Assessing the Effects of Conservation Practices and Fertilizer Application Methods on Nitrogen and Phosphorus Loss from Farm Fields – A Meta Analysis
IPNI-2014-USA-4RM07
01 Mar 2014
Project Description
Summary
We propose a meta-analysis of the effects of water and soil conservation practices, as well as the effects of fertilizer application methods, in reducing nutrient loss from farm fields. An preliminary analysis of an existing database showed that the difficulty of a successful meta-analysis lies in how to properly handle confounding factors because studies of agriculture practices are carried out under different conditions (e.g., crops, tillage methods, and climate). As we may not have the full knowledge of these confounding factors, conventional statistical methods are often ineffective and potentially misleading. In this meta-analysis project, we propose to use two statistical causal analysis methods (propensity score and multilevel modeling) for quantifying the effects of water and soil conservation practices in reducing phosphorous loss from agricultural fields. The project will augment the existing database by (1) revising studies included in the existing database to update information about fertilizer application methods, as well as additional variables, and (2) updating the database with recent studies. The project will document the use of the propensity score method and the multilevel modeling approach in the context of meta-analysis. Using the resulting database, the project will publish a series of papers to document the need of the two statistical methods in analyzing cross-sectional data, and meta-analysis result in the effects of water and soil conservation practices and the effects of fertilizer application methods in reducing nutrient (P and N) losses from farm field and in crop yield. Results from the proposed study are applicable for improved assessment of agricultural practices and their effects on the environment and can be used for providing realistic parameter values for watershed-scale modeling.
1 Project Description
1.1 Introduction
Implementing soil and water conservation practices is an important part of sustainable agriculture. In response to increased funding of conservation programs in the Farm Security and Rural Investment Act of 2002 (FarmBill 2002), the U.S. Department of Agriculture initiated the Conservation Effects Assessment Program (CEAP) to provide scientific understanding of the impacts and benefits of conservation practices [Duriancik et al., 2008]. CEAP produced a comprehensive bibliography on available literature, a suite of mechanistic models, and a series of model simulations on the effects of conservation practices at a watershed level for many regions of the U.S. In addition, CEAP facilitated collection of field-scale measurements of nutrient loadings to provide basis for empirical assessment of various agricultural conservation practices.
To support CEAP and other national modeling and assessment efforts, the Measured Annual Nutrient loads from AGricultural Environments (MANAGE) database was created to provide readily accessible, easily queried watershed characteristic and nutrient load and concentration data to guide policy and management decisions based on comparative nutrient load information from various land management alternatives [Harmel et al., 2006, 2008]. MANAGE compiles and summarizes measured annual nitrogen (N) and phosphorus (P) load and concentration data representing field-scale transport from agricultural and forest land uses. Because the MANAGE database includes data from many studies from different locations with different land use and management options, conventional statistical methods such as analysis of variance and regression may not be appropriate for estimating the effects of specific practices in reducing nutrient loss. Specifically, alternative methods are preferred because nutrient loads were measured from fields with differing management and site conditions (e.g., crop, fertilizer application method, etc.), which confounds the conservation practice effects. For example, fields with one or more conservation practices in MANAGE database received on average more P fertilizer than fields without any conservation practices (Figure 1). As a result, a direct comparison of P loading from these two groups of fields can be misleading; therefore, statistical methods for estimating the causal effect using observational data are preferred.
In this proposal, we propose two such methods (propensity score and multilevel modeling) for quantifying the effects of conservation practices (CPs) and fertilizer application methods on reducing field-level P and N loads. In our preliminary analysis of the MANAGE database, we examined only the effect on TP load. With a limited sample size, the preliminary study lumped all conservation practices together and did not specifically examine the effect of fertilizer application methods.
1.2 Problem Statement
Quantifying the effects of agriculture management practices, specifically, conservation practices and fertilizer application methods, is challenging because planned (randomized) experiments are rarely feasible. Even when such studies are available, a management practice is typically assessed under specific conditions (e.g., crop, BMPs) that often prevents the extrapolation of the results to fields with a different condition. While studies documenting the association between nutrient loss and management practices are abundant, comparison of these studies for quantifying the effects of a specific practice is meaningful only if confounding factors are adequately accounted for. Furthermore, the effect of a specific conservation practices is likely associated with factors related to crop, crop management, and so on. A general statement of the effect of a conservation practice is often less informative.
1.3 Project Description
We propose a meta analysis of the effects of agriculture management practices on nutrient loss from fields. Our analysis will focus on two practices: conservation practices and fertilizer application methods. Based on our preliminary analysis of a small database, we propose to use two statistical methods for this meta analysis: the multilevel modeling approach and the propensity score method. These methods
are proposed to analyze the data from two different angles to ensure the accuracy of the outcome.
The project will use the USDA-ARS database MANAGE (2007 version) as a starting point. We will first conduct a thorough analysis of the existing MANAGE database to explore the potential features of the database. The results (e.g., important confounding factors) will be then used to guide the expansion and updating of the MANAGE database to include additional studies and additional variables. In a preliminary analysis of the 2007 version of MANAGE data base, we found that variables such as fertilizer application methods and type of drainage systems may be important in explaining the variation in nutrient loss among studies. The expanded database will be used for the meta-analysis using the two statistical methods described in this proposal. We propose to use two statistical methods because both methods have their appeals and their shortcomings. The propensity score method will result in an accurate estimate of the average effect, but cannot provide information of the variation of the effect under different conditions. The multilevel modeling approach will provide both the overall average and the variation, but the results are contingent on the method used for stratifying the data. With both methods, we can better ensure the accuracy of the outcome. The project will be used to develop a master’s thesis, in preparation for the student’s PhD research.
1.4 Goals and Objectives
The project is aimed at quantifying the effects of two agriculture practices on reducing nutrient (P and N) loss from fields. We are interested in the effects of water and soil conservation practices (grassed waterways, contour farming, terraces, riparian forest buffer, and filter strip), as well as the effects of fertilizer application methods (e.g., injected, incorporated, and surface applied). In our preliminary analysis of the 2007 version of MANAGE, we find that conservation practices are potentially more effective in reducing TP loading than most current assumptions [Qian and Harmel, in review]. This finding must be verified by a larger dataset and explained in a level of details that is necessary for the result to be credible. We were unable to estimate the effects of fertilizer application methods in our preliminary study using the 2007 MANAGE because of its limited sample size. The proposed project will incorporate changes made to the MANAGE database since 2007 and include additional studies and variables.
Because that randomized experiments on agriculture practices are rare, a meta analysis of agriculture conservation practices and fertilizer application methods cannot use conventional meta-analysis methods, which assumes studies use similar experimental conditions. We propose to use two appropriate statistical methods for the meta analysis. We will document these methods and illustrate their applications using the updated database.
1.5 Outcome
The project will compile a large cross-sectional database to document existing studies on agriculture management practices. The project will also document the use of two statistical methods for meta analysis, as well as the effects of various conservation practices and fertilizer application methods in reducing nitrogen and phosphorus loss from farm fields.