic response to rain or snowmelt events and may not be contiguous.Runoff, erosion, and nutrients associated with runoff and erosion are explicitlytracked for each HRU and can be spatially mapped back to the HRU land areas.Fine resolution division of a watershed into many HRUs can produce model output

on a field scale Water, nutrient, and sediment loads are summed to provide shed-scale loading estimates as well

water-12.3 PREDICTING INFILTRATION AND RUNOFF

Runoff and infiltration are predicted in GWLF using the Soil Conservation Service(SCS) curve number (CN) method (Soil Conservation Service 1972) Daily runoff

depth Q is calculated by

(12.1)

where P (mm) is the depth of rain and snowmelt, Ia (mm) is the initial abstraction

of rain and snowmelt retained by the watershed prior to the beginning of runoff

generation, and S (mm) is a parameter that represents the potential maximum soil water retention when runoff begins Ia is estimated as an empirically derived fraction

of available storage (typically assumed to be 0.2 S).

Potential soil water retention, S, depends on the moisture status of the soil of the HRU and varies daily between a maximum Smax (mm) when the HRU soil is dry

and a minimum Smin (mm) when the HRU soil is wet Effective soil water retention

for average watershed moisture conditions Savg (mm) is calculated from the SCS CN

U.S Department of Agriculture (USDA) (Soil Conservation Service 1986) for ferent combinations of land use and soil hydrologic group Soil hydrologic groupsrank soils by their infiltration characteristics, and are used to qualify the propensity

dif-of an HRU to generate excess rundif-off

The upper and lower limits of S are estimated in relation to Savg, based on empiricalanalysis of rainfall and runoff data for experimental watersheds (Hawkins 1978):

and

The daily value of S is determined in the original GWLF model by the

watershed antecedent moisture (am) condition, determined by the sum of

precip-itation occurring during the previous five days (P5-day) S is set to Smax for the dry

condition (P5-day = 0) and then declines linearly to Savg and Smin as P5-day increases,

as given in Figure 12.1 The relationship of S to P5-day is different for the dormant

vs the growing season Breakpoint values (from Ogrosky and Mockus 1964) for

the dormant and growing season curves, respectively, in Figure 12.1 are am1 = 1.27

FIGURE 12.1 Variation in soil water retention parameter, S, as a function of 5-day antecedent

precipitation for growing and dormant seasons am1 and am2 represent breakpoint values

between different antecedent precipitation conditions.

and 3.56 cm, and am2 = 2.79 and 5.33 cm When snowmelt occurs, it is assumed

that the HRU soils are at their wettest condition; hence, S is set to Smin, irrespective

of P5-day

An alternative method for calculating S as a direct function of soil moisture

content is used in VSLF (Schneiderman et al 2006) Using the method of Arnold

et al (1998), S varies from storm to storm as

(12.5)

where SW is the average soil water content (cm3/cm3) and w1 and w2 are shape

coefficients The shape parameters w1 and w2 are calculated by

(12.6a)

and

(12.6b)

where FC is the amount of water in the soil at field capacity (cm3/cm3) and SAT is

the amount of water in the soil when saturated (cm3/cm3) When the top layer of thesoil is frozen, the available storage is modified by

(12.7)

where Sfrz is the available storage adjusted for frozen ground, S is the available storage for a given soil moisture content calculated with Equation 12.5, and R2frzx

is a parameter that adjusts for frozen ground conditions R2frzx is set to –0.000826

in the SWAT model but can be calibrated

Stormwater runoff is the primary mechanism for transporting soluble phosphorus (P)from the point where it accumulates on or near the ground surface to the stream andoutlet of the watershed Accurate model predictions of P loads and effects of water-shed management depend on realistic prediction of runoff source areas GWLF, likemany CN-based water-quality models, uses the SCS CN method in a way thatimplicitly assumes that infiltration excess is the runoff mechanism Each HRU in awatershed is defined by land use and a hydrologic soil group classification via a CNvalue that determines runoff response CN values for different land use andhydrologic soil group combinations are provided in tables compiled by the USDA

(e.g., Soil Conservation Service 1972, 1986) The hydrologic soil groups used toclassify HRUs are based on infiltration characteristics of soils (e.g., Natural

Resources Conservation Service 2003) and thus clearly assume infiltration excess

as the primary runoff-producing mechanism

The traditional infiltration-excess-based CN method for runoff estimation in GWLFlimits the original model’s utility to watersheds where infiltration excess is the dominantrunoff-generating mechanism In humid, well-vegetated areas with shallow soils, such

as in the northeastern U.S., infiltration excess does not explain observed storm runoffpatterns On shallow soils characterized by highly permeable topsoil underlain by adense subsoil or shallow water table, infiltration capacities are generally greater thanrainfall intensity, and storm runoff is usually generated by saturation excess on VSAs(Beven 2001; Dunne and Leopold 1978;Needelman et al 2004; Srinivasan et al 2002)

To improve the accuracy of runoff source area predictions in watersheds wheresaturation excess is the dominant runoff-generating mechanism, Schneiderman et al.(2006) created a new version of GWLF (VSLF) that simulates runoff from VSAs

In VSLF the watershed is subdivided into wetness index classes by mapping awetness index (e.g., the topographic index ln a/tan b of the TOPMODEL; Bevenand Kirkby 1979) and by defining discrete classes ordered along an axis of increasingavailable moisture storage Steenhuis et al (1995) and Schneiderman et al (2006)showed that the CN Equation 12.1, when interpreted as representing a saturationexcess runoff generation process, gives rise to a characteristic relative soil moisturedistribution that is invariant from storm to storm:

Runoff q (mm) at a given point location in the watershed is simply

q = P – Ia – se for P > σe+ Ia (12.9)Given the CN-based relative moisture distribution (Equation 12.8), runoff for

any point location along the As fraction of watershed area with lower local moisturestorage axis is calculated as

Runoff qi for a discrete wetness index class, bounded on one side by the

fraction of the watershed that has lower local moisture storage, As,i, and on theother side by the fraction of the watershed that has greater local moisture storage,

As,i+1, is given by

(2.11)

Runoff and nutrient loads from each wetness and land-use HRU are trackedseparately in the model A wetness index class may coincide with multiple landuses Whereas runoff depth within an index class in VSLF will be the same irre-spective of land use, the concentration of pollutant in runoff may vary by both landuse and index class Wetness index classes are thus subdivided by land use to defineHRUs with unique combinations of wetness class and land use

In the original GWLF model, runoff is calculated for each defined soil and landuse HRU using Equation 12.1 In VSLF, runoff is calculated for each wetness and

FIGURE 12.2 Distribution of effective local moisture storage, σ e , normalized to the

water-shed average potential soil water retention parameter, S, along an axis of increasing fraction

of watershed area with lower local moisture storage, As.

0 8

1 1

land use HRU with Equation 12.11 For the entire watershed, runoff depth Q is the areally weighted sum of runoff depths qi for all discrete wetness and land-usecontributing areas:

(12.12)

Total runoff depth, Q, calculated by this equation is the same as that calculated

by Equation 12.1 (Schneiderman et al 2006), so runoff volume estimates for thewatershed as a whole with VSLF are compatible with the original GWLF and othermodels that use the traditional SCS CN equation The main hydrological difference

is that VSLF distributes storm runoff according to a moisture storage distributionrather than by land use and soil type This has important implications for predictions

of chemical constituents of runoff

12.4 PREDICTING PHOSPHORUS IN RUNOFF

Dissolved P loads in runoff from each HRU are calculated daily in GWLF as theproduct of simulated runoff and empirically derived HRU-specific nutrient concen-trations Haith et al (1992) compiled runoff concentrations for different land usesfrom the literature for rural land uses Urban dissolved nutrient concentrations werecompiled in the Nationwide Urban Runoff Program (U.S Environmental ProtectionAgency 1983) Literature concentration values from these and other sources provide

an initial basis for determining these parameters Runoff concentration data forspecific watersheds can be used when available In the original GWLF, dissolvednutrient concentrations are input as constants, with the exception of agricultural landuses on which winter spreading of manure or fertilizer occurs For these land uses,seasonal variability in nutrient concentrations is introduced, with elevated concen-trations applied to snowmelt and rain on snow

Use of literature-based concentrations of nutrients in runoff in GWLF presents

a number of challenges Literature-based concentrations generally provide a range

of values for a given land use, and the choice of an appropriate value for a givenwatershed requires an act of judgment Since the GWLF parameters represent con-centrations in runoff as expressed at the outlet of a watershed, scale differencesbetween study sites on which literature values are based and the watershed beingmodeled may affect the translation Even concentration data for runoff in the studywatershed may not translate directly to GWLF parameters if the data are sampled

at the plot or field scale Model calibration may be necessary and is recommendedwhere loading data are available Schneiderman et al (2002) calibrated nutrientconcentrations with a single adjustment factor that was applied to all nutrient con-centrations In this way the relationships between concentrations associated withdifferent land uses are maintained, as all concentrations shift up or down with themultiplicative factor In effect, the literature values are used to establish the relativeconcentrations for different land uses, and calibration establishes the absolute values

The use of constant concentrations for nutrients in runoff is deserving of somediscussion On the local plot or field scale, P concentrations would be expected

to vary with changes in soil P content in the upper soil layer (Sharpley 1995) orwith the timing and intensity of manure or fertilizer applications (Walter et al.2001) On the watershed scale, however, where the simulation unit (in GWLF andother lumped-parameter HRU models) is the HRU composed of many plots orfields at different stages of fertility and soil P cycles, the temporal variations thatexist on the fine scale may cancel out on the broad scale The New York CityDepartment of Environmental Protection (2005) used the Soil Water AssessmentTool (SWAT) (Bicknell et al 2001) model, which simulates daily P concentrations

in soil and runoff by keeping a mass balance of P, to investigate temporal variation

in soil and runoff P concentrations for agricultural land uses in New York Citywatersheds The results of an extensive sensitivity analysis suggested that with theexception of agricultural fields that are subject to tillage, predicted soil and runoff

P concentrations at a watershed scale were fairly constant Tilled agricultural fieldsdemonstrated a distinct pattern of reduction of concentrations in runoff (due tomixing of high P upper-layer soil with lower P lower-layer soils) at the time oftillage, followed by a gradual increase in concentrations with time after tillage.VSLF was modified to allow seasonally varying concentrations for cases like thiswhere such variations occur on a watershed scale

12.5 PREDICTING PHOSPHORUS LEACHING

GWLF assumes that the major pathways by which P is exported from a watershedare in runoff from different HRUs and in base flow GWLF does not predict leaching

of P explicitly In actuality, the importance of P leaching as a pathway for export of

P on a watershed scale is uncertain Observations of elevated P concentrations inbase flow in P-enriched watersheds could be considered evidence for transport ofleached P from P-saturated soils, but this has not been demonstrated conclusively.Elevated P concentrations in base flow could just as well be caused by high Pavailability at groundwater discharge sites, including seeps and stream banks.The effects of factors that influence base flow P levels, including leaching of Pinto shallow groundwater and P entrainment at groundwater discharge zones, isaccounted for in GWLF by model and input adjustments to the concentration of P

in base flow In the original GWLF model, the P concentration in base flow was anempirical function of the aerial percentage of active agriculture in the watershed(Figure 12.3) VSLF has an option to allow user input of a base-flow concentrationvalue that is representative of the watershed being studied

12.6 SIMULATING MANAGEMENT AND BMPS

The general approach to simulating management and BMPs with GWLF is totranslate the effects of BMPs, either individually or combined, into model parameteradjustments, which are then applied in subsequent scenario runs of the model.For example, hydrologic effects of watershed management and BMPs may be

expressed through modifications in curve number (affecting runoff), soil water capacity(affecting percolation), melt coefficient (affecting snowmelt), or vegetative covercoefficients (affecting ET) Water-quality effects of BMPs are expressed by modi-fying HRU-specific P concentrations in runoff, P concentration in base flow, point-source concentrations, septic system failure rates, Universal Soil Loss Equation(USLE) parameters that control erosion rates, sediment delivery ratio, and HRU-specific P concentrations in soils GWLF-VSA and AVGWLF have reduction factorsbuilt into the model to streamline the application of BMP effects on model parameters.The reduction factor approach to simulating effects of watershed management

is supported by an ever increasing body of knowledge in the literature on theeffectiveness of BMPs Phosphorus removal efficiencies of urban BMPs are mea-sured and compiled for stormwater treatment practices (Winer 2000) Gitau et al.(2005) compiled a database of BMP effectiveness for agricultural BMPs Gitau andVeith cover the effect of P control BMPs in Chapter 15 of this volume The USLEmethodology (Wischmeier and Smith 1978) provides coefficient values for variousmanagement practices Results of watershed-specific field studies on BMP effec-tiveness can be utilized where available

12.7 SIMULATING IN-STREAM PROCESSES

GWLF simulates in-stream processes as a lumped statistical process where thestream is treated as a single unit The effect of channel length and physical properties

on the timing of base-flow discharge at the outlet of the watershed is treated as asimple exponential time delay that can be calibrated from stream flow data Thoughthe original GWLF model does not permit a time delay for runoff — runoff at the

FIGURE 12.3 Dissolved P concentration in base flow as a function of percent of agricultural

land use within a watershed.

Percent Agriculture

0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0