Skip to main content

Parameterizing farmer tweets to predict seasonal corn-planting activity

The increasing popularity of social networking sites among farmers presents an opportunity to relay high-quality information into farm management optimization models and on-farm decision-support. Online social networks contain abundant near real-time, high-resolution, and previously unavailable data that could replace delayed, anecdotal, and low-resolution data from traditional sources to provide timely, reliable, and localized advice. The recent evidence cumulative Twitter activity related to corn planting correlates with National Agricultural Statistics Survey corn progress reports but becomes available an average of five days sooner exemplifies how tweets capture agricultural tasks currently occurring. Based on this correlation, we hypothesized trends in tweet occurrence rate can predict future Twitter activity, and by proxy, farm activity, thereby expanding decision-support windows of opportunity further. To explore this possibility, we employed five unique methods to parameterize records of tweets related to corn planting to produce Poisson probability distributions that can predict corn-planting progress. Relatively lower percent bias and higher Nash-Sutcliffe Efficiency values for the discrete and continuous parameter estimates support a time-dependent tweet occurrence rate. Prediction methods did not achieve acceptable accuracies for long-term estimates and were highly variable for short-term estimates. Prediction inaccuracy coupled with low chi-square p-values evidence a Poisson distribution may not describe tweet occurrence, so future research needs to utilize more complex predictive models that consider tweet content.

Author: 
Chelsea Peterson
School: 
Cornell University
Department: 
Environmental Engineering
Research Advisor: 
Dr. Luis Rodriguez
Department of Research Advisor: 
Agricultural & Biological Engineering
Year of Publication: 
2017