|Title||Spatial distribution and risk assessment of Johnsongrass (sorghum halepense) in Big Bend National Park, Texas|
|Publication Type||Government Report|
|Year of Publication||2014|
|Authors||Young K, Schrader T. S|
|Section||Scientific Invesigations Report 2012-5162|
|Publisher||US Department of Interior; US Geological Survey|
|ARIS Log Number||316429|
We used Landsat 7 ETM+ imagery to illustrate how remotely sensed data can model predicted Johnsongrass habitat. We used spectral reflectance values for three seasons of data across 5 years (fall 1999, summer and fall 2000, spring and fall 2001, spring 2002, and spring 2003) to capture Johnsongrass vegetation phenology. Johnsongrass occurrences in BIBE consisted of 147 georeferenced localities. We used the program Maxent to predict Johnsongrass distributions. Maxent is a general approach for modeling species distributions using presence-only data sets. Maxent estimates a target probability distribution by finding the probability distribution of maximum entropy (the distribution that is most spread out, or closest to uniform). Maxent uses pixels with known species occurrence records to constitute the sample points. Remotely sensed and GIS data sets can provide environmental variables measured at each sample point. Analysis output includes a probabilistic interpretation, grading from least to most suitable habitat conditions. We evaluated the Johnsongrass predicted habitat model in Maxent by withholding 10 percent of the occurrence locations for testing. This approach, considered threshold-dependent, used a binomial test based on omission and predicted area. Model performance is evaluated using extrinsic omission rate and the proportion of all the pixels that are predicted as suitable habitat. The second approach to evaluating model performance is considered a threshold-independent procedure and uses receiver operating characteristic (ROC) curves. The advantage of ROC analysis is that area under the ROC curve (AUC) provides a single measure of model performance, independent of any particular choice of threshold. The AUC can be interpreted as the likelihood that habitat quality is correctly classified by the predictive model at randomly selected sites.