Unintended Errors Research Articles

SPREAD PATTERN ANALYSIS TOOL (SPAT): II. EXAMPLES OF AIRCRAFT PATTERN ANALYSIS

The quality of aerial spread patterns was studied based on over 400 test runs in Arkansas during the period1992-1997. A custom developed computer program called SPAT (Spread Pattern Analysis Tool, Grift, 2000) was used tocompute overlapped spread patterns and to determine the pattern quality. The quality of a spread pattern is traditionallyexpressed by the average application rate (the mean of the overlapped pattern) and uniformity [expressed in the statisticalcoefficient of variation (cv)]. Both parameters depend highly on the swath width. Inspired by studying the cv-swath widthrelationships, a new measure for pattern quality called robustness was conceived. This parameter indicates theflexibility of a spread pattern shape, or the ability of the applicator to vary the swath width (and, hence, the applicationrate), either purposely or due to unintended flight path errors, and be confident that the overlapped pattern will have anacceptable uniformity. Grift (2000) stated that, as a rule of thumb, patterns with a robustness factor lower than 5% can beconsidered robust. From studying spread patterns of unadjusted spreaders in Back&Forth mode, it was concluded thatonly 3.45% of them are robust and after adjustments had been made, 2.75%. In RaceTrack mode, the percentage of robustpatterns was 8.59% and after adjustments it increased to 10.28%. The majority of all patterns (approximately 70%), inboth modes, had robustness factors between 5% and 15%. The results of this study imply a definite need to improve thequality of aerial application spread patterns by (1) avoiding pattern shapes that are sensitive to transforming to non-robust,and (2) periodic calibration. The magnitude of the problem indicates that periodic adjustment of spreaders willnot be sufficient, redesign of equipment and spreading procedures may be necessary to accomplish overall high qualityspread patterns.

USE AND MISUSE OF COMPLEX MODELS: EXAMPLES FROM WATER DEMAND MANAGEMENT1

ABSTRACT: The power of computers has increased in recent decades, and one might expect improved management to result because decisions can be made with understanding available only via models. However, there is potential for quite the opposite: poor decisions due to unrealistic model output generated by users without access to appropriate training in the use of models. We discuss and, by reference to water demand models (IWR‐MAIN, MWD‐MAIN), illustrate three areas in which unintended errors of judgment by untrained personnel may cause difficulty: Attributes of management models; if output from any type of model has no measure of confidence, then results may be over‐ or undervalued Input data; with complex models, problems here typically will be difficult to detect. Calibration and history‐matching (verification); if these steps or data are combined, then users should be less trustful of model output than otherwise. Because all models have weaknesses and because there always is uncertainty about output from any model, we end with suggestions for coping with complex models. Monitoring programs play a central role in such efforts because they can identify discrepancies between model predictions and actual events and because they can ensure time is available to develop solutions for problems unanticipated in the modeling effort.