Modeling phosphorus in the environment - Chapter 17 (end) pptx

17.2 RUNOFFMost of the current models use the Soil Conservation Service SCS curve number approach McCuen 1982 to estimate runoff.. With a fully distributed model, digital elevation model

Section IV

Modeling in the Future

Modeling of Phosphorus

David E Radcliffe

University of Georgia, Athens, GA

Miguel L Cabrera

University of Georgia, Athens, GA

CONTENTS

17.1 Introduction 405

17.2 Runoff 406

17.3 Soil Phosphorus Pools and Model Parameters 406

17.4 Leaching of Phosphorus 408

17.5 Best Management Practices 409

17.6 In-Stream Processes 410

17.7 Phosphorus Indices 411

17.8 Calibration and Uncertainty 411

17.9 Data to Support Phosphorus Modeling 412

17.10 Summary 413

References 413

17.1 INTRODUCTION

Much progress has been made in the science of modeling phosphorus (P) fate and transport: from field-scale, event-based, empirical models to dynamic, watershed-scale, process-based models Nonetheless, some improvements are clearly needed

in today’s models This chapter’s purpose is to identify the most necessary changes

to improve the modeling of P in the environment Earlier chapters in this book have been drawn on heavily in developing the ideas presented here Suggestions are grouped into eight general areas: (1) runoff; (2) soil P pools and model parameters; (3) leaching; (4) best management practices (BMPs); (5) in-stream processes; (6)

P indices; (7) calibration and uncertainty; and (8) data for modeling

17.2 RUNOFF

Most of the current models use the Soil Conservation Service (SCS) curve number approach (McCuen 1982) to estimate runoff This approach is widely proven and sensitive to land use and soils, but it limits models to a daily time step Some models use, or have the option of using, infiltration equations such as Green and Ampt (1911) These equations can be used for shorter time steps, but they do not readily reflect differences in land use Furthermore, the infiltration equations are based on

a one-dimensional (vertical) approach to water flow and as a result estimate infil-tration excess runoff in contrast to saturation excess runoff (Beven 2001) Research

in humid regions has shown that saturation excess runoff is generated during storms

in areas low in the landscape that vary in size (variable source area, or VSA) depending on the antecedent soil water content because of lateral subsurface water flow (Hewlett and Troendle 1975) It is essential not only that the correct amount

of runoff from a watershed be estimated but also that the area where runoff is generated be identified Combined with the areas that are sources of high concen-trations of P in runoff, these are the critical source areas where management can be focused to reduce P losses (Gburek and Sharpley 1998)

Models need to incorporate the VSA concept into their estimations of where runoff occurs This may move P simulation toward fully distributed watershed models such as AnnAGNPS or ANSWERS-2000 With a fully distributed model, digital elevation model (DEM) data can be used to generate a wetness index (Beven 2001) for each cell, which in turn can be used to identify areas that are likely to receive lateral subsurface flow of water Alternatively, it may be possible to incor-porate the VSA concept into semidistributed models such as Soil Water Assessment Tool (SWAT) and HSPF According to Steenhuis et al (1995), the SCS curve number approach for estimating the amount of runoff is consistent with the VSA concept Schneiderman (Chapter 12, this volume) describes a new version of the Groundwater Loading Function (GWLF) known as VSLF that uses the wetness index to distribute runoff estimated by the SCS curve number method among hydrologic response units (HRUs)

In the present authors’ opinion, models must consider both infiltration excess and saturation excess processes in estimating where runoff will occur Recent work has shown that both soil properties and landscape topography are important in predicting runoff (Soulsby et al 2005; Tetzlaff et al 2005)

17.3 SOIL PHOSPHORUS POOLS AND MODEL

PARAMETERS

Many of today’s watershed-scale models have evolved from field-scale models developed in the 1980s The soil P routines in these models are derived from concepts current in the 1980s and implemented in the EPIC model (Williams 1995) These routines need to be updated to reflect recent research The most important improve-ment will be to add a separate manure pool for P (Vadas et al 2004) Current models assume that P added in manure instantly becomes part of the soil P pools However,

in grasslands where manure is not incorporated this is not the case, and models can

underestimate P losses in runoff from storms shortly after manure is applied (Pierson

et al 2001) How these pools interact is an important question DeLaune et al (2004) showed that the manure pool dominated soil test P as a source for up to 14 days after application of broiler litter, but it is not known how much longer these pools remain separate Vadas et al (2005a) developed equations for a separate manure pool based on laboratory-derived relationships similar to what has been used earlier for soil P pools (see Chapter 3) Vadas is also developing a full model of the manure pool and its interaction with other pools (Vadas 2006) This model estimates dis-solved organic P in runoff in addition to disdis-solved inorganic P Disdis-solved organic P

in manure has been found to increase as the pH of the manure decreases after application from its original pH of around 8 to a pH of 6, which is typical of agricultural soils (Tasistro et al 2006) It may also be important to model the contribution to particulate P that comes from the manure pool as well as the P derived from vegetation or residue pools (Sharpley and Smith 1992) Further research is needed to determine how much P loss is due to losses from these pools

Nelson and Parsons (Chapter 4, this volume) note that the EPIC approach leads

to an assumption of linear P sorption in soils, but sorption isotherms show that sorption is highly nonlinear Assuming linear sorption could lead to error in predict-ing runoff and leachpredict-ing losses, especially in soils with high concentrations of sorbed P Nonlinearity also arises in the data used to develop the laboratory-derived relation-ships because of the effective depth of interaction and water-to-sediment ratio terms (Chapter 3, this volume)

For many models, it is not clear what form of measurable soil P (e.g., total, soil test, bioavailable, anion-exchangeable) corresponds to the different model soil P pools (e.g., labile, active, residual) In the original papers by Jones et al (1984) and Sharpley et al (1984) describing the EPIC approach, it was clear that the labile P pool corresponded to anion-resin-exchangeable P (Sharpley 2000) Sharpley et al (1984) recognized that the most commonly measured forms of soil P were the different soil test extractions (Bray-1, Mehlich-1, and Olsen P) and provided regres-sion equations for converting soil test P concentrations to labile P concentrations This clarity has been lost in the documentation of some of the current models In

the SWAT User’s Manual (Neitsch et al 2002), a dissolved P pool is described, but

the equations are identical to those in the EPIC routines, so this is most likely labile

P In HSPF, two soil P pools — dissolved and adsorbed — are simulated, but no guidance is given on what forms of measurable P correspond to these pools Models have been found to be sensitive to the initial storages of soil P pools (Chapter 9, this volume) Wherever possible, soil P pools in models should be related to mea-surable forms of P in soils

The rate of transfer between soil P pools in soil can be important Vadas et al (2006) showed that SWAT does not include a delay that was part of the original EPIC routine in transfer between the labile and active P pools when fertilizer is added As noted by Chaubey et al (Chapter 7, this volume), several improvements to the P routines in SWAT are under consideration, including slowing the transfer of added (fertilizer) P from labile into active pools and adding a separate manure pool Most models estimate the dissolved inorganic P concentration in runoff by mul-tiplying the concentration of dissolved or labile P in the topsoil by an extraction

coefficient The National P Project (Vadas et al 2005b) resulted in a very large number

of rainfall simulation studies performed across the U.S where dissolved inorganic P concentration in runoff was regressed on different forms of P in soil This information

on extraction coefficients needs to be incorporated into model databases

HSPF uses a number of soil parameters in the soil P routines — and in the hydrology routines — that are unique to the model and not commonly measured in soils A database of parameter values used in models by experienced users is provided

in HSPFParm (Donigian et al 1999), which is very useful A technical note for selecting values for hydrology parameters is available (EPA 2004), as well as a new technical note for sediment parameters (EPA 2006) A similar technical note for selecting P parameters is needed

17.4 LEACHING OF PHOSPHORUS

Concern over leaching of P is relatively new, as recent research has shown it is an important pathway for subsurface losses to nearby streams, especially in deep sands and in systems with tile drains (Nelson et al 2005; Sims et al 1998) As noted by Nelson and Parsons (Chapter 4, this volume), leaching also affects the estimation of runoff in that it removes P from the topsoil that would otherwise be available for runoff loss They point out that in soils with low P sorption capacities, leaching can potentially remove more P from the root zone than crop uptake Leaching is also important in long-term modeling studies where it becomes a significant component of P mass balances However, current models either ignore P leaching or treat it in a simplistic manner AnnAGNPS, ANSWERS-2000, and WEND-Pdo not model leaching of P

or losses by tile drains SWAT uses a field-capacity water balance (FCWB) approach

to estimate vertical flux of water within the soil profile (up to 10 layers) and includes preferential crack flow The concentration of P in leaching water is determined by the labile P concentration in the top soil layer HSPF uses an empirical equation to determine percolation of water, and user-specified reduction factors are used to determine how much P is transported with the percolating water GWLF does not model leaching explicitly, but the concentration of P in groundwater, which affects stream base-flow P concentration, is a user-specified input or a function of the percent

of agricultural land use in the basin

A better description of the P leaching process needs to be included in models Modeling may benefit from some of the same improvements mentioned for improv-ing the modelimprov-ing of runoff — a separate manure pool because manure P, especially when surface applied, will not be immediately available for leaching and because adsorption is likely to be nonlinear Models need to be able to estimate tile drain flow and the concentration of P in tile drains It may be possible to estimate the P concentration in leachate by using a relationship with soil test P in the topsoil Nathan and Parsons (Chapter 4, this volume) refer to these as quantity–intensity relationships Preferential flow should also be incorporated into future models if it

is not already present Research has shown that it is especially important in fields with tile drains (Kladivko et al 1991) Particulate P as well as dissolved P may move via preferential flow (Chapter 4, this volume) Model validation should include comparisons of model estimations with P distributions in the soil profile

Model developers seem aware of the need for improving the simulation of P leaching SWAT2003 has improved tile drain simulation, and further testing of the preferential flow model in SWAT is stated goal (Chapter 7, this volume)

ANSWERS-2000 is undergoing major revision to link the model’s subsurface flow components with the groundwater model MODFLOW (Chapter 10, this volume)

17.5 BEST MANAGEMENT PRACTICES

Today, watershed-scale models are being used to test scenarios for reducing P loads

to sensitive water bodies because of limits imposed by Total Maximum Daily Loads (TMDLs) (Reckhow et al 2001), source water assessment (EPA 1997), or lawsuits (Blackstock 2003) This is usually done in part by reducing loads from nonpoint sources through the use of best management practices (BMPs) Despite the wide-spread use of models for this purpose, BMPs are usually modeled in a simplistic manner (Chapter 15, this volume)

All the models in this volume allow for simulation of BMPs to some degree The most extensive list of BMPs occurs in SWAT and AnnAGNPS Common BMPs simulated are crop rotations, tillage, and nutrient management (i.e., timing, type, and rate) However, in most cases a simple reduction factor is used to simulate the effect of BMPs Often the reduction factor for a given BMP is the same for different pollutants For example, in SWAT the reduction factor for filter trips is specified by the user, but it is the same for sediment and dissolved P Research has shown that filter strips are not as effective in removing dissolved P as they are in removing sediment (Chapter 15, this volume)

One of the most important BMPs that needs improvement in modeling is the use of riparian buffers, since these are commonly used in implementing TMDLs Semidistributed models such as SWAT and HSPF do not model overland flow or subsurface flow from one land use to another, so it is difficult to model the effect

of buffers beyond using a simple reduction factor Buffers can be more easily modeled in fully distributed models such as AnnAGNPS or ANSWERS-2000 This, along with identification of variable source areas of runoff, may be another reason for moving toward fully distributed models Including buffers as separate land uses may require improvements in the leaching and lateral flow predictions as well (Chapter 4, this volume) The SWAT model developers are considering the addition

of vegetated filter strips and riparian zones, but it will require reconfiguration of HRUs to allow more detailed variation in topography and management practices (Chapter 7, this volume) The advantages of this approach are being weighed against the fact that increased complexity will make it less user friendly Lowrance et al (2000) developed the Riparian Ecosystem Management Model (REMM), which simulates all of the important P and hydrologic processes thought to occur in riparian buffers, given the runoff and subsurface lateral flow inputs from adjoining fields It may be possible to use REMM to parameterize HRUs that consist of a field with a riparian buffer and thereby avoid developing a fully distributed model

Gitau and Vieth (Chapter 15, this volume) recommend that the cost of BMPs

be included in models They also suggest that the interaction between BMPs needs

to be simulated since management scenarios will consider more than one type of BMP

Interesting work is being done on using optimization methods to determine the most effective suite of BMPs to achieve a given reduction (Chapter 15, this volume) Multi-objective functions can be used that include cost and the effect on other pollutants such as nitrogen and sediment This type of capability is also needed in calibration and uncertainty analysis (see Chapter 6)

The long-term effectiveness of BMPs must also be considered WEND-P, which uses a long-term mass balance approach at an annual time step, reminds readers that

to be effective in the long term BMPs must affect P storage in the watershed

17.6 IN-STREAM PROCESSES

A clear trend in the modeling of P has emerged to scale up Models have moved from field scale models such as EPIC and GLEAMS towatershed scale models such

as AnnAGNPS, ANSWERS-2000, GWLF, HSPF, SWAT, and WEND-P At first the watershed models were used for single events on ephemeral or small, first-order streams, but now they are being used for basins that are thousands of square kilo-meters in area In modeling P, the objective is often to predict the load to a sensitive lake, reservoir, or estuary Modeling of P processes in large lakes, reservoirs, and estuaries is usually done with separate, specialized models, and for the near future

it does not seem likely that this will change

To estimate accurate loads at this scale, it is essential that in-stream processes affecting the transport of P and sediment be included in models (see Chapters 1 and water body and the bioavailability of this P Processes change with flow (i.e., storm

vs base flow) and stream order (Vanotte et al 1980) Stream water column P con-centrations may be heavily buffered by elevated bed sediment (benthic) P concentra-tions in watersheds with a history of large P losses This may delay the reduction in

P loading to lakes or reservoirs in response to the implementation of BMPs Most current models ignore or poorly model in-stream processes Only three of the models discussed in this volume include in-stream processes: HSPF, SWAT, and GWLF HSPF provides the most comprehensive modeling of in-stream processes, taking into consideration benthic release of dissolved P, sorption and desorption of

P to suspended sediment in the water column (which varies for different sediment size fractions), settling of suspended sediment, scouring of bed and bank sediment, uptake by phytoplankton or benthic algae and respiration and inorganic excretion

by zooplankton Atmospheric deposition is also considered SWAT includes settling

of organic P, benthic release of P, and uptake by algae but does not simulate interaction between dissolved P and suspended sediment GWLF models in-stream processes as a single, lumped stream segment It includes channel erosion and transformation between dissolved and particulate P under base-flow conditions

It is necessary to know the initial stores of benthic P for models, and studies that have measured equilibrium P concentrations (EPC0) of bed sediments may be useful in this regard (Sharpley et al 2002) They may also provide an integrated measure of historical local inputs (Page et al 2005)

It may be possible to use measurements of P uptake lengths in streams to determine if models are correctly estimating losses of P in streams As described by

5, this volume) Stream processes also affect the form of P that reaches a sensitive

Haggard and Sharpley (Chapter 5, this volume), uptake length is measured by applying a constant source of dissolved P to a stream and measuring concentrations

of dissolved P in the stream at various distances below the input point Concentra-tions typically decay exponentially with distance, and the inverse of the exponential decay coefficient is the uptake length (Newbold et al 1981, 1983) It should be possible to do a similar experiment with a watershed model by including a point source and examining the concentrations of P in downstream reaches Concentrations estimated by the model should have a decay coefficient that is on the order of that measured in streams of the region

17.7 PHOSPHORUS INDICES

In the present authors’ opinion, P indices (PIs) can be considered a type of model representing one end of the spectrum in that they are usually used at the field scale and estimate long-term averages (one to five years) In most cases, they estimate

a risk index of P loss, not an actual loss, but there are three states where the PI

in fact estimates P loss (Chapter 13, this volume) PIs are probably weakest in their treatment of transport, and in this area they need improvements in some of the same areas as the more complex, dynamic, watershed-scale models These include identification of critical source areas, better treatment of P leaching, and better quantification of the effects of BMPs Better methods for rapidly calculating erosion under different scenarios are needed, perhaps by linking PI software to Revised Universal Soil Loss Equation, version 2 (RUSLE2) programs PIs also need to be linked to databases of soils, hydrography, and land use

Other needed improvements include the development of a preliminary risk assessment that only considers a field’s inherent transport risk factors such as soil type, slope, and distance from water body (Chapter 13, this volume) PIs need to be integrated into comprehensive manure management software such as the Manure Management Planner (Chapter 14, this volume)

17.8 CALIBRATION AND UNCERTAINTY

Dynamic watershed-scale P models are necessarily calibrated because of the large number of parameters, many of which operate at a scale beyond which measurements

can be made — called noncommensurability by Beven et al (Chapter 6, this volume) Sensitivity analysis can be used to reduce the number of calibrated parameters to only those that have a large effect on estimations Because watershed models are frequently used in the regulatory arena for TMDLs (Reckhow et al 2001), source water assessment (EPA 1997), and lawsuits (Blackstock 2003), some measure of the uncertainty of estimations must be included

Methods and software have been developed for performing sensitivity analysis, for automating calibration, and for quantifying uncertainty, but these tools are usually not integral parts of today’s models In fact, some model users oppose the use of autocalibration tools because they feel that model users should be limited to expert hydrologists (Gupta et al 2003) The present authors disagree and recommend that calibration and uncertainty tools be integrated into modern models

A good example of this type of approach is contained in the latest version of SWAT, which incorporates an autocalibration capability described by van Griensven (2002) It includes sensitivity analysis using the one-factor-at-a-time approach Once the significant parameters are identified, the Shuffled Complex Evolution (SCE) algorithm is used for autocalibration (Chapter 7, this volume) Another example is

a new version of GWLF that has a built-in calibration tool (Chapter 12, this volume) Automated calibration tools may be easier to use with stream-flow data (where daily or more frequent measurements are frequently available) than with P data, which may be collected at monthly intervals in many TMDL data sets For the near

future it may be necessary to accept the idea of equifinality (Chapter 6, this volume);

no unique set of parameter values probably provides the most accurate model estimation But it may be possible to separate parameter values into those that are plausible (behavior) and those that are not (nonbehavior) All of this points to the importance of monitoring data

17.9 DATA TO SUPPORT PHOSPHORUS MODELING

Edge-of-field and stream-monitoring data are essential for calibrating and evaluating watershed-scale models (Chapter 16, this volume) One of the most difficult variables

to measure accurately is flow, but it is essential for calculating loads and for cali-brating and evaluating models Advantage should be taken wherever possible of U.S Geological Survey (USGS) sites that measure daily discharge on large streams Usually, these sites do not have automated storm samplers installed, so they are an ideal location for partnering among universities, the U.S Department of Agriculture (USDA), and USGS

The development of automated storm samplers represents a huge advance in the ability to collect stream-water-quality data Frequently when models are calibrated for the estimation of sediment and P concentrations in streams, there are only one

or two samples during a storm In that situation, the model should estimate concen-trations that are at least as large as those measured Only with a well-defined chemograph can a comparison be made to the model estimations and can there be some assurance that the model processes are correct Optimum sampling schemes for automated samplers need to be further developed and tested (Chapter 16, this volume) The type of sampling will likely be different depending on whether the aim is to calculate an annual load or to calibrate a dynamic model

Even when monitoring data are not used for calibrating dynamic models, annual loads calculated from monitoring data from small streams with a predom-inant land use are very useful for comparison with unit area loads calculated from dynamic models as well as export-coefficient loading models Edge-of-field data should also be used to assess the accuracy of and to improve PIs Ideally, studies should employ a nested design where edge-of-field and low-order stream-moni-toring sites are located within the watershed where the high-order stream is monitored Last, better quantification of the uncertainty (error) in monitoring data

is needed

17.10 SUMMARY

Much progress has been made in the modeling of P in the environment over the past

30 years To continue this progress, the following improvements to modeling P are suggested

• The estimation of the areas where runoff occurs must be improved through the incorporation of saturation excess as well as infiltration excess pro-cesses Better information on where these areas occur will allow identi-fication of the critical source areas where runoff and P sources coincide and where management needs to be focused

• Modeling of soil P pools needs to be improved by adding a separate manure pool, relating the forms of P in model pools with forms of P measured in soils, and refining the rates of movement between P pools

• Modeling of leaching of P needs to be improved by using a more process-based approach to estimating the flux of percolating water and the con-centrations of P in leachate This is especially important in soils with ditch

or tile drainage

• A more process-based approach is needed in modeling the effects of BMPs and their interactions Riparian buffers are especially important in this regard

• Simulation of in-stream processes must be improved in most of the current models if they are to be used to predict P loading to sensitive lakes, reservoirs, and estuaries in large watersheds Stream measurements of equilibrium P concentrations and P uptake lengths offer opportunities to assess model in-stream estimations

• Calibration and uncertainty tools need to be an integral part of future models Uncertainty in model estimations must be quantified as models are increasingly used for regulatory purposes such as TMDL development

• Monitoring data are critical for modeling Long-term studies with a nested design of automated samplers are needed in partnership with the USGS stream-monitoring network

REFERENCES

Beven, K.J 2001 Rainfall-Runoff Modeling: The Primer New York: John Wiley & Sons.

Blackstock, J.D 2003 Using computer models in court: challenges for expert witnesses Presented at the Total Maximum Daily Load Environmental Regulations II Confer-ence, November 8–12, 2003, Albuquerque, NM.

DeLaune, P.B., P.A Moore, Jr., D.K Carman, A.N Sharpley, B.E Haggard, and T.C Daniel.

2004 Development of a phosphorus index for pastures fertilized with poultry litter —

factors affecting phosphorus runoff J Environ Qual 33:2183–2191.

Donigian, A.S., Jr., 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.