A disease-specific spectral index tracks Magnaporthe oryzae infection in paddy rice from ground to space

Abstract

Rice blast (RB, caused by the fungus Magnaporthe oryzae) is the most devastating disease in global rice production, and can cause significant yield losses and increasingly threaten global food security. Accurate detection of RB occurrence with a universal metric is crucial to facilitate early disease prevention and curtailing the disease expansion but has not been addressed to date. This study aimed to design a rice blast index (RIBI) for quantifying the disease index (DI) and tracking the smallholder rice blast dispersal over multiple spatial scales. To achieve this goal, a large dataset including leaf- and canopy-scale reflectance spectra and satellite imagery was acquired within the framework of seven independent campaigns over four years (2018–2021). Specifically, an extensive collection of Magnaporthe oryzae infected samples were analyzed to examine the specific spectral response to pathogen infection in paddy rice from leaf to near-ground canopy scales. Two variants of the RIBI were developed, which were RIBInir = (R753-R1102)/(R665 + R1102) and RIBIred = (R753-R1102)/(R665 + R1102) based on the single-band separability and exhaustive search of band combinations. They were subsequently evaluated for quantifying the RB occurrence from ground to space. Spatial cluster analysis was then integrated with the superior RIBI adjusted for Sentinel-2 imagery to explore the spatio-temporal dynamics of pathogen infection, and to reveal the within-field hotspots of potential rice blast dispersal in smallholder farms.The results demonstrated that both RIBInir and RIBIred exhibited high overall accuracies for the classification of infected and healthy samples at the leaf scale under greenhouse in 2018 (RIBInir: 81.41%; RIBIred: 84.62%) and 2019 (RIBInir: 81.30%; RIBIred: 90.37%) and field conditions in 2020 (RIBInir: 86.36%; RIBIred: 89.39%). Compared with traditional VIs (Near-ground: R2 < 0.47, satellite: R2 < 0.54), the RIBInir yielded improved R2 in quantifying the DI from in situ spectra to satellite imagery (Near-ground: R2 = 0.73, satellite: R2 = 0.78). The strongest DI-RIBInir relationship was attributed to the use of two near-infrared (NIR) bands that helped enhance the unique spectral responses in the NIR region induced by pathogen infection, in contrast to the extensively studied visible region. Multi-temporal analysis of Sentinel-2A derived RIBInir successfully captured the temporal dynamics of RB infection and recovery and yielded compelling maps showing the spatial propagation and attenuation of disease over time. This research opens new opportunities towards quantifying field disease occurrence and detecting the within-field hotspots of potential disease dispersal in a timely manner from publicly available satellite imagery.