A Predator Susceptibility Model of Juvenile Salmon Survival and a Voronoi Tessellation-based Approach for Generating Hypothetical Forest Landscapes
MetadataShow full item record
In relating juvenile salmonid size to adult survival, past observational studies have shown mixed results, while modeling has been limited to regression methods lacking in ecological mechanism. This paper develops a simple 4-parameter model relating the survival of juvenile Chinook salmon (Oncorhynchus tshawytscha) to their size based on susceptibility to predation. The model is parameterized to test the hypothesis that there is a critical size for smolts beyond which they experience significantly lower mortality due to a reduction in gape-limited predatory pressure. The model is validated on spring-run Chinook salmon tagged in the Columbia River Basin, tracking either in-river survival during out-migration from hatcheries to Lower Granite Dam or adult returns for fish tagged at Lower Granite Dam and at the mouth of the Grande Ronde River returning to Bonneville Dam. The predator susceptibility model outperforms a generalized linear model for adult return data and for in-river survival when mortality is high, while providing interpretable parameters and a mechanistic structure. These results show strong support for the critical-size critical-period hypothesis of Beamish & Mahnken (2001) while giving a plausible functional form for the response. There is also evidence that larger fish realize greater survival increases from barging than smaller fish. This provides a framework through which interannual variations in size-dependent survival can be separated from overall survival, which could help guide management targets in both hatchery and dam management as well as wild habitat restoration. Optimization models used for forest management are highly complex and the demand for data to test them far exceeds the supply of real data sets, thus hypothetical test forest landscapes are often used. However, current hypothetical landscape generators are limited in their ability to match the characteristics of real forests and offer little or no control over important landscape metrics such as average vertex degree. This paper identifies four characteristic landscape statistics that could affect the efficiency of spatially explicit optimization algorithms and can be compared between simulated and real forest landscapes. Using these as guidelines, we describe a method for generating hypothetical landscapes capable of generating landscapes of specified characterization. A generated landscape is based on a Voronoi tessellation, created from points chosen by a mixture of random point processes, which is then edited to include gaps and non-convex polygons, overcoming the shortcomings of Voronoi tessellations in this application. Through a series of multiple regressions the algorithm determines appropriate control parameters so that the output landscape will have high probability of matching targeted characteristic statistics within a tolerance. The method produces landscapes with a wide range statistics, covering the characteristics of real forests and extending into extreme cases unlikely to be encountered in real data, while providing greater flexibility and control over the generated landscapes than previous methods for generating hypothetical forests.