The reliability of visual estimates of disease severity on cereal leaves

Visual estimates of wheat disease severity were compared with actual severities determined using image analysis of tracings of diseased leaves. Septoria tritici, leaf senescence and Erysiphe graminis were studied. Observer estimates were widely scattered about the actual severities (i.e, they were imprecise), differed substantially from the actual severities even after averaging (i.e. they were inaccurate), and varied considerably over short time-scales. Relative bias decreased with increasing disease severity. In a comparison of three seed treatments to control powdery mildew on winter barley, visual assessment errors altered the conclusions of the experiment. Two treatments were statistically indistinguishable on the basis of visual severity estimates, while estimates obtained by image analysis showed that one seed treatment was twice as effective as the other.