Remote sensing of rice leaf folder damage using ground-based hyperspectral radiometry

• Multinomial logistic regression analysis (MLR) identified the sensitive bands (396, 675, 764, 1123 and 1458 nm) specific to leaf folder damage. • Principal component analysis identified the optimum combination of these five sensitive bands. • Principal components were used to build MLR models for assessing the rice leaf folder severity. • These hyperspectral indices helps in assessing the area-wide pest damage in rice. Rice is the staple food for more than 60 % of the world's population and pest damage is one of the major limiting factors of rice production in India. Rice leaf folder, Cnaphalocrocis medinalis (Guenee) (Lepidoptera: Pyralidae) is assuming the major pest status in view of its severe damage. There was severe outbreak of rice leaf folder in three different states of India during different periods between 2010 and 2022. Extensive field surveys were conducted from the locations during this outbreak period and the ground-truth data was collected. A total of 272 rice plants with varying levels of leaf folder damage symptoms were sampled across 3 major rice growing regions of India to collect the spectral reflectance data using FieldSpec-3 Hi-Res hyperspectral spectroradiometer (spectral range: 350–2500 nm, ASD Inc., USA) during 3 seasons. The spectral data was interpolated using ASD ViewSpecPro software in the post-processing to produce values at each nanometer. Multinomial logistic regression analysis (MLR) was performed to identify the sensitive bands (396, 675, 764, 1123 and 1458 nm) specific to leaf folder damage. Principal component analysis (PCA) was performed to identify the optimum combination of these five sensitive bands. The new hyperspectral indices identified in this study specific to leaf folder damage were found to perform better than the relevant spectral indices published earlier. The identified principal components (PCs) were used to build MLR models for assessing the rice leaf folder severity. Model outputs were validated using independent data sets. The classification accuracy of the model using the first four PCs as independent variables was in the range of 33 to 72. This study suggests a new set of hyperspectral indices specific to leaf folder damage to assess the area-wide pest damage in rice.