This project seeks to quantify the performance of the Maryland Winter Cover Crop Program (MWCCCP) from multiple criteria. First, we will estimate the amount of cover crop biomass on growers’ fields at two critical timings during winter: at the onset of dormancy and prior to termination. The dormancy-onset estimate serves as a surrogate for measuring soil N prevented from leaching into the Chesapeake Bay, which is the stated goal of the MWCCP. The at-termination estimate provides information about other ecosystem services that could potentially be provided by cover crops. These estimates will be generated from satellite imagery through our collaboration with the USGS and USDA-ARS Remote Sensing Laboratory.
Second, we will determine the association between agronomic factors of cover crop management and these biomass estimates. This association will be conducted by joining a database of enrollment in the MWCCP (provided by the Maryland Department of Agriculture) and our satellite imagery biomass estimates using geospatial analysis.
Third, we will test whether the incentive structure provided by the MWCCP provides a return-on-the-dollar commensurate with the performance of the cover crop. Additionally, we will test whether the incentives have increased adoption of management practices associated with improved performance over the lifetime of the program.
1) Connect existing datasets to develop a model that predicts cover crop biomass (as estimated from satellite imagery) based on agronomic management (as reported to the Maryland Department of Agriculture by farmers) as a function of thermal time (as calculated from historical weather data).
2) Given the model developed in Objective 1), determine how much the variability in cover crop biomass is due to each agronomic management category relative to landscape variables: soil texture, drainage class, and local topography.
3) Develop a return-on-investment index that scales cover crop biomass to the incentives paid to growers and evaluate each agronomic management category based on performance on this index.
Loss of nitrogen to ground water represents a significant water quality concern. Cover crops are one of the primary mechanisms for scavenging residual fall soil inorganic nitrogen and are being promoted in the Chesapeake Bay watershed to conserve water resources. Beginning in 2005, the Maryland Agricultural Water Quality Cost-Share Program has incentivized farmers to adopt cover crops through direct payments. Payments are structured in a tiered system whereby management affects payment (i.e., species choice, planting date and method, fertilizer source, and field history). State stakeholders are interested in quantifying the success of the program. A study area of farms was identified on the Eastern Shore of Maryland in Talbot, Queen Anne’s, and Caroline Counties within the Choptank River Watershed; biomass was collected from 2005-2011 at two points each season: winter dormancy, and prior to cover crop termination. The farms in the study planted cereal rye (Secale cereale L., 23%), wheat (Triticum aestivum L., 46%), and barley (Hordeum vulgare L., 25%), with less than 6% of the study sites planting other cover crops, following corn (Zea mays L., 72%), soybeans (Glycine max (L.) Merr., 25%), or vegetables (3%). These data were used to calibrate vegetative indices for ongoing analysis of satellite imagery (2005-present), which is used to estimate performance of cover crops across those counties. This proposal seeks to assess the agronomic management of cover crops by linking the program enrollment data from the Maryland Department of Agriculture with the estimated biomass as measured using remote sensing techniques.
The data were collected from two primary sources:
- satellite imagery at field-scale resolution, used to calculate cover crop biomass estimates from NDVI for Talbot County and some adjoining parts of the Lower Choptank River watershed
- enrollment database of the Maryland Department of Agriculture, listing management practices and geospatial references for each farmer in the program
Farms in Talbot County enrolled in the Program from 2005-2011 had their geospatial extents collected along with their management plans (n=9,384); these fields represented 85,273 hectare-years. Satellite imagery over this region was collected twice per cover cropping season: in December-January, representing fall growth and crop dormancy, and in March, representing spring growth prior to termination. A calibration dataset had been used to develop equations to predict cover crop biomass from the Normalized Difference Vegetation Index in the same region (n=224). The calibration dataset contained 107 unique management factor levels (combinations of 6 categorical variables), while the remote imagery represented farms with 390 unique management factor levels. These factors were highly unbalanced, with many levels having less than 10 observations, while some levels had over 600 observations. Log-transformed linear models were used to estimate each coefficient and its standard error for the calibration dataset, which provided informative priors for the remote dataset. However, for factor levels that did not appear in the calibration dataset, a larger sample population (i.e. pooling multiple levels) had to be used for the prior, increasing the variance of the estimate. This procedure was conducted hierarchically, first pooling commodity program, then previous cash crop, then establishment method, then planting date category, and finally species (the final variable, sample timing, was always used). Using these priors, a Bayesian generalized linear mixed model was fit, using year as a random effect, and the management factors as main effects.
The dominant species used in the program in Talbot County was winter wheat, representing ~80% of fields enrolled. A majority of fields were enrolled following corn (~58%) and soybeans (~38%). Planting dates were approximately evenly divided among the three payment categories of the program (Early, Standard, and Late). A majority of fields were planted using no-till drilling (~51%). The dominant method of termination was by herbicide application in the spring (~62%).
While the number of calendar days between sampling times were consistent at approximately 100 days each year, the number of accumulated growing degree days (base 4.4°C) between the winter and spring observations ranged from 104 to 320.
We found that the overall data could be well-explained by a model that used 6 management variables as categorical factors, explaining over half of the variance in the raw data (median pseudo-R2: 0.538, 95%CI: 0.531-0.546). Coefficients for cover crop biomass ranged from 64.1 to 2,453 kg ha-1. However, there was a wide range of estimates for uncertainty in these predictions, where the span of the 95%CI ranges from 28.4 to 3525 kg ha-1.
When comparing coefficient estimates for the same combination of management factors between winter and spring sampling, we found that approximately 2/3 of them represented a significant contrast (i.e. growth occurred between winter and spring), while ~1/3 were not significant.
Education & Outreach Activities and Participation Summary
The outreach conducted in 2017 on this project was a pair of talks at the Northeast Cover Crop Council’s Annual Meeting, held in Ithaca NY. Approximately 30 people attended the talks, a combination of growers, extension agents, and researchers. These talks presented the data available for the project, our plans for analysis, and initial findings.
The outreach activities in 2018 were two talks at the American Society of Agronomy, and a PhD entrance seminar at the Department of Environmental Science and Technology at the University of Maryland. These talks discussed our methodology for the project, with a particular focus on the statistical challenges associated with this data. The methods presented were used to illustrate both the value of this dataset and this project, as well as provide an outline for Bayesian mixed models to researchers in other areas of agriculture and sustainability. Approximately 45 people attended the three talks.