#### Abstract

Field surveys for common crupina, as part of an eradication program, are time intensive and could be made more efficient if common crupina habitat could be predicted. Slope, aspect, and vegetation data were used as generalized plant community variables to predict common crupina habitat using a transformed logistic regression. Models were constructed using either aspect or slope as an explanatory variable such that one model predicted the overall effect of either slope or aspect and a set of models predicted the effect of slope or aspect at each of three vegetation classes. A second data set was used to validate the prediction equations for slope and aspect. The proposed models fit the data well and validations were successful as indicated by analysis of residual plots. The probability of finding common crupina was highest for southeast to southwest aspects. In addition, common crupina was most likely to occur, overall, at 25 to 30% slope with decreasing probability at gentler and steeper slopes. Slope models fitted at each vegetation class indicated maximums at 25 to 30% slope for forest and mesic grassland areas but the maximum for arid grasslands was 50% slope. A field detection survey of common crupina that was directed according to probability of occurrence differences along aspect and slope gradients could reduce the area surveyed to 34 to 42%, respectively, of the study area (using a probability cutoff of 30% of the model's maximum). Detection surveys directed according to slope models would find 14% more common crupina than aspect models but would survey 8 to 11 % more area. Models that considered vegetation class, when contrasted with models that did not consider vegetation class, did not decrease the total area surveyed while maintaining the same percentage of common crupina found.

#### Keywords

Logistic regression, common crupina, geographic information system, weed eradication

#### Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

PREDICTING COMMON CRUPINA HABITAT WITH GEOGRAPHIC AND REMOTE SENSING DATA

Field surveys for common crupina, as part of an eradication program, are time intensive and could be made more efficient if common crupina habitat could be predicted. Slope, aspect, and vegetation data were used as generalized plant community variables to predict common crupina habitat using a transformed logistic regression. Models were constructed using either aspect or slope as an explanatory variable such that one model predicted the overall effect of either slope or aspect and a set of models predicted the effect of slope or aspect at each of three vegetation classes. A second data set was used to validate the prediction equations for slope and aspect. The proposed models fit the data well and validations were successful as indicated by analysis of residual plots. The probability of finding common crupina was highest for southeast to southwest aspects. In addition, common crupina was most likely to occur, overall, at 25 to 30% slope with decreasing probability at gentler and steeper slopes. Slope models fitted at each vegetation class indicated maximums at 25 to 30% slope for forest and mesic grassland areas but the maximum for arid grasslands was 50% slope. A field detection survey of common crupina that was directed according to probability of occurrence differences along aspect and slope gradients could reduce the area surveyed to 34 to 42%, respectively, of the study area (using a probability cutoff of 30% of the model's maximum). Detection surveys directed according to slope models would find 14% more common crupina than aspect models but would survey 8 to 11 % more area. Models that considered vegetation class, when contrasted with models that did not consider vegetation class, did not decrease the total area surveyed while maintaining the same percentage of common crupina found.