Guidelines for Data Use
The purpose of this document is to highlight specific qualities and limitations of the data presented on this website in the interest of preventing errors in the application of the data. The GRTS-based sampling designs employed to collect data on spawner abundance of adult and juvenile salmonids and their habitat in Oregon provide an excellent foundation to answer the questions posed on this site. However, whenever extrapolations are made from a sample, attention must be paid to the assumptions inherent in the calculations and the confidence of the result.
The Oregon Coastal Coho and Lower Columbia Coho sampling designs were intended to make estimates at three different spatial extents: ESU, Strata and Independent Populations. For spawner abundance, ODFW aims to sample a minimum of 30 sites or 30% of the miles of spawning habitat (whichever is smaller) in each independent population. Because of the spatially-balanced nature of this sample it may be possible to make estimates for subsets of larger populations provided that the sample size is large enough to ensure reasonable confidence in the result. If the area of interest is too small and a 30% sample results in too few sample sites then a census would be needed. Samples for juvenile fish and fish habitat are much less dense and only allow for estimates at the Strata and ESU scales. A single sample point should never be used to characterize a stream. The confidence interval of the estimate should always be presented along with the estimate.
If there is interest in making estimates of the response by land use or land ownership, the same restrictions on sample size should be considered. Making estimates in this way will also compromise the spatial balance of the sample which will make the confidence intervals wider. Researchers should be careful not to argue causality based on this correlation since many attributes important to salmon production co-vary with land ownership and management.
Care should also be taken in calculating sample weights for making expansions. The weight multiplied by the response at each sample point should be the original weight (the inverse of the selection probability) multiplied by the ratio of the number of sites successfully surveyed over the total number of sites selected (target:response / (target:response + target:non-response). Data on the number of sites selected, whether the sample point was target or non-target, and whether we were able to obtain a response (successfully survey a site) can be found in the Site Category table available via download. The sample weight can be calculated by dividing the number of miles of habitat specific to the life stage in the region of interest by the number of sites selected. Sites selected between 1998 and 2006 should reference the stream coverages used in the 1998 site selection. Sites sampled in 2007 and after should reference the GIS coverages used in the 2007 site selection. For those interested in conducting this type of analysis please use the “Contact Us” link on the website to request the required stream coverages. See Stevens (2002) and Jacobs et al. (2001) for more details about how spawner abundance estimates and trends are calculated.
The sampling designs employed for monitoring of Oregon Coastal Coho and Lower Columbia Coho employ rotating panels of sample sites that bolster sensitivity for trend detection. In any given year, 25% of the sites drawn will be annual samples, 25% will be sampled on a three-year interval, 25% will be sampled on a nine-year interval, and 25% will be unique to that year (see Firman and Jacobs (2001)) for more details). We can make use of the multi-phase aspect of the sample to utilize data from the other panels. From our sample, we can calculate the 1-year slope only for the sites that are observed every year, but we can calculate the 3-year slope for both annual and 3-year sites. The idea behind multi-phase regression is that we use the 3-year slope to predict the 1-year slope for the 3-year sites. Essentially, the technique allows us to fill in trend data for sites that were not observed every year. If the correlation between the 1-year and 3-year slopes is high, then we can expect a substantial increase in precision, since we will have effectively quadrupled our sample size. See Stevens (2002) for more details.
Overlap Between Sampling Projects
The sampling design employed to monitor Oregon Coastal Coho and Lower Columbia Coho forces overlap of sites sampled for adult spawners, juvenile fish, and fish habitat. However, due to the different sampling extents and densities for different projects, not every site sampled for one project is sampled for all of the others. Researchers should also keep in mind that surveys for juveniles and habitat are generally 1000 meters long (occasionally landowner denials or barriers will shorten a survey to 500 meters), whereas surveys for adult spawners tend to be about a mile long, so a survey for juveniles or habitat at the same site as a spawning survey will rarely encompass the entire spawning survey. Also, habitat surveys are centered on the sample point, whereas spawning and rearing surveys encompass the point but are not usually centered on the point. Even if there was complete overlap between the areas sampled, juvenile fish may rear outside of the reach in which they were spawned so 1:1 relationships should be approached with caution.
Surveyors tend to see 0.75 of the adult coho that are present in a stream segment (Solazzi 1984). Spawner estimates should be expanded to adjust for undetected adults. Estimates for Populations, Strata and ESUs presented on this website have been adjusted to reflect the probability of detection.
Non-Response and Data Revisions
Because the AUC calculation used to estimate the number of adult coho spawning in a stream segment requires visual counts at regular intervals, weather conditions that decrease visibility can greatly increase the proportion of sites at which we are not able to obtain a response. Methods to determine which sites have valid surveys that can be used for expansions have varied over time. When sample sizes were smaller, sites were evaluated on a case-by-case basis to determine if any gaps in visits occurred during the period when most fish spawned in that stream or nearby stream. In more recent years large sample sizes have made this approach too time intensive and determination of whether a survey is valid or not has been automated and the criteria are thus much more rigid. Conditions for surveying were particularly bad in 2006 and a large proportion of the sites were excluded from analysis. It is likely that data from 2006 and other years with poor survey conditions will be revisited and estimates may be revised at that time.
Juvenile Fish Data
It is not possible to make estimates of the total number of juveniles present using the juvenile data presented on this site. The protocol employed for these surveys only samples fish in pools >40 cm deep and of 6m2 surface area. The relationship between the density of fish in pools and the density of fish in other habitat unit types (riffles, glides etc.) varies with the seeding level and is not linear. When few adults spawn in a stream (low seeding) almost all juveniles reside in pools. When the stream approaches full seeding more juveniles spill over into less attractive habitats like riffles and glides. This is one reason why the juvenile data do not appear among the six measurable criteria used to describe the population category. The juvenile data can be used as presence absence data, and are useful in describing trends in the distribution and density of juvenile fish. Although data on several species of fish are collected during sampling for juvenile fish, the sampling design for the coast is specific to coho for all years except 2002 to 2008, and extrapolations should not be made for species other than coho outside of those years. Sampling was done on the full distribution of coho and steelhead in the Lower Columbia and the Rogue, so all species sampled except cutthroat can be extrapolated in those regions. In these instances, the sample weight will differ depending on the species of interest.
- 1. All sites have an equal probability of selection for sampling.
- The site selection technique forces an equiprobable selection that is uniformly distributed (Stevens 2002).
- 2. Selected sites provide an unbiased sample of habitat.
- This assumption implies that our site selection methods provide a representative sample of habitat and fish. A random selection will generate an unbiased sample if: (a) our database of habitat is representative of the available habitat, and (b) no differences exist between the quality of habitat between accessible and inaccessible sites.
- 3. We are accurate in assuming zero use for sites judged to be devoid of habitat
- Sites are assumed to be devoid of spawning habitat if there is no spawning gravel present within the survey, if the survey is located upstream of an impassable barrier, or if the survey is tidally influenced. Based on the results of surveys on verification sites (Jacobs and Cooney 1992), we are fairly confident of our ability to make correct assumptions of zero spawning density using the criteria listed in Jacobs and Cooney (1990).
- 4. AUC methodology provides an unbiased estimate of the spawning density of coho salmon in spawning surveys.
- The assumptions implicit in the AUC methodology are discussed in detail in Ganio et al. (1986).
- We believe that this is the best method of determining spawning density estimates in Oregon coastal streams. Solazzi (1984) demonstrated that surveyors tend to underestimate the number of spawners present by a factor of 1.75. We use the equations generated by Solazzi (1984) to adjust spawner estimates. Recent work comparing different estimation methods (Jacobs, 2002) provides further evidence in support of the accuracy of AUC-based estimates.
- 5. Spawning density estimates should be adjusted to exclude naturally spawning hatchery fish.
- Hatchery fish that stray from the point of release or that are released directly into natural spawning areas should not be included in estimates of population spawner abundance. The proportion of hatchery fish in spawner counts is estimated for each major basin or subbasin, and the counts from that area are adjusted accordingly.
Firman, J.C., and S.E. Jacobs. 2001. A survey design for integrated monitoring of salmonids. In Nishida T. and C.E. Hollingworth. Proc. of First. Int. Symp. On GIS in Fishery Science, Saitama, Japan.
Ganio, L.M., L.D. Calvin, and C.B. Pereira. 1986. Estimating coho salmon escapement in Oregon streams. Final Report of Oregon State University, Department of Statistics, to the Oregon Department of Fish and Wildlife, Portland.
Jacobs, S.E. 2002. Calibration of estimates of coho spawner abundance in the Smith River Basin, 2001. Monitoring program report number OPSW-ODFW-06, Oregon Department of Fish and Wildlife, Corvallis, OR.
Jacobs, S.E. and C.X. Cooney. 1990. Improvement of methods used to estimate the spawning escapement of Oregon Coastal Natural coho salmon. Oregon Department of Fish and Wildlife, Fish Research Project F-145-R-1, Annual Progress Report, Portland.
Jacobs, S.E. and C.X. Cooney. 1992. Improvement of methods used to estimate the spawning escapement of Oregon Coastal Natural coho salmon. Oregon Department of Fish and Wildlife, Fish Research Project F-145-R-1, Annual Progress Report, Portland.
Jacobs, S., J. Firman and G. Susac, 2001. Status of Oregon coastal stocks of anadromous salmonids. 1999-2000. Monitoring Program Report Number OPSW-ODFW-2001-3, Oregon Department of Fish and Wildlife, Portland, Oregon.
Stevens, D.L. 2002. Sampling design and statistical analysis methods for integrated biological and physical monitoring of Oregon streams. OPSW-ODFW-2002-07, Oregon Department of Fish and Wildlife, Portland, Oregon