UCLA

Accurate estimates of wetlands and stream extent provide context for scientific investigations, enable informed management, and measure progress towards no-net-loss policy goals. However, the default approach for monitoring extent, comprehensive mapping, is prohibitively resource intensive over large areas, making it both impractical and statistically unreliable. Therefore, a number of national and state-level programs have begun to employ probabilistic approaches to monitor wetland and stream extent. These programs have proven practical for their intended applications but little information exists about the ability of the designs to meet diverse, state-level information needs such as accurately capturing rare wetland types or detecting small-scale changes in extent or spatial distribution. Here, I utilized a simulated sampling approach to empirically model and evaluate probabilistic methods for monitoring wetland and stream extent in California. The optimized design was then directly validated against comprehensive mapping through a pilot-scale implementation. The flexibility of the simulated sampling method enabled assessment of performance for a variety of objectives, beyond statistical precision. By employing this unique approach to sampling design, the results could be customized for the specific information needs of California. Results indicate that generalized random tessellation stratified (GRTS) sampling, a spatially balanced selection methodology, provides significant advantages over simple random sampling or stratification. In addition, fixed sampling locations over time provide the best power for both estimating the current extent and detecting changes in extent over time. The pilot-scale implementation was conducted in two separate regions of California. By using regions with existing, comprehensive aquatic resource maps, a direct comparison was made between the probabilistic and comprehensive mapping efforts. The pilot produced precise estimates of aquatic resource extent but suggested that mapping and classification methodologies may require more standardization in order to compare estimates between probabilistic approaches and comprehensive maps. In addition to systematic methodological differences, additional simulated sampling suggests that the expected accuracy of the 95% confidence interval is, in fact, dependent on the sample size. Final design conclusions from this dissertation will be recommended to the State of California for implementation in a California S&T program for aquatic resource extent.