PARENT SESSION
Oral Session - Spatial Analysis - Morning Session Chair(s): Nash, Maliha¹, ¹ Landscape Ecology Branch, Las Vegas, NV
Friday, April 2, 2004 9:00 AM - 10:40 AM Apollo Room 3

Quantifying habitat fragmentation: Do remote sensing classification errors influence landscape metrics? LANGFORD, BILL ¹ and ^*GERGEL, SARAH E. ², ¹ National Center for Ecological Analysis and Synthesis, Santa Barbara, CA² Centre for Applied Conservation Research, Vancouver, BC

ABSTRACT- Remote sensing imagery is routinely classified into different land cover types and then used as base maps to measure fragmentation using landscape metrics such as mean patch size, number of patches, etc. No classified map, however, is ever completely accurate, and classification errors are often greater at boundaries between cover types. We used a set of replicated synthetic landscapes with varying proportions of habitat and clumpiness to determine whether different maps with the same classification error rates could have different errors in landscape metrics. First, we created a set of simulated correct base landscapes with two cover types, representing a map classified into habitat and background with no error. We then simulated classification errors on these maps to create incorrect or misclassified maps with larger errors at patch edges. We then smoothed the misclassified images using a process routinely used to correct salt-and-pepper classification errors on images. Relatively minor classification error rates often resulted in large magnitude errors in landscape metrics. Percent error in metrics (relative to the correct image) was often extremely high, sometimes as high as 5000%. While reducing classification error, image smoothing often increased the error in several metrics and reversed the direction of bias in metrics introduced by the misclassification errors. Classification error is not always a good predictor of errors in landscape metrics, and some types of image post-processing (e.g., smoothing) might result in under-estimation of habitat fragmentation.

KEY WORDS: pattern analysis, landscape metrics, fragmentation, accuracy assessment