Modeling phosphorus in the environment - Chapter 1 pptx

This introduction chapter previews the general principles of how models repre-sent soil P release and transport, effects of mineral fertilizer and manure management on P loss, spatial re

Section I

Basic Approaches

Movement from

Agriculture to Surface Waters

Andrew N Sharpley

U.S Department of Agriculture-Agricultural Research Service, University Park, PA

CONTENTS

1.1 Introduction 3

1.2 Types of Models 4

1.2.1 Process–Based Models 5

1.2.2 Export Coefficient Models 5

1.2.3 Statistical or Empirical Models 6

1.3 How Models Simulate P Transport 6

1.3.1 Dissolved P 6

1.3.2 Particulate P 8

1.4 Fertilizer and Manure Management 10

1.5 Spatial Data Requirements for Modeling 11

1.6 Defining Future Best Management Practices 12

1.7 How Models Simulate Fluvial Processes and Impact of P in Surface Waters 12

1.7.1 Fluvial Processes 12

1.7.2 Surface Water Impacts 14

1.8 Summary 14

References 15

1.1 INTRODUCTION

Phosphorus (P), an essential nutrient for crop and animal production, can accelerate freshwater eutrophication, which is the most ubiquitous water quality impairment

in the U.S., with agriculture a major contributor of P (Sharpley 2000; U.S Geological Survey 1999) Environmental concerns from harmful algal bloom outbreaks

(Burkholder and Glasgow 1997) and regulatory pressure to reduce P loadings to surface waters via implementation of Total Maximum Daily Loads (TMDLs) (U.S Environmental Protection Agency 2000) have increased the urgency for information

on the impacts of agricultural management, specifically conservation practices and best management practices (BMPs) on P loss Because of the time and expense involved in assessing P loss, models are often a more efficient and feasible means

of evaluating management alternatives In their most comprehensive form, models can integrate information over a watershed scale to identify BMPs and critical source areas where BMPs are most likely to affect watershed-scale P losses

A common limitation to model application is the lack of detailed parameteriza-tion data on soil physical, chemical, and biological properties as well as on crop and tillage details Thus, existing databases are increasingly being linked to nonpoint source models, often via geographical information systems (GIS) Generally, key input data for nutrient transport models involve land use, soil texture, topography, and management practices Once these data are in digital form, GIS techniques can

be used to combine them with experimental or model results to extrapolate other properties needed for model application

This introduction chapter previews the general principles of how models repre-sent soil P release and transport, effects of mineral fertilizer and manure management

on P loss, spatial resolution, and channel processes that translate edge-of-field losses

to water body inputs Future modeling efforts needed to address these issues are presented

1.2 TYPES OF MODELS

Models that simulate the runoff and water quality from watersheds can be categorized

in several ways, but for purposes of this brief review they are segregated into three groups:

1 Process-based models: Models that explicitly simulate watershed

pro-cesses, albeit usually conceptually These models typically involve the numerical solution of a set of governing differential and algebraic equa-tions that are a mathematical representation of processes such as rainfall runoff; infiltration leaching; P application method, rate, and timing; land management; and fate and chemical transformation of added P in soil

2 Export coefficient models: Models that rely on land-use categorization —

sometimes through a linkage to a GIS evaluation — coupled with export coefficients or event mean concentrations (EMCs), loosely categorized as spreadsheet approaches, although highly sophisticated in many cases These models rarely, if ever, involve solution of a differential equation and almost always rely on simple, empirical formulations, such as the use

of a runoff coefficient for generation of runoff from rainfall

3 Statistical or empirical models: Models that involve regression or other

techniques, which relate water-quality measures to various characteristics

of the watershed These models range from purely heuristic regression equations (e.g., Driver and Tasker 1990) to relatively sophisticated

derived-distribution approaches for prediction of the frequency distribu-tion of loadings and concentradistribu-tions (e.g., DiToro and Small 1984; Driscoll

et al 1989)

All of the model types have their drawbacks related to availability of required data, scaling up from pedon input parameters, for example, to a watershed scale, and quantifying system functionality For more detailed information on the approaches used in models described in the sections following and in other models, reviews are given as separate chapters in this publication

1.2.1 P ROCESS –B ASED M ODELS

The Agricultural Nonpoint Source (AGNPS) Pollution model (Young et al 1989, 1995) was originally developed to provide estimates of runoff water quality from watersheds of up to 20,000 hectares and to quantify the effects of BMPs targeted

to specific areas To make model output more meaningful to decision makers, such

as conservationists and farmers, AGNPS, which ran on a storm or flow event basis, was recently superseded by an annualized version, Annualized AGNPS (AnnAG-NPS) (Bingner et al 2001; Croshley and Theurer 1998) The model operates on a cell basis that makes it possible to analyze spatially discrete management units (fields) within a watershed, thereby enabling identification of individual fields that may serve as critical source areas of nutrient export AnnAGNPS is described in Chapter 9 of this book

The Soil and Water Assessment Tool (SWAT) was developed to assess the impact

of land management on water quality in watersheds and large river basins (Arnold

et al 1998) The model runs on a continuous time step and is currently being utilized

in a variety of large-scale studies to estimate the off-site impacts of climate and management on water use and nonpoint source loadings SWAT is described in Chapter 7 of this book

Other process-based nutrient transport models include, but are not limited to Areal Nonpoint Source Watershed Environment Response Simulation 2000 (ANSWERS-2000) (Beasley et al 1985; Bouraoui and Dillaha 1996), the Guelph Model for Evaluating the Effects of Agricultural Management Systems on Erosion and Sedimentation (GAMES) (Cook et al 1985), Hydrologic Simulation Program-Fortran (HSPF) (Johanson et al 1984), Agricultural Runoff Model (ARM) (Donigian

et al 1977), Erosion Productivity Impact Calculator (EPIC) (Sharpley and Will-iams 1990), Groundwater Loading Effects of Agricultural Management Systems (GLEAMS) (Leonard et al 1987), Watershed Ecosystem Nutrient Dynamics-Phosphorus (WEND-P) (Cassell et al 1998), and CENTURY (Parton et al 1993) HSPF, ANSWERS-2000, and WEND-P are described in Chapters 8, 10, and 11, respectively

1.2.2 E XPORT C OEFFICIENT M ODELS

Export coefficient models have also been widely used to predict P loading of receiving water bodies (Beaulac and Reckhow 1982; Hanrahan et al 2001;

Johnes et al 1996) Export coefficients define P loss from a particular source or land use in a watershed and are usually derived from actual field measured losses of P

or from EMC values, if runoff volumes are known (Johnes 1996; Johnes and Heathwaite 1997) Both export coefficients and EMCs fit easily into spreadsheet formats for watershed loading estimates An advantage of EMCs is that they may be coupled with any hydrologic simulation model to produce loads

Export coefficient models calculate watershed export of P as the sum of indi-vidual loads from each source in the watershed This approach accounts for the complexity of land-use systems, spatial distribution of data from various sources (point and nonpoint), and permits scaling up from plot to watersheds As export coefficients are empirical, these types of models are as accurate as input data, as are process-based models (Hanrahan et al 2001) Coefficients derived from short-term monitoring of small drainage areas, however, can contribute to predictive variability (Lathrop et al 1998) The Generalized Watersheds Loading Functions (GWLF) model (Haith and Shoemaker, 1987) is an example of an export coefficient model and is described in Chapter 12

1.2.3 S TATISTICAL OR E MPIRICAL M ODELS

Statistical models are empirical Although they are derived from observations, the relationship described must have a basis in our underlying understanding of pro-cesses if we are to have faith in the predictive capabilities of the model (National Research Council 2000) Furthermore, extrapolation from empirical data is known

to be fraught with danger For example, scaling problems can occur when one extrapolates the results of scaled experiments to full-sized natural systems One must, of course, always remain cognizant of the fact that system function may be scale dependent Thus, these models are most judiciously used in the range of observational situations used to derive the model

Statistical or empirical models are most useful when they are based on first principles The ability to describe system functions in terms of mathematical equa-tions often gives the impression that the underlying principles are fully understood,

as might be the situation in basic physics Unfortunately, empirical coefficients introduced into these equations often hide the degree of uncertainty concerning these principles This publication does not include reviews of any statistical models per

se, but many P indices include statistical relationships and might be considered a type of statistical model P indices are described in Chapter 13

1.3 HOW MODELS SIMULATE P TRANSPORT

1.3.1 D ISSOLVED P

Most nonpoint source models simulate dissolved P transport in overland flow as a function of the extractability of P in the surface 5 cm of soil [e.g., Chemicals, Runoff and Erosion from Agricultural Management Systems (CREAMS), AGNPS] This can be represented by

(1.1)

where dissolved P is orthophosphate loss in overland flow (kg ha−1), available soil

P is the amount of P in a unit depth of surface soil — usually 5 cm — (Sharpley 1985b)

as estimated by recommended soil test P methods (STP) (kg ha−1 5 cm−1), and

extraction coefficient is the fraction of STP that can be released to overland flow

for a given flow event volume (cm) Extraction coefficients can be determined as the slope of the linear regression of STP and overland flow dissolved P (Figure 1.1a) A similar relationship holds for subsurface flow P and surface STP, although the slope of the relationship (0.93) is almost half that for overland flow (slope of 1.98) (Figure 1.1b) The dependence of dissolved P transport in subsurface flow as well as overland flow suggests the importance of preferential flow pathways, such as earthworm burrows and old root channels, in the downward movement of

P through the soil profile (Kleinman et al 2003; McDowell and Sharpley 2001a; Sims et al 1998)

FIGURE 1.1 Relationship between the concentration of dissolved P in overland (a) and

subsurface flow (b) from 30-cm-deep lysimeters and the Mehlich-3 extractable soil P con-centration of surface soil (0 to 5 cm) from a central Pennsylvania watershed (Adapted from

R.W McDowell and A.N Sharpley, J Environ Qual 30, 508–520, 2001; and A.N Sharpley, P.J.A Kleinman, R.J Wright, T.C Daniel, B Joern, R Parry, and T Sobecki, in International

Conference on Agricultural Effects in Ground and Surface Waters, J Steenvooreden (Ed.),

Wageningen, The Netherlands, International Association of Hydrologic Sciences.)

Mehlich-3 extractable soil P (mg kg -1 )

0 200 400 600 800

-1 )

2000 1500 1000 500

0 1000 750 500 250 0

y = 1.98x + 79 R2 = 0.78

y = 0.93x + 60 R2 = 0.79

b Subsurface flow from lysimeters

a Overland flow

Most models use a constant extraction coefficient value, assuming that STP extractability is similar among soils A re-analysis of data published by McDowell and Sharpley (2001a), Pote et al (1999), and Sharpley and Smith (1994) relating STP and overland flow dissolved P revealed a range of extraction coefficient values (Figure 1.2) Extraction coefficients were much greater for cropped (8 to 17) than grassed watersheds (1 to 4) Using erosion as a surrogate for land cover, extraction coefficients increased with greater erosion or decreased soil cover (Figure 1.2) Although erosion is influenced by other factors such as slope and soil structure, the sites used in this example were similar in slope (~4%) A larger soil P extraction coefficient represents a greater release of P as overland flow dissolved P per unit STP increases This can be attributed to a lower degree of interaction between surface soil and overland flow with a protective grass cover than for a cropped situation, where the soil is more exposed to overland flow Other factors that influence P release among soils are the dominant forms of P in soil, texture, aggregate diffusion, degree

of interaction between soil and water, organic matter content, vegetative soil cover, and P sorption capacities

1.3.2 P ARTICULATE P

As the sources of particulate P in overland flow and stream flow include eroding surface soil, stream banks, and channel beds, processes determining erosion also

FIGURE 1.2 Extraction coefficient — the slope of the relationship between soil test P and

dissolved P in overland flow — as a function of erosion to represent soil vegetative cover for sites in Arkansas, Oklahoma, New York, and Pennsylvania (Data adapted from D.H Pote, T.C Daniel, D.J Nichols, A.N Sharpley, P.A Moore, Jr., D.M Miller, and D.R Edwards,

J Environ Qual 28, 170–175, 1999; McDowell and A.N Sharpley, J Environ Qual 30,

508–520, 2001; and A.N Sharpley and S.J Smith, Soil Tillage Res 30, 33–38, 1994.)

Decreasing soil cover Erosion (tonnes ha-1yr-1)

10 1

0.1 0.01

0.001

Native grass / pasture

No till Reduced till Conventional till

20

15

10

5

0

y = 1.25x 0.30

R2 = 0.90

control particulate P transport In general, eroded particulate material is enriched with P compared to source surface soil, due to the preferential transport of finer (i.e., clay size), more sorptive soil and organic particles of greater P content than coarser inorganic particles (i.e., sand size) Sharpley (1985a) found that the plant available P content of sediment in overland flow was on average three times greater — or more enriched — than that of source soil and 1.5 times greater for total, inorganic, and organic P The degree of P enrichment is expressed as a P enrichment ratio (PER), that is, the P concentration of sediment discharged divided by that of source soil In assembling enrichment ratio information for the CREAMS model, Menzel (1980) concluded that for particulate P, a logarith-mic relationship as in Equation 1.2 seemed to hold for a wide range of soil vegetative conditions

where sediment discharge is in kg ha−1 Most nonpoint source models adopted this approach to predicting particulate P transport in overland flow This relationship is based on the well-documented increase in particulate P loss with increasing erosion (Figure 1.3) Based on the total P concentrations of source soils for each of the watersheds represented in Figure 1.3, PER decreases with an increase in erosion As erosion increases, there is less particle-size sorting by overland flow, relatively less clay-size particles are transported, and P enrichment thus decreases

FIGURE 1.3 Particulate P loss and enrichment ratio of eroded sediment as a function of

erosion in overland flow from watersheds in El Reno, Oklahoma (Adapted from A.N Sharpley,

S.J Smith, J.R Williams, O.R Jones, and G.A Coleman, J Environ Qual 20, 239–244, 1991; and S.J Smith, A.N Sharpley, J.W Naney, W.A Berg, and O.R Jones, J Environ.

Qual 20, 244–249, 1991.)

Erosion (tonnes ha -1 )

10 100 1

0.1 0.01

0.001

-1 )

7.5

5.0

2.5

0

10 8 6 4

2

1

Particulate P

P enrichment ratio

Once an appropriate PER is obtained from sediment discharge, particulate P loss can be calculated as

× PER × Overland flow volume (1.3)

where particulate P is the loss in overland flow (kg ha−1), total soil P is the amount

in a unit depth of surface soil (usually kg ha−1 5 cm−1), sediment concentration is g

sediment L−1 overland flow, and PER is calculated from Equation 1.2, for a given

flow event volume (cm)

1.4 FERTILIZER AND MANURE MANAGEMENT

Fertilizer and manure management, as it affects P availability to overland flow over the near term, can profoundly affect prediction of P transport in overland flow Although soil P represents a source of P enrichment in overland flow, the application of fertilizer and manure to soil — including type, method, timing, and rate of P application — can temporarily overwhelm relationships derived between STP and P in overland flow (Sharpley and Tunney 2000) As such, accounting for fertilizer and manure management in P models is essential to their accuracy under certain conditions However, most models do not directly address the effect of applied P, either as fertilizer or manure, on P transport in overland flow Rather, added P is incorporated into the soil P pool, and the extraction coefficient is adjusted Thus, P transport in overland flow as affected

by the amount, type, method, and time after applying P is, in general, poorly represented and predicted

Mineral fertilizer and manure represent concentrated sources of soluble P that can greatly increase dissolved P losses in overland flow Consequently, the concen-tration of soluble P in these sources may provide effective predictions depending

on the solubility of the P source, method of application, rate of application, and timing of application relative to the overland flow event (Figure 1.4) (Kleinman et al 2002) Surface application of manure and mineral fertilizer concentrates P at the extreme soil surface where it is vulnerable to removal by overland flow (Eghball and Gilley 1999; McDowell and Sharpley 2001b; Sharpley et al 1984) Although injection, knifing, and immediate incorporation of manure and fertilizer may decrease P losses, cultivation may increase site vulnerability to particulate P loss due to greater erosion potential (Andraski et al 1985; Romkens et al 1973) Modifying the effect of P source and application method on P concentrations

in overland flow is the timing of application relative to when an overland flow event occurs (Sharpley 1997; Westerman and Overcash 1980) Immediately following application of a P source, the potential for P loss peaks and then declines over time,

as applied P increasingly interacts with the soil and is converted from soluble to increasingly recalcitrant forms (Edwards and Daniel 1993) As a result, fertilizer and manure management effects on overland flow P are predictable

1.5 SPATIAL DATA REQUIREMENTS FOR MODELING

Models that assess nonpoint sources of P loss from agricultural lands rely on spatial data as input Land use, soil properties, and topographic data that include stream locations and watershed boundaries are commonly required inputs However, with an expansion in the geographical scale at which watershed processes are to be modeled, there is a great increase in the size of associated spatial databases Data and parameter requirements also increase rapidly as models become more mechanistic to better represent physical and chemical processes and spatial interactions involved in P loss The complexity of managing these large databases in support of a watershed model can limit the degree of spatial resolution of existing models Spatial parameters are frequently lumped so that units having similar soil, land use, and topographic characteristics respond the same to driving variables, such as those used to simulate runoff generation However, spatially lumped parameters can pose a problem when responses from lumped units cannot distinguish between relative spatial locations of individual units, which can be critical in determining P export from a watershed to

a water body

To overcome the spatial data limitations thus far identified, a nested modeling approach is recommended Field and farm scale models that incorporate the knowledge

of P source and transport processes involved in P loss can be supported with highly detailed spatial databases that are already available in some areas or could be easily developed in others Results and generalizations from these models could be aggre-gated to represent sub-basins in a simpler, less mechanistic model that requires lower spatial resolution Similarly, results from sub-basin models could be further aggregated

to represent whole watersheds of several hundreds of square kilometers in size Beyond that scale and with enough knowledge of processes operating in individual subwatersheds,

FIGURE 1.4 Relationship between water extractable manure P and the dissolved P in

over-land flow one week after manure or mineral fertilizer was broadcast (100 kg total P ha−1) on

a Hagerstown silt loam soil (7 cm hr−1 rainfall for 30 min) (Adapted from P.J.A Kleinman,

A.N Sharpley, B.G Moyer, and G.F Elwinger, J Environ Qual 31, 2026–2033, 2002.)

8 6

4 2

0

Water extractable manure P (g kg -1 )

-1 ) 6000

4000

2000

0

Dairy manure

Dairy compost

Poultry litter Poultry compost

Swine slurry Poultry manure