COS 22-1 - Error and uncertainty in mapping vegetation zonation in the Prairie Pothole Region of Alberta, Canada

Monday, August 7, 2017: 1:30 PM
B115, Oregon Convention Center
Matthew T Bolding1, Rebecca Rooney1 and Derek Robinson2, (1)Biology, University of Waterloo, Waterloo, ON, Canada, (2)Geography and Environmental Management, University of Waterloo, Waterloo, ON, Canada

Shallow open-water wetlands in the Prairie Pothole Region of Alberta provide many essential ecosystem services. Agricultural and industrial expansion has led to a loss of these wetlands, prompting the Government of Alberta to introduce policy aimed at mitigating wetland loss by requiring the replacement of destroyed wetlands. For effective replacement, a tool is needed to assess the condition of wetlands prior to removal and after restoration. To meet this need we created an assessment tool that measures wetland condition based on the spatial arrangement of wetland vegetation communities (i.e., zonation). Using high-precision GPS to delineate and map vegetation communities, our tool generates a suite of metrics from the spatial arrangement of these communities. However, before implementation, we first needed to quantify the error and uncertainty in this assessment tool. Twenty-four wetlands were visited over two years and mapped using an SX Blue II+ GPS/GNSS receiver. A subset of these sites was mapped twice within one year to provide an estimate of user-interpretation error. All mapping was conducted by the same technician following a comprehensive protocol. Spatial metrics based on the vegetation community maps were created using FRAGSTATS software.


Error and uncertainty can arise from equipment, user-interpretation, data- processing, or inter-annual variation in zonation. Equipment error was assessed by comparing the accuracy and precision of our GPS unit (2.5m 2dRMS) with a Leica CS15GNSS rover/base station unit (10mm RMS). Fifty-nine sample points were recorded with both devices and the Euclidean distance between each point was calculated. The actual error of our GPS unit was 1.86 m 2dRMS, meaning there was a 95% chance that a measured point would be within 1.86 m its actual location. Data-processing error was controlled with a comprehensive protocol and quality control on all GIS work. To estimate error in mapping attributed to the field technician interpretation, a paired t-test was carried out on the values for 17 Class-level and 24 Landscape-level metrics obtained from mapping sites that were visited less than a month apart. Only one metric differed significantly between visits, while most metrics were insensitive to variation introduced by field interpretation of vegetation zonation. To evaluate toll sensitivity to interannual variation in zonation, we compared zonation in 24 wetlands mapped in two consecutive years. Thirteen metrics were sensitive, indicating that interannual variation is responsible for the bulk of uncertainty in tool application. Quantifying error is an important step in validating this zonation-based assessment tool.