Machine Learning for Asymptomatic Ratoon Stunting Disease Detection With Freely Available Satellite Based Multispectral Imaging

TL;DR

Ratoon Stunting Disease (RSD) is asymptomatic and hard to detect in sugarcane. The authors evaluate several classifiers on freely available Sentinel-2 multispectral data and vegetation indices to detect RSD across five sugarcane varieties, identifying SVM-RBF as the top performer (up to 96.6% accuracy) and strong results from RF. They show that including variety information and vegetation indices improves detection, while NDVI is less predictive for asymptomatic disease. The findings demonstrate a cost-effective, scalable satellite-based approach for large-scale sugarcane health monitoring with practical implications for management decisions.

Abstract

Disease detection in sugarcane, particularly the identification of asymptomatic infectious diseases such as Ratoon Stunting Disease (RSD), is critical for effective crop management. This study employed various machine learning techniques to detect the presence of RSD in different sugarcane varieties, using vegetation indices derived from freely available satellite-based spectral data. Our results show that the Support Vector Machine with a Radial Basis Function Kernel (SVM-RBF) was the most effective algorithm, achieving classification accuracy between 85.64% and 96.55%, depending on the variety. Gradient Boosting and Random Forest also demonstrated high performance achieving accuracy between 83.33% to 96.55%, while Logistic Regression and Quadratic Discriminant Analysis showed variable results across different varieties. The inclusion of sugarcane variety and vegetation indices was important in the detection of RSD. This agreed with what was identified in the current literature. Our study highlights the potential of satellite-based remote sensing as a cost-effective and efficient method for large-scale sugarcane disease detection alternative to traditional manual laboratory testing methods.

Figures (7)

• Figure 1: Visual representation of the field sampling for both block types where the red marker indicates a sampling location.
• Figure 2: Sentinel-2 multispectral bands for an example block, with each panel showing a different band. Pixel values have been normalised to highlight within-field variation.
• Figure 3: Reflectance distributions across Sentinel-2 bands for five sugarcane varieties and an overall panel. Outliers are omitted for clarity. Healthy and diseased crops are colour-coded (green = healthy, orange = diseased).
• Figure 4: Nine vegetation indices for a sample block that are historically significant for vegetation monitoring or were found to be important for detecting RSD. Pixel values have been normalised to highlight within-field variation.
• Figure 5: Visual guide to the machine learning workflow. The dataset was split into training and testing sets. The training set underwent 10-fold cross-validation to select the best model. Model performance was then assessed by bootstrapping the test set 5,000 times, followed by permutation testing.