Northern Prairie Wildlife Research Center
Kevin L. Sallee, Independent private consultant, 3154 53rd Street, Sacramento CA 95820 USA
Misclassification of habitat occurs when the geographical coordinates of an estimated location are plotted within a different habitat type than the actual location. It is possible to use maximum likelihood procedures to estimate the amount of uncertainty associated with the habitat assignment of an estimated location. If the bearing errors for a study site are sampled from known transmitter locations maximum likelihood probabilities for each habitat type near an estimated location can be calculated using simulations. However, the locational errors from simulations create a bivariate normal distribution with a variance that will vary depending on several factors including the distance between the transmitter and receiver, and between the different receiver locations. Therefore, one must also know the unique receiver locations associated with each estimated location's triangulation to properly calculate the maximum likelihood probabilities. It quickly becomes very cumbersome to implement this method with mobile receivers that may never use the same locations to calculate the bearings for a triangulation. In addition, one must often move estimated locational points from one program designed to perform a maximum likelihood simulation into a Geographical Information System (GIS) software that contains the associated habitat polygons.
This may represent moving and manipulating thousands of simulated data points. As a consequence, habitat reliability estimates continue to be ignored because there has not been an easy method to estimate reliability of habitat use. As an alternative to simulating the maximum likelihood estimators for habitats near each estimated location I performed simulations that estimated the expected range of distances between randomly located "actual" locations and estimated locations found by using normally distributed random angular errors and randomly placed receiver locations. From this simulated sample I selected the 95th percentile distance as a radius length that formed circles using each estimated data point as the center of the circle. By comparing this method to maximum likelihood procedures, I found that the habitat proportions within the circular area were similar to the proportional estimates from maximum likelihood simulations at each point. The accuracy of each method at estimating the habitat type associated with the actual location did not differ. However, accuracy was related to the shape of the landscape features used by an animal, which varied between landscapes. If an animal used a greater preponderance of narrow habitats, such as water courses, or habitat edges, the accuracy of both methods was lowered. Only one simulation is necessary when using a circular area instead of performing separate simulations at each estimated data point. In addition, there is no need for mass data transfer between programs. One can directly apply habitat analysis to GIS software that can place circular buffers around a point and calculate the habitat proportions within this area. Estimates of probable habitat use for each location may then be calculated and used directly in habitat use analysis or used to select the group of habitat types most likely associated with any location. The former may be a more appropriate method when dealing with edge species where the combination of habitats may be more informative in describing habitat use.