Prediction of turfgrass quality using multispectral UAV imagery and Ordinal Forests: Validation using a fuzzy approach

Authors: Hernandez, Alexander; Bushman, Shaun; JOHNSON, PAUL - Utah State University; Robbins, Matthew; Patten, Kaden

Submitted to: Meeting Proceedings
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/29/2024
Publication Date: 11/1/2024
Citation: Hernandez, A.J., Bushman, B.S., Johnson, P., Robbins, M.D., Patten, K.T. 2024. Prediction of turfgrass quality using multispectral UAV imagery and Ordinal Forests: Validation using a fuzzy approach. Meeting Proceedings. https://doi.org/10.3390/agronomy14112575.
DOI: https://doi.org/10.3390/agronomy14112575

Interpretive Summary: There is a need to evaluate turf grasses quality in order to evaluate their performance under different ecological situations. Usually, these quality assessments are conducted with visual raters that go to the field and based on several criteria such as color, density, and coverage of the ground determine the quality of a turf grass variety using a gradient from 1 to 9 with 1 being the lowest and 9 the highest qualities, respectively. These visual assessments are oftentimes subjective and they can vary with each rater, the point of observation, time of the day, and other factors. We are presenting a protocol to utilize imagery from sensors on-board unmanned aerial vehicles (aka. "drones") to build predictive models of turf grass quality that are less subject to the drift observed in visual ratings, and that include seasonal (spring, summer and fall) variability across two years of data collection. Our developed models can serve as proof-of-concept that can be implemented in other instances where rapid turf grass evaluations are needed. In addition, the methods and results presented can initiate the development of a digital library of predictive models that can encompass different varieties of turf grass as well as other landscapes with different ecological characterizations.

Technical Abstract: Protocols to evaluate turfgrass quality rely on visual ratings that depending on the rater’s expertise can be subjective, and susceptible to positive and negative drifts. We developed seasonal (spring, summer and fall) as well as inter-seasonal machine learning predictive models of turfgrass quality using multispectral and thermal imagery collected using unmanned aerial vehicles for two years as a proof-of-concept. We chose ordinal regression to develop the models in-stead of conventional classification to account for the ranked nature of the turfgrass quality assessments. We implemented a fuzzy correction of the resulting confusion matrices to ameliorate the probable drift of the field-based visual ratings. The best seasonal predictions were rendered by the fall (multi-class AUC: 0.774, original kappa 0.139, corrected kappa: 0.707) model. However, the best overall predictions were obtained when observation across seasons and years were used for model fitting (multi-class AUC: 0.872, original kappa 0.365, corrected kappa: 0.872), clearly highlighting the need to integrate inter-seasonal variability to enhance models’ accuracies. Vegetation indices such as the NDVI, GNDVI, RVI, CGI and the thermal band can render as much information as a full array of predictors. Our protocol for modeling turfgrass quality can be followed to develop a library of predictive models that can be used in different settings where turfgrass quality ratings are needed.