FIELD VALIDATION OF WEED-CONTROL RECOMMENDATIONS FROM HERB AND SWC HERBICIDE RECOMMENDATION MODELS

Field validation studies were conducted in seven Mississippi environme nts at three application timings to confirm postemergence (POST) recom mendations generated by the computer herbicide decision aids HERB and SWC. HERB and SWC agreed on herbicide treatments in only 14% of the lo cation-application timing combinations, Weed scientists involved in th e study agreed on treatment recommendations approximately 33% of the r ime. The HERB model agreed with a faculty member on only one herbicide treatment, while the SWC model was slightly more agreeable in this re gard. Subsequent weed flushes, varied production practices, and delaye d weed emergence accounted for a majority of the underestimated predic tions given by HERB. Only 55% of the predicted values presented for es timated weed control ratings were similar to actual weed control ratin gs. Over 75% of the predictions that differed from actual weed control values were underpredictions. Recommendations from both computer mode ls were effective in reducing yield loss below that of the untreated c heck, and recommendations from the HERB model generally improved yield more than those from the SWC model in most instances. HERB and SWC pr edictions of yield losses with no weed control were not significantly different from the actual yield loss from the untreated check in nine of the 12 instances at Starkville, seven of the 12 instances at Brooks ville, six of the 12 instances at Newton, and three of the six instanc es at Hollandale. The HERB model estimated yield loss similar to that of the actual yield loss 83% of the time; while predictions from the S WC model were accurate 76% of the time. HERB overestimated yield loss in six of 21 application timing-experiment combinations and underestim ated yield loss only once. Yield loss was overpredicted as high as 78% . SWC overpredicted yield loss in live of 21 instances and also undere stimated in five instances. SWC did not overpredict yield loss to the same magnitude as HERB in many instances.