A comparison between visual estimates and image analysis measurements to determine Septoria leaf blotch severity in winter wheat

Methods to estimate disease severity vary in accuracy, reliability, ease of use and cost. Severity of Septoria leaf blotch (SLB, caused by Zymoseptoria graminicola) was estimated by four raters and by image analysis (assumed actual values) on individual leaves of winter wheat in order to explore accuracy and reliability of estimates, and to ascertain whether there were any general characteristics of error. Specifically, (i) we determined the accuracy and reliability of visual assessments of SLB over the full range of severity from 0 to 100%, and we investigated (ii) whether certain 10% ranges in actual disease severity between 0 and 100% were more prone to estimation error compared with others, and (iii) whether leaf position affected accuracy within those ranges. Lin's concordance correlation analysis of all severities (0 to 100%) demonstrated that all raters had estimates close to the actual values (agreement: ρc = 0.92-0.99). However, agreement between actual SLB severities and estimates by raters was less good when compared over short 10% subdivisions within the 0-100% range (ρc = -0.12 to 0.99). Despite common rater imprecision at estimating low and high SLB severities, individual raters differed considerably in their accuracy over the short 10% subdivisions. There was no effect of leaf position on accuracy or precision of severity estimate on separate leaves (L1-L3). Pursuing efforts in understanding error in disease estimation should aid in improving the accuracy of assessments, making visual estimates of disease severity more useful for research and applied purposes.