Landscape and Watershed Processes

Landscape and Watershed Processes Modeling Phosphorus Concentrations in Irish Rivers Using Land Use, Soil Type, and Soil Phosphorus Data Karen Daly,* Paul Mills, Brian Coulter, and Martin McGarrigle ABSTRACT

held diffuse pollution precepts to establish a framework of key components based on field and catchment studies. Broadly, the conceptual model of P transfer identifies hydrology and source as two key compartments, which expand to embrace other related factors. While the hydrological parameters of rainfall intensity, overland flow, and subsurface and macropore flow are recognized factors controlling diffuse P transfer (Haygarth and Jarvis, 1997; Stamm et al., 1998; Heathwaite, 1997; Pionke et al., 1997), the source compartment describes the influence of soil and agronomic practices. Soil chemical and biological processes such as P sorption and desorption, and microbial mineralization, contribute to the retention and release of soil P (Sharpley et al., 1993, 1995, 1996; Sharpley, 1995; Heckrath et al., 1995). Modeling diffuse P losses to surface waters typically follows one of two approaches; deterministic modeling, which seeks to describe an environmental process, or empirical equations, which seek the simplest relationships between variables. Deterministic models can simulate a process such as overland flow generation or kinetics of soil P desorption, and provide a fundamental understanding of a physical process. However, they are often data intensive and too complex as a tool for environmental managers. In contrast, the empirical model offers a more holistic approach that establishes relationships between variables for the identification and management of P loss from the landscape. The export coefficient and risk assessment techniques are examples of such an approach that predict losses from defined areas. The export coefficient method has calculated P and N loads from UK catchments based on export rates that are assigned to land use categories, fertilizer usage, livestock numbers, and human population (Johnes and O’Sullivan, 1989; Johnes, 1996; Johnes and Heathwaite, 1997; Johnes et al., 1996; McGuckin et al., 1999). The risk assessment technique assigns an arbitrary weighting to similar criteria or site characteristics (runoff, erosion, soil P test, fertilizer, and manure loading) that are either summed (Lemunyon and Gilbert, 1993; Magette, 1998) or multiplied (Gburek et al., 2000) toward a single score or risk value for a catchment. Both approaches embrace a multicriteria analysis of catchment data, weighting parameters either arbitrarily or statistically. A recent amendment to the Irish Water Pollution Act (Department of Environment and Local Government, 1998) has provided a statutory instrument outlining reg-

Modeling diffuse phosphorus (P) loss may indicate management strategies to minimize P loss from agricultural sources. An empirical model predicting flow-weighted phosphorus concentrations (MRP) was derived using data collected from 35 Irish river catchments. Monitoring records of riverine P and stream flow data were used to calculate MRP values averaged for the years 1991–1994. These data were modeled using land use, soil type, and soil P data. Soil type in catchments was described using soil survey classifications weighted according to their P desorption properties from laboratory results. Soil test P concentrations for the studied watersheds were obtained from a national database. Soil P levels were weighted based on the results of field experiments measuring P losses in overland flow from fields at different soil test P levels. The 35 catchments were statistically clustered into two populations (A and B) based on differences in soil type, specifically, soil hydrology. Catchments in Cluster A had predominantly poorly drained soils and comparatively higher MRP concentrations (0.03–0.17 mg L⫺1 ) than Cluster B areas (0.01–0.7 mg L⫺1 ) with mostly well-drained soils. Regression equations derived for A and B type catchments predicted MRP values with 68 and 62% of the variation explained in the models, respectively. Data extracted for the rest of the country were applied to the models to delineate areas at risk on a national scale. While the models were only moderately accurate they highlighted the influence of land management, specifically, high production grassland receiving high P inputs, in conjunction with the effect of soil type and soil hydrology on the transport of P to surface waters.

P

hosphorus (P) inputs from agricultural runoff are typically a major component of nonpoint-source (NPS) surface water pollution (Daniel et al., 1998). Water quality reports published by the Irish Environmental Protection Agency for the period 1991–1994 indicate a fivefold increase in the incidence of slight pollution and a threefold increase in moderate pollution since monitoring began in 1971 (Bowman et al., 1996). Monitoring agencies have concluded that this increase is attributed to diffuse pollution originating from agricultural activities. Attempts to quantify and assess land related P losses on water quality usually involve a multidisciplinary approach considering soil and hydrological processes and their spatial and temporal variations. Haygarth and Jarvis (1998) provide a conceptual model of P transfer from soil to water, which incorporates widely K. Daly and B. Coulter, Teagasc, Johnstown Castle Research Centre, Wexford, Ireland. P. Mills, Compass Informatics, 19 Nassau Street, Dublin, Ireland. M. McGarrigle, Environmental Protection Agency Regional Inspectorate, The Mall, Castlebar, Mayo, Ireland. Received 5 Dec. 2000. *Corresponding author ([email protected]).

Abbreviations: dmf, daily mean flow; MRP, flow-weighted phosphorus concentrations; ortho-P, molybdate reactive orthophosphate.

Published in J. Environ. Qual. 31:590–599 (2002).

590

DALY ET AL.: MODELING P CONCENTRATIONS IN IRISH RIVERS

ulations for specific improvements in surface water quality based on P concentrations and biological parameters. There is now a pressing need for simple models or indices that predict P concentrations or loadings to surface waters based on catchment characteristics and activities, so that the Irish Environmental Protection Agency, local government, and agricultural bodies can address the current legislation. This study strove to develop tools (i.e., models) that could be used to improve catchment

591

scale water quality management, using statistical techniques, but with a view toward combining some of the principles of risk assessment in terms of indexing spatial data on soil type and soil P levels. The research described here looked beyond simply modeling land use and land management, by combining these spatial datasets with results from laboratory and field experiments. Models were constructed that combined experimental data on the effect of soil type and soil P levels on poten-

Fig. 1. Location and identification of 35 subcatchment areas used in the model development.

592

J. ENVIRON. QUAL., VOL. 31, MARCH–APRIL 2002

tial losses to water, with land use, land management, climate, and river water quality data. These data were collected for 35 Irish river subcatchments and modeled to predict flow-weighted P concentrations, and the models extrapolated to other areas of the country to build a national picture of P loss in Irish rivers.

date within the season, and then divided by the sum of the dmfn values for that season. Summer and winter MRP concentrations were calculated for each year during the 1991–1994 period and then averaged over this period. These averaged seasonal concentrations are represented in the model by the variables sumMRP (averaged summer flow-weighted phosphorus concentrations) and winMRP (averaged winter flowweighted phosphorus concentrations).

MATERIALS AND METHODS Land Use

Introduction The models developed in this work used water quality and stream flow data collected for 35 river subcatchments of the Boyne, Barrow, Slaney, Blackwater (Munster), Erne, Suir, Maigue, Deel, Bonet, and Ballysadare rivers. The selection of catchments in this study was based on the availability of suitable stream flow and P concentration data and their location on a national scale are depicted in Fig. 1. The independent variables and the sources of data used in deriving them are listed briefly as follows: Land use (CORINE landcover database for Ireland) Fertilizer P use (Central Statistics Office, Teagasc) Estimates of runoff from long-term rainfall patterns (Met Eireann) Soil types (The General Soil Map of Ireland, Teagasc) Soil test P levels (Teagasc).

Water Quality and Stream Flow Data River P levels were expressed as flow-weighted P concentrations using stream flow data and orthophosphate concentrations collected at 35 river monitoring stations, for the period 1991–1994. Daily mean flow (dmf) data was obtained from records processed by the Office of Public Works from water level recorders along a main channel or stream in hydrometric areas around the country. Orthophosphate concentrations, measured as molybdate-reactive P (ortho-P), were obtained from the Irish Environmental Protection Agency at physico– chemical stations on main channels and tributaries of selected catchments. Flow stations and physico–chemical stations for which data was available were used, giving a total of 35 catchments for model development. Flow-weighted P concentrations (MRP) were calculated using stream flow data and MRP data provided by the physico–chemical stations that coincided with each flow station. To calculate an annual flow-weighted MRP value for a river or stream, river P concentrations on each sampling date (ortho-Pn ) were multiplied by the daily mean flow on that date (dmfn ). These data (ortho-Pn ⫻ dmfn ) were summed for the year and divided by the sum of the dmfn values for that year. Hence, MRP for a given year was derived from flow-weighted river P concentrations (MRP) calculated from river ortho-P monitoring data (ortho-P) and daily mean flow data (dmf), as described by Eq. [1]:

MRP ⫽

⫻m 兺(ortho-Pn ⫻ dmfn) 兺(mg 3 ⫺1 兺dmfn 兺m s L⫺1

3

s⫺1) [1]

Annual MRP concentrations were calculated for 1991–1994 and then averaged over this 4-yr period. The variable AnMRP (averaged annual flow-weighted phosphorus concentrations), the units of which are mg L⫺1, represents these averaged concentrations in the model. To calculate seasonal concentrations, ortho-P and daily mean flow data for summer (May to October) and winter (November to December and January to April) were extracted from the records. Seasonal ortho-P data were multiplied by the corresponding dmfn values for each

Land-use patterns were derived from a digital land cover database of Ireland known as CORINE (Co-ORdination of INformation on the Environment), which provides a national overview of landcover based on the interpretation of LANDSAT satellite imagery recorded in 1989 and 1990. The CORINE dataset, which was part of a Pan-European project, uses a standard scheme of 44 landcover types. For the purposes of the model, nine classes outlined in Table 1 were selected and expressed as proportion of the subcatchment area.

Soil Phosphorus Level In Ireland, Morgan’s reagent (Morgan, 1941) is used in the national soil P testing procedure for agronomic recommendations for soil samples taken at 10-cm depths. Soil P concentrations in catchments were described by combining existing soil test P results from a Teagasc client database, with the results from a field experiment where P concentrations in overland flow were monitored from three fields at different soil test P levels. Soil samples submitted by client farmers in 1996 produced more than 50 000 records of soil test P, which were geocoded, linking each client address to an existing GIS database of some 1200 towns and villages. While the geo-coding of soil test P data did not explicitly record the exact farm location, it provided a generalized GIS map of soil test P based on a mathematical interpolation of the geo-coded soil test data. This data set was summarized into four soil P indices according to soil test P level (Table 2), and GIS used to calculate the area in a catchment within each of the four indices. Instead of assigning arbitrary weightings to each soil test P index, monitoring data from a field experiment provided a more objective approach. The field experiment was designed to monitor and calculate P losses from three grassland fields at soil test P levels of 4 (P1), 7 (P2), and 15 mg L⫺1 (P3) Morgan’s P for soil samples taken at 10-cm depths. This experiment was part of an overall study on P loss from grassland (Tunney et al., 2000) and the initial results covering a monitoring period of 144 d were available for use in the model. Dissolved reactive phosphorus (DRP) concentrations of filtered overland flow samples were collected from automatic samplers placed at the edge of the three sites. The initial results from the experiment suggested Table 1. Land use descriptions and abbreviated terms of CORINE (Co-ORdination of INformation on the Environment) land cover classes used in the model. Abbreviated term Artificial Arable Higrass Lograss Mixgrass Other Ag Forestry Seminatural Peats

Land use description urban and suburban areas, industrial zones, dump sites arable high-productivity grasslands low-productivity grasslands indeterminate mixture of high- and low-productivity grassland area mixed areas of pasture and scrub, agroforestry coniferous, broadleaf, and mixed forestry natural grasslands, heathlands, transitional woodland-scrub peat bogs, peatland areas

593


Table 2. Soil P indices and corresponding ranges of soil test P (Morgan’s) and estimated ratios of dissolved reactive phosphorus (DRP) concentrations collected in an overland flow experiment at Johnstown Castle, Wexford, Ireland.

P index P1 P2 P3 P4

Soil test P (Morgan’s) range

Ratio of DRP concentrations in overland flow monitored from sites for the period 8 Nov. 1997 to 31 Mar. 1998

mg P L⫺1 0–5 6–10 11–20 ⬎20

1 4 10 30

Mean Pfeo per unit Pm (Pfeo/Pm) using mean values in each soil P index Desorption class

P1 (0–5)

[2]

Soil Type A digital soil type dataset based on The General Soil Map of Ireland (Gardiner and Radford, 1980) was used to describe soil types in each of the 35 catchments. These general soil survey descriptions of soil types were combined with laboratory data on soil P desorption characteristics of soil types (Daly et al., 2001) to weight soil types similar to the method used for weighting soil test P data described above. A laboratory based investigation on 90 soil samples encompassing 11 soil types from The General Soil Map of Ireland (Gardiner and Radford, 1980) were analyzed for P desorption using the iron-oxide strip phosphorus (Pfeo) method (Menon et al., 1988). Soil samples from fields at each of the P indices outlined in Table 2 were collected for each soil type at a sample depth of 10 cm. This study concluded that soils with large amounts of extractable Al and Fe and low organic matter levels were more heavily saturated with P and desorbed greater amounts

P2 (6–10)

P3 (11–20)

P4 (⬎20)

mg L⫺1 S1 S2 S3 S4

an approximate ratio of DRP concentrations (relative to the lowest soil P site) of 1:4:10. This ratio was used for weightings and assigned to soil P indices P1, P2, and P3. Using the linear regression relationship between these weightings and soil P levels, a weighting for soils at index P4 (P ⬎ 22 mg L⫺1 ) was extrapolated, and weightings of 1, 4, 10, and 30 were assigned to soils at indices P1, P2, P3, and P4, respectively. To derive a soil P variable for each of the 35 subcatchments, the areas within each catchment in soil indices P1 to P4 were multiplied by their assigned weightings (Table 2). These values were then summed and divided by the total area of the catchment, producing an area-averaged soil P value for each subcatchment, denoted by the variable SoilP. Equation [2] describes the area averaged soil P variable (SoilP) calculated from soil P indices and soil P weightings:

(P1 ⫻ 1)ha ⫹ (P2 ⫻ 4)ha ⫹ (P3 ⫻ 10)ha ⫹ (P4 ⫻ 30)ha Soil P ⫽ Total catchment area (ha)

Table 4. Phosphorus desorption (Pfeo) expressed per unit soil test P (Pm) at each P index for each of the four soil groups (S1 to S4).

4.5 3.8 3.2 2.0

3.3 3.3 2.9 1.8

2.3 2.8 2.1 1.5

no data 2.1 1.1 0.9

of P to solution than high organic matter soils at similar ranges of soil test P level. The experimental methods and soil data are presented in Daly et al. (2001). The 11 soils selected in this work were reduced to four categories of soil type (S1 to S4) based on organic matter levels and the range of Pfeo concentrations extracted (Table 3). To estimate the relative differences between the four soil groups, a soil desorption weighting was derived using average Pfeo and Morgan’s P (Pm) data at each P index for each of the four soil groups. First, the average Pfeo at each P index was divided by the average Pm concentrations in that index, expressing P desorption (Pfeo) per unit Morgan’s P (Table 4). Then, for each P index, these values were expressed as ratios, relative to the lowest values in the S4 group of soils (Table 5). These ratios were then averaged across P indices to give one ratio expressed for each soil group (Table 6), which was used as weighting assigned to soil types. To use this information as an input parameter or variable that described soil type for the model, the soil types in each of the subcatchment were classed according to the soil groups in Table 6. The catchment areas in each of the four soil groups S1 to S4 were multiplied by their corresponding weight. For each catchment, these values were summed and divided by the total catchment area, providing an area-weighted value for soil type in catchments, denoted by the term Soil Type and expressed by Eq. [3]:

(S1 ⫻ 1.9)ha ⫹ (S2 ⫻ 1.9)ha ⫹ (S3 ⫻ 1.4)ha ⫹ (S4 ⫻ 1)ha [3] Soil Type ⫽ Total catchment area (ha) Land Management Mineral fertilizer P use data obtained from the Central Statistics Office (1991) were mapped on the basis of the District Electoral Divisions and are represented as fertilizer use per catchment in Mg P ha⫺1. This variable is denoted by the term Fertilizer.

Table 3. Soil types split into four soil groups S1 to S4 according to percent organic matter, with mean P desorption (Pfeo) and mean soil test P (Pm) presented at each P index for each soil group. Soil group

Organic matter

Principal soil types

Mean

P1 (0–5)

P2 (6–10)

P3 (11–20) mg

S1

⬍12

S2

13–20

Gleys

S3

21–30

Peaty Gleys, Peaty Podzols

S4

⬎30

Brown Earths, Podzolics

Peats

mean mean mean mean mean mean mean mean

Pfeo Pm Pfeo Pm Pfeo Pm Pfeo Pm

12.2 2.7 14.6 3.8 13.8 4.3 4.4 2.2

22.9 6.9 24.7 7.6 21.1 7.3 13.6 7.5

P4 (⬎20)

L⫺1 34.2 14.7 34.9 12.6 39.1 18.5 28.2 18.6

no data no data 65 30.7 33.2 30.5 28.2 28.7

594


Table 5. Phosphorus desorption (Pfeo) per unit soil test P (Pm) expressed as a ratio relative to the lowest values in S4, at each P index, in each soil group. Ratio of mean Pfeo per unit Pm relative to S4 values Desorption class

P1 (0–5 mg

S1 S2 S3 S4

L ⫺1 )

P2 (6–10 mg L⫺1 )

P3 (11–20 mg L⫺1 )

P4 (⬎20 mg L⫺1 )

1.8 1.8 1.6 1

1.5 1.8 1.4 1

no data 2.1 1.1 1

2.2 1.9 1.6 1

Climate Rainfall data from 700 Met Eireann weather stations for the period 1960–1990 were used to build a GIS database of long-term average rainfall patterns on a national scale. Effective rainfall (rainfall less assumed evapotranspiration of 455 mm) was calculated from these long-term averages to derive an estimate of runoff or overland-flow patterns for the country and the term Runoff described this parameter for each of the subcatchments.

Statistical Analysis The datasets described above were combined for each of the 35 catchment areas to derive an empirical model of river P levels. Data were analyzed using SAS procedures (SAS Institute, 1988) and Datadesk (Data Description, 1997). Testing for differences about the mean and median of data was carried out using the Mann–Whitney and t tests in Datadesk. Correlations were calculated using a Pearson Product– Moment Correlation and the correlation coefficient R refers to linear correlations only. The significance of R and R2 values are denoted by symbols *, **, and *** for significance levels of p ⫽ 0.05, 0.01, and 0.001, respectively.

RESULTS Relationships between Catchment Characteristics The datasets on water quality, land use, soil type, soil P levels, fertilizer use, and runoff were extracted from a GIS for each of the 35 catchment areas. To examine the relationships between variables, these data were combined in a correlation analysis. Linear regression statistics are presented in Table 7. The land use category Higrass correlated positively with Soil Type (R 2 ⫽ 0.58***) but was negatively associated with the Runoff parameter (R 2 ⫽ 0.69***), indicating that high-production grassland areas are on predominantly well-drained soils such as those of the S1 component of the soil type variable. The weak, but significant, positive correlation (R 2 ⫽ 0.26**) between SoilP and the land use class Artificial, describing mostly urban areas, was probably derived from the geo-coding of soil P data, which used nearest towns and villages to assign a location to soil P values. Although the correlation between annual flow-

weighted MRP values (AnMRP) and Higrass was weak (R 2 ⫽ 0.13*), the scatter plot between these two variables depicted stronger but separate trends within two clusters of data points. A cluster analysis of AnMRP and Higrass was carried out using Datadesk’s Complete Linkage Clustering. The cluster “tree” that emerged displayed a similar grouping of catchments as depicted in the scatter plot (Fig. 2) between these two variables. This pattern of clustering in the data was examined by separating the two clusters in Fig. 2 denoted as Clusters A and B. To examine the cause of this effect the catchment characteristics of each cluster were explored.

Testing for Differences in Catchment Characteristics between Clusters A and B An initial inspection of the water quality data between the two clusters indicated that MRP values in Cluster A (n ⫽ 17) exceeded those in Cluster B (n ⫽ 18). The difference was significant when the data were subjected to hypothesis testing using a t test (t ⫽ 4.2**) and a Mann–Whitney (U) test (U ⫽ 243***). In addition, summer MRP values in Cluster A exceeded winter values, indicating some seasonality in the data, but no seasonal effects were apparent in the MRP data of Cluster B. A paired t test (tp) on summer and winter data in Cluster A indicated that the seasonal differences were significant (t ⫽ 4.7**) but when similar data in Cluster B were tested no significant differences were found. Other statistically significant differences between the two clusters were found in the land use and soil type data. The proportion of land use categorized as Lograss, representing low-productivity grassland, was higher in Cluster A type catchments compared with Cluster B areas and significant differences about the mean (t ⫽

Table 6. Soil types arranged according to percent organic matter and associated desorption weighting. Soil group S1 S2 S3 S4

Principal soil type Brown Earths, Podzolics, Grey Brown Podzolic Gleys Peaty Gleys, Peaty Podzols Peat

Organic matter

Desorption weighting

% ⬍12

1.9

12.1–20 20.1–30 ⬎30

1.9 1.4 1

Fig. 2. Relationship between proportion of high production grassland (Higrass) in a catchment and the annual flow-weighted phosphorus concentrations (MRP) values for 35 catchments, where the two clusters of data, A and B, are highlighted and separated by the broken line.

595


Table 7. Linear regression equations between catchment variables using all subcatchments. Y

Variable

Higrass

Soil Type Constant Runoff Constant Artifical Constant Higrass Constant

Higrass SoilP AnMRP

Coefficient

Standard error coefficient

Probability

R2

0.192 0.341 0.123 0.074 46.4 0.248 0.033 0.019

⬍0.0001 ⬍0.0001 ⬍0.0001 ⬍0.0001 ⬍0.0001 0.0019 0.03 0.12

0.58

1.301 ⫺1.771 ⫺1.052 1.143 157.3 3.498 0.073 0.030

2.5*) and median (U ⫽ 191*) existed between the two clusters. Low-production grasslands are generally on poor soils with impeded drainage, and this was reflected in differences in soil types between the two clusters. Originally, the variable Soil Type was derived to represent a soil P desorption index of some grassland soils, using four categories S1 to S4 (Table 6). However, the distributions of each of the soil desorption categories varied between clusters. Cluster A catchments had a greater proportion of S4 soils as confirmed by the t test (t ⫽ 2.1*) and Mann–Whitney test (U ⫽ 88*) while Cluster B catchments had a greater proportion of welldrained S1 soils (t ⫽ ⫺2.6*). The differences between the two clusters appeared to be reflected in land use and soil type, indicating that the cause for clustering could lie in soil hydrology. To investigate this hypothesis further the soil type data were reorganized into wet and dry soil categories. The S1 soils were comprised of mostly well-drained Brown Earths and Podzolics and renamed Dry Soils, while the poorly drained Gleys (S2), Peaty Gleys, Peaty Podzols (S3), and Peats (S4) were grouped and renamed Wet Soils. Test statistics confirmed that significant differences about the mean (t ⫽ 2.7**) and median (U ⫽ 199**) existed in the wet soil data between the two clusters. Thus, A type catchments were characterized by predominantly wet soils, while B type catchments had mostly dry, well-drained soil coverage. Summary statistics of data in each cluster are presented in Table 8.

Modeling River Phosphorus Levels in Each Cluster For data in Cluster A, river MRP values were positively associated with Higrass (R 2 ⫽ 0.69***) and negatively associated with Peat (R 2 ⫽ 0.84***) and Seminatural areas (R 2 ⫽ 0.57***). The land use class Higrass correlated positively with fertilizer use (R 2 ⫽ 0.79***),

0.69 0.26 0.13

indicating that fertilizer P applied to this land use class may contribute to MRP levels in these catchments. The soil type variable was derived by weighting soil categories according to P desorption. However, since A type catchments had mostly wet soils (S2, S3, and S4 soils) the correlation between Soil Type and river MRP values, in particular, winter MRP values (R 2 ⫽ 0.68***) probably reflects the influence of soil hydrology rather than soil P chemistry, on river MRP levels. While the Soil Type variable was constructed to describe soils in terms of potential losses due to desorption, the hydrology of soils was the more dominant factor. An empirical model describing river MRP levels for Cluster A catchments was constructed using statistical techniques and selecting variables that reflect diffuse P losses to water. Four variables were selected that might contribute to river MRP levels in Cluster A type catchments. The land use category Higrass was selected, since it receives applications of artificial fertilizer. The categories describing seminatural areas and peat land were combined into one variable denoted as Semipeat, since these areas are similar and normally associated with each other and correlated with MRP data. Given the strong relationship between soil P levels and P losses in overland flow from the field experiments, Soil P was also included. The soil type parameter, while originally constructed to describe soil types in terms of P desorption, was selected, but it was considered in the context of soil hydrology. Due to the high intercorrelation among variables, a principal components analysis (PCA) was carried out using four variables: Higrass, Semipeat, SoilP, and Soil Type, so that MRP data could be described using fewer and uncorrelated dimensions. Four components were generated and the first component (L1) selected since it accounted for 79% of the variability in the data. This component of variables and component weightings, L1, was used as a new variable in a

Table 8. Summary statistics of annual and seasonal flow-weighted phosphorus concentrations (MRP) and flow data, proportion of wet and dry soils, and low-production grassland areas in the catchments of Clusters A and B. Win ⫽ winter; Sum ⫽ summer; An ⫽ annual. Statistic

WinMRP

SumMRP

AnMRP

WinFlow

mg L⫺1 Mean Median Min. Max.

0.09 0.07 0.02 0.16

0.12 0.10 0.03 0.25

0.09 0.08 0.03 0.17

5 368 5 392 2 283 10 510

Mean Median Min. Max.

0.05 0.05 0.02 0.09

0.04 0.05 0.01 0.09

0.04 0.05 0.01 0.07

4 241 4 465 1 967 6 057

SumFlow m3 ha⫺1 Cluster A 1 837 1 476 792 4 756 Cluster B 1 395 1 272 605 2 864

AnFlow

Wet Soils proportion

Dry Soils proportion

Lograss proportion

7 206 6 694 3 548 15 265

0.66 0.58 0.38 1

0.34 0.42 0 0.62

0.12 0.11 0 0.33

5 636 5 972 2 894 8 921

0.43 0.38 0.05 1

0.57 0.62 0 0.95

0.05 0.03 0 0.17

596


Table 9. Regression model for Cluster A catchments using winter, summer, and annual flow-weighted phosphorus concentrations (MRP) values using the principal component L1. Y

Variable

Coefficient


Probability

R2

WinMRP

Constant L1 Constant L1 Constant L1

0.086 0.041 0.117 0.049 0.097 0.040

0.007 0.007 0.009 0.009 0.007 0.007

⬍0.0001 ⬍0.0001 ⬍0.0001 ⬍0.0001 ⬍0.0001 ⬍0.0001

0.69

SumMRP AnMRP

0.64 0.68

regression step with MRP data in Cluster A catchments and is described by Eq. [4]. The principal component L1 accounted for 69% of the variation in winter MRP data and 68% of the variation in annual MRP values in Cluster A (Table 9). Equation [4] describes the first principal component (L1) of variables and component weightings, used in the regression model in Table 9: L1 ⫽ 1.21 ⫻ Higrass ⫹ 1.86 ⫻ Soil Type ⫹ 0.15 ⫻ Soil P ⫺ 2.81 ⫻ Semipeat ⫹ 4.08

[4]

A model of MRP data in Cluster B catchments was derived by first considering some of the characteristics of these catchments and also examining the correlations among data. Hypothesis testing elucidated the differences in soil hydrology between the two Clusters A and B, where soil type in Cluster B areas was described by predominantly well-drained soils (Brown Earths and Podzolics). As most of the soils in Cluster B catchments fell into the S1 group, and did not vary in hydrological conditions, the soil type variable was not included in a model for Cluster B areas. The correlations among data in Cluster B were compared with Cluster A data, with the most significant R values at the p ⫽ 0.05 level. However, annual river MRP data correlated positively with land use categories Higrass (R 2 ⫽ 0.30*) and artificial surfaces (R 2 ⫽ 0.31*), while summer MRP values correlated negatively with low-production grassland, Lograss (R 2 ⫽ 0.24*). The associations between water quality data and land use in these catchments suggest P loss from high-production grassland areas with some inputs from point sources or urban areas. To model MRP data for Cluster B areas the variables selected were Higrass, Lograss, and SoilP. Soil type was omitted since these are drier catchments with predominantly S1 group soil cover. As these variables were uncorrelated with each other, a multiple regression approach was used to predict river MRP. However, summer MRP gave the best fit to the data (R 2 ⫽ 0.62) in contrast to winter MRP (R 2 ⫽ 0.33), and the model derived to predict summer MRP data in Cluster B catchments is presented in Table 10. Table 10. Regression model for Cluster B catchments using summer flow-weighted phosphorus concentrations (MRP). Y

Variable

Coefficient


SumMRP

Constant Higrass Lograss SoilP

⫺0.056 0.074 ⫺0.211 0.018

0.031 0.032 0.082 0.006

Probability

R2

0.0924 0.0349 0.0215 0.0115

0.62

Predicting River Flow-Weighted Phosphorus Concentrations on a National Scale The models developed for Cluster A and B type catchments predicted river MRP values using land use, soil P data, and soil type information. However, to predict these values for other areas not used in the model development, the first criterion was to categorize other areas according to their catchment characteristics as being representative of Cluster A or B type catchments. In order to do this on a national scale, a geographical template was required to provide units from which input data (land use, soil P, and soil type) could be extracted and the appropriate model applied. A national map of river catchment areas was considered initially and later rejected on the basis that the large scale of these catchments might mask variations in land cover and soil type required to assign the appropriate model. Thus, the country was partitioned into 5-km2 grid cells and the data extracted from each cell was used to classify these areas as either A or B catchments before applying the relevant model to construct a national picture of river MRP levels (Fig. 3). The values of predicted MRP for both categories of catchments or clusters were categorized into four classes A1/B1 to A4/B4 according to the range of values in each class. These classes were constructed to reflect the range of median ortho-P concentrations used by the Irish Environmental Protection Agency for water quality classification (Table 11). Figure 3 depicts predicted MRP values for Cluster A and B type areas on a national scale. A limited validation of the models was carried out using some available data from 15 subcatchments of the Lough Conn and Mask regions in the west of Ireland. The areas were initially classed as either A or B type, depending on the dominant soil types, and the appropriate model was run to predict flow-weighted MRP values. These data were then compared with observed total P (TP) data provided from an ongoing study in these catchments. The results suggested that the model was reasonably linear accounting for 67 and 29% (Fig. 4) of variation in P exports. The models however, appeared to overpredict observations, especially as TP would normally be some two to six times the ortho-P concentrations (McGarrigle, personal communication). Stevens and Smith (1978) estimated that soluble reactive P accounted for half the total P entering the River Main in County Antrim. The models may typically have predicted values that were three to four times too high. Table 11. Classification of predicted values of flow-weighted phosphorus concentrations (MRP) for Cluster A (A1–A4) and B (B1–B4) type areas corresponding to Irish Environmental Protection Agency water classification using median ortho-P values. MRP class

MRP range

Irish EPA median ortho-P range and associated pollution class

A1/B1 A2/B2 A3/B3 A4/B4

0.000–0.030 0.031–0.060 0.061–0.100 ⬎0.1

mg L⫺1 0.015–0.03 (unpolluted) 0.045 (slightly polluted) 0.007 (moderately polluted) ⬎0.1 (heavy/gross pollution)


597

Fig. 3. Map of predicted flow-weighted phosphorus concentrations (MRP) on a national scale using the models developed for Cluster A and B type areas.

DISCUSSION The derivation of separate models for catchments was prompted by the clustering of data in the scatter plot between MRP and Higrass, in Fig. 2. Hypothesis testing of the data in each cluster indicated that data clustered according to soil type, in particular, soil hydrology. The hydrological differences between the two clusters of catchments were also reflected in river flow data where catchments with predominantly poorly drained soil cover had comparatively higher river flows, although these differences were not statistically significant. In addition, flow-weighted P concentrations in these catchments were also higher, indicating that poorly drained soils produce greater amounts of overland flow and carry more P to surface waters compared with drier well-drained soils. The significantly higher MRP in Cluster A catchments can be explained further by examining the components

of the model derived for these areas. These were land use (Higrass, Semipeat), soil type, and soil P level. If the mathematical sign (i.e., ⫹ or ⫺) of a coefficient in the model is an indication of whether a variable is a source or a sink for P, then for Cluster A, Higrass, SoilP, and Soil Type were sources of P while Semipeat was a sink. The contribution of soil P and soil hydrology to overland flow P concentrations in Ireland has be documented by Tunney et al. (2000) while the Higrass variable in these catchments was associated with fertilizer P use, which could exacerbate P loss to water. However, that Semipeat is a sink for P in this model is less easily explained in the same way that forestry is often cited as a sink for nutrients (Ryden et al., 1973; Tufford et al., 1998). The negative weighting assigned to Semipeat would suggest that low MRP concentrations are associated with seminatural and peatland areas. These areas are associated with peat and peaty gley soil, which have

598


the stronger correlations among catchment characteristics in Cluster A data indicate that including the soil type parameter, and hence soil hydrological conditions, is central to modeling diffuse P losses to water. ACKNOWLEDGMENTS This study was undertaken as part of the Environmental Monitoring R&D sub-programme of the Operational Programme for Environmental Services, 1994–1999. The financial and technical assistance of the Environmental Protection Agency and Teagasc is gratefully acknowledged. The authors thank Ms. Isabelle Kurz, Prof. David Jeffrey, and Dr. William Magette for their thoughtful comments and suggestions during the course of this research.

REFERENCES

Fig. 4. Relationship between observed total P and predicted flowweighted phosphorus concentrations (MRP) for (a ) Cluster A and (b ) Cluster B type areas of the Lough Conn and Mask catchments.

a poor capacity for storing P and are usually located in high rainfall areas (Tunney et al., 2000). If this land cover class is located near the stream or river, then partial area theory suggests that its contribution to overland flow into a waterbody will be significant. In addition, if nutrients are moving from a neighboring cultivated or high production grassland area, then seminatural or peat land areas will be unable to trap nutrients in the same way that forested areas do. If this was the case in Cluster A catchments, then the higher MRP values could be due to a combination of catchment hydrology and a predominant soil cover incapable of trapping or absorbing nutrients applied to or moving from the landscape to the streams. The catchments in Cluster B were defined as comparatively drier catchments with well-drained soil cover, with lower river flow and MRP levels. The model derived for these areas used multiple regression with Higrass, Lograss, and SoilP, indicating that Higrass and Soil P acted as sources of P while Lograss may have been a sink if Lograss areas are on nutrient-deficient soils. The critical step in the development of models is validation of the equations with independent data. This was moderately successful (R 2 ⫽ 0.67) when predicted data from the model derived for Cluster A type catchments were compared with observed data from the Loughs Conn and Mask regions in the west of Ireland. However, predicted and observed values for Cluster B type catchments were poorly correlated (R 2 ⫽ 0.29). The different levels of agreement between predicted and observed data between the two models and also

Bowman, J.J., K.J. Clabby, J. Lucey, M.L. McGarrigle, and P.F. Toner. 1996. Water quality in Ireland 1991–1994. Irish Environmental Protection Agency, Wexford, Ireland. Central Statistics Office. 1991. Census of agriculture. Gov. Publ., Dublin, Ireland. Daly, K., D. Jeffrey, and H. Tunney. 2001. The effect of soil type on phosphorus sorption capacity and desorption dynamics in Irish grassland soils. Soil Use Manage. 17:12–20. Daniel, T.C., A.N. Sharpley, and J.L Lemunyon. 1998. Agricultural phosphorus and eutrophication: A symposium overview. J. Environ. Qual. 27:251–257. Data Description. 1997. Datadesk 6.0. Data Description, Ithaca, NY. Department of Environment and Local Government. 1998. Water quality standards for phosphorus. Regulations S.I. No. 258. DELG, Dublin, Ireland. Gardiner, M.J., and T. Radford. 1980. Soil associations of Ireland and their land use potential: Explanatory bulletin to soil map of Ireland. National Soil Survey of Ireland, An Foras Talu´ntas. Gburek, W.J., A.N. Sharpley, L. Heathwaite, and G.J. Folmar. 2000. Phosphorus management at the watershed scale: A modification of the phosphorus index. J. Environ. Qual. 29:130–144. Haygarth, P.M., and S.C. Jarvis. 1997. Soil derived phosphorus in surface runoff from grazed grassland lysimeters. Water Res. 31: 140–148. Haygarth, P.M., and S.C. Jarvis. 1998. Phosphorus transfers from agricultural soils described by a conceptual model. p. 2–5. In 3rd Int. Conf. on Diffuse Pollution. Proc., Edinburgh, Scotland. 31 Aug.–4 Sept. 1998. IAWQ, Edinburgh. Heathwaite, A.L. 1997. Sources and pathways of phosphorus loss from agriculture. p. 205–223. In H. Tunney et al. (ed.) Phosphorus loss from soil to water. CAB Int., Wallingford, UK. Heckrath, G., P.C. Brookes, P.R. Poulton, and K.W.T. Goulding. 1995. Phosphorus leaching from soils containing different phosphorus concentrations in the Broadbalk experiment. J. Environ. Qual. 24: 904–910. Johnes, P.J. 1996. Evaluation and management of the impact of land use change on the nitrogen and phosphorus load delivered to surface waters: The export coefficient modelling approach. J. Hydrol. 183:323–349. Johnes, P.J., and A.L. Heathwaite. 1997. Modelling impact of land use change on water quality in agricultural catchments. Hydrol. Processes 11:269–286. Johnes, P., B. Moss, and G. Phillips. 1996. The determination of total nitrogen and total phosphorus concentrations in freshwaters from land use, stock headage and population data: Testing of a model for use in conservation and water quality management. Freshwater Biol. 36:451–473. Johnes, P.J., and P.E. O’Sullivan. 1989. Nitrogen and phosphorus losses from the catchment of Slapton Ley, Devon—An export coefficient approach. Field Studies 7:285–309. Lemunyon, J., and R.G. Gilbert. 1993. The concept and need for a phosphorus assessment tool. J. Prod. Agric. 6:483–486. Magette, W.L. 1998. Factors affecting losses of nutrients from agricultural systems and delivery to water resources. p. 6–31. In O.T. Carton (ed.) Draft guidelines for nurient use in intensive agricultural enterprises. Teagasc, Johnstown Castle, Wexford, Ireland. McGuckin, S.O., C. Jordan, and R.V. Smith. 1999. Deriving phospho-

WHICKER ET AL.: WIND EROSION IN UNBURNED AND BURNED SHRUBLAND

rus export coefficicents for CORINE land cover types. Water Sci. Technol. 39:47–53. Menon, R.G., L.L Hammond, and H.A. Sissingh. 1988. Determination of plant available phosphorus by the iron-oxide impregnated filter paper (Pi) soil test. Soil Sci. Soc. Am. J. 53:110–115. Morgan, M.F. 1941. Chemical soil diagnosis by the Universal Soil Testing System. Bull. 450. Connecticut Agric. Exp. Stn., New Haven. Pionke, H.B., W.J. Gburek, A.N. Sharpley, and J.A. Zollweg. 1997. Hydrological and chemical controls on phosphorus loss from catchments. p. 225–242. In H. Tunney et al. (ed.) Phosphorus loss from soil to water. CAB Int., Wallingford, UK. Ryden, J.C., J.K. Syers, and R.F. Harris. 1973. Phosphorus in runoff and streams. p. 1–45. In N.C. Brady (ed.) Advances in agronomy 25. Academic Press, New York. SAS Institute. 1988. SAS/STAT user’s guide. Release 6.03 ed. SAS Inst., Cary, NC. Sharpley, A.N. 1995. Dependence of runoff phosphorus on extractable soil phosphorus. J. Environ. Qual. 24:920–926. Sharpley, A.N., T.C. Daniel, and D.R. Edwards. 1993. Phosphorus movement in the landscape. J. Prod. Agric. 6:492–500. Sharpley, A.N., T.C. Daniel, J.T. Sims, and D.H. Pote. 1996. Determin-

599

ing environmentally sound soil phosphorus levels. J. Soil Water Conserv. 51:160–166. Sharpley, A.N., M.J. Hedley, E. Sibbesen, A. Hillbricht-Ilkowska, A.W. House, and L. Ryszkowski. 1995. Phosphorus transfers from terrestrial to aquatic ecosystems. p. 171–200. In H. Tiessen (ed.) Phosphorus in the global environment: Transfers, cycles and management. Wiley Press, Chichester, UK. Stamm, C., H. Flu¨hler, R. Ga¨chter, J. Leuenberger, and H. Wunderli. 1998. Preferential transport of phosphorus in drained grassland soils. J. Environ. Qual. 27:515–522. Stevens, R.J., and R.V. Smith. 1978. A comparison of discrete and intensive sampling for measuring the loads of nitrogen and phosphorus in the River Main, County Antrim. Water Res. 12:823–830. Tufford, D.L., H.N. McKellar, and J.R. Hussey. 1998. In-stream nonpoint source nutrient prediction with land-use proximity and seasonality. J. Environ. Qual. 27:100–111. Tunney, H., B. Coulter, K. Daly, I. Kurz, C. Coxon, D. Jeffrey, P. Mills, G. Kiely, and G. Morgan. 2000. Quantification of phosphorus (P) loss to water due to soil P desorption. Environmental Protection Agency, Wexford, Ireland.

Temporal and Spatial Variation of Episodic Wind Erosion in Unburned and Burned Semiarid Shrubland Jeffrey J. Whicker,* David D. Breshears, Piotr T. Wasiolek, Thomas B. Kirchner, Rebecca A. Tavani, David A. Schoep, and John C. Rodgers ABSTRACT Redistribution of soil, nutrients, and contaminants is often driven by wind erosion in semiarid shrublands. Wind erosion depends on wind velocity (particularly during episodic, high-velocity winds) and on vegetation, which is generally sparse and spatially heterogeneous in semiarid ecosystems. Further, the vegetation cover can be rapidly and greatly altered due to disturbances, particularly fire. Few studies, however, have evaluated key temporal and spatial components of wind erosion with respect to (i) erosion rates on the scale of weeks as a function of episodic high-velocity winds, (ii) rates at unburned and burned sites, and (iii) within-site spatial heterogeneity in erosion. Measuring wind erosion in unburned and recently burned Chihuahuan desert shrubland, we found (i) weekly wind erosion was related more to daily peak wind velocities than to daily average velocities as consistent with our findings of a threshold wind velocity at approximately 7 m s⫺1; (ii) greater erodibility in burned vs. unburned shrubland as indicated by erosion thresholds, aerodynamic roughness, and nearground soil movement; and (iii) burned shrubland lost soil from intercanopy and especially canopy patches in contrast to unburned shrubland, where soil accumulated in canopy patches. Our results are among the first to quantify post-fire wind erosion and highlight the importance of accounting for finer temporal and spatial variation in shrubland wind erosion. This finer-scale variation relates to semiarid land degradation, and is particularly relevant for predictions of contaminant resuspension and redistribution, both of which historically ignore finer-scale temporal and spatial variation in wind erosion.

Jeffrey J. Whicker, Pitor T. Wasiolek, Rebecca A. Tavani, and John C. Rodgers, Health Physics Measurements, Los Alamos National Laboratory, Mail Stop G761, Los Alamos, NM 87545. David D. Breshears, Environmental Dynamics and Spatial Analysis, Earth and Environmental Sciences Division, Los Alamos National Laboratory, Mail Stop J495, Los Alamos, NM 87545. Thomas B. Kirchner and David A. Schoep, Carlsbad Environmental Monitoring and Research Center, 1400 University Drive, Carlsbad, NM 88220. Received 27 Mar. 2001. *Corresponding author ([email protected]). Published in J. Environ. Qual. 31:599–612 (2002).

S

oil is redistributed and transported by wind erosion in semiarid ecosystems. Rates of wind erosion fundamentally depend on the characteristics of wind in a complex fashion (Bagnold, 1941). For example, wind erosion exhibits a threshold-like response to increasing wind velocities (Bagnold, 1941; Gillette et al., 1980; Nicholson, 1993; Belnap and Gillette, 1998). Similarly, concentrations of airborne dust and soil contaminants are predicted to increase as a power function with velocity (Woodruff and Siddoway, 1965; Anspaugh et al., 1975). Because of these nonlinear relationships, wind erosion may occur primarily during episodic, high-wind events (Helgren and Prospero, 1987; Cahill et al., 1996; Goossens and Offer, 1997; Godon and Todhunter, 1998; Stout, 2001). However, few studies have evaluated how wind erosion over longer time frames (e.g., weeks) relates to net soil redistribution and transport of associated nutrients and contaminants during episodic, highvelocity wind events. Wind erosion also fundamentally depends on landsurface characteristics of vegetation structure and associated ground cover (Fryrear, 1985). Characteristics of the vegetation matrix that are particularly influential on wind erosion include the amount, type, and spatial pattern of vegetation (Raupach et al., 1993; Wolfe and Nickling, 1996). One of the key characteristics of this matrix is the proportion and spacing of two fundamentally different patch types: the canopy patches associated with woody plants (trees and shrubs) and the intercanopy patches that separate them (Belsky and Canham, 1994; Scholes and Archer, 1997; Breshears and Barnes, 1999). These two patch types differ in a number of Abbreviations: DOE, United States Department of Energy; TSP, total suspended particulate; WIPP, Waste Isolation Pilot Plant.