MODELING CROP LOSSES DUE TO DISEASE

Crop loss is a function of many stress factors; one such factor is plant disease. Contemporary models for describing and predicting losses caused by plant diseases were selected, usually with ordinary least-squares regression, to best fit the data and not necessarily represent biological reality. These models also were developed for specific cases without inherent flexibility. A generalized nonlinear model has been developed to characterize the relationship between crop loss and plant disease. In addition to providing loss predictions at various levels of disease, this model also incorporates: (i) a threshold disease level below which no loss occurs; (ii) a maximum level of loss which may occur prior to the maximum amount of disease; and (iii) a large family of shapes to depict disease-loss relationships. The model contains three parameters which are estimated from loss data. A FORTRAN computer program was written to estimate the parameters and their respective variances, and also provides summary statistics such as error mean square and coefficient of determination.;Once it was determined that the general loss model could accurately describe loss data, a second study was conducted to: (i) compare the use of final disease severity with area under the disease progress curve (ADPC) as the representation of disease in the general model; and (ii) compare the general model with the multiple-point model. The multiple-point model used a weighted sum of disease levels throughout the epidemic to predict loss. Data from the two disease-host systems were used for the analyses. Error mean square and coefficient of determination were used to compare models. The general loss model provided an equal or better fit to the data than the multiple-point model. The use of ADPC in the general model did not improve the fit of the data over simply using final disease severity. The superiority of the general model was attributed to the theoretical considerations used in developing the model and to the high correlation among successive disease variables in an epidemic. The ability to use a single model has the advantage that loss caused by different diseases on different crops can be compared directly through an analysis of the model parameters.;The general loss model was evaluated to determine if it could numerically approximate the basic loss models reported in the literature and also if it could describe actual loss data. The general model, therefore, was fit to simulated and actual loss data sets. The model numerically approximated the loss models reported in the literature as indicated by coefficients of determination greater than 0.95 in all cases. The general model also described losses for the following disease-host systems over a 2 yr period: (i) Cercospora leaf spot of peanut (Arachis hypogeae) caused by Cercospora arachidicola and C. personata, and (ii) potato (Solanum tuberosum) early blight caused by Alternaria solani. Coefficients of determination were greater than 0.80 for the fit of the general model to these actual loss data sets.