|Abstract:||Nitrogen (N) is an essential and often a limiting nutrient in corn production systems. Fertilization with N is critical to achieving high yields and quality. However, because of N's highly mobile nature in the soil, the unpredictability of weather, and the complex interactions between weather, soil, and crop processes, matching N supply and crop N demand to minimize N losses is difficult to achieve. Sub-applications result in low yields, and over-applications result in environmental losses that harm water quality. These difficulties create what is known as the Nitrogen challenge. This challenge's importance is enormous since multiple sectors are affected (agriculture, fertilizer industry, cities water districts, environmental groups, food processors, etc). Additionally, since it is related to food production and food affects lives, it is a sensitive topic.
For years, we solved the N challenge by prioritizing food production and overlooking the environmental impact. Industrial agriculture and the green revolution were successful at increasing yields all over the world. They made a lot of lives possible. Now, we can do more and better; there are growing calls to start using technologies and policies that protect the environment and continue feeding the world.
The problem with those technologies and policies is that it is difficult to know how they will finally affect yield, farm profits, and environmental losses. Farmers are sometimes overwhelmed by offers of new technologies that promise to increase yield or reduce their costs. Policy-makers lack precise long-term estimations to decide what practices they need to advance to help solve the problem. Such long-term estimates are essential to support the negotiations among the different stakeholders that would be impacted by any of the policies.
In this thesis, we designed a crop modeling approach to tackle the N challenge from multiple angles. Chapter 1 describes the development of an extensive database of crop response to multiple N rates over 4,270 fields in a 30-year long period. The database accounted for multiple variables measured in each simulation, including yield, N leaching, soil, and weather characteristics. We calibrated and validated the simulations using actual data from a state trialing network in the region. Using crop modeling allows leveraging those trials over more fields and years and adding extra variables that were not measured in the original fields -i.e. N leaching, soil, and weather conditions. This results in robust conclusions that are not misguided by particular conditions. The developed crop modeling architecture is innovative and can be applied at scale for solving other challenges seen in agriculture.
In Chapter 2 we evaluate different N management strategies that have been proposed as a solution to the N challenge. Several comparisons were made. First, we compare using a regional approach that consists of running trials in research stations and using the data to optimize tools that provide recommendations in other fields versus using a local approach and making trials in the same field where we will provide the recommendations. Second, we examine two approaches to making predictions. The static approach recommends one N rate that maximizes long-term profits, versus the dynamic one that uses time-dynamic variables to adjust N rates based upon particular soil and weather conditions. Third, we evaluated two sets of predictor variables, a low-information set based on sensor information and a high-information set requiring more time and effort to collect the variables. Fourth, we compared using uniform rate and variable rate technology on the fields. We found that regional approaches allow for capturing more temporal variation in the recommendation models and perform better than local approaches. The dynamic approach needs high information to perform better than the static approach; nevertheless, that performance did not translate into higher profits. Variable-rate technology did not provide value in the region. This chapter provides for the first time a comparison of all those strategies together, allows us to see the big picture, and provides meaningful insights about how we can use them the best way possible.
It has been proposed that complex N recommendation tools that use predictive analytics to recommend N rate with higher accuracy will solve the N challenge, increasing profits for farmers and reducing N leaching. Nevertheless, some research shows that more straightforward methods that recommend static N rates, constant over time, provide a similar performance (including our findings in Chapter 2. As a result, in Chapter 3 we analyzed this comparison more deeply. In the first part, we compared multiple N recommendations tools, some of them belonging to the static group and some to the dynamic group. In the second part, we selected each group's best tool and zoomed in to the situations that make dynamic recommendations succeed or fail relative to the static ones.
We found that dynamic tools face disadvantageous economic conditions. Even though they are more accurate, they are not perfect. Around half of the time, they recommend N rates below the observed EONR at harvest. In that situation, the yield penalty is higher, and it is usually not compensated by other more successful recommendations. Static tools that tend to over-predict, avoid that penalty, and provide similar value as that of the more accurate tools. From an environmental point, both static and dynamic tools can be optimized to reduce N leaching (5% reduction) without significant harm to the profits. This chapter's findings are essential to understand why complex tools struggle to provide consistent economic performance and re-organize future research to solve the Nitrogen issue with greater success.
So far, we evaluated options to solve the N challenge in current market conditions, using present corn and N fertilizer prices and not considering any particular economical support from the government. Chapter 4 addressed government intervention and compared four policies that would induce farmers to reduce their N rates further. The first policy was an increase in N price. The second policy was an N leaching fee that charges farmers a fee for the N leaching occurring in their fields. The third policy is an N balance fee, which charges farmers a fee for the extra amount of N applied as fertilizer that was not finally part of the N exported by grains. The fourth policy is a voluntary reduction of N rates, in exchange for compensation. In all policies, we considered that the money collected by the government (by fees or taxes) was returned to farmers, as well as an additional payment that covers the loss due to the policy. We evaluated how farmers would react to the policies and how that will affect their profits and reduce N leaching. The reported results show that all policies allow achieving reductions in leaching by 20%, and that would have a cost of 30-37 $/ha. This cost is comparable with current programs that provide subsidies to farmers and make the policies viable options for reducing N leaching in the area.