NYCSGA Precision Ag Project Update

Authored by: Savanna Crossman, Research Coordinator, CCA
Statistical Analysis by: Margaret Krause, Cornell University PhD Student

 *This article is part of an ongoing series.  Previous articles can be found at www.nycornsoy.org/research*

crossman-fig-1New York State has always presented a unique challenge to grain growers due to the large amount of in field variability.  In recent years, growers have also added adverse weather conditions to that list.  From the project’s perspective, two of the past three growing seasons have fallen far outside the conditions of a normal year.  The 2015 season brought early precipitation amounts far above than the historical average while the 2016 season is setting up to be one of the driest in decades.  These conditions have resulted in significantly lower, less uniform yields than a typical year such as 2014 (Figure 1).  Variable rate seeding technology is one of the many tools that NYS growers can use to help overcome these challenging conditions.  However, mainstream companies have yet to design a prescription writing software that is developed to meet the unique conditions of New York State and the Northeast.  This project seeks to address this void by developing a software that will do just that.

The project has been collecting data on a large scale since 2014 in order to create a model that will select hybrids and population rates given certain soil properties and characteristics.  To do this, six major data types are being examined; seeding rate, hybrid, topographical information, NRCS soil survey maps, Veris soil sampling data, and grid soil sampling data.  Each data type consists of many variables which are analyzed individually and as interacting networks.

Figure 2. This example random forest regression analysis demonstrates that phosphorus is the variable with the largest effect on yield in this field.
Figure 2. This example random forest regression analysis demonstrates that phosphorus is the variable with the largest effect on yield in this field.

To examine the effect that each variable has on yield, a statistical approach called random forest regression is being used.  This method essentially ranks each variable based on its importance to yield.  The greater the importance number that is assigned to a variable, the larger effect that variable has on yield (Figure 2).

The project has seen that the variables can rank very differently given the field, crop type, or year.  Each field location is unique and thus has a unique combination of variables influencing yield.  Some fields exhibit a very strong yield response to seeding rate, while others exhibit a strong yield response to fertility factors or topography.

Figure 3. 2014 and 2015 resulted in similar population curves on this corn field in Clyde, NY.
Figure 3. 2014 and 2015 resulted in similar population curves on this corn field in Clyde, NY.

Though each field may be different, it is important to see stability within a field across years.  For example, this 80 acre corn field in Clyde, NY produced similar population curves in two drastically different seasons.  The first year, 2014, resulted in high and uniform yields across the field.  The second year, 2015, yielded dramatically lower with a large variance in yield uniformity.  Though the two seasons were very different, both demonstrated a negative yield response to increased seeding rate (Figure 3).  The lowest rate of 27,000 sds/ac yielded the highest across the two years and which was 5,000 sds/ac lower than the grower’s typical rate.  The random forest regression confirmed that seeding rate was the most important variable influencing yield across both years.

This year to year stability in yield response to seeding rate has been seen between crop types as well.  This 60 acre field in Pavilion, NY is managed as a conventional till field in a corn-soybean rotation.  In 2014, its soybean crop exhibited a strong positive yield response to increased seeding rate.  The random forest regression confirmed that seeding rate held a dramatically greater importance than any of the other variables.  The next year, 2015, the field was planted with corn and again exhibited a positive yield response to seeding rate.  This time, the analysis showed that while seeding rate was still the most important variable, many other factors were also important.  This difference could be due physiological preferences between the two crops or the different weather conditions between the two years. (Figure 4)

Figure 4. Random forest regression analysis of 2014 soybean and 2015 corn of a sixty acre field in Pavilion, NY.
Figure 4. Random forest regression analysis of 2014 soybean and 2015 corn of a sixty acre field in Pavilion, NY.

To explore the idea of physiological differences between corn and soybean, some further analysis was conducted.  In this same Pavilion field, soybean exhibited positive relationships with calcium and pH, while corn exhibited negative relationships with the same variables.  These observations are likely related to the differences in crop preference for pH.  Soybeans grow best in more neutral soils where the rhizobia bacteria that provide the soybean plant nitrogen are most active.  Whereas the corn plant is known to prefer a slightly acidic soil where some key micronutrients, such as zinc and manganese, are more available.  It is understanding relationships such as these from an agronomic and a statistical perspective that will result in a reliable model for NYS growers.

This year has marked the first infield testing of the model which will provide side by side comparison of grower practice to the model’s prescriptions.  Each year of additional data collected will serve to further the development of the model into a robust and reliable resource to growers of the State.

The project is currently looking to bring on additional participants for the 2017 season and encourages any interested growers to contact Savanna Crossman at (802) 393-0709 or savanna@nycornsoy.com .

screen-shot-2016-09-12-at-1-42-09-pm

Print Friendly, PDF & Email