OSRI Final Technical Report
Forage fish and near-shore habitat associations
PI Rob J. Bochenek
Axiom Consulting & Design, LLC
Anchorage, AK 99504
Co-PI Ryan Nielson
Western EcoSystemsTechnology
Laramie, WY 82070
OSRI Contract Number: 08-10-18
Contract Period July 2008 – December 2009
Abstract:
Pacific herring (Clupeapallasi)has been identified as a critical species for North Pacific marine ecosystemsand historically has been a commercially important species. In order to improve understanding of this key species,datasets which describe herring spawn activities were integrated with known habitat information and analyzed to determine which habitat has been critical for herring survival and recruitment. Investigatorscoupled herring spawn observation data to Environmental Senstivity Index (ESI), Bioband, and localized ocean depthinformation available for Kamishak Bay, Alaska tolook for relationships between habitat characteristicsand observed herring spawn events.A standard use-availability resource selection analysis (Manly et al. 2002) was used to identify habitats typically selected by herring for spawning. The final resource selection function (RSF) indicated herring have selected shoreline segments for spawning that have shallow ocean depths,and that herring prefer to spawn along shoreline that have certain ESI and Bioband(Shorezone) habitat classifications.The final RSF was used to predict the relative probability of habitat being selected by herring for spawning based upon the shoreline ocean depth and habitat characteristics. The resulting RSF allowed investigators to produce a series of spatial datasets and maps that can be used by resource managers and scientists to better understand what factors describe ideal herring spawning habitat and plan for future management/policy decisions in Kamishak Bay.Source data and predictive maps produced from this project have been integrated into the emerging Prince William Sound Science Center - Headwaters to Ocean (H2O) framework. H2O allows visual and granular access to source data and analytical products through the project web site(
Progress Summary:
The following details the progress and impediments encountered by the study team in regard to the project objectives.
Objective 1.Process, standardize and integrate herring and Shorezonehabitat datasets for Prince William Sound and Kamishak Bay into a coherent geospatial framework.
This objective was completely met by the study team. This involved processing Shorezone datasets and metadata for both Prince William Sound (PWS) and Kamishak Bay. Most herring data for PWS had already been processed (work performed in 2007 for Exxon Valdez Oil Spill Trustee Council [EVOSTC] project 070822 and in 2008 for EVOSTC project 08082).Only Kamishak herring data needed to be processed and integrated with other local habitat data. In addition, localized ocean depth characteristics were worked into the analysis by including the NOAA 50-1000 meter resolution multi-beam vector ocean depth dataset available for the Cook Inlet region.
The Shorezone dataset is a spatially explicit classification of nearshore coastal habitat characteristics. The dataset is composed of a series of ESRI Shapefiles, aMicrosoft Access database, and digital photo and video files. The resource is highly descriptive and details specific classifications of biological zones in addition to geological characteristics along the shoreline. The Access database is in a normalized form and contains explicit FGDC metadata. The dataset delineates biological sections of the Shorezone into the following theme areas: Splash Zone, Marsh Grass + Herbs + Sedges, Dune Grasses, Sedges, Barnacles, Rockweed, Green Algae, Bleached, Red Algae, Blue Mussels, Red Algae, Kelp, Soft Brown Kelps, Dark Brown Kelps, Eelgrass, Dragon Kelp, Giant Kelp and Bull Kelp. Additionally, the dataset delineates the following geological characteristics of the shore zone: Dominant Morphology, Sediment Type, Wave Exposure, Oil Residence Index and Shore Modifications. More information regarding these classifications and the methodology used is available in the Alaska ShorezoneProtocol (
The Shorezone data was much more complex than originally assessed and required extensive data manipulation in addition to some involved spatial analysis to prepare the data for statistical analysis. The Shorezonedataset unit segments were highly variable in terms of their length and topological characteristics, which presented problems in the development of the resource selection function (RSF). Single Shorezoneunits could be composed of one continuous shoreline segment or an amalgamation of multiple discontinuous shorelines and islands. Additionally, the datasetwas further complicated by the lack of uniform segment length. In order to proceed with the statistical analysis portion of the project, the study team developed a data normalization process to standardize the lengths of the shoreline segments while keeping the Shorezone habitat characteristics intact. This process entailed breaking each complex geometric instance (single Shorezone units composed of multiple ‘linestrings’) into individual simple geometries. Once the Shorezone spatial data was topologically consistent the resulting segments were further processed to create a relatively consistent sampling length with a target of 50 meters. This second step in the normalization process involved fractioning Shorezone units into subunits while propagating the original Shorezone characteristics of theoriginal unit. The resulting normalized spatial dataset, though much larger in terms of number of features than the original, provided a spatially consistent dataset for the subsequent spatial/statistical analysis.
All datasets were integrated into a spatially enabled database to produce secondary analytical datasets to be used in the statistical analysis. These secondary datasets delineated the occurrence of spawn events within a 200 meter buffer, extending perpendicular from the shoreline and surrounding the normalized segment, and the average ocean depth within a 200 meter buffer of the normalized segment. The spatial processing and analysis are further detailed in Appendix I.
Objective 2.Evaluate Shorezonehabitat factors contributing to herring recruitment success through the analysis of spatially and temporally explicit herring spawn, age, sex, length, weight, and biomass datasets coupled with Shorezonehabitat features.
This objective was partially met by the study. This objective was impeded by issues of data inconsistency between Kamishak Bay and PWS and limitations of the underlying data. The study team refined their analysis to focus on habitat selection by spawning herring as predicted by Shorezonetype and ocean depth data in KamishakBay. The KamishakBay data was selected as the best candidate for the statistical analysis due to the consistent and well documented sampling efforts of the aerial surveys. The PWS data was determined to be inadequate for robust statistical analysis due the inconsistent and often undocumented sampling effortsduring the aerial surveys. Figure 1 provides a GIS visualization of historical aerial survey effort in PWS and Kamishak Bay. Investigators postulated that a resource selection model developed using Kamishak Bay bathymetric profiles and Shorezone characteristics could also be applied to the PWS area. This proved problematic due to the discrepancies between regions in terms of their different Shorezone habitat compositions and localized ocean depth characteristics.
Figure 1. Comparison of aerial survey effort (yellow lines) between Kamishak Bay (left photo) and PWS (right photo). The Kamishak Bay survey effort is uniform and consistent through time making it an ideal candidate for the subsequent statistical analysis. The PWS survey effort is highly irregular, inconsistent through time, and poorly documented.
Due to poor source data and the unforeseen complexity of the Shorezone dataset, the study team was forced to focus on habitat characteristics (Shorezone and ocean depth) and herring spawn selectivity of these habitats for the development of the RSF. The analysis undertaken during this project did not take into account subsequent herring recruitment or Age Sex Length (ASL) datasets. After consulting with experts in the herring biological community and reviewing the literature it was concluded that herring recruitment is primarily a function of physical ocean conditions related to survival, forage abundance, physical ocean conditions, and spawning biomass (Stokesbury et al. 2002, Norcross et al. 2001, Williams and Quinn 2000, Zheng 1995).
Objective 3. Identify Shorezonehabitat types and locations which are critical for herring reproductive and recruitment success.
This objective was fully met by the study team. It was determined that habitat selection was inversely proportional to localized ocean depth when all other habitat characteristics (Shorezone) were held constant. Additionally, it was determined that the ESI classification of sheltered rocky shores/scarps and wave cut platforms were consistently selected over other Shorezone types in KamishakBay. Furthermore, the bleached red algaeBiobandwas highly selected for spawn by herring in Kamishak Bay. The resulting RSF, which was developed to model and predict habitat selection, was applied to the Kamishak region to develop a spatially explicit spawn prediction GIS data layer for Kamishak Bay. The resulting mapis shown in Figure 2.
Figure 2. Spawn prediction map generated from output of resource selectivity function. Red segments are predicted to be spawned upon less frequently than blue and violet segments. Predictions represent relative probability of selection and were scaled so the maximum prediction was equal to 1000.
The normalized Shorezone dataset for KamishakBay was then inspected to determine which Bioband and ESI characteristics were viable for the statistical analysis based upon their frequency of occurrence. Resource selection functions are subject to estimation problems when considering rare or outlier characteristics. Bioband data (which is classified as patchy, continuous or absent on the supratidal, subtidal and intratidal plane) was further simplified by ignoring supratidal characteristics and lumping intertidal and subtidal into Bioband presence or absence along each Shorezone segment. For example, if soft brown kelp were recorded as patchy or continuous in the subtidal or intertidal zones we classified soft brown kelp as present along that shoreline segment. Some ESI classifications were combined based on their rarity and sharing of similar characteristics, and others that were absent or rare in Kamishak Bay were dropped from the analysis.
An RSF was then developed from the underlying data to model selectivity across known habitat variables in Kamishak Bay. The specifics of the model development are detailed in Appendix 2: Statistical Analysis.
The final RSF indicated positive relationships between observed herring spawning habitat selection and shorelines characterized as having wave-cut platforms, sand and gravel beaches, sheltered rocky shores, bleached red algae, alaria, green algae, and barnacles (Table 1). According to the final RSF, there were negative relationships between spawning habitat selection and marshes, rocky shores, rockweed, and water depth (Table 1).
Table 1.Estimates of model coefficients, standard errors (SE), selection ratios, and 95% confidence intervals for selection ratios based on the final RSF.A selection ratio represents the proportional increase in relative probability of selection when a Shorezone type is present, or when ocean depth is increased by 1 meter. A selection ratio for the intercept term (ESI = tidal flats was reference level) was not calculated because of difficulty in interpretation.
Selection / 95% CIVariable / Estimate / SE / Ratio / LL / UL
Intercept / -5.581 / 0.262 / NA / NA / NA
ESI: Rocky shores / -0.176 / 0.114 / 0.838 / 0.670 / 1.049
ESI: Wave-cut platforms / 1.014 / 0.044 / 2.756 / 2.529 / 3.003
ESI: Sand/gravel beaches / 0.262 / 0.057 / 1.299 / 1.162 / 1.453
ESI: Sheltered rocky shores / 1.156 / 0.150 / 3.177 / 2.367 / 4.264
ESI: Salt- and brackish-water marshes / -1.424 / 0.294 / 0.241 / 0.135 / 0.428
Bioband: Bleached red algae / 1.558 / 0.038 / 4.747 / 4.405 / 5.116
Bioband: Rockweed / -1.563 / 0.067 / 0.210 / 0.184 / 0.239
Bioband: Alaria / 0.655 / 0.053 / 1.925 / 1.737 / 2.134
Bioband: Green algae / 0.218 / 0.048 / 1.243 / 1.132 / 1.365
Bioband: Barnacle / 0.187 / 0.033 / 1.205 / 1.129 / 1.287
Average water depth (m) within
200 m from shore / -0.003 / 0.001 / 0.997 / 0.995 / 0.998
Figure 3 shows the estimated relative probability of selection based as a function of water depth (m), with all other variables held constant. Figures 4 and 5 compare the estimated relative probability of selection as a function of various ESI and Bioband classifications, holding all other variables constant. The relative probability of selection is inversely propositional to localized ocean depth.
Figure 3. Estimated relative probability of spawning herring selecting shoreline segments of various depths, given all other shoreline characteristics are held constant. The relative probabilities of selection were scaled so the maximum was equal 1.0.
Figure 4. Selection ratios and 95% confidence intervals (CI) for Environmental Sensitivity Index (ESI) classifications in the RSF for spawning herring.A selection ratio represents the proportional increase in the relative probability of selection by spawning herring given a Shorezone type was present. Since the ESI classification of tidal flats was the reference level (model intercept term), selection ratios for other ESI values represent the proportional increase or decrease in relative probability of selection compared an ESI classification of tidal flats. A 95% CI that encompasses 1.0 indicates selection was not considered statistically significant (equivalent to a significance level of α = 0.05).
Figure 5. Selection ratios and 90% confidence intervals (CI) for Bioband classifications in the final resource selection model for spawning herring. A selection ratio represents the proportional increase in the relative probability of selection by spawning herring given a Shorezone type was present. A 95% CI that encompasses 1.0 indicates selection was not considered statistically significant (equivalent to a significance level of α = 0.05).
Accomplishments:
Complicated Shorezone data was acquired, synthesized, and integrated with other ecological data (herring spawn observations and ocean depth) to develop a resource selection function to predict relative spawn probability in Kamishak Bay. Additionally, the study team has developed a standardized methodology for normalizing and preppingShorezone datasets so that they may be utilized in the development of other RSFs in addition to other spatial analysis efforts in other regions and ecological contexts. All datasets utilized and produced from this effort have been made available for public access in both visual and raw formats via the Prince William Sound Headwaters to Ocean Program in addition to the project website ( Figure 6 provides a screenshot of the OSRI Herring Habitat web application which provides access to underlying datasets utilized and created during this project.
Figure 6. Project website interface to datasets produced and utilized during this project.
This investigation has identified relationships between ESI and Bioband classifications and habitat selection by spawning herring in Kamishak Bay. These relationships can be used as a basis for future analyses in other regions. Provided assumptions about data quality are met, the methods developed for this analysis can also be applied to data from other regions.
Conclusions:
This project explored methods for the development of resource selectivity models which predict what habitat will be selected for spawn by herring based upon Shorezone characteristics and localized ocean depth profiles in Kamishak Bay. The analytical methods produced from this projects efforts can be applied to other regions where Herring are known or believed to spawn. Furthermore, the resource selectivity function could be made more robust with the incorporation of additional habitat and physical ocean condition parameters.
Data Citations
Alaska Department of Fish and Game, Division of Commercial Fisheries. 2002.Kamishak Bay Data Synthesis, Version 1.0, 13 December 2002.CD-ROM. Homer, Alaska.
Coastal & Ocean Resources Inc. 2009.Cook Inlet and PWS ShoreZone Data, May 2009.DVD.Sidney, BC.
NOAA-NMFS. Steve Lewis. 2008.Alaska Multibeam Vector Bathymetry (50-1000M), September 2008.DVD. Juneau, Alaska.