Soil, landscape, and weather affect spatial distributions of corn population and yield

Authors: Sudduth, Kenneth - Ken; CONWAY, LANCE - University Of Missouri; Kitchen, Newell

Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 5/13/2022
Publication Date: 6/26/2022
Citation: Sudduth, K.A., Conway, L.S., Kitchen, N.R. 2022. Soil, landscape, and weather affect spatial distributions of corn population and yield. In: Proceedings of the 15th International Conference on Precision Agriculture, June 26-29, 2022, Minneapolis, Minnesota. Available: https://ispag.org/proceedings.

Interpretive Summary: As more planters are equipped with the technology to vary seeding rate, farmers need to know what seeding rate is appropriate in different parts of their fields. One important piece of information to consider in that decision process is how plant stand and yield are related to soil and landscape factors, and how these relationships vary from year to year under different weather conditions. Therefore, this research examined corn yield, harvest population, and soil and landscape data obtained for a central Missouri, USA field over nine growing seasons. In every year, population variability was large across the field, and population at harvest was as much as 40% lower than seeding rate. Using “random forest” machine learning models, harvest population and per-plant yield could be estimated for a single year using a combination of soil and landscape data but estimates valid across multiple years were not obtained. However, field regions where population and yield were less variable across years could be identified with a combination of easily obtained landscape data and soil data from proximal sensors. A next step will be to investigate more complex models that can incorporate year-to-year changes in weather-related growing conditions to predict population and yield more accurately in those different conditions. If validated in multiple growing environments, the results of this study will provide information helpful for determining variable crop seeding rate recommendations for variable soils and landscapes.

Technical Abstract: As more planters are equipped with the technology to vary seeding rate, evaluation of the within-field relationships between plant stand density (or population) and yield is needed. One aspect of this evaluation is determining how stand loss and yield are related to soil and landscape factors, and how these relationships vary with different weather conditions. Therefore, this research examined nine site-years of mapped corn yield, harvest population, and soil and landscape data obtained for a central Missouri, USA field. Mechanical population sensors collected data during combine harvesting and provided information at the same scale as yield monitor measurements. Results showed spatial population variability at harvest was large in all site-years, with populations as much as 40% lower than seeding rate. Random forest machine learning models were created to relate population and yield ratio (or per-plant yield) to landscape data, proximal soil sensor data, and data from laboratory analysis of grid soil samples, both for a single year and multiple years. Single-year harvest population modeled very well (test set R2 = 0.84), with most important predictors including landscape and proximal sensor variables as well as soil-test phosphorus. Yield ratio also modeled well (test set R2 = 0.65), with the most important predictors being landscape properties. Direct modeling of multi-year population and yield ratio was not successful; however, models representing the temporal standard deviation in spatial population and yield ratio were moderately successful (test set R2 = 0.50-0.52), with the most important predictors being landscape variables, soil apparent electrical conductivity, soil organic matter, and cation exchange capacity. This case-study analysis showed the potential for explanatory modeling of spatial variability in harvest population and yield ratio, as well as their across-year temporal variability. Further research should investigate additional machine learning approaches more capable of modeling weather information in multiple-year analyses.