Implementation of landscape dynamics monitoring at the North Coast and Cascades Network: initial assessment of change detection methods

Antonova, Natalya; Beirne, Katherine; Thompson, Catharine

In this report, we summarize the changes made to the original protocol methodology during its implementation. We also provide the results of the 2004/2005-2006 change detection analysis and validation for MORA, NOCA and OLYM. Lastly, we evaluate the ability of this protocol to address the NCCN’s landscape dynamics monitoring goals. This report describes the first implementation of the “Protocol for Landsat-Based Monitoring of Landscape Dynamics at North Coast and Cascades Network Park” (Kennedy et al. 2007a). The Protocol contains a detailed description of the methodology for the NCCN landscape dynamics monitoring and provides the context for this report. In this report, we summarize the changes made to the original protocol methodology during its implementation. We also provide the results of the 2004/2005-2006 change detection analysis and validation for MORA, NOCA and OLYM. Lastly, we evaluate the ability of this protocol to address the NCCN’s landscape dynamics monitoring goals. The North Coast and Cascades Network (NCCN) selected landscape dynamics as a Vital Sign for monitoring. A list of targeted disturbance types for yearly monitoring included: avalanche chutes, landslides, fire, insect defoliation in forests, windthrow, riparian disturbance, clearcuts, and rural development. A remote-sensing based two date change detection protocol was developed by cooperators at Oregon State University (OSU, Kennedy et al. 2007a). Several key modifications were made to the protocol’s initial “startup” phase. First, baseline images from 2004 and 2005, rather than circa 1994 were used in order to link to recent NPS field data. Second, the OLYM baseline image development method had to be substantially altered to accommodate the fact that OLYM is covered by multiple satellite scenes.In the change detection phase, the 2005 baseline image (2004 for OLYM) was then compared to a 2006 image. The results from the two-date change detection demonstrated that the current protocol could detect the types of change of interest to the NCCN. However, the results were dominated by a large amount of real but “uninteresting” changes due to differences in phenology, cloud cover, shadows and snow, and image to image registration. Techniques for reducing the uninteresting change were explored, including developing cloud, shadow, and snow masks and increasing the minimum mapping unit. False change due to differences in phenology remained particularly intractable.Difficulties were encountered during change validation and accuracy assessment step. Individual interpreters varied in their approach to the validation process, both in delineating polygons and handling fuzzy boundaries. The accuracy assessment became more of a measure of how well the interpreter mimicked the computer algorithm process, rather than an assessment of how well the algorithm mapped real change. The reported accuracy results are therefore not a reliable representation of the accuracy of the change detection.The NCCN protocol incorporated the best remote sensing methods for two date change detection available at the time. In the interim, the OSU group has developed a new change detection method that takes advantage of the entire Landsat archive and minimizes the number of “uninteresting” changes present in the final product (Kennedy et al. 2007b). The existence of improved methods makes it difficult to envision continuing with this current protocol.

Implementation of landscape dynamics monitoring at the North Coast and Cascades Network: initial assessment of change detection methods

Series

Collections