A novel phenotyping approach for reconciling precision and variance in disease severity estimates from high-resolution imaging

Abstract Accurate quantification of plant disease is essential for resistance breeding, variety testing, and precision agriculture, yet visual ratings are limited by subjectivity, low precision, and restricted throughput. Image-based phenotyping can address these limitations, but field applications are challenging due to spatial heterogeneity and miniature diagnostic features. High-resolution imagery with a limited field of view and a shallow depth of field is therefore needed, restricting the analyzed leaf area. This introduces a fundamental trade-off: high-resolution images provide precise local measurements of disease, but spot-level estimates can be highly variable within experimental units. A key challenge is therefore to understand this variability and to develop sampling strategies that translate precise local observations into reliable plot-level measurements. Here, we analyzed spot-level estimates of septoria tritici blotch and brown rust severity derived from 16,650 images acquired in 15 experimental plots over two consecutive days to quantify the statistical distribution and spatial dependence of local estimates of disease. High-resolution macro-scale images were obtained at multiple positions around a plot and at multiple heights. Focus bracketing was evaluated as a practical strategy to increase the sampled leaf area without additional operator effort. Autocorrelation in severity estimates within focal image stacks and across positions within a plot was comparable, with 10 focal stack images or 10 positions per plot contributing approximately 2.5 independent observations each. Modeling plot-level severity as a latent Beta-distributed variable provided robust estimates of mean disease severity and its associated uncertainty, highlighting the importance of dense spatial sampling, particularly at advanced stages of disease development. Obtaining a precise understanding of the correlation structure and an accurate quantification of achievable precision gains from repeated measures will depend on larger experiments. Nevertheless, our results indicate that combining dense spatial sampling with focus bracketing and appropriate hierarchical statistical models can substantially improve the precision of plot-level disease estimates without sacrificing throughput.