Rice leaf pathogen infection alters the spectral–biochemical relationships revealed by radiative transfer model inversion

The PROSPECT model, grounded in generalized plate theory, provides a mechanistic framework linking leaf spectral reflectance to its chemical composition and structural properties. Such physically-based approaches enable the widespread retrieval of plant photosynthetic pigments (C_a+b and C_x+c). However, its effectiveness is compromised in diseased leaves, where pathogen-induced collapse of intercellular structure violates the model's structural assumptions and consequently, alter spectral–biochemical relationships. To address this issue, this study presents a methodology for accurately retrieving within-lesion photosynthetic pigments by integrating Multiple Endmember Spectral Mixture Analysis (MESMA) with PROSPECT (MESPECT). Using MESPECT together with multi-temporal proximal hyperspectral imagery (HSI: 400–850 nm) of rice blast (RB)-infected leaves, we examined how infection alters spectral–biochemical coupling within disease lesions, and generated spatio-temporal biochemical maps to reveal pre-visual disease symptoms. The sensitivity and importance of MESPECT-derived C_a+b and C_x+c were evaluated to quantify their contributions to random forest classification for detecting RB infection. Our results showed that pathogen infection disrupted the spectral–biochemical relationships captured by PROSPECT in infected leaf tissues caused by irreversible structural collapse, whereas this decoupling was effectively mitigated by MESPECT. The MESPECT-derived leaf C_a+b and C_x+c exhibited high sensitivity to pathogen infection and captured disease signals three days before visual lesions became apparent. Notably, incorporating MESPECT_C_a+b into machine learning significantly improved discrimination of infected leaves, yielding an overall accuracy of 75% (к = 0.48) and 91% (к = 0.80) at the asymptomatic and symptomatic stages, respectively. Our findings underscore the disruption of the spectral–biochemical relationships modeled by PROSPECT under pathogen infection, advancing our understanding of spectral biology in response to pathogen infection and enabling spectroscopic approaches for detecting pre-visual disease symptoms.