Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Abstract. Severe weather events impact many crop-growing regions annually and bring with them heavy rain and high winds, which can cause lodging. Evaluating the presence and level of damage has traditionally been a task that is time-consuming and only provides a qualitative estimate of in-field damage. A preliminary proof-of-concept determined that unmanned aerial system (UAS) imagery can be used to detect simulated damage. This damage was simulated by breaking a maize crop either at the root or just below the first ear over the course of five weeks around the relative maturity of the crop. A randomized complete block study was conducted to determine if simulated crop damage can be reliably detected using vegetative index (VI) and digital elevation model (DEM) methods. This study utilized the extracted VIs and DEMs as inputs for a machine learning model to predict the presence and level of damage. A comparison of Random Forest and Artificial Neural Networks was conducted to determine which approach could more accurately predict the level of crop lodging in a field. The modeling approaches were evaluated using a Kappa coefficient to determine the level of agreement between model prediction and actual damage mode in the field. Testing was conducted using k-folds cross validation over the treatment course and the best performing fold for each week was used for comparison between models. In the best week, the Random Forest model performed the best (Kappa = 0.944) with the Artificial Neural Network still performing in a satisfactory manner (Kappa = 0.842).