Nitrogen Fertilizer Management Based on Site-Specific Maximized Profit and Minimized Pollution

Progress report for GW19-190

Project Type: Graduate Student
Funds awarded in 2019: $24,992.00
Projected End Date: 07/31/2022
Grant Recipient: Montana State University
Region: Western
State: Montana
Graduate Student:
Major Professor:
Dr. Stephanie Ewing
Montana State University
Major Professor:
Expand All

Project Information

Summary:

The loss of nitrogen (N) from crop fields is a national issue that plagues not only terrestrial and aquatic ecosystems, but human and animal health as well. The over application of fertilizer on farms across the country results in N that is left unused by crops, acidifying soils and flowing into waterways in addition to loss of revenue for the producer. Using variable rate N application (VRA) to site-specifically apply fertilizer, rather than application of a uniform rate, has potential to reduce the amount of N applied to fields and increase grower profits. This study seeks to determine the amount of nitrogen fertilizer to place at each point in the field to maximize profit and at the same time minimize nitrogen lost to leaching through the soil profile. I will use full-field variable N rate experimentation conducted with precision agriculture technologies (VRA, yield monitors and protein sensors) numerous GIS, climate, and remote-sensing based covariates and spatiotemporally dense soil and plant sampling and N analysisto construct equations that will allow prediction of optimum N rates in dryland small grain agroecosystems. The predictive equations will be used in a software developed for this project to create site-specifically optimized nitrogen fertilizer application rates based on maximized net-return and minimized nitrate pollution under different climatic and weather scenarios and to generate nitrate-loss potential maps at subfield scales. Site-specific N fertilizer management will be compared with other application approaches based on net returns to producers and pollution potential tradeoffs. Annual workshops and reports for producers and the public will happen every year in order to disseminate research results. If site-specific management indicates a decrease in nitrogen pollution and is economically efficient, it could spur a major shift of growers statewide on the path to more efficient and sustainable agriculture.

Project Objectives:

My primary research goal is to estimate the N fertilizer loss (N not taken up by plants) from each point in a wheat field where yield is reported by the combine yield monitor and grain protein analyzer. This will require measurement of N loss under the range of conditions encountered and extrapolating the losses using predictive models that I will develop. To achieve this goal, my specific objectives are:

  • Quantify the relationship of crop (dryland winter wheat) yield and grain protein with N fertilizer rates from variable rate application (VRA) on-farm experiments (OFE) utilizing grain yield combine-harvester monitor data and protein analysis along with GIS, climate, and remote sensing-based covariates.
  • Determine N loss as the amount of N unaccounted for after soil and plant inventories by calculating a N mass balance using intensive soil and plant tissue sampling.
  • Extrapolate N loss to each yield/protein monitored point using geostatistical techniques.
  • Quantify and package site-specific optimized N fertilizer application rates and loss potential maps based on maximized net return and minimized pollution potential by leaching, surface runoff and atmospheric loss into a field-specific software application available for use by growers.

Cooperators

Click linked name(s) to expand

Research

Materials and methods:

Initial Methods:

Two fields in Montana representing different climatic and soil regions were selected from a larger population of fields for which data is being collected for the OFPE project led by my advisor Dr. Bruce Maxwell (Fig. 1). The benefit of using producers who have prior experience and participation in site-specific management with OFPE is their experience in research management on their fields as well as our access to their historical data. Fields were selected based on the crop grown in 2020 (winter wheat) and on the quality and promptness of data collection by the producers.

Figure 1. Map of OFPE farm boundaries of collaborators with Montana State University. Colors represent different farmers, while shapes represent areas in which their respective fields fall within. The letter identifiers match those in Table 1.

Information on both fields can be found in Table 1. Both fields were under OFPE experimental management in 2016 and 2018. Fields selected were required to have experimental N data available from at least two years, as well as observed yield and protein measurements from across the entire field. Nitrogen fertilizer rates were experientially applied in the field in blocks varying from 400 – 600 feet long and as wide as the farmer’s fertilize sprayer or spreader width, varying from 70 – 120 feet wide. Rates were randomly placed throughout the field, stratified on previous yield, protein, and N applied.

Table 1. Collaborator farm ID (Fig. 1), field names, field size (acres), crop harvest history, years of N experimentation. The letter identifier for the Farm corresponds to Figure 1.

Farm

Field

Field size

Crop History1:

2014 / 2015 / 2016 / 2017 / 2018 / 2019

Years N rate treatment

B

sec35middle

158

WW / CF / WW / CF / WW / CF

2016, 2018

I

henrys

118

WW / SW / WW / CF / WW / CF

2016, 2018

1 WW = winter wheat, CF = chemical fallow, P = peas, SW = spring wheat, SF = safflower, NA = Not Available

The response variables of interest are crop productivity (yield in bushels per acre) and quality (grain protein percent), both of which are gathered from monitors mounted in the cab of the farmer’s combine harvesters. Yield is collected about every three seconds while protein is measured about every ten seconds. Yield is related to price received for wheat based on volume (bu), however protein in winter wheat is judged based on a nonlinear scale. On typical years, protein content above approximately 12% is rewarded with increases on the base price received, however if protein is below 12%, the price received is decreased exponentially as protein decreases. Thus, the combination of quantity and quality of the winter wheat crop influences farmer net returns. Farm equipment used in this study varied between farmers, with one using John Deere and the other using Case. All yield monitors are calibrated every spring by the farmers, according to their respective manufacturing instructions. Grain protein content was measured with Next Instrument’s CropScan 3300H near infrared monitor which was installed, maintained, and calibrated by Triangle Ag Crop Consultants from Fort Benton, MT. Both yield and protein data are subjected to proprietary cleaning practices according to the software of the machine used to make the measurement. In addition to the yield, protein, and as-applied N data that are collected from the machines on the field, remotely sensed covariate data from open sources was gathered (Table 2). These data were gathered or derived from Google Earth Engine (Gorelick et al. 2017) and aggregated together at the locations of the observations in the yield and protein datasets.

Table 2. Table of covariates gathered from Google Earth Engine to enrich the yield and protein datasets gathered from on-farms. In some cases, multiple sources are used, however only one data source is used when aggregating to yield and protein.

Data Type

Data Sources

Resolution

Years Collected

Description

Reference

Normalized Difference Vegetation Index (NDVI)

Sentinel 2, Landsat 5/7/8

10m, 30m

S2: 2016-present

L5: 1999-2011

L7: 2012-2013

L8: 2014 – present

Sentinel 2 is from the European Space Agency as part of the Copernicus program, Landsat is an ongoing USGS and NASA collaboration.

S2: Uses bands B8 for NIR and B4 for red

L5/L7: B4 and B3

L8: B5 and B4

Sentinel 2

Landsat 5

Landsat 7

Landsat 8

Normalized Difference Red Edge (NDRE)

Sentinel 2, Landsat 5/7/8

20m

S2: 2016-present

Bands B5 and B6

Sentinel 2

Red Edge Chlorophyll Index (CIRE)

Sentinel 2, Landsat 5/7/8

20m

S2: 2016-present

Bands B7 and B5

Sentinel 2

Elevation

USGS NED, NRCan,

~10m (1/3 arc second), ~23m (3/4 arc second)

1999-present

USGS National Elevation Dataset, Natural Resources Canada’s Canadian Digital Elevation Model (CDEM), Shuttle Radar Topography Mission from the NASA Jet Propulsion Laboratory

USGS NED

NRCan

SRTM

Aspect

USGS NED, SRTM Global DEM

~10m (1/3 arc second), 30m

1999-present

Direction the surface faces, function of neighboring elevations, in radian. For non-USA fields, SRTM elevation dataset used to calculate else USGS NED

API

USGS NED

SRTM

 

Slope

USGS NED, SRTM Global DEM

~10m (1/3 arc second), 30m

1999-present

Rate of change of height from neighboring cells, in degrees. For non-USA fields, SRTM elevation dataset used to calculate else USGS NED

API

USGS NED

SRTM

 

Topographic Position Index (TPI)

USGS NED, SRTM Global DEM

~10m (1/3 arc second), 30m

1999-present

See equation above, measure of divots and low spots. For non-USA fields, SRTM elevation dataset used to calculate else USGS NED

API

USGS NED

SRTM

 

Precipitation

DaymetV3, GRIDMET

1km, 4km

1999-present

Estimates from the NASA Oak Ridge National Laboratory (ORNL), University of Idaho GRIDMET combo of PRISM and ground based measurements

Daymet V3

GRIDMET

GDD

DaymetV3, GRIDMET

1km, 4km

1999-present

Estimates from the NASA Oak Ridge National Laboratory (ORNL), University of Idaho GRIDMET combo of PRISM and ground based measurements

Daymet V3

GRIDMET

The first objective will be achieved by characterizing the response and variation in yield and protein data to N fertilizer application using multiple statistical methods such as linear and non-linear regression models and Bayesian non-linear regression as well as more advanced techniques such as random forests and neural networks. These models are fit using the large amounts of data collected from yield harvesters and protein analyzers mounted on the combine, and from aggregated 10m scale covariate data aggregated from the earth observation satellites. Additionally, economic data on the base price of winter wheat and premiums related to protein levels from the elevators that the producers take their harvest have been gathered, as well as other associated costs of site-specific management such as equipment and labor costs. The response functions of yield and protein will be used in a model that incorporates the economic data in order to determine site-specific N rates that maximize the return on investment (ROI) for each segment (~900m2) of the field.

For every field, the available data was representatively split across both available years into training and validation datasets. Sixty percent of data were used to fit models, while forty percent was reserved to validate the ability of the models to predict responses in a dataset that is new to the model. For each field, all the models used the same training and validation datasets. The first model is a universal nonlinear logistic model based on the expected response of yield and protein to N fertilizer (Equation 1);

Where R is yield (bu/ac) or grain protein content (%), alpha represents parameters that influence the response when no N fertilizer is present, beta contains parameters that influence the maximum asymptotic response, gamma is the rate at which the value of responses increase with added N fertilizer (N), and delta is an inflection point (Table 3). This equation was chosen because of the expectation that there is a N fertilizer rate at which N is not the most limiting resource and additional fertilizer does not contribute to increased crop productivity or quality.

Table 3. Covariate names and component of the nonlinear logistic model in which they apply. No covariates were assigned to the and model components.

Covariate Name

Nonlinear Logistic Model Component

Aspect (cosine)

Alpha

Aspect (sine)

Alpha

Slope

Alpha

Elevation

Alpha

TPI

Alpha

Previous N

Alpha

Precipitation (Most recent year)*

Beta

Precipitation (Year prior)

Beta

Growing Degree Days (Most recent year)*

Beta

Growing Degree Days (Year prior)

Beta

NDVI (Most recent year) *

Beta

NDVI (Year prior)

Beta

NDVI (Two years prior)

Beta

* Data constraints were imposed on variables collected in the most recent year. These data were gathered up to the March 30th decision point or September 1st hypothetical harvest.

The second model used to predict the response of dryland winter wheat yield and protein to variable N fertilizer rates was a generalized additive model (GAM) using thin plate shrinkage splines that uses the same parameters as the logistic model above. A GAM was chosen due to its flexibility and ability to characterize the response of the observed data. Thin plate shrinkage splines are used for all variables in order to allow the estimated degrees of freedom (EDF’s) of parameters to shrink to zero, combining the process of model fitting and selection. The parameter exceptions to thin plate shrinkage splines was the use of a gaussian process for the interaction of latitude and longitude in order to account for spatial autocorrelation between observed points within the field.

The second objective will be approached using intensive soil and plant tissue sampling along with managerial control over the N fertilizer rates applied. Previous work and ongoing investigations led by Dr. Stephanie Ewing in the Judith River Watershed (JRW) of Montana has instigated reactive transport modeling undertaken with nitrate data collected with lysimeters in the JRW. This model, developed in R, assesses N loss pathways and their relationship with soil characteristics, weather, and management decisions. Insights from this model indicate that in a typical Montana dryland agroecosystem, soil water holding capacity has the greatest influence on soil regulation of N loss. This work informed the investigation of spatial relationships between data collected on-farms and from remote sensing will be investigated in OFPE fields to indicate portions of the field that exhibit characteristics of N loss. Under the assumption that soil water storage drives nitrate dynamics and N loss, spatial relationships between historic productivity and precipitation will be evaluated in OFPE fields to identify subfield units that consistently experience high, low, and variable productivity over time, indicative of potential N loss. For every 10m pixel across each field, the mean and coefficient of variation for NDVI between 1999 and 2019 was calculated and grouped based off of the Jenk’s Natual Breaks method to classify the field spatially into areas of low, medium, and high estimated water holding capacity. Stratified random selection of points based on these classifications were created to measure potential N loss.

Potential N loss is defined as any N that is not assimilated by the crop during the growing season. While N may be held in the soil for the following year, this N is “lost” to the crop and the producer during the growing season. Prior to fertilization, soil cores will be taken at five depths (0-15, 15-30, 30-60, 60-90, and 90-180 cm or as deep as possible). Soil cores will be analyzed for total and inorganic N (NO3, NH4+) to determine N stored in the soil profile prior to fertilization using Costech and Lachat products (LECO Corp., St. Joseph, MI; Hach Corp., Loveland, CO). Applications of N will vary at each sampling location depending on location within the experimental layout, and tissue samples of the crop will be taken at each sampling location when the wheat begins to harden and dry out for harvest and will be analyzed for total N content. Soil cores will be resampled at the same locations post-harvest to determine a N inventory as done prior to fertilization. Potential N loss will thus be calculated as;

This represents a liberal estimate of potential N loss because of the assumption that any N not utilized by the crop will be lost to the environment over some timescale and unavailable for crop use in the following crop year, however takes into account all potential loss through volatilization, surface runoff, lateral flow, denitrification, and leaching. Calculating potential N loss in this way is also of use for producers, as “optimal” fertilizer management aims to replace nutrients taken up by the crop in the previous cropping year. However, due to the experimental design of the fields, the contribution of mineralized N can be determined when no fertilizer is applied in subunits of the field and the crop’s N uptake is solely derived from the soil.

At each sampling site, data on soil characteristics, such as texture and depth to coarse substrate, will be recorded to investigate the relationship between soil parameters dictating dominant N loss pathways identified from modeling in the JRW to stored soil N in the OFPE fields. It is impractical for a farmer to take intensive soil samples in a field annually, due to time and budget constraints, so a function between stored soil N and previous N applications, soil characteristics, moisture, and freely available environmental variables such as slope, elevation, aspect, and topographic position index will be developed. Additionally, farmers do not directly measure N content of their crop, so using measured N in straw and grain at sampling sites, a function between N content, yield, protein, and applied N will be developed for use in future years after experimentation has ceased.

The third objective is to build models to predict within-field spatial variability in N loss based on environmental variables (same as those for the crop response models). Using the data collected at sampling points, associations between environmental variables and N loss will be investigated to characterize a relationship that can be used to make predictions of N loss at other yield/protein points. In the case that Bayesian and random forest techniques are not sufficient to develop a function between N loss and environmental covariates, ordinary kriging will be used to extrapolate N loss values to yield/protein points across each field.

The fourth objective will be achieved through the creation of a machine learning based optimization model developed to generate management plans that optimize between ROI and N pollution potential and to create maps of site-specific potential N loss. The software will be presented in a web-based R-Shiny application in which the automated program will receive, and store farmer collected and remotely sensed data to generate the site-specific fertilizer management plans under various potential climatic and weather scenarios plus subfield scale maps of potential N loss. Data will be stored in a spatial database designed in a specific OFPE format to facilitate automated data organization from various farmers, farming implements, and open source data. The storage and utilization of historical data year after year increases the predictive ability of the models trained annually. Management plans will be generated that optimize N fertilizer applied to maximize net returns and minimize N loss from each field by adding economic data on the prices and costs associated with each field as well as climate and weather data using the models from objectives 1-3. The app will incorporate automation in the collection of remotely sensed data, model fitting and subsequent parameterization, validation, and optimization of management plans for maximized net returns with minimized nitrate loss and will present various management plans in an easy-to-use format for easy comparisons by the producer. The R-Shiny web application will be driven by the OFPE R-package that contains functions and algorithms written to facilitate the OFPE data workflow.

Research results and discussion:

Initial Results:

The experimental nitrogen fertilizer rates applied to each field are shown below. The size of the treatments varied between farmers based on their requests for minimum treatment lengths. The widths are determined by each farmer’s sprayer boom width.

Figure 1. Nitrogen fertilizer rates applied to Broyles’ sec35mid field. The rate surrounding the experiment is a base rate that the producer would have applied to the whole field.

Figure 2. Nitrogen fertilizer rates applied to Wood’s henrys field. The rate surrounding the experiment is a base rate that the producer would have applied to the whole field.

The following are preliminary figures related to objective 1. At this point of the project, two functional forms are being tested and compared for their ability to predict winter wheat yield and protein, the nonlinear logistic model and a generalized additive model. Both models use the same covariate data, including as-applied fertilizer data collected from the farmer’s equipment while responding to a field trial protocol map that we provide to control the VRA, and remotely sensed data from satellites and available from open sources. See the Materials and Methods section for more detailed information. In the figures below, observed yield or protein data is plotted as the black points, while the colored points indicate the predicted yield or protein data. The plotted points are from the 40% of data withheld as a validation dataset and used to test the accuracy of the model fit with the training data.

Figure 3. Scatterplots of observed and predicted yield (left) and protein (right) vs. as-applied nitrogen fertilizer for Broyles’ sec35mid (top) and Wood Henrys (bottom). Black points represent observed observations from the indicated year, and colored points show the predicted points from the nonlinear regression model. AIC is reported in the subtitle.

The effect of temporal variation caused by annual differences in precipitation and growing degree days is demonstrated by the clear definition between lines in the yield plot for henrys above.

Figure 4. Scatterplots of observed and predicted yield (left) and protein (right) vs. as-applied nitrogen fertilizer for Broyles’ sec35mid (top) and Wood Henrys (bottom). Black points represent observed observations from the indicated year, and colored points show the predicted points from the generalized additive model (GAM). AIC is reported in the subtitle.

Prior to nitrogen fertilization in each of the fields, normalized difference vegetation index data from 1999 to 2019 was used to classify each field into zones of low, medium, and high-water holding capacity zones. The following figures are maps of the resulting classification upon which stratified random soil and tissue samples are taken each crop year.

Figure 5. Map of Broyles’ sec35mid field, with black points indicating soil core and plant tissue sampling points and colored areas representing the distribution of water holding capacity levels classified with the Jenk’s Natural Breaks method.

Figure 6. Map of Wood’s henrys field, with black points indicating soil core and plant tissue sampling points and colored areas representing the distribution of water holding capacity levels classified with the Jenk’s Natural Breaks method.

Soil samples were collected in the second week of March in both fields and weighed wet and dry. The volume of water held in the soil is determined by the difference in mass between the wet and dry samples. The same method was used to calculate the volume of water in the plant samples taken.  

Figure 7. Frequency distribution of the volume of water in soil (top) and plant (bottom) across Broyles sec35mid (left) and Wood henrys (right).

The distribution of water volumes observed in both fields indicates that there were some measurement errors associated with the soils data, as negative soil water volumes are reported. There is evidence of right tailed distributions in all figures except for plant water volume in sec35mid.

Figure 8. Boxplot of the volume of water in soil (top) and plant (bottom) by water holding capacity zone in Broyles sec35mid (left) and Wood henrys (right).

The conclusion based off of initial visual assessments is little to no difference between the volume of water held in plants or soil from either field across water holding capacities. The only exception would be that in areas of ‘henrys’ with high water holding capacities, the volume of water in the plant appears lower compared to areas with a low water holding capacity. This could be due to areas of the field with higher clay contents that have a higher plant wilting point.

Figure 9. Boxplot of the volume of water in soil (top) and plant (bottom) by SSURGO soil types in Broyles sec35mid (left) and Wood henrys (right).

There appears to be little variation in soil water volumes between soil types in both fields, however there is much more observed variation in plant water volume across soil types in both fields. While the range varies the most in ‘sec35mid’, in ‘henrys’ the variations are relatively constant with differing means. Note that variation could be inflated due to uneven sample sizes between soil types.

Participation Summary
2 Farmers participating in research

Educational & Outreach Activities

1 Consultations
2 Workshop field days

Participation Summary

3 Farmers
1 Ag professionals participated
Education/outreach description:

A. Consultations

In late February of 2020, Paul Hegedus met with Neil Fehringer, of Fehringer Agricultural Consulting to inquire about soil sampling best practices in spring weather in Montana. The ground can be frozen into mid-March while quickly thawing and becoming muddy on a rapid diel scale, necessitating soil sampling to occur efficiently within a small-time window. Mr. Fehringer was consulted with because he is the long-term crop consultant for one of our producers, Mr. Broyles and his experience soil sampling in February and March at the field this project is conducting research. Mr. Fehringer indicated that his success soil sampling in frozen soils is due to a center mounted truck probe, leveraging the weight of the truck to push through the frozen ground, and that all operations are conducted inside of the cab. The truck mounted hydraulic soil sampling probe used for this project was mounted in the bed of a truck and thus required less than 2 inches of frozen soil. Mr. Fehringer’s services to Mr. Broyles and his operation include generating prescriptive maps for fertilizer rates. Mr. Fehringer explained that he manually adjusted fertilizer rates based on zones that were classified based off of elevation and previous yield. Paul Hegedus shared the On-Farm Precision Experiments methodology of generating fertilizer prescriptions optimized on maximum net-returns using experimental variation of fertilizer rates. Mr. Fehringer then explained his fertilizer rate trials conducted in small plots in Mr. Broyles fields, concurrent and separate from this project’s fields.

B. Field Days

In March of 2020, prior to fertilizer applications, soil samples were taken within each field. While Mr. Wood was unavailable, a field day at Mr. Broyles’ operation was used to demonstrate the soil sampling process using a truck mounted hydraulic probe and to explain the reasoning on gathering soil data at multiple points during the season and at such a high density. In return, the Broyles producers showed Paul Hegedus areas of the field that were accidentally burned from combine ignited fire during the prior year harvest, which were subsequently mapped. This field day was imperative to gathering information on the areas of the field that were influenced by fire, and were easily distinguishable by the fine, sooty, soil texture and lack of significant winter wheat emergence. During this visit, Mr. Broyles shared historic soil sampling and nutrient inventories, listened to the methodology for our nitrogen mass balance approach, and shared and discussed his and our expectations for the results.

In June of 2020, a second field day was conducted at the Broyles’ farm. Due to the Covid-19 pandemic, masks were worn at all times, all social distancing protocols were followed, and all Montana University System, state, and federal guidelines were adhered to. This trip was to conduct plant tissue sampling at the same locations in the field that soil samples were collected, and at the request of Mr. Broyles. There were marked differences between areas within the field that received high and low experimental fertilizer rates that Mr. Broyles requested we observed before the crop proceeded further to harvest.

Figure 1. Picture of a portion of ‘sec35mid’ taken June 25th, 2020 with 80 lbs N/acre fertilizer applied in the foreground, 0 lbs N/acre in the middle treatment, and 50 lbs N/acre beyond, below the road in the middle of the picture.

During this field day, an open discussion with the Broyles family was had about the mechanisms and timing of nitrogen uptake into the grain of the wheat plant. In addition, we were taken on a detailed tour of the experiment and shown areas of the field typically productive or unproductive, the years in which exceptions occurred, and participated in a discussion on the causes behind observations that deviated from Mr. Broyles’ historical perspective. There was no talk about site-specific fertilizing rates for net return. This producer wants the highest crop yield and protein and net return is secondary

C. In Progress/Upcoming

A third, less formal field day for soil sampling following the producer’s harvest is planned for mid-August. This trip will likely not be participatory with the producers, as they will be busy harvesting other fields. The first week of December is when the annual Precision Agriculture Research Association (PARA) meeting is held in Great Falls, that includes both Mr. Broyles and Mr. Wood. At the PARA meeting, both researchers and producers have equal time to share and discuss field specific observations and results.

 

There are multiple papers in progress relating to this project or data collected during funding for this project. Most directly related to the project is a paper that describes the process for estimating soil water holding capacity based off of historical normalized difference vegetation index data, the calculation of a nitrogen mass balance, and development of a model of nitrogen loss using covariates collected on the field or from remotely sensed resources. This paper requires data collected next year in the spring and summer of 2021 and is directly related to objectives 2 and 3. Also requiring the second year of data collected on this project is a paper demonstrating the proof of concept for developing nitrogen management plans that optimize environmental and economic benefits as a step on the path from industrial to sustainable agriculture, site-specific optimization of variable rate nitrogen fertilizer rates based on maximized net-return and minimized nitrogen loss using on-farm and open source data. This paper is the culmination of all objectives. The first paper scheduled for publication will use data collected in the harvest of 2020, along with the years of experimental data from prior years to fit models of yield and protein to variable nitrogen fertilizer rates. However, this paper focuses on the On-Farm Precision Experiments methodology, and uses the on-farm data collected for this project to demonstrate the premise that fertilizer recommendations should be based on models fit with training data collected up to a decision point compared to training data collected through the harvest of the winter wheat. These models are those developed in satisfaction of objective 1. Finally, the web-based decision support application is in development, driven by the development of the OFPE R package to satisfy objective 4.

Project Outcomes

Did this project contribute to a larger project?:
Yes
3 New working collaborations
Knowledge Gained:

This project is still in a preliminary phase, however some initial insights can be made. First, multiple years worth of data are required to adequately fit crop response (yield and protein) models, as indicated by the preliminary model fits using data collected from prior OFPE work on the fields in this study. Second, while zero rates are required to fit regression models, they are disliked by producers because of the difference in appearance compared to well fertilized plants, however judging by the distribution of crop response observations at zero rates, nitrogen fertilizer does not have undue influence on yield and protein. Third, from discussions with both our producers, they are committed to reducing their environmental impact, however reiterate that, in their eyes, their livelihood relies on their inputs. This emphasizes that the steps towards sustainable agriculture from modern conventional practices will be incremental, with convincing evidence required at every step to shift producer mindsets from the current input centric paradigm.

Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture or SARE.