Applying Statistical Causal Analyses to Agricultural Conservation: A Case Study Examining P Loss Impacts

Estimating the effect of agricultural conservation practices on reducing nutrient loss using observational data can be confounded by factors such as differing crop types and management practices. As we may not have the full knowledge of these confounding factors, conventional statistical meta‐analysis methods can be misleading. We discuss the use of two statistical causal analysis methods for quantifying the effects of water and soil conservation practices in reducing P loss from agricultural fields. With the propensity score method, a subset of data was used to form a treatment group and a control group with similar distributions of confounding factors. With the multilevel modeling method, data were stratified based on important confounding factors, and the conservation practice effect was evaluated for each stratum. Both methods resulted in similar estimates of the conservation practice effect (total P load reduction avg. ~70%). In addition, both methods show evidence of conservation practices reducing the incremental increase in total P export per unit increase in fertilizer application. These results are presented as examples of the types of outcomes provided by statistical causal analyses, not to provide definitive estimates of P loss reduction. The enhanced meta‐analysis methods presented within are applicable for improved assessment of agricultural practices and their effects and can be used for providing realistic parameter values for watershed‐scale modeling.     

A Multilevel Model of the Impact of Farm‐Level Best Management Practices on Phosphorus Runoff1

Abstract: Multilevel or hierarchical models have been applied for a number of years in the social sciences but only relatively recently in the environmental sciences. These models can be developed in either a frequentist or Bayesian context and have similarities to other methods such as empirical Bayes analysis and random coefficients regression. In essence, multilevel models take advantage of the hierarchical structure that exists in many multivariate datasets; for example, water quality measurements may be taken from individual lakes, lakes are located in various climatic zones, lakes may be natural or man‐made, and so on. The groups, or levels, may effectively yield different responses or behaviors (e.g., nutrient load response in lakes) that often make retaining group membership more effective when developing a predictive model than when working with either all of the data together or working separately with the individuals. Here, we develop a multilevel model of the impact of farm level best management practices (BMPs) on phosphorus runoff. The result of this research is a model with parameters which vary with key practice categories and thus may be used to evaluate the effectiveness of these practices on phosphorus runoff. For example, it was found that the effect of fertilizer application rate on farm‐scale phosphorus loss is a function of the application method, the hydrologic soil group, and the land use (crop type). Further, results indicate that the most effective method for controlling fertilizer loss is through soil injection. In summary, the resultant multilevel model can be used to estimate phosphorus loss from farms and hence serve as a useful tool for BMP selection.     

Evaluation of a Watershed Model for Estimating Daily Flow Using Limited Flow Measurements1

Abstract: The Soil and Water Assessment Tool (SWAT) model was evaluated for estimation of continuous daily flow based on limited flow measurements in the Upper Oyster Creek (UOC) watershed. SWAT was calibrated against limited measured flow data and then validated. The Nash‐Sutcliffe model Efficiency (NSE) and mean relative error values of daily flow estimations were 0.66 and 15% for calibration, and 0.56 and 4% for validation, respectively. Also, further evaluation of the model’s estimation of flow at multiple locations was conducted with parametric paired t‐test and nonparametric sign test at a 95% confidence level. Among the five main stem stations, four stations were statistically shown to have good agreement between predicted and measured flows. SWAT underestimated the flow of the fifth main stem station possibly because of the existence of complex flood control measures near to the station. SWAT estimated the daily flow at one tributary station well, but with relatively large errors for the other two tributaries. The spatial pattern of predicted flows matched the measured ones well. Overall, it was concluded from the graphical comparisons and statistical analyses of the model results that SWAT was capable of reproducing continuous daily flows based on limited flow data as is the case in the UOC watershed.     

EVALUATION OF ROSGEN'S STREAMBANK EROSION POTENTIAL ASSESSMENT IN NORTHEAST OKLAHOMA1

ABSTRACT: In this study three components of the Rosgen Level III Stream Reach Condition Assessment were tested for their ability to predict short-term erosion rates. Rosgen's bank erosion potential (BEP) ratings and near bank stress (NBS) estimates and the Pfankuch channel stability ratings were evaluated. Thirty-six banks with a range of BEP ratings and NBS estimates were selected on the 101 km Upper Illinois River in northeast Oklahoma. The Upper Illinois River is a meandering, gravel-dominated, riffle/pool channel. Cumulative erosion data measured with bank pins after four 2.0 to 2.5-year return period flows from September 1996 to July 1997 were used in the analyses. When integrated as indicated in Rosgen (1996), the BEP indices and NBS estimates were poor predictors of bank erosion. Individually, the grouped BEP ratings and Pfankuch ratings performed relatively well compared to grouped NBS estimates in predicting erosion; however, the variability of erosion was large within each rating group. Linear regression between erosion and BEP numerical indices and Pfankuch scores was significant (a = 0.05), but variability was high (illustrated by low r² values). Regression between erosion and NBS estimates was not significant.     

The MANAGE database: nutrient load and site characteristic updates and runoff concentration data

The "Measured Annual Nutrient loads from AGricultural Environments" (MANAGE) database was developed to be a readily accessible, easily queried database of site characteristic and field-scale nutrient export data. The original version of MANAGE, which drew heavily from an early 1980s compilation of nutrient export data, created an electronic database with nutrient load data and corresponding site characteristics from 40 studies on agricultural (cultivated and pasture/range) land uses. In the current update, N and P load data from 15 additional studies of agricultural runoff were included along with N and P concentration data for all 55 studies. The database now contains 1677 watershed years of data for various agricultural land uses (703 for pasture/rangeland; 333 for corn; 291 for various crop rotations; 177 for wheat/oats; and 4-33 yr for barley, citrus, vegetables, sorghum, soybeans, cotton, fallow, and peanuts). Across all land uses, annual runoff loads averaged 14.2 kg ha(-1) for total N and 2.2 kg ha(-1) for total P. On average, these losses represented 10 to 25% of applied fertilizer N and 4 to 9% of applied fertilizer P. Although such statistics produce interesting generalities across a wide range of land use, management, and climatic conditions, regional crop-specific analyses should be conducted to guide regulatory and programmatic decisions. With this update, MANAGE contains data from a vast majority of published peer-reviewed N and P export studies on homogeneous agricultural land uses in the USA under natural rainfall-runoff conditions and thus provides necessary data for modeling and decision-making related to agricultural runoff. The current version can be downloaded at http://www.ars.usda.gov/spa/manage-nutrient.     

Development of Sediment and Nutrient Export Coefficients for U.S. Ecoregions

Water quality impairment due to excessive nutrients and sediment is a major problem in the United States (U.S.). An important step in the mitigation of impairment in any given water body is determination of pollutant sources and amount. The sheer number of impaired waters and limited resources makes simplistic load estimation methods such as export coefficient (EC) methods attractive. Unfortunately ECs are typically based on small watershed monitoring data, which are very limited and/or often based on data collected from distant watersheds with drastically different conditions. In this research, we seek to improve the accuracy of these nutrient export estimation methods by developing a national database of localized EC for each ecoregion in the U.S. A stochastic sampling methodology loosely based on the Monte‐Carlo technique was used to construct a database of 45 million Soil and Water Assessment Tool (SWAT) simulations. These simulations consider a variety of climate, topography, soils, weather, land use, management, and conservation implementation conditions. SWAT model simulations were successfully validated with edge‐of‐field monitoring data. Simulated nutrient ECs compared favorably with previously published studies. These ECs may be used to rapidly estimate nutrient loading for any small catchment in the U.S. provided the location, area, and land‐use distribution are known.     

A model for phosphorus transformation and runoff loss for surface-applied manures

Agricultural P transport in runoff is an environmental concern. An important source of P runoff is surface-applied, unincorporated manures, but computer models used to assess P transport do not adequately simulate P release and transport from surface manures. We developed a model to address this limitation. The model operates on a daily basis and simulates manure application to the soil surface, letting 60% of manure P infiltrate into soil if manure slurry with less than 15% solids is applied. The model divides manure P into four pools, water-extractable inorganic and organic P, and stable inorganic and organic P. The model simulates manure dry matter decomposition, and manure stable P transformation to water-extractable P. Manure dry matter and P are assimilated into soil to simulate bioturbation. Water-extractable P is leached from manure when it rains, and a portion of leached P can be transferred to surface runoff. Eighty percent of manure P leached into soil by rain remains in the top 2 cm, while 20% leaches deeper. This 2-cm soil layer contributes P to runoff via desorption. We used data from field studies in Texas, Pennsylvania, Georgia, and Arkansas to build and validate the model. Validation results show the model accurately predicted cumulative P loads in runoff, reflecting successful simulation of the dynamics of manure dry matter, manure and soil P pools, and storm-event runoff P concentrations. Predicted runoff P concentrations were significantly related to (r2=0.57) but slightly less than measured concentrations. Our model thus represents an important modification for field or watershed scale models that assess P loss from manured soils.     

Evaluation of regression methodology with low-frequency water quality sampling to estimate constituent loads for ephemeral watersheds in Texas

Water quality regulation and litigation have elevated the awareness and need for quantifying water quality and source contributions in watersheds across the USA. In the present study, the regression method, which is typically applied to large (perennial) rivers, was evaluated in its ability to estimate constituent loads (NO(3)-N, total N, PO(4)-P, total P, sediment) on three small (ephemeral) watersheds with different land uses in Texas. Specifically, regression methodology was applied with daily flow data collected with bubbler stage recorders in hydraulic structures and with water quality data collected with four low-frequency sampling strategies: random, rise and fall, peak, and single stage. Estimated loads were compared with measured loads determined in 2001-2004 with an autosampler and high-frequency sampling strategies. Although annual rainfall and runoff volumes were relatively consistent within watersheds during the study period, measured annual nutrient and sediment concentrations and loads varied considerably for the cultivated and mixed watersheds but not for the pasture watershed. Likewise, estimated loads were much better for the pasture watershed than the cultivated and mixed landuse watersheds because of more consistent land management and vegetation type in the pasture watershed, which produced stronger correlations between constituent loads and mean daily flow rates. Load estimates for PO(4)-P were better than for other constituents possibly because PO(4)-P concentrations were less variable within storm events. Correlations between constituent concentrations and mean daily flow rate were poor and not significant for all watersheds, which is different than typically observed in large rivers. The regression method was quite variable in its ability to accurately estimate annual nutrient loads from the study watersheds; however, constituent load estimates were much more accurate for the combined 3-yr period. Thus, it is suggested that for small watersheds, regression-based annual load estimates should be used with caution, whereas long-term estimates can be much more accurate when multiple years of concentration data are available. The predictive ability of the regression method was similar for all of the low-frequency sampling strategies studied; therefore, single-stage or random strategies are recommended for low-frequency storm sampling on small watersheds because of their simplicity.     

Expansion of the MANAGE Database with Forest and Drainage Studies

The “Measured Annual Nutrient loads from AGricultural Environments” (MANAGE) database was published in 2006 to expand an early 1980s compilation of nutrient export (load) data from cultivated and pasture/range land at the field or farm scale. Then in 2008, MANAGE was updated with 15 additional studies, and nitrogen (N) and phosphorus (P) concentrations in runoff were added. Since then, MANAGE has undergone significant expansion adding N and P water quality along with relevant management and site characteristic data from: (1) 30 runoff studies from forested land uses, (2) 91 drainage water quality studies from drained land, and (3) 12 additional runoff studies from cultivated and pasture/range land uses. In this expansion, an application timing category was added to the existing fertilizer data categories (rate, placement, formulation) to facilitate analysis of 4R Nutrient Stewardship, which emphasizes right fertilizer source, rate, time, and place. In addition, crop yield and N and P uptake data were added, although this information was only available for 21 and 7% of studies, respectively. Inclusion of these additional data from cultivated, pasture/range, and forest land uses as well as artificially drained agricultural land should facilitate expanded spatial analyses and improved understanding of regional differences, management practice effectiveness, and impacts of land use conversions and management techniques. The current version is available at www.ars.usda.gov/spa/manage-nutrient .     

Nutrient load generated by storm event runoff from a golf course watershed

Turf, including home lawns, roadsides, golf courses, parks, etc., is often the most intensively managed land use in the urban landscape. Substantial inputs of fertilizers and water to maintain turf systems have led to a perception that turf systems are a major contributor to nonpoint source water pollution. The primary objective of this study was to quantify nutrient (NO(3)-N, NH(4)-N, and PO(4)-P) transport in storm-generated surface runoff from a golf course. Storm event samples were collected for 5 yr (1 Apr. 1998-31 Mar. 2003) from the Morris Williams Municipal Golf Course in Austin, TX. Inflow and outflow samples were collected from a stream that transected the golf course. One hundred fifteen runoff-producing precipitation events were measured. Median NO(3)-N and PO(4)-P concentrations at the outflow location were significantly (p < 0.05) greater than like concentrations measured at the inflow location; however, median outflow NH(4)-N concentration was significantly less than the median inflow concentration. Storm water runoff transported 1.2 kg NO(3)-N ha(-1) yr(-1), 0.23 kg NH(4)-N ha(-1) yr(-1), and 0.51 kg PO(4)-P ha(-1) yr(-1) from the course. These amounts represent approximately 3.3% of applied N and 6.2% of applied P over the contributing area for the same period. NO(3)-N transport in storm water runoff from this course does not pose a substantial environmental risk; however, the median PO(4)-P concentration exiting the course exceeded the USEPA recommendation of 0.1 mg L(-1) for streams not discharging into lakes. The PO(4)-P load measured in this study was comparable to soluble P rates measured from agricultural lands. The findings of this study emphasize the need to balance golf course fertility management with environmental risks, especially with respect to phosphorus.