Skip to main content
ARS Home » Plains Area » Miles City, Montana » Livestock and Range Research Laboratory » Research » Publications at this Location » Publication #170236


item Rinella, Matthew - Matt
item Sheley, Roger

Submitted to: Weed Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/15/2005
Publication Date: 9/1/2005
Citation: Rinella, M.J., Sheley, R.L. 2005. Models that predict invasive weed and grass dynamics: 2. Accuracy evaluation. Weed Science 53:605-614.

Interpretive Summary: In the companion manuscript, we developed models that predict how weeds and plants that grow with them will respond to weed control strategies. In this manuscript, we tested the predictive capability of these models using data from grass seeding experiments and herbicide experiments and concluded that the models predict weed and grass abundances acceptably accurately. Our models will help land managers make better invasive weed management decisions.

Technical Abstract: How invasive weed management strategies impact plant community compositions is highly dependent on local conditions and temporal variability. Therefore, research that studies management strategies at/during one set of sites and years will not necessarily predict composition at/during other sets of sites and years. In theory, grass and leafy spurge models we've developed will partially overcome this problem (see companion manuscript). Testing this theory is the goal of this manuscript. We formulated our models to predict selective plant removal, herbicide and grass seeding experiment data from a wide array of sites and years. The models predicted selective plant removal and herbicide experiment data acceptably accurately, and residual error appears randomly distributed about model predictions. The leafy spurge model predicted seeding experiment data acceptably accurately (leafy spurge densities often exceed 200 stems m-2 and the mean predicted - observed was 10 stems m-2), but in some cases residual error is not randomly distributed about predicted values. This systematic deviation, as well as all random deviations, may reflect factors the models don't consider (i.e. random error), improper model structure and/or variable parameters, all of which would cause error if our models were used for predicting management's impact on plant community composition. Alternatively, measurement error, unintended herbicide injury and/or inaccurate allometric relationships may have caused the deviations, and these factors would not cause prediction error if the models were applied. Though our analysis did not conclusively identify the major sources of deviations, it did allow for informed speculation. We conclude that, because the models predicted data fairly accurately, they could help managers predict leafy spurge and grass responses to management, but because models predictions are imperfect and sometimes biased, it will be very beneficial to quantify prediction confidence, especially if the major sources of error turn out to reflect inadequacies in the models as apposed to inadequacies in the data used to test the models.