MAMU Collision Risk Assessment for 2018 & 2019

Marbled Murrelet Collision Risk Assessment Associated with the Humboldt Wind Project Proposed for Humboldt County, California: 2-Year Report

Project #3980

Prepared for:

Humboldt Wind, LLC 11455 El Camino Real, Suite 160 San Diego, CA 92130

Prepared by:

H. T. Harvey & Associates

Richard Golightly, Ph.D., Senior Associate Ecologist Stephanie Schneider, M.S., Wildlife Ecologist Scott B. Terrill, Ph.D., Principal

September 2019

983 University Avenue, Building D  Los Gatos, CA 95032  Ph: 408.458.3200  F: 408.458.3210

Executive Summary

The marbled murrelet (Brachyramphus marmoratus) is a federally threatened and state endangered seabird that flies between the ocean and inland sites used for nesting. In southern Humboldt County, development of a wind energy facility has been proposed by Humboldt Wind, LLC., which would require construction of a line (or “string”) of turbines along the top of Bear River Ridge and Monument Ridge. There is potential for murrelets to encounter the wind turbines associated with this Project if they cross these ridges as they transit to and from nesting habitat. Marbled murrelet activity at the site of this Project was assessed using radar surveys conducted during the 2018 and 2019 murrelet breeding seasons (April through September). We used ridge passage rates determined by radar to develop two Collision Risk Models (CRMs) that assess potential collision risk for marbled murrelets: a deterministic model that provided a single conservative estimate of risk each year and a probabilistic model that quantified the probability of all possible estimates of risk which facilitated expression of uncertainty around the most likely estimate and produced an upper limit of potential risk each year.

For a CRM to accurately estimate collision risk, it is essential to quantitatively assess variability in passage rate across time and space. For murrelets, inland flights are non-random and vary in predictable ways seasonally: with the majority of flights occurring during the breeding season, and topographically, with primary flyways being valleys rather than ridgelines. To appropriately capture this variability in inland activity, radar-based observations of murrelet flights were made across the length of the turbine string for the duration of the breeding season at least once each month. These radar surveys provided an accurate estimate of passage rates (birds/day/km) for the breeding period at monthly intervals. Passage rates were calculated for each month from the radar surveys. Calculating passage during the non-breeding period (October-March) was complicated because observations were extremely rare and required some extrapolation between neighboring months. Spatially, the turbine strings were divided into 11 zones based on large gaps (>800 m) between adjacent turbines and topographical differences (peaks, low points, and elevational drops into the adjacent river valleys).

Both CRMs used passage rate, murrelet size, flight speed, and turbine characteristics to calculate the probability that individuals crossing the turbine string would pass through a rotor swept area of a turbine and, upon passage, collide with a rotor blade. To the maximum extent possible, input data was based on empirically derived values to ensure that estimates reported here were as realistic as possible. Flight speed, avoidance, and interannual variability in passage rate were each identified as having a range of possible values. For the deterministic model we assumed an average flight speed, a conservative value of avoidance (the probability that a bird will detect and avoid a turbine), and the best-fit relationship between inland activity and a reliable predictor of inland activity (ocean productivity). For the probabilistic model, we did not make these assumptions; rather, all possible flight speeds, avoidance rates, and interannual variability in passage rate (as informed by oceanic conditions) were considered using a Bayesian framework. Probabilities of occurrence were calculated for each potential value of these three variables which enabled estimation of collision risk while accounting for inherent variability and uncertainty needed to produce an upper estimate of risk.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates i Humboldt Wind Project September 2019

Because turbine types have not yet been finalized by the Project, we assumed a worst-case scenario in terms of specifying the maximum turbine size and rotational speed in the CRMs. These assumptions were unrealistic and resulted in an overestimation of collision risk.

In this maximum turbine size and speed scenario, the deterministic model estimated that 0.175 birds per year in 2018 and 0.246 birds in 2019 would collide with turbines associated with the proposed facility; if extrapolated over the 30-year life of the Project, there would be potential for 5.26 and 7.38 murrelet collisions (approximately 1 individual every 3 to 4 years) based on 2018 and 2019 data, respectively. When considering long-term (30- year) variability in ocean conditions associated with inland activity of murrelets, the number of collisions estimated over a 30-year period was 5.27 or 6.85 birds based on 2018 or 2019 data, respectively. Considering the year-specific outcomes of the deterministic CRM, this would be an average of 6.1 birds during the 30 years life of the Project.

Like other deterministic CRMs, the model presented here was sensitive to change in avoidance probability. The deterministic CRM was also sensitive to changes in the annual passage rate of murrelets as well as operational factors specific to the turbines (e.g., blade speed) that were assumed to be maximal for estimation of collision risk. A benefit of the deterministic model was that differences in passage rates could be assessed along the length of the turbine strings. In 2018 the lowest estimate of passage rate among the 11 zones was 0.00 birds/km/day and the greatest estimate was 0.683 birds/day/km (푥̅ = 0.2136 birds/day/km for the entire length of the turbine string). In 2019 the lowest estimate of passage rate among the 11 zones was 0.078 birds/km/day and the greatest estimate was 0.973 birds/day/km (푥̅ = 0.2871 birds/day/km for the entire length of the turbine string). Generally, passage rate was greatest for zones that descended into adjacent river valleys, which is logical given that low-elevation valleys serve as primary travel corridors for inland flying murrelets. The deterministic CRM in 2018 was used to identify the three highest risk zones and all turbines were removed from those zones; this resulted in a 36% decrease in overall risk from the original 60-turbine Project layout. The updated 47-turbine layout was used for all results presented in this Collision Risk Assessment Report.

The probabilistic model estimated an upper bound of 6.17 collisions over the 30-year life of the Project based on data collected in 2018 and 7.77 collisions over the 30-year life of the Project based on data collected in 2019; this upper bound represents the one-tailed threshold that contains 95% of the probability distribution. Thus, the maximum take given uncertainty in the measurements is estimated to be 7.77 birds over 30 years, with only a 5% chance of being greater. Like the deterministic CRM, these estimates represent a worst-case scenario in that the calculations assume operational factors specific to turbines (e.g., turbine size and blade speed) were maximal. Estimates produced by the probabilistic CRM corroborate estimates produced by the deterministic CRM. The benefit of a probabilistic approach was that it allowed for accounting of uncertainty and variability associated with model inputs, particularly those that the deterministic model was most sensitive to (specifically avoidance probability, interannual variability in passage rate, and flight speed). Although flight speeds were well documented, this input was relatively variable and accounting for all potential flight speeds provided a more realistic assessment of the range of risk rather than basing the assessment on the average flight speed. Unlike

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates ii Humboldt Wind Project September 2019

the deterministic CRM, collision risk was not able to be estimated by zone in the probabilistic model. Since both CRMs ultimately produced similar estimates of risk, we also assumed the deterministic model provided a reliable assessment of zone-specific risk. The results presented in this report are consistent with earlier reports of collision risk and consistent with conclusions of significance and associated mitigation measures provided within the Draft EIR for the Project.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates iii Humboldt Wind Project September 2019

Table of Contents

Executive Summary ...... i Tables ...... vi Figures ...... viii List of Preparers ...... ix Section 1.0 Introduction ...... 1 1.1 Objectives ...... 4 Section 2.0 Methods ...... 5 2.1 Overview ...... 5 2.1.1 Project description ...... 5 2.1.2 Collision risk approach ...... 5 2.2 Collision risk assessment ...... 12 2.2.1 Radar data ...... 12 2.2.2 Sampling area ...... 13 2.2.3 Passage rate...... 14 2.2.4 Collision probability ...... 16 2.2.5 Avoidance probability ...... 16 2.2.6 Interannual variability ...... 20 2.2.7 Analysis scenarios ...... 21 2.3 Sensitivity analysis for the deterministic model ...... 22 Section 3.0 Results ...... 23 3.1 Overview ...... 23 3.2 Collision risk assessment ...... 24 3.2.1 Deterministic model...... 24 3.2.2 Probabilistic model ...... 30 Section 4.0 Discussion ...... 33 4.1 Passage rates ...... 35 4.1.1 Sources of error ...... 36 4.1.2 Seasonal variability in passage rate ...... 37 4.1.3 Daily variability in passage rate ...... 37 4.1.4 Topographical variability ...... 38 4.2 Collision rates ...... 38 4.3 Avoidance rates ...... 39 4.4 Interannual variability in inland activity ...... 39 4.5 Potential indirect effects ...... 40 4.6 Conclusions ...... 41 Section 5.0 Acknowledgments ...... 43 Section 6.0 References ...... 44 Appendix A. Avian avoidance of wind turbines ...... A-1 Appendix B. Relating inland activity of murrelets to ocean conditions ...... B-1 Appendix C. Bayesian likelihoods, priors, and posteriors...... C-1 Appendix D. Limited evidence of avian fatalities caused by rotor turbulence at wind energy facilities .. D-1 Appendix E. Comparison of collision risk estimated for oblique approach angles relative to a simplified method for marbled murrelets ...... E-1

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates iv Humboldt Wind Project September 2019

Appendix F. The influence of fog on navigation and ridge-crossing by murrelets ...... F-1

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates v Humboldt Wind Project September 2019

Tables

Table 1. Type and description of each input used to calculate collision risk in the Probabilistic Collision Risk Model. The input characteristics match that of the Deterministic Collision Risk Model with the difference being that inputs that exhibit variability and were described here as probabilistic were not assigned a singular value as was the case for a deterministic approach; rather a probabilistic range of values for these inputs were derived using a Bayesian framework...... 10 Table 2. Monthly and annual passage rates (birds/km/day) of marbled murrelets through the proposed Humboldt Wind Energy Project by zone (A-K; Figure 2) and across the entire wind energy facility in 2018 and 2019 including Zones C,G, and K...... 26 Table 3. Monthly and annual collision risk estimates for marbled murrelets by zone (A-K; Figure 2) and across the entire proposed Humboldt Wind Energy Project in 2018 and 2019. The Deterministic Collision Risk Model assumed a layout of 47 large turbines (turbine height = 180 m, blade radius = 75 m, blade width = 4.2 m, and a maximal rotation speed = 17 rotations per minute)...... 26 Table 4. Summary of values for each parameter associated with the Deterministic Collision Risk Model inputs (A, B, D, G-J, N-W, AA, AB, AF, AH, AJ), intermediate calculations (C, E, F, K, L, M, X, Y, Z, AC, AD, AE, ), and outputs (AI, AK, AL, AM)...... 28 Table 5. Collision estimates for the entire range of potential avoidance rates, with avoidance of 98% being the most conservative estimate supported for the literature for birds similar to the murrelet and values lower than 98% being the most conservative estimates supported by the literature for all birds studied to date (see Appendix A)...... 29 Table 6. Sensitivity analysis of the Collision Risk Model for 47 large turbines installed as part of the Humboldt Wind Energy Project based on 2019 data only. Changes in input parameter characterized changes in collision estimates expressed as a percent of the unmodified collision estimate as well as the number of birds at 30-year intervals. The relationship between the change in the input and the change in the output was linear for all variables except blade width, flight speed, and blade speed. For blade width, the average change in output across 10% increments was used. For flight speed, the relationship was exponential and increased rapidly for flight speeds <7 m/s but the change in output was <1 for speeds greater than 12 and <0.2 for speeds greater than 22 m/s. For blade speed, the relationship was approximately linear for speeds between 6 and 17 rotations per minute (the maximum rotation speed for turbines being considered for use) and was equally low between 0 and 6 rotations per minute...... 30 Table A-1. The 35 estimates of avoidance that were selected from a total of 152 estimates because they met criteria detailed in text for use in a Bayesian analysis that generated a model of avoidance to be input into the Probabilistic Collision Risk Model. The types of avoidance that the estimate represents was indicted as 1 for micro-avoidance, 2 for meso-avoidance, and 3 for macro-avoidance. Behaviors were categorized as either local (L), meaning these individuals may have spent time near turbines engaging in various behaviors in addition to navigation, or migratory (M), for individuals whose sole intent was navigational and they quickly passed through areas occupied by turbines...... A-1 Table A-2. The name, location, and general year of installation for each wind energy facility from which avoidance estimates input into the Probabilistic CRM were derived. Avoidance was estimated in either Cook et al (2014), Cook et al. (2018), or Furness (2013) and was calculated by dividing the number of collisions observed by the number of collisions expected in the absence of avoidance (see Table A-1). The expected number of collisions in the absence of avoidance were calculated using the Band (2012) model and were based on the species-specific passage rate individuals through each facility. The observed number of collisions were derived from carcass searches within each facility, where the total number of carcasses observed were corrected to account for missed carcasses (e.g., searcher efficiency and carcass persistence). For each, the duration of monitoring, method used to determine passage rate, and method used to correct carcasses found in fatality searches was indicated. The ID corresponds with the ID in Table A-1...... A-3

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates vi Humboldt Wind Project September 2019

Table C-1. Likelihood functions, parameters, prior specifications (distribution and value), and posterior estimates (and their 95% highest density interval) for each parameter. Posterior estimates for each parameter represent the outcome of Bayesian inference, which mathematically combined the distribution of prior assumptions (which were generally uninformed) with the distribution of data observed for each input. These posterior estimates are then put into the associated likelihood function to generate a posterior predictive distribution (PPD) that could be input into the Probabilistic Collision Risk Model. A corresponding figure depicting the PPD for each of the variables of interest was specified in the PPD Figure column...... C-5 Table C- 2. For all parameters detailed in Table C-1, the quantiles of the posterior distribution and information relevant to posterior predictive checks have been provided, specifically the Gelman-Rubin Statistic (푹) and the effective sample size (ESS)...... C-9

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates vii Humboldt Wind Project September 2019

Figures

Figure 1. A flow diagram indicating structure of the Deterministic Collision Risk Model for the turbines proposed by the Humboldt Wind Project...... 7 Figure 2. Equation used to quantify collision risk for marbled murrelets (MAMU) transiting through the Humboldt Wind Project. The primary equation is displayed at the top. Components of the primary equation are color-coded to facilitate expansion of the primary equation for each component. Blue text summarizes the calculation of probability that a murrelet will collide with a turbine blade upon entry into a rotor swept zone (RSZ); this is the Band (2012) Model. Green text summarizes the calculation of the number of MAMU expected to pass through a single RSZ given their rate of passage through the proposed turbine string. Orange text summarizes the adjustment of the collision estimate produced by multiplying components 1 and 2 to account for micro, meso, and macro avoidance. The deterministic and probabilistic CRMs are both based on this equation. However, specific inputs that were modeled using a Bayesian framework using the probabilistic approach are in red text...... 8 Figure 3. Map of topographically delineated zones used to assess collision risk for marbled murrelets flying through the proposed Humboldt Wind Energy Project as they transit between the ocean and inland nesting habitats. This includes Zones C,G, and K which no longer have any turbines because they descended into adjacent river valleys and consistently had greater passage and collision rates in both 2018 and 2019...... 25 Figure 4. Collision risk probabilities for marbled murrelets over a 30-year period based on the Probabilistic Collision Risk Model that incorporated variation and uncertainty in murrelet passage rate, flight speed and avoidance behavior. The most likely estimate for each model run is indicated by a dashed vertical line and the 95% highest density interval is indicated by the solid red line on the horizontal axis...... 31

Figure B-1. Univariate regressions depicting the relationship between inland activity of murrelets in the Eel River Valley over a 10-year period (2002-2011; HRC 2018) for each of the variables indicative of ocean productivity considered for inclusion into a multivariate regression...... B-3 Figure B- 2. Visual representation of the correlations between ocean productivity predictor variables, with the size of dots indicating the strength of the correlation and the color representing the direction of the association...... B-4 Figure C-1. Normal quantile plots for marbled murrelet flight speed, avian avoidance of turbines, inland activity of murrelets and ocean productivity measures at 10-year and 120-year scales. Datasets for which most points fall within the 95% confidence interval (represented by the gray shading around the line of fit) were well approximated by the normal distribution...... C-2 Figure C-2. Distributional comparisons for marbled murrelet flight speed and avian avoidance data used to generate posterior predictive distributions using Bayesian inference. For flight speed and avoidance, the density distribution of the actual data (red line) and the various distributions considered to serve as a model for those underlying data were depicted to facilitate visual assessment of fit for each.C- 3 Figure C-3. Probability distributions specified for the priors of (A) the mean of flight speed [log(µ)], (B) the slope [β1] and intercept [β0] of the relationship between ocean productivity and inland activity of murrelets, and (C) the shape parameter [ν] that described the distribution of residuals about the best-fit line relating measures of ocean productivity to inland activity of murrelets. Although the priors for mean of flight speed were specified on a logarithmic scale, for intuitive purposes it is visualized here on the actual scale...... C-5 Figure C-4. Raw data, a representative subset of credible distributions or regression lines resulting from 100000 successive random selections from four Markov Chains (in blue), and the final posterior predictive distribution (in black) of (A) murrelet flight speed, (B) avoidance, (C) the relationship between ocean productivity and the inland activity of murrelets, and (D) variability in ocean productivity (specifically, the Pacific Decadal Oscillation or PDO; see Appendix B). Only a subset (n=1500) of

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates viii Humboldt Wind Project September 2019

credible distributions or regression lines were plotted as additional lines would overlay what has already been depicted...... C-7 Figure C-5. A representative example of the plots used to confirm that the Markov Chain Monte Carlo (MCMC) values were representative of the posterior distribution of a single parameter and that the effective sample size (ESS) was large enough to produce a stable and accurate estimate of the distribution...... C-8

Figure E-1. Estimated probability that a marbled murrelet will be struck by a blade of a large turbine as it flies through the rotor swept area for all approach angles possible in the 360-degree sphere around a turbine. The blue arrow indicates the direction that wind is blowing towards. For each of the three scenarios, risk was lowest when birds transited with the wind (90 to 270 degrees) and greatest when birds transited against the wind (0-90 degrees and 270-360 degrees). The red line (BAND) depicts collision probabilities generated by the Band Model (2012). The open circles (OBLIQUE RAW) represents the exact collision probability generated by the Oblique Model (Christie and Urquhart 2015) for each of the 55 marbled murrelet ridge crossings. To facilitate direct comparison of collision risk profiles for the Band and Oblique Model, the black line (OBLIQUE) was generated using empirically-derived averages for flight speed and wind speed for the 55 ridge crossings. .. E-3 Figure E-2. Circular histogram generated using a 10-degree bin that depicts the proportion of flight directions (MAMU) and wind directions (WIND) determined for the 55 marbled murrelet ridge crossings detected during radar surveys. To facilitate comparison, murrelets and wind both move towards the indicated direction...... E-4

List of Preparers

Richard T. Golightly, Ph.D. Stephanie R. Schneider, M.S. Scott. B. Terrill, Ph.D.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates ix Humboldt Wind Project September 2019

Section 1.0 Introduction

The marbled murrelet (Brachyramphus marmoratus, hereafter “murrelet”) is a small seabird found along the Pacific Coast of western North America, ranging from central California to northern Alaska (Carter and Erickson 1992, Piatt and Naslund 1995). This species was listed as Federally Threatened in 1992 (Federal Register 57 FR 45328) and State Endangered in California in 1992 (USFWS 1997). Like other members of the family Alcidae, murrelets are long-lived, with the oldest known individual being at least 21 years old (R. Golightly, personal observation), require 2 to 3 years to reach sexual maturity (Peery and Henry 2010), and do not nest every year (Peery et al. 2004a, Hébert and Golightly 2006). They lay a single egg on the branch of an old tree as far inland as 88 km from the ocean (Raphael et al. 2016). Paired birds are thought to persist through time, and pairs often reuse the same nest or nest stand for repeated years (Golightly and Schneider 2011; summary in Plissner et al. 2015). This low fecundity, which is typical of alcids, results in slow population growth with the population’s capacity for growth being most sensitive to adult survival (Sæther and Bakke 2000, Peery and Henry 2010).

Humboldt Wind, LLC has proposed to build a wind energy facility in southern Humboldt County, CA (hereafter “Project”). The Project is within the range of the murrelet and there is potential for murrelets to fly through the area where the turbines will operate. The Project has projected a maximum capacity of 155 megawatts that may be generated by up to 47 turbines, depending on turbine type (which has yet to be finalized). The turbines will be installed on Bear River Ridge and Monument Ridge, two prominent ridges that are both bordered by the Eel River to the northeast and Bear River to the southwest. Old-growth forest available to murrelets for nesting in southern Humboldt County occurs up-river and to the southeast of the Project. Murrelets travel inland along waterways and low-lying valleys and tend to restrict ridge-crossings to passes (Burger 1997, Rodway and Regehr 2000, Harper et al. 2004); thus, it is likely that murrelets primarily use the Eel River Valley and, to a lesser extent, the Bear River valley, as corridors for transiting between foraging grounds at sea and inland nesting habitat. These ridges parallel the valley floors and are 750 to 915 m above the valley, but there is still potential that some murrelets will use passes to cross these ridges during their daily transits while nesting and, on occasion, outside of the breeding season. To assess interaction between the Project and murrelets required a quantitative assessment of murrelet movements across the ridges where turbine installation has been proposed.

In contrast to most seabirds which nest on offshore rocks and islands, murrelets typically nest on large branches of old-growth and late-successional coniferous trees. Murrelet nests are infrequently found due to the difficulty of accessing the canopy of old-growth forests and the cryptic behaviors exhibited by murrelets in and around their nests (Nelson and Hamer 1995, Golightly and Schneider 2009). Murrelets do not build nest structures; rather they lay and incubate their egg on a large, flat tree branch with natural depressions or moss, lichen, and tree litter. In northern California, redwoods (Sequoia sempervirens) are the most common conifer available for nesting although they also nest in Douglas Fir (Pseudotsuga menziesii; Golightly et al. 2009).

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 1 Humboldt Wind Project September 2019

Murrelets spend much of their life at-sea and must transit between foraging areas at-sea and inland nesting habitat throughout the year, but most inland flights occur during the breeding season as murrelets incubate their egg and, upon hatching, feed young (Naslund 1993, Hébert and Golightly 2006, Sanzenbacher et al. 2014). Generally, in Oregon and northern California, egg laying begins between mid-April and June depending on ocean conditions (Hamer and Nelson 1995). Chicks hatch between late May and July. Incubation lasts for 28 to 30 days (Nelson and Hamer 1995, Hébert and Golightly 2006) with each parent taking a 24-hour shift sitting on the egg and alternating with its mate until the chick hatches (De Santo and Nelson 1995). During incubation, each nesting murrelet makes a single transit to land or sea each morning around sunrise (45 minutes prior to and 75 minutes following sunrise; Evans Mack et al. 2003). The chick is fed by the parents for another approximately 28 days until it fledges and flies to the ocean. After their chick hatches, adult murrelets continue to make this early morning transit each day to deliver a fish to their chick and they often embark on an additional flight to feed chicks in the evening around sunset (with the average evening flight occurring 23 minutes before sunset; Hébert and Golightly 2006). Outside of the breeding period, murrelets will periodically travel inland at dawn (Naslund 1993, Sanzenbacher et al. 2014) for unknown reasons that may include maintaining pair bonds, examining future nesting areas, or engaging in other social activities probably associated with breeding (Carter and Sealy 1986, Carter and Erickson 1992, Naslund 1993). Throughout the year, murrelets restrict transits between the ocean and inland nesting habitat to times of limited lighting (Naslund and O’Donnell 1995, Stumpf et al. 2011)—sunrise and sunset—to minimize discovery of eggs and chicks by predators and minimize risk of predation to the flying adult by birds of prey (Nelson and Hamer 1995, Hébert and Golightly 2006). At sunrise breeding birds may be accompanied by additional non-breeders that may be seeking nesting opportunities or socializing (Naslund 1993, Peery et al. 2004b, Hébert and Golightly 2006).

Murrelets may fly as far inland as 89 km to reach their nest (Raphael et al. 2018) and, to do so, requires that they be strong flyers. However, they must also successfully capture small fish and invertebrates underwater and to do so requires that they are adept at swimming. Their body form is a compromise that allows them to fly efficiently through both the air and water; specifically, murrelets have wings that are small relative to their mass (217 g; Hébert and Golightly 2006) and shaped for maneuverability (Nelson 1997). When transiting between foraging and nesting habitats murrelets can attain cruising speeds that average 81 km per hour and have been documented flying as fast as 154 km per hour (Burger 1997, Elliot et al. 2004).

Murrelet flights over Bear River Ridge and Monument Ridge are thought to be rare relative to flights in the river valleys and nesting habitats (Rodway and Regehr 2000). Radar is the most effective way to ensure that rare events are adequately detected for purposes of objectively assessing movements across a space; a single radar station can acquire data on bird movements as far as 1.5 km in all directions (Stantec 2018), which is unlike audio-visual surveys that have a very limited detection range (100-200 m). Unlike audio-visual surveys, radar can accurately measure flight speed, altitude, and direction which are all important for development of Collision Risk Models (CRMs). Therefore, the Project commissioned Stantec Consulting Services, Inc., to establish radar sampling stations on Bear River and Monument Ridges where turbines will be installed and begin surveys for murrelets. Radar surveys have been used to quantify passage rates of murrelets (typically over the span of two breeding seasons) upon which to base a risk assessment (Nations and Erickson 2009, Cooper and Mabee 2010,

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 2 Humboldt Wind Project September 2019

Sanzenbacher et al. 2015), including on Bear River Ridge (Sanzenbacher and Cooper 2010). However, the intensity of radar sampling for the risk assessment associated with this Project has been greater than previous radar sampling in the range of the murrelet (Stantec 2018; see also Cooper and Sanzenbacher 2006, Hamer Environmental 2009, Sanzenbacher and Cooper 2010, Sanzenbacher et al. 2015); this larger dataset created by more frequent sampling facilitated the stratification of ridge crossings both spatially (by topographically defined zones) and temporally (by month).

In the case of murrelets, the Project turbines will not directly damage or degrade murrelet nesting habitat; however, the turbines do represent a collision risk hazard for individuals that cross Bear River Ridge or Monument Ridge in the vicinity of the Project. The earliest physics-based model to estimate collision risk for birds encountering wind turbines was developed by Tucker (1996). Today, CRMs are frequently developed to predict likely collisions between sensitive species and proposed wind energy developments. Multiple approaches for modeling collision risk have been developed (Masden and Cook 2016) and used to assess risk posed by both onshore and offshore wind facilities in a variety of situations including migrating flocks of shorebirds (Gordon and Nations 2016), eagles (USFWS 2013), seabirds (Cooper and Day 2004), sea ducks (Desholm et al. 2006), terns (Everhart and Stienen 2007) and murrelets (Nations and Erickson 2009, Sanzenbacher and Cooper 2010, Sanzenbacher et al. 2015).

CRMs generally have two goals: (1) estimation of the number of birds that fly through the rotor swept area, and (2) estimation of collision probability for birds passing through a rotor swept area (Masden and Cook 2016). CRMs require the input of a considerable amount of data including site-specific information about passage rates, feeding and flight behaviors, size and speed of the bird, bird flight paths, wind conditions, characteristics of individual wind turbines, and wind turbine layout (Tucker 1996, Holmstrom et al. 2011, Band 2012, Christie and Urquhart 2015). It is essential to ensure that input parameters are, to the greatest extent possible, based on empirically-derived data and that reliance on assumptions not supported by data are limited to ensure that the CRM produces estimates that approximate reality (Holmstrom et al. 2011, Band 2012, Masden and Cook 2016).

To assess the risk posed by installing turbines on Bear River and Monument Ridges for murrelets, we developed two CRMs following the approach detailed by Band (2012). The Band model was conceived of in 1995 (Masden 2015) and developed in conjunction with the Scottish Natural Heritage (SNH 2000), an organization created by the Scottish Government in 1991 and responsible for securing the conservation and enhancement of Scotland's natural heritage (Scotland 1991). Since inception, the Band Model has undergone several revisions and refinements (SNH 2000, Band et al. 2007, Band 2012) following critical review (Chamberlin et al. 2005, Masden 2015) and international input from the Strategic Ornithological Support Service (Brittan) and the Joint Nature Conservation Committee (United Kingdom; Band 2012, JNCC 2014). The Band Model guidance has been applied extensively in Europe to assess collision risks at various onshore and offshore wind energy projects that are currently operational (Furness 2013, Cook et al. 2014), thus allowing for validation. The exact type and number of turbines that will be installed have yet to be finalized, although maximum limits have been established. Collision risk was therefore quantified under the unrealistic assumption that the largest turbine being considered will be installed at each of the 47 currently identified sites. However, installing a

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 3 Humboldt Wind Project September 2019

turbine of this size at all 47 sites would produce more energy than planned (by as much as 127%) and when decisions regarding turbine type are finalized, the actual number and/or size of turbines may be less than has been used for purposes of the risk estimates provided herein. As such, the estimates detailed here should be regarded as a worst-case turbine scenario, representing the upper bounds of what should be expected.

1.1 Objectives

The first objective of this collision risk assessment was to estimate the potential number of collisions between turbines and murrelets expected over the 30-year life of the Project. A secondary objective was to partition estimates of risk across the ridge into topographically distinct zones such that areas of high-use by murrelets could be identified to inform decisions regarding turbine placement such that collision risk is reduced.

We developed two models of risk: one was deterministic and the other was probabilistic. The deterministic model allowed for setting some sensitive parameters to be maximally conservative. It also informed variation in parameters such as spatial (within-string) and temporal (monthly) distribution of ridge crossings. The probabilistic model allowed for direct calculation of uncertainty (in the form of credible intervals), both for specific parameters and for the overall projected take over the life of the Project. The probabilistic model used a Bayesian approach (Kruschke 2015) to quantify the likelihood of all possible estimates of collision risk using data that was either empirically derived from the radar surveys at Bear River and Monument Ridges, or informed by the scientific literature. This probabilistic approach allowed for assessment of the magnitude of collisions as well as the uncertainty and the upper limit of the risk estimate (less than a 5% chance that the real collision value could be greater).

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 4 Humboldt Wind Project September 2019

Section 2.0 Methods

2.1 Overview

2.1.1 Project description

The Project identified 47 potential locations suitable for turbine installation, with the actual number of turbines that will be installed ultimately depending on size and generating capacity. Turbines sites are restricted to two adjacent ridgetops, Bear River Ridge and Monument Ridge, such that the layout is best described as a line (or “string”) of turbines in contrast to an array of turbines. Initially the Project identified 60 locations, but the number of turbines was reduced to 47 consistent with calculations informed by the 2018 passage rates that identified relatively high risk areas within the strings for murrelets. The 47-turbine layout was used in this Collision Risk Assessment Report.

2.1.2 Collision risk approach

2.1.2.1 Radar data

Murrelet flight activity was measured 1-2 times each month of the 2018 and 2019 breeding seasons (April to September, which encompasses most flights associated with breeding) using ornithological radar (Stantec 2018, 2019). These surveys encompassed the full duration of morning and evening activity periods that have been well established and described by the Inland Survey Protocol for murrelets (Evans Mack et al. 2003). Telemetry studies (Hébert and Golightly 2006, Lorenz et al. 2019) reported an absence of murrelet flights outside low- light periods associated with sunrise and sunset; however, to validate this observation along the ridges associated with the Project, mid-day radar surveys were also conducted during the breeding season of 2018 to ensure murrelet passage was comprehensively documented. Additionally, murrelet activity was measured 4 times over the winter of 2018-2019 (October, December, February, and March). Radar-based observations of murrelets flying over and in the vicinity of these two ridges were used to determine ridge crossing locations, passage rates, flight direction, flight heights, and flight speeds.

Radar-based techniques are a reliable means of observing murrelets flying over and in the vicinity of these two ridges and can be used to quantify ridge crossing locations, passage rates, flight direction, flight heights, and flight speeds. Radar-based techniques have also been the standard for assessing collision risk to murrelets (Hamer Environmental 2009, Nations and Erickson 2009, Sanzenbacher and Cooper 2010, Sanzenbacher et al. 2015) due to the accuracy of this technique in terms of reliably detecting targets within a 1.5 km range if there are no topographically induced blind spots. All radar surveys by Stantec were digitally recorded to ensure repeatability and consistency in identification of murrelet targets. Thus, sampling errors associated with radar- based detections of murrelets were small. This presumption has been justified by other radar-based studies of bird movement. For example, Cooper et al. (2001) concluded that all variability in passage rates determined by radar surveys of murrelets was due to actual variations in murrelet passage rate rather than arising from

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 5 Humboldt Wind Project September 2019

detection errors. Also, due to the accuracy of radar-based counts of avifauna, radar counts have served as the standard by which other methods of counting are typically calibrated (Mateos et al. 2010).

2.1.2.2 Deterministic model

Overall collision risk to marbled murrelets was assessed using a deterministic model (Figure 1); the equation for this model is presented in Figure 2. To the maximum extent possible, this risk assessment relied on empirical data to inform model inputs rather than relying on assumptions about passage rates and behaviors. First, we used the number of murrelets observed crossing the ridge for the extent of the turbine string (including buffer areas that extended beyond the edge of first and last turbines). Because data provided by Stantec indicated that murrelets cross these ridges in a spatially predictable and non-random fashion such that they differentially used low-points and passes relative to other topographical features, the turbine string was sub-divided into zones based on large distances between adjacent turbines (>0.8 km) and topography defined by sections of the ridge lacking a low-point or saddle, sections of the ridge including a prominent low-point or saddle, or areas that descended in elevation towards the Eel or Bear River valley. This partitioning of risk by zone provided a tool to optimize placement of turbines such that areas of high-use might be avoided and thereby minimize risk to murrelets.

To facilitate estimation of the collision risk posed to murrelets by turbines associated with the Project, annual passage of murrelets through the Project area was estimated. Passage rates during the breeding season were derived directly from radar-based observations of murrelets passing through each zone and were calculated at a monthly interval. Flight direction and speed were known for each of these detections; for many of the detections, flight height (± 30 m) was also determined using the vertical radar.

Passage in the winter months was directly measured, but not in every month. The detections were so few that those measurements had to be extrapolated and missing values imputed across the low passage months (October to February). Further, for comparative purposes, we also derived passage rates outside of the breeding season using a series of assumptions to forecast murrelet passage rates across these ridges. Other murrelet- specific CRMs have typically relied on similar assumptions to estimate inland activity of murrelets during the non-breeding period (Sanzenbacher and Cooper 2010, Sanzenbacher et al. 2015), except for the Radar Ridge project (Hamer Environmental 2009, Nations and Erickson 2009). There was little difference in effect on the overall outcome using inferenced based input or using the empirical input; we used the empirical data for model calculations.

The potential for interaction between murrelets and turbines was determined by modifying zone-specific passage rates by the proportion of airspace occupied by turbine rotors in each zone. We then estimated collision probability for a bird as it flies through a rotor swept area using the kinematic model specified in Band (2012). The Band model estimated a collision probability for a single passage through a turbine rotor by calculating the probability that one of the turbine blades will occupy the same space as a bird during the time that it takes the bird to transit through the rotor swept area based on bird speed, size, and various turbine characteristics.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 6 Humboldt Wind Project September 2019

Figure 1. A flow diagram indicating structure of the Deterministic Collision Risk Model for the turbines proposed by the Humboldt Wind Project.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 7 Humboldt Wind Project September 2019

Figure 2. Equation used to quantify collision risk for marbled murrelets (MAMU) transiting through the Humboldt Wind Project. The primary equation is displayed at the top. Components of the primary equation are color-coded to facilitate expansion of the primary equation for each component. Blue text summarizes the calculation of probability that a murrelet will collide with a turbine blade upon entry into a rotor swept zone (RSZ); this is the Band (2012) Model. Green text summarizes the calculation of the number of MAMU expected to pass through a single RSZ given their rate of passage through the proposed turbine string. Orange text summarizes the adjustment of the collision estimate produced by multiplying components 1 and 2 to account for micro, meso, and macro avoidance. The deterministic and probabilistic CRMs are both based on this equation. However, specific inputs that were modeled using a Bayesian framework using the probabilistic approach are in red text.

The collision estimate produced by the Band model (2012), and all kinematic models of collision for that matter, does not factor in behaviors and therefore fails to account for the ability of birds to detect and avoid collision (see Appendix A for a comprehensive review of avoidance abilities of birds). The accuracy and predictive power of any mathematical model (Morris et al. 1999, Keedwell 2004), including CRMs (Band 2012), depends on the use of realistic estimates of avoidance. After accounting for avoidance, the total predicted collisions in each zone were summed by month to reflect the chronology of the nesting season and the associated changes in inland activity and ridge use.

This collision risk was estimated using the breeding season passage rates measured by radar in 2018 and 2019 and from the non-breeding season passage rate calculated from the period between the two breeding seasons. The resulting collision risk estimates were then extrapolated over 30-year periods as an estimate of collision risk. However, it is likely that the passage rate of murrelets varies from one year to the next as a function of the proportion of individuals of breeding age that actually nest (breeding effort); nesting is the most energetically demanding phase of a seabirds lifecycle and, for many, the decision of whether to nest in a given year is related to ocean conditions associated with primary production that forms the base of the marine food web and determines subsequent availability of prey to seabirds. Therefore, we related interannual variability in inland activity over a 10-year period in the Eel River Valley to local, regional, and basin-wide indicators of ocean productivity and characterized the magnitude of variability in these conditions since 1983 to account for the potential variability in passage rate over the 30-year period (Appendix B). This relationship facilitated a better understanding of whether the passage rates that were likely below, at, or above what would be expected on average each year as well as to predict the upper and lower expectations of annual passage rate based on environmental conditions and subsequent collision risk over a longer (30-year) period.

Lastly, the sensitivity of the collision risk estimates produced by the deterministic model were assessed by perturbing various input values and quantifying the impact of each perturbation on the collision risk estimate. This was necessary to identify which inputs had the greatest influence over estimates produced by the deterministic CRM.

2.1.2.3 Probabilistic model

The probabilistic model generally followed the same structure of the deterministic model (Figure 1, Figure 2), but was based on the full spatial extent of the turbine string (including topographically-defined buffer areas that extended beyond the edge of the first and last turbines) rather than a zone-by-zone analysis. Of the many inputs that were considered by these CRMs, the deterministic model was most sensitive to variability in three specific inputs parameters: murrelet flight speed, the ability of murrelets to avoid collision, and interannual fluctuations in passage rate. Because potential variability, or uncertainty, associated with these three inputs could substantially change the estimate of collision risk, a Bayesian approach was used to provide a probabilistic estimate of risk rather than a single value representative of average conditions (see Table 1 for how various input parameters were treated in the probabilistic model). Quantifying uncertainty associated with these three inputs using a Bayesian approach served two purposes: it was essential to assign a likelihood to the entire range

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 9 Humboldt Wind Project September 2019

of potential outcomes and it was a means of validating the single-value estimate derived from a simpler deterministic approach. Three other inputs (number of turbines, turbine size, blade speed) were set at maximum values as described in the approach for the deterministic model (Table 1).

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 10 Humboldt Wind Project September 2019

The goal of the probabilistic model was to use a Bayesian approach to estimate collision risk to marbled murrelets over the 30-year life of the Project in a way that accounted for inherent variability or uncertainty associated with critical inputs of the CRM. Bayesian inference allows an allocation of credibility (expressed as probability) across all possible outcomes for each parameter of input variables for which the exact value may fluctuate or the exact value is unknown. In a Bayesian analysis, the distribution of credibility initially reflects prior knowledge about the possibilities (called priors) and, as new data are observed, credibility is re-allocated to possibilities that are consistent with the data (called posterior) using mathematics that are logically coherent and precise (Kruschke 2015). If posterior predictions mimic the data with reasonable accuracy (based on posterior predictive checks), then representative data can be generated using the posterior predictive distribution (PPD) and input into the CRM in a way that allows for consideration of and assignment of probability to all possible values of collision risk. The collision risk estimate derived from this probabilistic approach was in the form of a probability distribution and has been described here as the most likely outcome (the value of risk with the greatest probability) and mean along with the upper bound of a one-tailed threshold that encompasses 95% of the probability distribution. An 80% highest density interval following a “risk- adverse” approach was recommended by the USFWS (2013) in association with their Bayesian CRM for eagles to ensure a conservative evaluation of collision risk “in the face of considerable uncertainty”. Herein we use a stricter measure of no more than a 5% chance of risk greater upper bound of the highest density interval.

To generate a PPD using Bayesian inference requires technical mathematical specifications. Technical details regarding the likelihood functions and priors used for each Bayesian variable (flight speed, avoidance, and interannual variability in inland activity) are detailed in Appendix C. In general, likelihood functions were selected to best represent the distribution of the data and priors were selected to be generally uninformed so that they did not influence the shape and spread of the posterior distribution.

The following steps outline the process of generating PPDs for murrelet flight speed, avoidance, and interannual variability passage rate: (1) Data were gathered and used to assess the credibility of parameter values specified by the priors; a. Flight speed data were derived from accurate measures of speed associated with each of the ridge crossings detected by radar surveys associated with this Project; b. Avoidance data were derived from a comprehensive literature review (see Section 2.2.5.2 and Appendix A); c. Interannual variability in inland activity data were derived from long-term radar studies of murrelets in the Eel River Valley and various ocean indices associated with inland activity (see Section 2.2.6 and Appendix B). (2) The form of likelihood function for the PPD was specified based on the likelihood function of the data, given that these two distributions should approximate each other (Appendix C; Kruschke 2015); (3) Broad and noncommittal priors were specified for parameters that define the location, shape, and scale of the likelihood function such that the priors exerted virtually no influence on the resulting posterior distributions for those parameters (Appendix C);

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 11 Humboldt Wind Project September 2019

(4) Posterior distributions were generated for each parameter using a Markov Chain Monte Carlo (MCMC) algorithm that re-allocated credibility across parameter values, from vague prior distributions to values that were consistent with the data using Bayesian inference (Appendix C); a. For each parameter that dictated the shape and scale of the PPD, four independent chains were generated using RJAGS in R (R Core Development Team 2019, Plummer 2018) where: i. The first 2,000 steps were allocated for the tuning of MCMC samplers to ensure the best performance of the algorithm; ii. The next 50,000 steps allowed the chain to move from an unrepresentative initial value to the modal region of the posterior; iii. The final 100,000 steps were compiled into a database. (5) Confirmation that the posterior distribution for each parameter mimicked the data with reasonable accuracy was done by: a. Ensuring that the final 100,000 steps of each chain were well mixed and converged on the same posterior distribution (Appendix C); b. Assessing the stability and accuracy of the chains by examining the effective sample size (Appendix C). (6) For each parameter, the average value of the 100,000 estimates generated using four MCMC was used to define the shape and scale of the PPD, and 100,000 values were randomly sampled from this average PPD for input into the CRM.

Each of the 100,000 values sampled from the 3 PPDs, in combination with the fixed parameters that were also used in the deterministic model, ultimately facilitated derivation of a robust probability distribution of collision risk for murrelets over a 30-year period based on 100,000 estimates. This process was repeated 10 times and the outcomes of these 10 model runs were averaged to produce the final estimates, this ensured stability in the outcome; the stochastic nature of Bayesian inference causes each model run to produce a slightly different estimate of collision risk.

2.2 Collision risk assessment

2.2.1 Radar data

Stantec (2018, 2019) provided detailed descriptions of murrelet flights intersecting or passing over Bear River Ridge and Monument Ridge in the Project area. Seven radar stations were positioned along these ridges; position was dictated by topography to facilitate radar vision out to 1.5 km distance. Each survey used two radar units: one that was positioned to detect targets in the horizontal plane and another that was positioned to detect targets in the vertical plane to facilitate measurement of the altitude of murrelet-like targets.

Radars were operated during the murrelet breeding period in the mornings, from 90 minutes prior to and 75 minutes following sunrise, and in the evenings, from 60 minutes prior to and 60 minutes following sunset. This comprehensive sampling schedule ensured that the daily activity period of murrelets transiting between the ocean and inland nesting habitat were captured in their entirety. Although murrelets were not expected to transit

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 12 Humboldt Wind Project September 2019

between at-sea foraging areas and inland nesting habitat outside of their well-established activity periods (Burger 1997, Evans Mack et al. 2003), to be sure an additional 5 radar surveys occurred mid-day in 2018 to assess if flights occurred outside the usual activity period. Surveys were recorded as videos and target analysis occurred in a laboratory setting. All flights that occurred during the observation periods were identified. Each murrelet- like target detected via radar was tracked for periods adequate to discern flight direction, flight speed, and possible flock size. Other data recorded with each detection were date, time relative to sunrise, the presence of fog, and wind speed and wind direction.

Radar operators had high confidence in their ability to distinguish murrelets targets from other targets both prior to and following dawn. There can be problems with contamination after sunrise, specifically where non- murrelet targets that could be mistaken for murrelets and included in calculations of passage rate (Cooper et al. 2001); consequently, some CRMs only count inland-flying murrelet-like targets before sunrise and adjust the before sunrise passage to account for after sunrise flights (Sanzenbacher and Cooper 2010, Sanzenbacher et al. 2015). However, others have chosen to use direct counts of murrelet-like targets even following sunrise for development of a CRM (Hamer Environmental 2009, Nations and Erickson 2009). There is a tradeoff between these two approaches, with the former relying on assumptions that murrelets flying seaward will do so along similar pathways and at similar heights as inland flying individuals and the latter potentially including non- murrelet targets observed post-sunrise in their calculation of passage rate. We chose to use direct counts of murrelet-like targets to minimize reliance on assumptions. In any case, where there was ambiguity in the identity of a murrelet-like target, it was treated as a murrelet. Consequently, any error in this approach would likely overestimate the true rate of passage through the proposed turbine string.

Radar operators also reported whether the radar target signature had characteristics indicative of a multi-bird flock or not, allowing for a more refined reporting of the number of birds detected. Some studies report being unable to reliably ascertain the number of birds in a target and used a modifier to account for targets containing more than one bird (e.g., Sanzenbacher et al. 2015). However, methodologies described by Urmy and Warren (2016) allowed them to reliably distinguish individual terns (similar in size to murrelets) flying to and from a large breeding colony in flocks ranging from 2 to 200 individuals. Urmy and Warren (2016) used a Furuno marine radar that had been recently recalibrated to ensure accurate radar cross section measurements and was capable of target resolution capabilities (in terms of beam width and pulse length) identical to units calibrated and used by Stantec to count murrelets. Radar equipment was calibrated each breeding season. For purposes of the CRM, we treated each bird present in a multi-bird target as a distinct murrelet flight (i.e., a two-bird target was counted twice).

2.2.2 Sampling area

The radar defined the area sampled; the area covered by the seven radar stations along the ridges encompassed approximately 72.2% of the length of the proposed turbine string. This measure included areas between radar stations as well as areas where radar vision was obscured due to vegetation, clutter, or topography (known as “blind spots”). For the remaining 27.8% of the proposed string that was not explicitly sampled, murrelet activity

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 13 Humboldt Wind Project September 2019

had to be assigned based on passage rates observed within the sampling area. So, for example, if a passage rate (birds/day/km) was observed in the sampling area, then this same passage rate would be applied to adjacent areas of the proposed turbine string that had not been explicitly measured. The spatial extent of this assessment extended a maximum of 0.8 km (0.5 mi) beyond the last turbine on either end of a string, depending on topography, using the logic that individuals crossing the ridge near the extent of the proposed turbine string would be likely to cross the within the proposed turbine string if these adjacent areas are topographically similar.

2.2.3 Passage rate

The radar surveys described in Section 2.2.1 identified and reported all murrelet-like targets detected in the sampling area, however only a subset of these murrelets crossed the ridgelines in the vicinity of the proposed turbine string. For purposes of estimating collision risk, it was necessary to determine which murrelets crossed the ridgelines in the vicinity of the proposed turbine string and were at risk. Targets at a distance or flying in directions that could not involve the proposed turbine string (e.g., paralleling the proposed turbine string) were not considered at risk and were therefore not included in the calculated passage rate for these CRMs.

In some instances, crossing events were directly recorded by the radar and each of these murrelets was considered at risk of collision. If murrelets were detected transiting near the proposed turbine string but the recorded flight path did not cross the proposed turbine string, then the flight paths were projected forwards and backwards for the length of the proposed turbine string to infer if and where the murrelet could have intercepted the string. For murrelets intercepting the string at more than one point, thus being exposed to the risk of collision more than once, each individual crossing was used in calculations of passage rate through the risk zone.

For each event in which a murrelet intercepted the turbine string and was considered at risk in terms of its horizontal position, we then evaluated the flight attitude to assess if these individuals were also at risk in the vertical dimension. Vertical position was not determined for every murrelet and, in instances where height was not known, we assumed they were flying at risk height. In instances where height was known, murrelets flying near the proposed turbine string below an altitude of 210 m relative to the ridgeline (i.e., 30 m above the upper extent of the rotor) were considered at risk of collision and were included in the calculated passage rate for these CRMs.

The passage rates of murrelets crossing the proposed turbine string were calculated independently for morning and evening periods and then summed to estimate daily passage. Mid-day surveys did not detect murrelets transiting inland between the morning and evening activity periods (consistent with the telemetry study by Hébert and Golightly 2006) and thus, did not contribute to the assessment of passage through the turbine string.

For the deterministic model, the turbine string was subdivided into zones based on discontinuous segments of turbines where there was >0.8 km between the rotor edges and where topographical features were relevant to ridge crossing murrelets (peaks, low-points, and drops into a neighboring valley). Each zone had a horizontal

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 14 Humboldt Wind Project September 2019

length (calculated in Google Earth) and a vertical height (based on the maximum height of a turbine rotor +30 m). Passage rate of murrelets (birds/day/km) was calculated for each zone using combined landward and seaward flying murrelets in morning and evening sampling periods. Zones were created independent of radar coverage and if the zone contained areas not covered by radar, a passage rate was calculated for the portion of the zone for which there were data. That rate was then applied as an attribute of the entire zone. Passage rates were also stratified by morning period, evening period, and month.

For the probabilistic model, we treated the turbine string as a single unit rather than being subdivided by zones due to the complexity of outputs and the need to report collision risk in a straightforward manner. Rather, we assumed that if both models produce overlapping estimates, then the deterministic model could reliably inform relative risk on a zone-by-zone basis. For the probabilistic model, passage rate of murrelets (birds/day/km) was calculated for the entire length of the turbine string rather than by zone (zone attributes were assigned to any blind spots). The annual passage rate was treated as a static variable for each year of radar surveys. As mentioned earlier, variability in this number would result from sampling error which is thought to be small using radar-based methods (Mateos et al. 2010).

2.2.3.1 Breeding season

Most inland flights by murrelets are known to occur during the breeding season (Naslund 1993, Sanzenbacher et al. 2014) due to the need of each nesting individual to transit between at-sea foraging areas and inland nests 1 to 2 times each day and, depending on if they are incubating an egg or provisioning a chick, for at least 60 days if they are successful. Each murrelet crossing of the proposed turbine string as measured via radar in April through September represented the breeding season passage rates.

2.2.3.2 Non-breeding season

Murrelet activity expected during the non-breeding period was estimated to assess the potential passage of murrelets through the turbine string for the period of October through March. Stantec (2019) conducted radar surveys in October, December, February and March to characterize the period. Detections of murrelets were rare (as expected) which made it necessary to impute values for missing months and extrapolate over the period, except for March which had adequate observations to treat in the same manner as months in the breeding season. To calculate values to be extrapolated into November and January, we averaged observations across surveys for October to February; the resulting average value was used in November and January. This nonbreeding estimate was used in both the 2018 and 2019 year estimates in this report.

For comparison to the method used in earlier reports of this collision risk model, we also estimated inland flight activity during the non-breeding period with the year-round radar flight estimates from a study at Epsa Lagoon (on the coast, 10 km north of Orick, CA; Sanzenbacher et al. 2014). Epsa Lagoon was used as a comparison because of the robust sample size associated with the study at this site. We calculated a ratio that was representative of each non-breeding month when compared to the average of June and July passage rates calculated by Sanzenbacher et al. (2014). That ratio was then used to generate non-breeding passage rates for

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 15 Humboldt Wind Project September 2019

Bear River and Monument Ridges by multiplying the ratio from Epsa Lagoon by the June and July passage rate calculated for each zone on Bear River Ridge and Monument Ridge. Once passage rates were calculated using the ratio estimate, numbers were subjected to the same calculations as the empirical data from the breeding season. This result did not differ substantively from the empirical data used to calculate collisions in the nonbreeding season; consequently, all aspects of the model reported herein are based on the empirically derived passage rates.

2.2.4 Collision probability

Collision probability is the likelihood that a bird will be struck by a rotor blade following entry into a rotor swept area. Birds that pass through a rotor swept area can do so without contacting the turbine blades and the probability of contact is partly a matter of physics. We calculated the probability that a murrelet will enter a rotor swept zone and be struck by a turbine blade based on physics alone using the Band Model (2012) which accounts for bird size and speed, turbine size and speed, and effects of flying with and against wind. The Band Model does not include the effect of turbulence, however there are no studies demonstrating that rotor- generated turbulence alters collision probability (Appendix D).

The Band Model treats bird passages through a rotor swept area as if flight paths are all perpendicular to a rotor. Band recognized that the approach angle of birds through the rotor swept area would vary but the positive and negative influences on collision probability would likely cancel. To ensure this was the case for murrelets, we calculated collision probability for each bird based on flight speed and angle off the wind using a refinement to the Band model developed for such purposes (Appendix E). In the case of murrelet flights, Band appears to be correct in assuming that the tradeoff between differing angles of approach would cancel.

The speed at which a bird travels through a single rotor swept zone greatly influences the probability of collision, with slower birds being much more likely to be struck before they can clear the rotor. For each murrelet intercepting the proposed turbine string, an accurate estimate of air speed was quantified. For the deterministic model, we used the average air speed of all crossings. Because a Bayesian approach allowed for each of these unique speeds to be considered when estimating collision risk, all possible flight speeds of crossings were considered when estimating collision risk for individuals that entered a rotor swept zone using the probabilistic model. The likelihood function and priors (see Kruschke 2015) used to generate a PPD for flight speed using these data are detailed in Appendix C.

2.2.5 Avoidance probability

Avoidance probability is the likelihood that an individual bird will detect turbines and adjust their flight to avoid a collision with the turbines. Avoidance is theoretically partitioned into macro-avoidance (entirely avoiding the area with turbines all together), meso-avoidance, and micro-avoidance. In practice, it is difficult to measure these separately (May 2015) and reporting of avoidance includes all three measures represented by a single value. For many seabirds, an avoidance of 0.99 or greater is typical (and has been used for CRMs when empirical evidence supports this). Unless there is species-specific evidence to support a greater or lower rate of avoidance,

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 16 Humboldt Wind Project September 2019

a value of 0.98 is empirically supported as a conservative estimate for birds in general (Furness 2013, Cook et al. 2014, Cook et al. 2018) and is recommended for use when generating a CRM for birds and wind energy facilities in the onshore environment (Appendix A; SNH 2017). Avoidance is often measured over a range of weather conditions, and if avoidance capabilities are altered by rain or fog, these are inherently incorporated into most empirically derived avoidance rates (see Appendix A and Appendix F).

There are no murrelet-specific studies of avoidance for wind turbines because there has been no opportunity to study this. The only murrelet-specific quantification of avoidance was for individuals at risk of collision along a transmission line (Cooper 2010); this study documented an avoidance of 100% based on a total observation of 423 murrelets using a corridor during the early morning activity period, 79 of which were at risk height and 10 of which were predicted to collide based on chance alone (Cooper 2010).

There are many reasons that murrelet avoidance of wind turbines will be on par with what has been documented in other avian species and particularly with seabirds interacting with onshore facilities. For a comprehensive justification see Appendix A. Briefly, nesting murrelets must be able to fly in and out of the forest canopy at high speed and their survival depends on their ability to detect and avoid collision hazards within this complex landscape during periods of low light. Furthermore, the amount of time a murrelet would be exposed to turbines associated with the Project will be very limited in duration and individuals passing through this area should be focused primarily on navigating between the ocean and nesting habitat. This contrasts with species that have been documented to avoid turbines at a rate lower than 0.98, such as raptorial species (Whitfield and Madders 2006, Smallwood and Thelander 2008, SNH 2017). Species for which individuals spend significant proportions of time in the vicinity of wind turbines are thought to exhibit lower avoidance, as they frequently shift their focus from avoiding structures to other behaviors (defending territory, capturing prey) while continuing to navigate across the landscape (Barrios and Rodriguez 2004, Thaxter et al. 2017). Distraction has been identified as a primary factor that contributes to the increased risk of collision for certain species (USFWS 2013).

2.2.5.1 Avoidance and the Deterministic Model

For murrelets, the deterministic CRM followed the evidence-based guidance of the Scottish Natural Heritage (SNH 2017; see Appendix A) and assumed a conservative avoidance probability of 0.98 for each month except April when an overly conservative value of 0.95 was used. In April, naïve young murrelets may accompany adults on their first flights inland since hatching. It is possible that these young murrelets exhibit lower micro- avoidance; however, data do not exist to corroborate this logic. Regardless, to be very conservative in the deterministic model, we used a lower avoidance probability in April (0.95) to consider the possibility of naïve individuals interacting with the turbines when first visiting nesting areas. For purposes of further understanding the possible outcomes of different fixed avoidance rates, the model was run with different avoidance values within the range of realistic avoidances for birds (0.99 to 0.95; see sensitivity analysis).

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 17 Humboldt Wind Project September 2019

2.2.5.2 Avoidance and the Probabilistic Model

In contrast to the deterministic model, where only two values of avoidance (0.98 for all months but April, which was 0.95) were considered in the CRM, the probabilistic model used a Bayesian approach to approximate the probability of all values of avoidance possibly exhibited by marbled murrelets, based on prior knowledge of avoidance of turbines by ecologically similar birds as they travel through wind energy facilities. Although avoidance has yet to be measured for murrelets, it has been empirically measured for various species at various locations in a range of weather conditions over the last two decades. Bayesian inference is ideal for situations when there is only partial knowledge of a given phenomenon, as it generates a distribution of probability for all possible outcomes (i.e., values of avoidance) conditioned on the observed data, in a systematic, repeatable way that directly accounts for uncertainty (Gelman et al. 2013, Kruschke et al. 2015). To this end, current understanding of avoidance was extracted from various sources and compiled into a database.

To ensure that avoidance rates compiled for the CRM would realistically approximate the avoidance likely to be exhibited by marbled murrelets as they travel through terrestrial habitats, we generated a set of criteria to facilitate the objective identification of avoidance estimates. Representative estimates of what may be expected for murrelets given their behavior (relative to other types of birds for which avoidance has been previously reported) served as inputs into the probabilistic model. To that end, the following criteria had to be met for each reported estimate of avoidance included in this analysis: 1. When only diurnal estimates were available, the estimate must come from a terrestrial environment; 2. Nocturnal estimates could come from any environment, including offshore marine habitats; 3. All estimates came from species with flapping flight (i.e., does not glide or use thermal lift for flight); 4. All estimates were derived from studies for which surveys of mortality and activity (birds/hour) spatiotemporally overlapped thereby eliminating the need for spatial or temporal extrapolations which could lead to an inaccurate assessment; 5. All estimates were independently calculated, if identical estimates for the same location, species, and year were reported by more than once source, then that estimate was treated as a single estimate; 6. All estimates were derived using the Band Model (Option 1) following the same procedure of calculating collision risk as used in this CRM for murrelets; 7. All estimates had a precision of at least 0.01 because very small changes in avoidance (on the order of 0.01) results in large changes in estimated fatality (Chamberlin et al. 2005); to ensure that an estimate achieved this level of precision required that an estimate be derived from situations where either the number of predicted collisions without avoidance was at least 100 or the number of individuals documented flying at turbine height exceeded 10000. Further, an inadequate sample with poor precision would be more likely to also be inaccurate. 8. Estimates came from species transiting through the wind facility rather than spending time foraging and engaging in other non-transitory behaviors; 9. Estimates were species-specific and location-specific rather than representing an average condition across various species or locations/habitat types.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 18 Humboldt Wind Project September 2019

10. Estimates were derived from research or monitoring designs that ensured that all estimates of fatality either were fully inclusive of fatalities or used corrections to adjust fatalities for searcher efficiency and carcass persistence (Table A2).

To the best of our ability, we strictly adhered to the criteria described above with the exceptions of 8 and 9 above. Consistent application of Criteria 8 and 9 would have been ideal, but this would have caused many otherwise robust estimates of avoidance to be excluded from our analysis. We made exceptions to Criterion 8 by including estimates of avoidance by gulls. Gulls likely have a greater propensity for distracted flight near turbines relative to murrelets given their adaptability in terms of food sources and time spent in urbanized landscapes (Belant 1997). However, because they are the most common seabird present in onshore environments, avoidance by gulls has been disproportionately studied due to a priority to assess the degree of impacts to seabirds in general that may result from installing turbines in marine environments rather than on land. Most studies that met Criteria 1-8 and 10 also met Criterion 9; however, if an estimate of avoidance met all but criteria 9, we decided to include it as this was just intended to maximize inclusion of very specific data rather than averages that could have the potential to mask the extent of variability in avoidance behaviors.

The primary source of avoidance estimates was a database created by the British Trust for Ornithology for the Scottish Government; this database contains the calculations that form the basis of Cook et al.’s (2014) comprehensive review of avoidance rates. Two additional reviews of avoidance were also used (Furness 2013, Cook et al. 2018). These later reviews were primarily focused on avoidance by seabirds (gulls, terns, and cormorants) at onshore wind facilities located in coastal bottoms, coastal hills, and rural or agricultural areas in the Netherlands, England, United Kingdom, Poland, France, and Belgium. An additional 7 studies were also incorporated that included diurnally active geese in hilly or mountainous terrain (Fernley et al. 2006), nocturnally migrating ducks and geese in an offshore environment (Desholm and Kalert 2005), and nocturnally migrating songbirds in mountainous terrain (Aschwanden et al. 2018).

These sources reported 152 estimates of avoidance, 35 of which met our selection criteria. The primary reason for disqualification was that the estimate did not meet the minimum precision requirements; this was the case for 61% of all estimates (n=93). To better illustrate the need for precision, 19% of the records in the Scottish Government’s database had predicted the number of collisions to be either 0 or 1 fatalities; however, if 1 or fewer fatalities were predicted, then there were only two possible estimates of avoidance: 0% or 100%. Because birds are generally adept at avoiding collision, each of these records had an estimated avoidance of 100% which, if included, would overinflate the probability estimates for avoidance. The remaining 24 disqualified estimates came from either an offshore location (n=10) or were reported elsewhere and had already been included in the database (n=14).

The 35 estimates of avoidance that met our selection criteria (see Appendix A) were then used as the data input for the Bayesian estimation of avoidance. The likelihood function and priors used to generate a PPD for avoidance based on these data are detailed in Appendix C.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 19 Humboldt Wind Project September 2019

2.2.6 Interannual variability

For a diversity of seabirds, from auks to shearwaters, interannual variability in breeding effort is often linked to oceanographic conditions and included ocean productivity and prey availability throughout the nesting period. In general, when conditions are such that overall productivity is low (such as a strong El Niño or weak Aleutian low-pressure system), breeding efforts by seabirds are reduced relative to average conditions (Schreiber and Schreiber 1984, Ainley and Boekelheide 1990). One mechanistic explanation for this response is that females need adequate energy to produce eggs and feed chicks; when the ecosystem has delayed or depressed productivity, the ability of females to produce eggs is also delayed or depressed (Reed et al. 2009, Schroder et al. 2009).

The inland activity of murrelets is associated primarily with nesting activities and, as such, it is logical that changes in passage rate across years would vary as a function of ocean conditions associated with productivity. However, because marbled murrelets are difficult to observe without the aid of radar or radio telemetry, long- term studies with sample sizes sufficient to statistically link variability in inland activity to environmental conditions have rarely been achieved. However, there is evidence that marbled murrelets also modify their breeding effort in response to ocean conditions (Lorenz et al. 2017).

A study on Vancouver Island found that detections of behaviors displayed near nests by murrelets (often called occupied behaviors) were significantly lower when nearshore ocean temperatures were warm (Burger 1999). A more recent and ongoing study has preliminarily linked the breeding effort of radio-tagged murrelets nesting in central Oregon to be linked to the Pacific Decadal Oscillation (PDO; Adrean et al. 2019). Further evidence that marbled murrelets may be sensitive to ocean conditions comes from a close relative, the Kittlitz’s murrelets (Brachyramphous brevirostrous). Breeding propensity of Kittlitz’s murrelets was best explained by the intensity of the Aleutian low-pressure system as indicated by the North Pacific Index (NPI), where more intense low- pressure results in earlier and stronger phytoplankton blooms (Kissling et al. 2016).

To assess how murrelet passage rate varied across a 30-year period, we created a dataset of murrelet passage through the Eel River Valley over a 10-year period (2002-2011) using the average radar counts of murrelets at the 6 reserve sites surveyed by the Humboldt Redwood Company and reported annually (HRC 2018). The dataset extended through 2018, however after 2011 there was a directed downward trend in the number of murrelet detections for unknown reasons (likely due to confounding factors not associated primarily with ocean condition). Thus, in our analysis we limited the data to the 2002-2011period.

We associated counts of murrelets from radar in the Eel River Valley to ocean conditions often correlated with biological activity in the north Pacific. The details of this analysis are presented in Appendix B. Various indices of ocean productivity have been used to explain variability in the breeding effort and success of various seabirds. As such, we considered six indicators often related with seabird breeding effort that operated at various spatial scales from local (sea surface temperature at the mouth of the Eel River), to basin-wide (Pacific Decadal

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 20 Humboldt Wind Project September 2019

Oscillation, North Pacific Gyre Oscillation, Northern Oscillation Index, Multivariate El Niño Southern Oscillation Index, Multivariate Ocean Climate Index).

Winter Pacific Decadal Oscillation (PDO), particularly in February, was identified as the best predictor of interannual variability in inland activity by murrelets. The winter PDO accounted for 56.9% of variability in inland activity across the 10-year period, and when using the month of February the PDO accounted for 70.7% of the variability (Appendix B). Notably, the PDO record extended back to 1900 (JISAO 2019, NOAA CPC 2019a) and variability in this parameter has been well established (Mantua and Hare 2002, Newman et al. 2016). This provided a solid foundation from which to better understand how collision risk may change across a 30- year period assuming a stable population.

For the deterministic model, we used a simple linear regression to estimate the best-fit equation of the line describing how inland activity in the Eel River Valley varied as a function of the PDO. To convert activity predicted in the valley to a passage rate on the ridge, we assumed a that passage rates along the ridge would be proportional to passage rates in the valley so that PDO-based estimates could be proportionally adjusted to reflect passage rates along ridge. Using this logic, the average, minimum, and maximum values of PDO observed over the previous 35-years (1983-2018) and inclusive of eight transitions between “strong” El Niño and “strong” La Nina conditions (NOAA CPC 2019b) were used to predict the average, minimum, and maximum passage rates expected along the ridge over a 30-year period. If the population using the Eel River Valley declines, then calculations described here would overestimate collision over the 30-year period. Alternatively, if the population using the Eel River Valley increased, then these calculations may underestimate collision over 30 years.

For the probabilistic model, interannual variability in passage rate was assessed using a similar method as described for the deterministic model. However, instead of generating a single best-fit line for inland activity and PDO, a Bayesian approach was used to assess uncertainty in the slope and intercept of the regression line. This resulted in many (100,000) unique equations that each approximated the relationship between PDO and inland activity. To predict how inland activity would change across many years, a PPD of PDO was generated using all values of winter (Jan-March) PDO since 1900 (JISAO 2019, NOAA CPC 2019a). From this PPD of PDO, 100,000 values were randomly selected as inputs for the equations predicting inland activity of murrelets in the Eel River Valley. Like the deterministic model, these estimates were proportionally adjusted for the ridges to estimate the interannual variability of passage along the ridge based on a 118-year dataset. The likelihood function and priors used to describe the relationship between PDO and inland activity of murrelets and generate a PPD for PDO are detailed in Appendix B.

2.2.7 Analysis scenarios

The type, number, and placement of turbines as well as their operational time and rotational speed directly relate to the likelihood that upon passing through a rotor swept area, murrelets may collide with a blade. Due to the need for flexibility in turbine type, number, and placement, the Project initially proposed 60 turbines;

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 21 Humboldt Wind Project September 2019

however, after obtaining radar and collisions estimates from the 2018 season, the number of turbines were reduced to 47. For comparison of the two turbine layout scenarios (60 and 47 turbines), we ran both the deterministic and probabilistic CRMs under these two scenarios. Specifically, we assumed that: (1) the largest turbines being considered for use by the Project would be installed; (2) this turbine would be installed at each of 60 or 47 specific locations; (3) the turbines would operate 82.45% in the morning and 84.15% in the evening; and (4) blades would rotate at their maximum speed. Because the rotor swept area will ultimately be less than modeled here and, as a result, fewer collisions; the magnitude of collision reduction will depend on the actual size and location of installed turbines. Thus, both CRMs overestimate collision risk due to these assumptions alone.

This turbine, identified here as “large”, had an overall height of 179.5 m, a blade radius of 74.5 m, a blade width at the widest point of 4.2 m, and a max speed of 17 rotations per minute. Blade pitch was not available in turbine specifications, so we assumed a blade pitch of 15 degrees, which is the default value for the Band Model in absence of more specific data.

2.3 Sensitivity analysis for the deterministic model

Since the deterministic model only produced a single estimate of collision, rather than a probability distribution, a sensitivity analysis was used to assess how variation in input parameters affected the outcome of the collision risk assessment. Each parameter that was deemed to vary in a way that could substantially alter the final estimate of collision risk was perturbed by a given amount and the overall change in estimated risk was quantified. Passage rate, murrelet flight speed, blade speed, and proportion of time that the turbines were operational were each changed by 10%. Because avoidance rates are less variable and are generally above 95%, this specific parameter was only changed by 1% (a 10% change would be uninformative in terms of providing a realistic estimate of collision risk). For ease of interpretation, we calculated output of 30-year collision estimates from the deterministic model in 1% increments of avoidance rate over the range of realistic empirically measured avoidance measures for avian species (99-95%).

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 22 Humboldt Wind Project September 2019

Section 3.0 Results

3.1 Overview

In 2018, Stantec conducted radar surveys to document murrelet activity along Bear River and Monument Ridges for a total of 209 hours in the morning and 132 hours in the evening between April and September 2018. In their radar report, Stantec identified 35 possible murrelet targets approaching or crossing the ridges between April and September 2018. After adjusting this to reflect the most current turbine layout, adding buffers on the ends of turbine strings (having the effect of increasing the area of intercept) and eliminating targets (n=3) that were flying at an altitude that would be well above any potential turbine (>210 m), these CRMs considered 57 murrelets to have crossed Bear River and Monument Ridges in the risk zone associated with the maximum proposed turbine string (60) in the 2018 breeding season. Of these 60 crossings (57 + 3), 75% occurred in the morning and 25% occurred in the evening; 62% were heading inland, 23% were heading seaward, and the direction for 15% could not be characterized. Of these crossings, 26.6% (n=16 birds) were reported to be within 2-bird flocks.

These 57 crossings were used to calculate passage rates of murrelets (birds/day/km) crossing the Project Area for each month of the breeding season in 2018. Altitude of targets potentially flying below the rotor was uncertain, and all flying murrelets below maximum rotor height were considered in the calculations of passage rate for the area. This made it impossible to directly model collision risk with stationary components. However, because all crossings at an altitude of 0-30 m were treated as if they were potentially exposed to a rotor swept area and collision risk is greater for rotor swept areas relative to stationary obstacles, these CRMs do account for that risk indirectly and do so in a conservative manner.

In 2019, Stantec conducted radar surveys similar in extent to 2018 during the breeding season. After eliminating targets (n=7) that were flying at an altitude that would be well above any potential turbine (>210 m), these CRMs considered 82 murrelets to have crossed Bear River and Monument Ridges during the 2019 breeding season in the risk zone associated with the maximum proposed turbine string (60). Of these 89 (82 + 7) crossings, 76% occurred in the morning and 24% occurred in the evening; 67.3% were heading inland, 33.3% were heading seaward. Of these crossings, 18% (n=16 birds) were reported to be 2-bird flocks. These 82 crossings were used to calculate passage rates for each month of the breeding season in 2019.

From October 2018 to March 2019, surveys were conducted in October, December, February, and March. Two murrelets were detected crossing the turbines at risk height in December, none in February, and two in March. All were below 210 m elevation above ground, 0.67 were headed inbound and 0.33 were headed seaward.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 23 Humboldt Wind Project September 2019

3.2 Collision risk assessment

3.2.1 Deterministic model

The turbine string and the associated buffers were divided into 11 zones (Figure 3). Each zone varied in length and number of turbines. Passage rates varied by month, with the greatest passage rates observed in June (Table 2), and by zone, with the greatest passage rates observed in zones G, C, and K (Table 2). Annual passage rate across all zones was 0.2136 murrelets/day/km in 2018, and 0.2871 in 2019.

Collision probability for a single passage through a rotor swept zone was 0.0433 in 2018 and 0.037 in 2019 for the largest turbine being considered for use by the Project (Table 4). Using the deterministic model, the estimated number of collisions across the entire turbine string in a single year was 0.1753 collisions in 2018 and 0.246 in 2019 (Table 3). The greatest number of collisions occurred during the breeding season, with April- September accounting for 87% and 91% of collisions for 2018 and 2019, respectively.

Overall, the number of collisions over the 30-year life of the Project, assuming 47 turbines when assessing the collision risk, was 5.26 murrelets in 2018 and 7.38 in 2019 (Table 4). If ocean conditions related to variability in inland activity of murrelets were considered, 2018 was average in terms of conditions conducive to ocean productivity, whereas 2019 was above average. Thus, 2018 passage rates were average based on values of the winter PDO over the last 35 years, whereas passage rates in 2019 were greater than the 35-year average. Based on the average condition, 5.27 murrelets would be impacted over the 30year life of the project using the passage data from 2018. Based on the greater than average condition in 2019, 6.85 murrelets would be impacted over the 30-year life of the Project. The difference between the PDO adjusted average estimates reflects unaccounted variance in the difference between years.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 24 Humboldt Wind Project September 2019

Figure 3. Map of topographically delineated zones used to assess collision risk for marbled murrelets flying through the proposed Humboldt Wind Energy Project as they transit between the ocean and inland nesting habitats. This includes Zones C,G, and K which no longer have any turbines because they descended into adjacent river valleys and consistently had greater passage and collision rates in both 2018 and 2019.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 25 Humboldt Wind Project September 2019

Table 2. Monthly and annual passage rates (birds/km/day) of marbled murrelets through the proposed Humboldt Wind Energy Project by zone (A-K; Figure 3) and across the entire wind energy facility in 2018 and 2019 including Zones C,G, and K.

2018 Zone Passage rate (birds/day/km) Length Name (km) Apr May Jun Jul Aug Sep Oct a Nov b Dec a Jan a Feb a Mar a Average

A 3.893 - 1.039 0.000 0.000 0.000 0.173 0.000 0.086 0.257 0.086 0.000 0.000 0.139 B 2.239 - 0.000 0.000 0.684 2.765 0.000 0.000 0.307 0.922 0.307 0.000 0.000 0.423 C 1.071 0.000 0.613 4.960 1.401 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.579 D 1.678 0.000 0.596 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.051 E 3.389 0.000 0.148 0.000 0.811 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.081 F 0.799 0.000 0.000 0.000 1.046 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.252 0.195 G 1.904 0.463 0.000 6.481 0.463 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.610 H 2.389 0.000 0.209 0.000 0.105 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.027 I 1.093 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 J 2.795 0.000 0.716 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.061 K 1.109 0.451 0.902 1.803 2.254 1.803 0.000 0.000 0.000 0.000 0.000 0.000 0.902 0.683 ALL 22.359 0.0618 0.4339 0.8790 0.4584 0.3663 0.0302 0.0000 0.0457 0.1370 0.0457 0.0000 0.0894 0.2136

2019 Zone Passage rate (birds/day/km)

26 Length a a a a a a Name (km) Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Average A 3.893 0.346 0.173 0.346 1.039 1.039 0.346 0.000 0.086 0.257 0.086 0.000 0.000 0.313 B 2.239 0.000 0.461 0.000 0.000 0.000 0.000 0.000 0.307 0.922 0.307 0.000 0.000 0.169 C 1.071 1.225 0.613 1.225 2.013 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.424 D 1.678 0.596 1.192 0.596 0.000 0.000 1.788 0.000 0.000 0.000 0.000 0.000 0.000 0.346 E 3.389 0.590 0.885 0.000 0.221 0.000 0.295 0.000 0.000 0.000 0.000 0.000 0.000 0.167 F 0.799 0.000 0.000 3.129 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.252 0.363 G 1.904 0.000 3.704 1.852 0.000 1.852 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.624 H 2.389 0.000 0.419 0.419 0.314 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.097 I 1.093 0.000 0.000 0.000 0.915 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.078 J 2.795 0.000 0.739 0.000 0.739 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.125 K 1.109 5.410 1.803 2.705 0.902 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.902 0.973 ALL 22.359 0.5215 0.8712 0.6121 0.5262 0.3386 0.2392 0.0000 0.0457 0.1370 0.0457 0.0000 0.0894 0.2871

a Based on empirical data collected between October 2018 and March 2019.

Table 3. Monthly and annual collision risk estimates for marbled murrelets by zone (A-K; Figure 3) and across the entire proposed Humboldt Wind Energy Project in 2018 and 2019. The Deterministic Collision Risk Model assumed a layout of 47 large turbines (turbine height = 180 m, blade radius = 75 m, blade width = 4.2 m, and a maximal rotation speed = 17 rotations per minute).

2018 Zone Collision estimates (birds/period)

Length Turbines Name (km) (#) Apr May Jun Jul Aug Sep Oct a Nov a Dec a Jan a Feb a Mar a Average A 3.893 12 - 0.0273 0.0000 0.0000 0.0000 0.0044 0.0000 0.0022 0.0067 0.0022 0.0000 0.0000 0.0428 B 2.239 8 - 0.0000 0.0000 0.0121 0.0484 0.0000 0.0000 0.0052 0.0161 0.0054 0.0000 0.0000 0.0872 C 1.071 0 0.000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 D 1.678 4 0.000 0.0052 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0052 E 3.389 10 0.000 0.0032 0.0000 0.0180 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0213 F 0.799 1 0.000 0.0000 0.0000 0.0023 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0027 0.0050 G 1.904 0 0.463 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 H 2.389 4 0.000 0.0018 0.0000 0.0009 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0027 I 1.093 1 0.000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 J 2.795 7 0.000 0.0110 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0110 K 1.109 0 0.451 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 ALL 22.359 47 0.0618 0.0486 0.0000 0.0334 0.0484 0.0044 0.0000 0.0074 0.0229 0.0076 0.0000 0.0027 0.1753

2019 Zone Collision estimates (birds/period) Length Turbines Name (km) (#) Apr May Jun Jul Aug Sep Oct a Nov a Dec a Jan a Feb a Mar a Average A 3.893 12 0.0184 0.0039 0.0075 0.0230 0.0228 0.0074 0.0000 0.0018 0.0056 0.0019 0.0000 0.0000 0.0923 B 2.239 8 0.0000 0.0067 0.0000 0.0000 0.0000 0.0000 0.0000 0.0043 0.0135 0.0045 0.0000 0.0000 0.0291 C 1.071 0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 D 1.678 4 0.0108 0.0087 0.0042 0.0000 0.0000 0.0126 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0363 E 3.389 10 0.0261 0.0162 0.0000 0.0041 0.0000 0.0052 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0516

27 F 0.799 1 0.0000 0.0000 0.0056 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0023 0.0079 G 1.904 0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

H 2.389 4 0.0000 0.0031 0.0030 0.0023 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0083 I 1.093 1 0.0000 0.0000 0.0000 0.0017 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0017 J 2.795 7 0.0000 0.0094 0.0000 0.0094 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0189 K 1.109 0 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 ALL 22.359 47 0.0552 0.0480 0.0203 0.0405 0.0228 0.0252 0.0000 0.0062 0.0191 0.0064 0.0000 0.0023 0.2460

a Based on empirical data collected between October 2018 and March 2019.

Table 4. Summary of values for each parameter associated with the Deterministic Collision Risk Model inputs (A, B, D, G-J, N-W, AA, AB, AF, AH, AJ), intermediate calculations (C, E, F, K, L, M, X, Y, Z, AC, AD, AE, ), and outputs (AI, AK, AL, AM). 60-turbines 47-turbines Parameter (2018/2019) (2018/2019) EXPOSURE AREA A Turbine string length (km) 22.4 18.3 B Turbine string height (m) 179.5 179.5 C Exposure area (km2) = A*B 4.01 3.28 D Rotor radius (m) 74.5 74.5 E Rotor swept area (m2) = pi*(D2) 17433 17422 F Proportion of area occupied by a rotor [weighted by zone for entire area] 0.0036 0.0049 EXPOSURE TIME G Proportion of time turbines rotate in morning a 0.8245 0.8245 H Proportion of time turbines rotate in evening b 0.8415 0.8415 MURRELET PASSAGE THROUGH EXPOSURE ZONE I Morning crossings (birds/km/morning) [weighted by zone for entire area] 0.175/0.242 0.109/0.173 J Evening crossings (birds/km/evening) [weighted by zone for entire area] 0.039/0.045 0.046/0.057 K Morning crossings adjusted for operational time (birds/morning/km) = G*I 0.144/0.199 0.090/0.142 L Evening crossings adjusted for operational time (birds/evening/km) = H*J 0.033/0.038 0.038/0.048 M Daily crossings adjusted for operational time (birds/day/km) = K+L 0.177/0.238 0.129/0.190 MURRELET CHARACTERISTICS N Length (cm) 25 25 O Wingspan (cm) 41 41 P Flapping or gliding flight Flapping Flapping Q Flight speed over ground (m/s) c 22.6/24.4 22.6/24.4 R Proportion of flights with wind (downwind) c 0.528/0.685 0.528/0.685 S Proportion of flights against wind (upwind) c 0.472/0.315 0.472/0.315 ROTOR CHARACTERISTICS T Blades (#) 3 3 U Blade pitch (deg) d 15 15 V Blade speed (rotations per minute) e 17 17 W Maximum blade width (m) 4.2 4.2 COLLISION PROBABILITY UPON PASSAGE THROUGH ROTOR [BAND MODEL] X Collision probability if flying with wind [D,N,O,P,Q,R,S,T,U,V,W] 0.025/0.025 0.025/0.025 Y Collision probability if flying against wind [D,N,O,P,Q,R,S,T,U,V,-W] 0.066/0.063 0.066/0.063 Z Average collision probability = (R*X)+(S*Y) 0.044/0.037 0.044/0.037 PROBABILITY OF AVOIDING TURBINES [MACRO/MESO/MICRO] AA Avoidance probability in April 0.95 0.95 AB Avoidance probability in May – March 0.98 0.98 PROBABILITY OF FATALITY AC Average exposure rate (birds/day) = M*A 3.95/5.31 2.35/3.48 AD Passage through rotor swept area (birds/rotor/day) = AC*F 0.015/0.019 0.012/0.017 AE Collisions following passage through swept area (birds/rotor/day) = AD*Z 0.00065/0.00070 0.00051/0.00062 AF Probability of fatality following collision with rotor 1 1 AG Fatalities per rotor (birds/rotor/day) = AE*AF 0.00065/0.0070 0.00051/0.00062 AH Rotor swept areas (#) 60 47 AI Daily fatality without avoidance (birds/day) = AG*AH 0.039/0.042 0.024/0.029 AJ Exposure duration (days) 365 365 AK Annual fatality without avoidance (birds/year) = AI*AJ 14.26/15.43 8.77/10.64 AL Annual fatality with avoidance (birds/year) = AK*(1-AA in Apr & 1-AB in May-Mar) 0.291/0.361 0.203/0.246 AM 30-year fatality estimate based on 2018/2019 ocean conditions = AL*30 8.73/10.81 5.26/7.38 AN 30-year fatality estimate based on the long-term average ocean condition 8.75/10.03 5.27/6.85 a Morning = -1.5 to +1.25 hr relative to sunrise; operational time derived from wind data along ridge during this time-period b Evening = -1 to +1 hr relative to sunset; operational time derived from wind data along ridge during this time-period c Empirically derived from radar surveys on ridge in 2018/2019 d Average pitch for a generic turbine e Maximum speed of rotation

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 28 Humboldt Wind Project September 2019

3.2.1.1 Sensitivity analysis

The deterministic model, like other CRMs, was sensitive to avoidance probability for the calculations of collision (Table 5). If the avoidance probability used was 99% instead of the 98%, depending on year the collisions would decrease by between 2.63 and 2.65 birds for a total of 2.63 to 4.19 murrelets over 30 years. However, if the calculations used 0.97 for the avoidance probability then the collisions over a 30-year period would rise to either 7.90 or 9.50 murrelets (2018 or 2019). After avoidance, the deterministic model was most sensitive to passage rate; a 10% change in passage rate resulted in a 10% change in estimated risk to murrelets (Table 6). Therefore, if passage increased by 10% the collision estimates for 2018 and 2019 would increase to 5.78 or 7.54 murrelets over 30 years and if passage rate decreased by 10% the collision estimate would decrease to 4.74 or 6.17 murrelets over 30 years, respectively. Murrelet flight speed also had the potential to change the outcome of the deterministic CRM, but considerably less so than avoidance and passage rate; a 10% change in flight speed resulted in a 3.5% change in estimated risk to murrelets (Table 6). Other variables that the CRM was sensitive to, in order of importance was: rotor size, proportion of time that rotors were actively spinning during the morning flight period, maximum width of the blades on the turbine, blade speed, and the proportion of time that rotors were actively spinning during the evening flight period (Table 6). Notably, this CRM used the maximum rotor size and blade size for all turbines being considered for installation and, when the blades are spinning, the speed was assumed to be at the maximum rotations per minute and thus overestimates risk.

Table 5. Collision estimates over 30 years for the entire range of potential avoidance rates, with avoidance of 98% being the most conservative estimate supported for the literature for birds similar to the murrelet and values lower than 98% being the most conservative estimates supported by the literature for all birds studied to date (see Appendix A).

Avoidance 2018 2019

99% 2.63 4.19 98% 5.27 6.85 97% 7.90 9.50 96% 10.54 12.16 95% 13.17 14.81

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 29 Humboldt Wind Project September 2019

Table 6. Sensitivity analysis of the Collision Risk Model for 47 large turbines installed as part of the Humboldt Wind Energy Project based on 2019 data only. Changes in input parameter characterized changes in collision estimates expressed as a percent of the unmodified collision estimate as well as the number of birds at 30-year intervals. The relationship between the change in the input and the change in the output was linear for all variables except blade width, flight speed, and blade speed. For blade width, the average change in output across 10% increments was used. For flight speed, the relationship was exponential and increased rapidly for flight speeds <7 m/s but the change in output was <1 for speeds greater than 12 and <0.2 for speeds greater than 22 m/s. For blade speed, the relationship was approximately linear for speeds between 6 and 17 rotations per minute (the maximum rotation speed for turbines being considered for use) and was equally low between 0 and 6 rotations per minute.

Unmodified Change in input Change in output Parameter parameter % Absolute % 30-year Passage rate (#/km/day) 0.19 10.0 0.02 10.0 0.685 Rotor radius (m) 74.49 10.0 7.45 9.9 0.675 Proportion of time turbines rotate in morning a 0.82 10.0 0.08 8.1 0.553 Maximum width of blade (m) 4.20 10.0 0.42 7.5 0.511 Bird flight speed (m/s) 24.36 10.0 2.44 3.5 0.240 Blade speed (rotations per minute) 17.00 10.0 1.70 2.7 0.185 Proportion of time turbines rotate in evening b 0.84 10.0 0.08 1.3 0.090 a AM activity period was 165 min (90 min before and 75 min after sunrise) b PM activity period was 120 min (60 min before and 60 min after sunset)

3.2.2 Probabilistic model

Annual passage rate across all zones with the exception of C, G, and K where all turbines have been removed was 0.129 murrelets/day/km in 2018 and 0.190 murrelets/day/km in 2019 (Table 7). Adjusted for the length of the turbine string without zones C,G, and K (18.28 km), the average exposure rate was 2.35 and 3.48 birds/day, respectively, for the entire string. Because the proportion of the turbine string occupied by a single rotor swept zone was 0.0049%, the passage rate through a single swept area across the entire turbine string was 0.0116 birds/rotor/day in 2018 and 0.0169 birds/rotor/day in 2019. This value was used for calculations of interannual variability in passage rate. The probabilistic model estimated collision risk for the entire string (rather than by zone) and for an entire year (rather than by month).

The mean estimate of collision, considering variability in ocean conditions over the last 118 years, was 2.94 in 2018 (most likely estimate: 1.05, one-tailed upper threshold that contains 95% of the probability distribution: 6.17). and 3.71 birds in 2019.(most likely estimate: 0.94, one-tailed upper threshold that contains 95% of the probability distribution: 7.77). The model outputs for the 30-year estimate of collision varied little between 2018 and 2019 (Figure 4).

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 30 Humboldt Wind Project September 2019

xx Figure 4. Collision risk probabilities for marbled murrelets over a 30-year period based on the Probabilistic Collision Risk Model that incorporated variation and uncertainty in murrelet passage rate, flight speed and avoidance behavior. The most likely estimate for each model run is indicated by a dashed vertical line and the 95% highest density interval is indicated by the solid red line on the horizontal axis.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 31 Humboldt Wind Project September 2019

Table 7. Summary of values for each parameter associated with the Probabilistic Collision Risk Model inputs (A, B, D, G-J, N-W, AA, AB, AF, AH, AJ), intermediate calculations (C, E, F, K, L, M, X, Y, Z, AC, AD, AE, ), and outputs (AI, AK, AL, AM).

60-turbines 47-turbines Parameter (2018/2019) (2018/2019) EXPOSURE AREA A Turbine string length (km) 22.4 18.3 B Turbine string height (m) 179.5 179.5 C Exposure area (km2) = A*B 4.01 3.28 D Rotor radius (m) 74.5 74.5 E Rotor swept area (m2) = pi*(D2) 17433 17422 F Proportion of area occupied by a rotor [weighted by zone for entire area] 0.0036 0.0049 EXPOSURE TIME G Proportion of time turbines rotate in morning a 0.8245 0.8245 H Proportion of time turbines rotate in evening b 0.8415 0.8415 MURRELET PASSAGE THROUGH EXPOSURE ZONE I Morning crossings (birds/km/morning) [weighted by zone for entire area] 0.175/0.242 0.109/0.173 J Evening crossings (birds/km/evening) [weighted by zone for entire area] 0.039/0.045 0.046/0.057 K Morning crossings adjusted for operational time (birds/morning/km) = G*I 0.144/0.199 0.090/0.142 L Evening crossings adjusted for operational time (birds/evening/km) = H*J 0.033/0.038 0.038/0.048 M Daily crossings adjusted for operational time (birds/day/km) = K+L 0.177/0.238 0.129/0.190 MURRELET CHARACTERISTICS N Length (cm) 25 25 O Wingspan (cm) 41 41 P Flapping or gliding flight Flapping Flapping Q Flight speed over ground (m/s) c Variable, empirically derived R Proportion of flights with wind (downwind) c 0.528/0.685 0.528/0.685 S Proportion of flights against wind (upwind) c 0.472/0.315 0.472/0.315 ROTOR CHARACTERISTICS T Blades (#) 3 3 U Blade pitch (deg) d 15 15 V Blade speed (rotations per minute) e 17 17 W Maximum blade width (m) 4.2 4.2 COLLISION PROBABILITY UPON PASSAGE THROUGH ROTOR [BAND MODEL] X Collision probability if flying with wind [D,N,O,P,Q,R,S,T,U,V,W] Varies due to Q Y Collision probability if flying against wind [D,N,O,P,Q,R,S,T,U,V,-W] Varies due to Q Z Average collision probability = (R*X)+(S*Y) Varies due to Q PROBABILITY OF AVOIDING TURBINES [MACRO/MESO/MICRO] AA Avoidance probability in April Variable, empirically derived AB Avoidance probability in May – March Identical to AA PROBABILITY OF FATALITY AC Average exposure rate (birds/day) = M*A Varies due to Q AD Passage through rotor swept area (birds/rotor/day) = AC*F Varies due to Q AE Collisions following passage through swept area (birds/rotor/day) = AD*Z Varies due to Q AF Probability of fatality following collision with rotor Varies due to Q AG Fatalities per rotor (birds/rotor/day) = AE*AF Varies due to Q AH Rotor swept areas (#) 60 47 AI Daily fatality without avoidance (birds/day) = AG*AH Varies due to Q AJ Exposure duration (days) 365 365 AK Annual fatality without avoidance (birds/year) = AI*AJ Varies due to Q AL Annual fatality with avoidance (birds/year) = AK*(1-AA in Apr & 1-AB in May-Mar) Varies due to Q, AA AM 30-year fatality estimate based on 2018/2019 ocean conditions = AL*30 Varies due to Q, AA AN 30-year fatality estimate based on the long-term average ocean condition Variable, empirically derived a Morning = -1.5 to +1.25 hr relative to sunrise; operational time derived from wind data along ridge during this time-period b Evening = -1 to +1 hr relative to sunset; operational time derived from wind data along ridge during this time-period c Empirically derived from radar surveys on ridge in 2018/2019 d Average pitch for a generic turbine e Maximum speed of rotation

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 32 Humboldt Wind Project September 2019

Section 4.0 Discussion

It is important to note that survey effort to quantify passage rates of murrelets flying along Bear River and Monument Ridges during the breeding season exceeded the quantity of effort used to develop similar estimates of collision risk for murrelets elsewhere. For example, the breeding season effort during the morning reported by Stantec (2018) was 14-times greater than the annual radar survey effort in the morning used for the CRM at Bear River Ridge (Sanzenbacher and Cooper 2010) and, if evening survey efforts for the Humboldt Wind Project are also considered, then the effort was 22-times greater. When the Humboldt Wind Energy Project’s radar sampling effort was compared to the Skookumchuck Wind Project (Sanzenbacher et al. 2015), the annual effort was 2.8-times greater and when evening flights reported by Stantec (2018) were included, it was 4.5-times greater. Finally, the effort reported by Stantec in 2018 in the morning during the breeding season was 5.5-times greater than the average effort for the Radar Ridge Wind Resource Area (Hamer Environmental 2009), and 9- times greater when evening surveys were included. The effort by Stantec in 2019 was similar. This greater level of effort was needed to stratify the sample spatiality so that the deterministic CRM could provide flexibility to develop appropriate siting of turbines and provide adequate data for development of a probabilistic model with acceptable levels of certainty.

The CRMs for Humboldt Wind differ from other CRMs for marbled murrelets in the following ways: 1. Morning radar sampling session started 45 minutes earlier than has been recommended by the Inland Survey Protocol for marbled murrelets in California (Evans Mack et al. 2003) and began 15 minutes earlier than previous radar surveys on Bear River Ridge (Cooper and Sanzenbacher 2006). This earlier start time ensured that inland flying murrelets were not missed if they happened to be flying in earlier than expected (Evans Mack et al. 2003). Mid-day radar surveys occurred on 5 occasions to assess the potential for murrelet movements at unexpected times of day. Furthermore, evening radar sampling occurred from 60 minutes before and after sunset although sampling for most murrelet CRMs do not sample in the evening. 2. Direct counts of murrelet-like targets during post-sunrise, mid-day, and evening radar surveys allowed for passage rate to be empirically measured rather than based on multipliers or assumptions typically used by CRMs. 3. Radar equipment was recently calibrated and consequently resolution could generally distinguish single murrelets such that each individual, even when traveling in close proximity to other murrelet(s), could be observed to facilitate measurement of approach angle and speed. Further, video of all radar surveys was recorded to facilitate laboratory-based examination and post-processing. 4. The robust radar sampling allowed division of the proposed turbine string into zones. The general pattern observed in each year was that passage rates of murrelets were greatest in zones that descended into adjacent river valleys that serve as primary travel corridors for murrelets accessing inland nesting habitat. 5. We used a traditional deterministic approach to estimate collision risk for murrelets as well as a more comprehensive probabilistic approach based on Bayesian inference. This latter approach allowed us to

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 33 Humboldt Wind Project September 2019

quantitatively account for variability and uncertainty inherent to certain inputs critical to the deterministic CRM and calculate an upper bound on the estimate of fatality. Although these were very distinct approaches, they produced equivalent results in that the range of estimates produces by the probabilistic CRM captured the single estimate produced by the deterministic CRM in both survey years. 6. There have been few CRMs developed explicitly for murrelets and they have primarily used the kinematic model of collision developed by Tucker (1996) or a modification of those calculations. There is a different kinematic model, the Band Model (2012), which was originally developed around the same time as Tucker (Masden 2015). Both approaches assumed that birds approach turbines in a perpendicular fashion, however only the Band Model accounts for birds flying against the wind (which will experience a higher risk of collision than birds flying with the wind due to the difference in the speed as they transit through the rotor swept area). Because the Band Model has been extensively used in Europe and has undergone more rigorous scrutiny and refinement through time relative to Tucker, we chose to use the Band Model. 7. We conducted a complimentary analysis to ensure that the assumptions made by the Band Model were valid (that the effect of approaching the turbine at various angles on collision probability would cancel out). We found that, in the case of murrelets using Bear River and Monument Ridges, the Band Model approximated the collisions probability that would have been estimated by explicitly accounting for oblique angles of entry (Appendix E). 8. The earliest CRMs had to guess regarding the ability of birds to detect and avoid collision with wind turbines. However, as more wind energy facilities have become operational through time, our understanding of the ability for birds to detect and avoid collision have improved considerably. We reviewed relevant literature as a means of conveying what is currently known about avian avoidance of wind turbines (Appendix A). Other, and older, murrelet CRMs have used a range of avoidance to estimate collision risk, including unrealistically low avoidance rates (0.9 and lower; Sanzenbacher and Cooper 2010). The conservative estimate of avoidance (0.98; SNH 2017) has been informed by a considerable number of empirical studies (see Cook et al. 2014), many in situations appropriate for extrapolation to murrelets. Because there is no evidence that a murrelet would exhibit such a drastically different avoidance relative to all other birds that have ever been studied, the deterministic CRM used 0.98. The probabilistic CRM incorporated all possible values of avoidance based on what has currently been measured for other appropriate species to identify the most likely avoidance probability and incorporate that into the overall measure of risk. 9. To facilitate collision risk estimation prior to making final decisions regarding turbine type the two CRMs assumed the largest turbine being considered will be installed at all 47 sites identified as suitable (thus overestimating risk in both models). This assumption is unrealistic, and cause collision risk to be overestimated. 10. Turbine layout was adjusted using insights from the 2018 survey year to avoid high passage areas (all turbines were removed from zones C, G, and K and the total turbines reduced from 60 to 47). This resulted in 21.6% fewer turbines, but an average (across both years) reduction in collision risk of 36%.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 34 Humboldt Wind Project September 2019

4.1 Passage rates

Murrelet passage rates in 2006 and 2007 reported in the deterministic CRM developed by Sanzenbacher and Cooper (2010) for Bear River Ridge northwest of the Humboldt Wind Project’s proposed location were estimated to be 0.4575 birds/morning/km based on an observation of 0.1525 landward birds/morning/km. We report a higher overall passage rate of 0.7052 birds/day/km averaged for June and July over 2 years (comparable modeling to Sanzenbacher and Cooper 2010). Location of radar stations 1-3 of the Sanzenbacher and Cooper (2010) study were closer to the ocean (between 9 and 13 km inland versus 19 and 33 km inland) and were lower in elevation (푥̅ = 592 m versus 759 m). Therefore passage rates observed by Sanzenbacher and Cooper (2010), although nearby, do not necessarily characterize murrelet use of ridges associated with the Humboldt Wind Project.

Additionally, Sanzenbacher and Cooper (2010) did not count murrelet-like targets following sunrise due to the risk that post-sunrise counts would be contaminated by non-murrelet targets. Instead, they doubled the landward count of pre-sunrise targets. However, murrelets moving from inland nesting habitats to the sea after sunrise may not travel along the same path or may fly at greater elevation. Thus, assuming the number of seaward flying murrelets matched the number of landward flying murrelets may have caused passage rates estimated by Sanzenbacher and Cooper (2010) to be inflated. Cooper and Sanzenbacher (2010) specifically noted in their discussion that most studies of murrelet movements have found that landward counts generally are higher than seaward counts (Cooper et al. 2001, 2005, 2006). Based on radar surveys by Stantec in 2018, only 23% of ridge crossing targets for which direction could be assigned with certainty were flying seaward. These data suggest that the appropriate correction factor for birds flying in the risk zone should have been 1.239 rather than the 2 assumed by Sanzenbacher and Cooper (2010).

Furthermore, Sanzenbacher and Cooper (2010) assumed that 50% of targets contained two murrelets. There is no evidence that this would be the case; during incubation, both members of a breeding pairs are required to travel between the sea and nests separately as one is arriving to the nest while the other is departing and, during chick-rearing, with less than 5% of fish-deliveries being made breeding pairs traveling to the nest together (Nelson and Hamer 1995). Therefore, assuming that 50% of targets contain two birds may greatly inflate passage rate. Based on radar surveys by Stantec in 2018, only 12.8% of all 47 targets observed along the ridges and in the Eel River Valley were 2-bird targets. For comparison purposes, we can use this percentage to generate a correction factor of`1.128 rather than the 1.5 assumed by Sanzenbacher and Cooper (2010).

If we were to apply these empirically derived correction factors of seaward flying birds and multi-bird targets to the 0.1525 landward birds/morning/km empirically measured by Sanzenbacher and Cooper (2010), then the passage rate would have been 0.2131 birds/morning/km instead of 0.4575. Differences in passage rate could be potentially explained by numerous factors including differences in ocean conditions, changes in the breeding population, differences in foraging areas at-sea, murrelets utilize different inland travel pathways, or simply elevation and proximity to the ocean consistent with murrelet flight patterns.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 35 Humboldt Wind Project September 2019

4.1.1 Sources of error

The use of high-frequency marine radar has become a standard protocol for counting murrelets moving between at-sea foraging location and inland nesting habitats (Burger 2001, Cooper et al. 2001, Manley 2006) and several studies show that radar is the most reliable method to assess local murrelet populations during the breeding season (Harper et al. 2004, Bigger et al. 2006, Cooper et al. 2006). There is very little within-site sampling error associated with marine radars in that they can reliably detect small mobile targets at distances up to 20 km distant (Beason et al. 2013), so long as they are each measuring the same air space. Because objective speed-based criteria were used distinguish murrelets or murrelet-like targets from other types of targets and very accurate measures of ground speed were paired with concurrent measures of wind speed, observer error associated with the passage rate calculated here should be small. Therefore, radar-based measures of passage along Bear River and Monument Ridges probably reflect true variation in passage rate rather than systematic within-study sampling error arising from methodology. For murrelets, variability in passage rates would result primarily by seasonal changes inland activity associated with breeding and the degree to which individuals utilize various flight paths to travel to and from inland nesting habitat.

Although within-study sampling error should be small, assuming consistent techniques were applied across surveys, there can be discrepancies between studies even when they occur at the same site. This is because the number of murrelets detected by radar can vary as a function of the tilt of the radar antenna (thus measuring different air space; Harper et al 2004). For this study, murrelets passing at altitudes greater than 210 m above the ridgeline were not considered at risk of colliding with a turbine and were not relevant to the calculations. Instead, radars were appropriately positioned to detect murrelets flying at or below the risk zone. This positioning of the radar to focus on the rotor swept zone may cause the total number of detections associated with this study to not be comparable to other studies intending to document total murrelet use of a given area. This is particularly true if the flight height distribution of murrelets using the study area are similar to what has been found elsewhere; more than 55% of 287 murrelets were observed flying above 225 m near the Queets River (Stumpf et al. 2011) and an average flight height of 307.9 m above ground was reported for Radar Ridge (Hamer Environmental 2009).

Estimates of passage can also be affected by the quality of the radar site (Hamer Environmental 2009); radars were positioned at the best available location to minimize blind spots due to topography and vegetation. Blind spots were distributed such that they were not concentrated in any one area of the ridge. We assumed that the passage rate measured in visible sections of each zone could approximate passage rate in non-visible sections of each zone.

Estimates of passage can be affected by ability of measurements to reflect the full activity period of murrelets. Murrelets have a very predictable and limited time of day in which they travel between the ocean and nesting habitat, with activity peaking around dawn and dusk (note that travel is not constant through the period; rather inland flight rate rises to a peak near dawn and rapidly decreases after dawn; Stumpf et al. 2011, Lorenz et al. 2019). The duration of surveys along Bear River and Monument Ridges in 2018, which was fully inclusive of

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 36 Humboldt Wind Project September 2019

the known activity periods as well as mid-day surveys to detect any potential aberrant movements, was specifically intended to completely sample inland activity of murrelets.

Estimates of passage can also be affected by the inability to distinguish other similarly sized fast-flying birds from murrelets that become active after sunrise (often called contamination). Cooper et al. (2001) reported 22% of targets that met radar criteria as a murrelet-type target (flying above the threshold speed of 64 km/hr) were visually confirmed to be other species (including cormorants, waterfowl, shorebirds, band-tailed pigeons and some raptors). Misidentification rates were greater for seaward flying targets relative to landward flying ones. This is likely the primary error associated with the passage rates of murrelets calculated for Bear River and Monument Ridges in 2018 and 2019 since any target that appeared to be a murrelet was counted as a murrelet. This would result in an inflated passage rate of murrelets and, consequently, an inflated collision risk estimate.

4.1.2 Seasonal variability in passage rate

Passage rates varied across the breeding season in a manner consistent with the telemetry study conducted 110 km north of the Project area in Redwood National and State Parks (Hébert and Golightly 2006). Thus, the frequency of radar surveys was adequate to produce a similar resolution in the data.

Passage rate was greatest in June 2018 and May 2019. There are several explanations for why this would be the case. For one, seabirds (and especially Alcids nesting in the Pacific Northwest) are known to adjust the timing of nest initiation in response to ocean conditions; for example, up to 71% of the variability in the timing of nest initiation by murres could be explained by the timing of the seasonal transition to upwelling at a nearby site in northern California (Schneider 2018). In 2018, the cumulative upwelling index estimated at 42oN (the latitude closest to the Eel River for which upwelling is measured; NOAA PFEL 2019) became net-positive on 29 January which is 42.2 days earlier than the 11-year average (2007-2017; Schneider 2018) and the second earliest record in this time-series. Thus, it is likely that nest initiation by many species of seabirds was relatively early in 2018. Early nesting may result in greater passage rates earlier in the season and very little passage late in the season once most individuals have either fledged (or failed to fledge) a chick. This will differ between years. Highest passage occurred earlier in 2019 which was consistent with 2019 being an above average year based on ocean conditions.

Most murrelet CRMs, except for Radar Ridge (Hamer Environmental 2009), did not measure murrelet passage rate in the winter and rather used a series of assumptions to estimate winter passage. For the collision risk estimated presented here, we used the empirical measurements but for comparison we also made calculations that followed the logic of previous CRMs regarding winter passage.

4.1.3 Daily variability in passage rate

As expected, passage rates across the project area were much greater during the morning activity period relative to the evening; evening passage rates here were 25% in 2018 and 24% in 2019 of the overall passage rate.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 37 Humboldt Wind Project September 2019

Interestingly, Hébert and Golightly (2006) also reported that 28% of flights occurred during the evening activity period.

Evening flights are thought to be associated with chick feeding events, and thus do not occur for the duration of the nesting period but are rather limited to months where chicks need to be fed. However, 19% of evening passage occurred in April, which does not align with the chick-rearing period. Evening flights were not detected in May, August, or September. Most of the evening flights occurred in June and July consistent with an early nest initiation (e.g., April-May). If nest initiation was early or nest failure was frequent such that only a few nests remained active into August and September, evening flights inland would be relatively rare when compared to earlier in the season when a greater percent of the population were actively feeding young.

4.1.4 Topographical variability

Passage over the proposed turbine string was non-random. Topography that facilitated reduced energy costs of transportation by birds can cause those birds to concentrate as they use passes to fly over ridges. There appears to be such concentrations associated with the topography in the 2018 and 2019 passages.

4.2 Collision rates

Based on the radar data collected during the breeding season and inland activity during the non-breeding season, the fatality estimated for this Project by the deterministic CRM was 5.27 or 7.38 birds over the 30-year life of the Project (or 1 bird every 5 or 4 years; 2018 or 2019 respectively). The average collisions based on the deterministic CRMs was 6.1 birds in 30 years.

For the probabilistic CRM, the one-tailed upper threshold that contains 95% of the probability distribution was 6.17 and 7.77 bird collisions in 30 years (2018 and 2019, respectively). The upper bound used here is more conservative than the 80% HDI (which corresponds to a 10% chance that the actual fatality will exceed the estimate rather than 5% as is used here) used by the USFWS Eagle Conservation Plan’s (2013) risk-adverse method of predicting eagle fatalities at wind energy facilities using Bayesian inference for purposes of permitting take and assessing mitigation needs (see New et al. 2015). Importantly, the estimates generated by the deterministic CRM are lower than the one-tailed upper threshold that contains 95% of the probability distribution resulting from the probabilistic CRM, indicating that the two models support each other. In addition to validating the outcome of each approach, this suggests that both models are probably reliable for their different purposes.

Estimates produced by both the deterministic and probabilistic CRMs are conservative and likely over-estimate risk to murrelets because they both account for a greater rotor-swept area than will be included in the final Project. They both assume that these turbines will operate at their maximum rotation rate of 17 rpms. Further, total targets may have included some contamination with birds that were not murrelets but were still included in the estimate of murrelet passage rate. Other scenarios such as smaller turbines or fewer turbines will reduce both estimates provided here.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 38 Humboldt Wind Project September 2019

4.3 Avoidance rates

The actual ability of murrelets to avoid collision with onshore turbines during their inland transits will remain unknown until they are exposed to a wind energy facility. However, what is known about murrelets combined with other appropriate birds that have been exposed to wind facilities suggests that the value used in the deterministic model is conservative. Murrelets are (1) very capable fliers, (2) restrict inland transits to periods when light intensity would enable visual detection of obstacles, (3) likely adapted to see well even in low-light conditions as they are visual predators that pursue prey underwater, and (4) their distractions from navigation should be few as they transit through the Project area (Appendix A). Thus, murrelets should exhibit avoidance similar to what has been documented elsewhere in other avian species. To this end, we conducted an extensive review of best available estimates of avoidance capabilities for birds, with a focus on seabirds at onshore wind facilities (Section 2.2.5.2 and Appendix A).

4.4 Interannual variability in inland activity

Data acquisition for other CRMs for murrelets have generally relied on two (or a maximum of 3) breeding seasons of sampling with radar to assess annual variation in passage rate. Passage rates determined over the 2- year period are typically averaged and input into a model to produce a single estimate of annual collisions. These other CRMs then multiplied the annual estimate by 30 to estimate the number of collisions expected over 30 years. For practical purposes, their 30-year estimate based on two breeding seasons was presumed to account for interannual variability in risk.

The CRMs presented here took a novel approach to estimate potential variability in interannual activity. Breeding effort of many seabirds is linked to ocean conditions indicative of productivity and prey availability. Because most inland activity by murrelets results from the need to access nests, inland activity is directly influenced by breeding effort. Annual variability in murrelet passage rate in the Eel River Valley over a 10-year period (2002-2011) was associated with ocean productivity (Appendix B), and winter PDO was the best explanatory variable (February PDO had the greatest ability to account for variability in murrelet inland activity). This relationship between murrelet activity and ocean conditions was then used to predict how activity would vary across a timeframe that was longer than 2 years. This predictive relationship facilitated adjustment of collision estimates produced by the deterministic CRM to reflect variability in relevant ocean conditions over a 35-year period and the probabilistic CRM to reflect variability in relevant ocean conditions over a 118-year period.

Because 2018 was average in terms of the February PDO, the average estimate of collision produced when considering long-term variability in ocean productivity was very similar to what was estimated by simply multiplying risk estimates from 2018 by 30. Conversely, the 2019 February PDO was indicative of above average conditions for reproduction, and thus greater numbers of inland flights (which were also documented in the radar measure of activity). Consequently, the CRM utilized an average number of flights for one year and an above average number of flights for the other. If the PDO had accounted for 100% of the interannual

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 39 Humboldt Wind Project September 2019

variability, the adjustment in the model would have calculated the two estimates of collision to be approximately the same. However, because the PDO did not explain 100% of the variability, both numbers become differing estimates of the actual collision rate. Still, the difference between these calculated numbers is small compared to the yearly differences for other CRMs. If averaged between years, as has been done elsewhere, the annual collision estimate would be 6.1 birds in 30 years.

The approach that incorporated long term measures of ocean productivity to assess potential interannual variability more accurately estimated longterm risk relative to the current practice of using two years of data try and assess interannual variability in passage rate. Ocean productivity is a cyclical phenomenon that operates over various time scales, including decades (Barber and Chavez 1993, Jacobs et al. 1994). If the present study had occurred during 2016 and 2017, a period of very warm and anomalous waters in northern California where many seabirds exhibited low breeding effort and poor reproduction, then passage rates calculated from these two years would likely be anomalously low and would have greatly underestimated overall risk.

4.5 Potential indirect effects

Direct effects are the loss of individuals from the population following a collision and are estimated by the CRMs presented here. Indirect effects are the impacts of this loss to other individuals in the population. The most recent population estimate for murrelets nesting in Zone 4, which encompasses the northern California population was 8546 individuals (95% CI: 6277-11311; Pearson 2018). The direct loss over 30 years of either 5.27 birds in 2018 (based on the deterministic CRM) or an upper estimate of 6.17 birds in 2018 (based on the upper 95% HDI of the probabilistic CRM) is small relative to the total population size and unlikely to impact the viability of the local population. The relative comparison to the population size was similar for the estimates from 2019.

We estimated the magnitude of additional indirect losses that could occur following a collision between a murrelet and the wind turbines associated with the Project. If a murrelet that collided with the turbine was incubating an egg or feeding a chick, it would be unlikely that the surviving mate would be able to successfully complete the nesting attempt. Thus, that egg or chick (hereafter, nesting attempt) would perish. Of the 5.27 or 6.85 (2018 or 2019) collisions forecast over 30 years by the deterministic CRM, about 80-86% of collisions would occur during the April to August reproductive period (Table 3). Using the 86% from 2019 and making the assumption that every bird struck by a turbine was a breeder with an active nest in 2019, then 5.89 nesting attempts at maximum would also be affected over 30 years.

The 5.89 nests greatly overestimates the potential indirect loss and must be modified to reflect the reality that only a proportion of inland flying murrelets participate in nesting. In central California, Peery et al. (2004a) reported that only 31% (range: 11-50% ) of inland flying murrelets had established nests. In far northern California (Redwood National and State Parks), Hébert and Golightly (2006) reported that 59% (range: 44- 73%) of inland flying murrelets had established a nest. Thus, the actual number of nesting attempts impacted

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 40 Humboldt Wind Project September 2019

over 30 years would be 1.82-3.47 (depending on whether murrelets that fly inland in the vicinity of the Project are more like birds in central California or in far northern California, respectively).

This adjusted estimate is still inflated because it does not account for the loss of nests due to predation. A primary cause of nest failure in California has been predation in recent decades as the majority of murrelet nesting habitat available in California has been protected. Based on the nest failure rates documented across California, predation rates in a given year could be greater than 90% (90.8-96.9%; Hébert and Golightly 2006) or as low as 67-81% (Peery et al. 2004a); due to the great range of potential predation rates observed by various studies in California, a moderate 71.4% failure rate due to predation reported in Golightly and Schneider (2011) was used as a realistic estimate of predation rate to account for current levels of nest predation to estimate the indirect effects of wind turbines. If we assume that 71.4% of nests failed due to predation, then the expected number of nesting attempts impacted over 30 years would be 0.38-0.99 (in central California or northern California, respectively).

This calculation of 0.38-0.99 nesting attempts impacted over 30 years is likely still inflated due to the risk of breeding individuals being wounded or killed by a predatory attack as they transit between the sea and inland nesting habitat on a daily basis while incubating eggs and feeding young; the proportion of inland flying birds that would be subject to predation while flying inland is unknown and cannot be accounted for in this analysis.

Finally, once a chick fledges, survivorship is considerably lower than adult survivorship and the new bird must survive for a minimum of 2-3 years before it is able to enter the breeding population and produce their own young. Assuming the survival rate of juveniles to age 1 is 0.605, survival from age 1 to 2 is 0.864 (Peery and Beissinger 2007), and the age of sexual maturity is age 2, then the 0.38-0.99 estimate would be further reduced to 0.20-0.52 nesting attempts that survive to reproductive age (for central California or northern California, respectively). If the age of sexual maturity is 3 rather than 2, the number of nesting attempts would be 0.17- 0.45. From this analysis we conclude that the total indirect impact to the production of new breeders could range from 0.17 to 0.45 breeding adults in 30 years.

A second potential consequence is the length of time that the surviving mate will take to find a new mate. There are no data on whether that can occur the next year or if it might take multiple seasons. Theoretically they can find their first mate by the time they reach sexual maturity, which is by 2 or 3 years of age, and this may be a reasonable estimate of the time it would take to replace a lost mate.

4.6 Conclusions

The two CRMs were designed to produce a conservative over-estimate of collision risk for murrelets transiting through the turbine string proposed along Bear River and Monument Ridges. In general, we conclude that there is concurrence among the two models and that the estimates produced by the probabilistic CRM overlapped estimates produced by the deterministic CRM. Indirect effects could add to the 30-year estimates, but their overall impact is very small.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 41 Humboldt Wind Project September 2019

The deterministic model predicted that, based on observations in 2018 and 2019, 5.26 and 7.38 murrelets would be impacted over the 30-year life of the Project. When interannual variability in at-sea conditions predictive of inland activity were accounted for, the deterministic model predicted that 5.27 or 6.85 murrelets (2018 or 2019) would be impacted over a 30-year period (Table 4). Over 30-years, the probabilistic model predicted that the mean number of murrelets impacted would be 2.94 or 3.71 individuals and the maximum would be 6.17 or 7.77 (this is the one-tailed upper bound threshold that contains 95% of the probability density for 2019 and 2019, respectively). Thus, the most precautionary estimate of take of marbled murrelets across the 30 years of the Project would be 7.77 birds in 30 years. This maximum expected loss will be 0.26 birds per year, which is very small compared to the at-sea population of 8574 birds (95% CI: 6358-11255 birds in 2017) estimated in the zone offshore of the proposed mitigation (Zone 4; numbers from McIver et al. 2019). Thus, there is little likelihood of population level effects from this amount of take, especially when coupled with a fully compensatory mitigation strategy.

The actual number of collisions will be less than what has been estimated here because of the use of highly conservative model inputs. When the passage rate data was stratified by month, it approximated other measures or indexes of murrelet activity using other methodologies (e.g. telemetry or audio-visual surveys); this observation is consistent with the model having good resolution of seasonal changes in passage rate. This good resolution was due in part to the intense effort of the radar sampling for murrelets. The results presented in this report are consistent with earlier reports of collision risk and consistent with conclusions of significance and associated mitigation measures provided within the Draft EIR for the project.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 42 Humboldt Wind Project September 2019

Section 5.0 Acknowledgments

We are thankful to Y. Akky, B. Roy, and E. DeVost of Stantec for providing insights into the collection of radar data and advice on the CRMs. We especially appreciate the regular discussion with and expertise of B. Roy for interpreting the radar images and in developing the sampling scheme. We also thank the U.S. Fish & Wildlife Service (L. Roberts, W. McIver, J. Norris, D. Everson, K. Brubaker, and G. Schrott) and California Department of Fish and Wildlife (J. Olson, M. van Hattem, and G. Leppig) for constructive input throughout the development of the conceptual framework of this approach. We are especially grateful for the advice and critical input of B. Furnas (California Department of Fish and Wildlife) as we developed the probabilistic model.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 43 Humboldt Wind Project September 2019

Section 6.0 References

Abernethy, R.B., J.E. Breneman, C.H. Medlin, and G.L. Reinman. 1983. Weibull analysis handbook. Report AFWAL-TR-83-2079 to the Wright Patterson Air Force Base Aero Propulsion Laboratory, Dayton, OH. 228 pp. Adrean, L.J., S.K. Nelson, M.S. Garcia-Heras, D.D. Roby, M.G. Betts, and J.W. Rivers. 2019. Factors associated with marbled murrelet (Brachyramphus marmoratus) nesting success in western Oregon. Pacific Seabird Group 46th annual meeting, Kaua’i, HI. Ainley D.G. and R.J. Boekelheide. 1990. Seabirds of the Farallon Islands: ecology, dynamics, and structure of an upwelling-system community. Stanford (CA): Stanford University Press. Aschwanden, J., H. Stark, D. Peter, T. Steuri, B. Schmid, and F. Liechti. 2018. Bird collisions at wind turbines in a mountainous area related to bird movement intensities measured by radar. Biological Conservation 220:228-236. Baerwald, E.F., G.H. D’Amours, B.J. Klug, R.M.R. Barclay. 2008. Barotrauma is a significant cause of bat fatalities at wind turbines. 18:R695-R696. Bakun, A. 1973. Coastal upwelling indices, west coast of north America 1946-71. NOAA Technical Report NMFS SSRF-671. Seattle, WA. 103 p. Band, W. 2012. Using a Collision Risk Model to assess bird collision risks for offshore windfarms. Strategic Ornithological Support Services Steering Group Report. 62 pp. Band, W., M. Madder, and D.P. Whitfield. 2007. Developing field and analytical methods to assess avian collision risk at wind farms. Pages 259-271 in Birds and wind farms: risk assessment and mitigation (M. de Lucas, G.F.E. Janss, M. Ferrer, Eds.). Quercus, Madrid. Barber, R.T. and F.P. Chavez. 1983. Biological consequences of El Niño. Science 222:1203-1210. Barrios, L. and A. Rodriguez. 2004. Behavioral and environmental correlates of soaring-bird mortality at onshore wind turbines. Journal of Applied Ecology 41:72-81. Beardall, J., E. Young, and S. Roberts. 2001. Approaches for determining phytoplankton nutrient limitation. Aquatic Sciences 63:44-69. Beason, R.C., T.J. Nohara, and P. Weber. 2013. Beware of the Boojum: Caveats and strengths of avian radar. Human-Wildlife Interactions 7:16-46. Becciu, P. M.H.M. Menz, A. Aurbach, S.A. Cabrera-Cruz, C.E. Wainwright, M. Scacco, M. Ciach, L.B. Pettersson, I. Maggini, G.M. Arroyo, J.J. Buler, D.R. Reynolds, and N. Sapir. Environmental effects on flying migrants revealed by radar. Ecography 42:942-955. DOI: 10.1111/ecog.03995. Belant, J.L. 1997. Gulls in urban environments: Landscape-level management to reduce conflict. Landscape and Urban Planning 38:245-258. Bevanger, K., F. Bernsten, S. Clausen, E.L. Dahl, O. Flagstad, A. Follestad, D. Halley, F. Hanssen, P.L. Hoel, L. Johnsen, P. Kvaloy, R. May, T. Nygard, H.C. Pedersen, O. Reitan, Y. Steinheim, R. Vang. 2008. Pre- and post-construction studies of conflicts between birds and wind turbines in coastal Norway. NINA Report 409. Norwegian Institute for Nature Research, Norway. 55 pp. Bevanger, K., F. Berntsen, S. Clausen, E.L Dahl, E.L., Ø. Flagstad, A. Follestad, D. Halley, F. Hanssen, L. Johnsen, P. Kvaløy, P. Lund-Hoel, R. May, T. Nygård, H.C. Pedersen, O. Reitan, E. Røskaft, Y. Steinheim, B. Stokke, and R. Vang. 2010. Pre- and post-construction studies of conflicts between birds and wind turbines in coastal Norway (BirdWind): Report on findings 2007-2010. NINA Report 620. Norwegian Institute for Nature Research, Norway. 152 pp. Bigger, D., M.Z. Peery, J. Baldwin, S. Chinnici, and S.P. Courtney. 2006. Power to detect trends in marbled murrelet breeding populations using audiovisual and radar surveys. Journal of Wildlife Management 70:493–504. Brook, M.D.L. 1989. Determination of the absolute visual threshold of a nocturnal seabird, the common diving petrel Pelecanoides urinatrix. Ibis 131:290-300. Burger, A.E. 1997. Behavior and numbers of marbled murrelets measured with radar. Journal of Field Ornithology 68:208-223.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 44 Humboldt Wind Project September 2019

Burger. A.E. 1999. Responses by marbled murrelets to variable ocean temperatures off southwest Vancouver Island. Proceedings of Biology and Management of Species and Habitats at Risk, Kamloops, B.C: 723- 732. Burger, A.E. 2001. Using radar to estimate populations and assess habitat associations of marbled murrelets. Journal of Wildlife Management 65: 696–715. Burnham, K.P. and K.P. Anderson. 2002. Model selection and multimodel inference: A practical information- theoretic approach (2nd edition). Springer-Verlag New York, NY. 488 pp. Carter, H.R. and R.A. Erickson. 1992. Status and conservation of the marbled murrelet in California, 1892- 1987. Pages 92-108 in Status and conservation of the marbled murrelet in North America (H.R. Carter and M.L. Morrison, Eds.). Proceedings of the Western Foundation of Vertebrate Zoology, Camarillo, CA. Carter, H.R. and S.G. Sealy. 1986. Year-round use of coastal lakes by marbled murrelets. Condor 88:473-477. Chamberlin, D.S. Freeman, and M. Rehfisch. 2005. Appraisal of Scottish Natural Heritage’s wind farm collision risk model and its application. British Trust for Ornithology Research Report 401. 53 pp. Chiaradia, A., J. McBride, T. Murray, P. Dann. 2007. Effect of fog on the arrival time of little penguins Eudyptula minor: a clue for visual orientation? Journal of Field Ornithology148. DOI: https://doi.org/10.1007/s10336-007-0125-5. Christie, D. and B. Urquhart. 2015. A refinement of the Band spreadsheet for wind turbine collision risk allowing for oblique entry. New Zealand Journal of Zoology 2015:1-8. Cook, A.S.C.P, E.M. Humphreys, E.A. Masden, and N.H.K. Burton. 2014. The avoidance rates of collision and offshore wind turbines. British Trust for Ornithology Research Report 656. 244 pp. Cook, A.S.C.P, E.M. Humphreys, F. Bennet, E.A. Masden, and N.H.K. Burton. 2018. Quantifying avian avoidance of offshore wind turbines: current evidence and key knowledge gaps. Marine Environmental Research 140: 278-288. Cooper, B.A. 2010. Memorandum regarding field and data summary for Duckabush avoidance study (Attachment B of Appendix J). In: TetraTech EC. Final Environmental Impact Statement for the Everpower Coyote Crest Wind Park. Chehalis, WA: Lewis County Community Development Cooper, B.A. and R.J. Blaha. 1999. Comparison of radar and audio-visual counts of marbled murrelets during inland forest surveys. Wildlife Society Bulletin 30: 1182-1194. Cooper, B.A. and R.H. Day. 2004. Modeling annual seabird use and fatality at the proposed Kaheawa Pastures Wind Energy Facility, Maui Island, Hawaii, fall 2004. Unpublished report from ABR, Inc., Forest Grove, OR to Kaheawa Windpower LLC, Makawao, HI and UPC Wind Management LLC, Newton, MA. Cooper, B.A. and T.J. Mabee. 2010. A radar study of marbled murrelets at the proposed Coyote Crest Wind Power Project, summer 2008 and 2009. Unpublished report from ABR, Inc., Forest Grove, OR to TetraTech EC, Portland, OR and Everpower Wind Holdings Inc, Portland, OR. 18 pp. Cooper, B.A. and P.M. Sanzenbacher. 2006. A radar study of marbled murrelets at the proposed Bear River Windpark, California, Summer 2006. Unpublished report from ABR, Inc., Forest Grove, OR to Shell Wind Energy, Inc., Houston, TX. 18 pp. Cooper, B. A. and P. M. Sanzenbacher. 2006. A radar study of marbled murrelets at the proposed Bear River Windpark, California, Summer 2006. Unpublished report prepared for Shell Wind Energy, Inc., Houston, Texas by ABR, Inc., Forest Grove, OR. 18 pp. Cooper, B.A., M.G. Raphael, and D.M. Evans. 2001. Radar-based monitoring of marbled murrelets. Condor 103: 219–229. Cooper, B. A., R. J. Blaha, and J. H. Plissner. 2005. Determining local population trends of marbled murrelets in that portion of Conservation Zone Four North of the Palco HCP Area. 2004. Unpublished. Report prepared for Pacific Lumber Company (Palco), U.S. Fish & Wildlife Service, and California Department of Fish and Game, Scotia, CA, by ABR, Inc., Forest Grove, OR. 26 pp. Corman, A.M., and S. Garthe. 2014. What flight heights tell us about foraging and potential conflicts with wind farms: A case study in Lesser Black-backed Gulls (Larus fuscus). Journal of Ornithology 155:1037-1043. Croll, D.A., B. Marinovic, S. Benson, F.P. Chavez, N. Black, R. Ternullo, B.R. Tershy. 2005. From wind to whales: Trophic links in a coastal upwelling system. Marine Ecology Progress Series 298:117-130.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 45 Humboldt Wind Project September 2019

Dahl, E.L., R. May, P.L. Hoel, K. Bevanger, H.C. Pedersen, E. Røskaft, and B.G. Stokke. 2013. White-tailed eagles (Haliaeetus albicilla) at the Smøla Wind-Power Plant, Central Norway, lack behavioral flight responses the wind turbines. Wildlife Society Bulletin 37:66-74. De Santo, T.L. and S.K. Nelson. 1995. Comparative reproductive ecology of the auks (family Alcidae) with emphasis on the marbled murrelet. Pages 33-47 in Ecology and Conservation of the Marbled Murrelet, General Technical Report PSW-GTR-152 (C.J. Ralph, G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station, Albany, CA. DelaSalla, D.A., P. Brant, M. Koopman, J. Leonard, C. Meisch, P. Herzog, P. Alaback, M.I. Goldstein, S. Jovan, A. MacKinnon, and H. von Wehrden. 2015. Climate change may trigger broad shifts in North America’s Pacific coastal rainforests. Reference Module in Earth Systems and Environmental Sciences. DOI: http://dx.doi.org/10.1016/B978-0-12-409548-9.09367-2. Delignette-Muller, M.L. and C. Dutang. 2015. An R Package for Fitting Distributions: fitdistrplus. Journal of Statistical Software 64:1-34. URL http://www.jstatsoft.org/v64/i04/ [Accessed 11 April 2019] Desholm, M. 2006. Wind farm related mortality among avian migrants – a remote sensing study and model analysis. Ph.D. Thesis. National Environmental Research Institute, Ministry of the environment, Denmark. Desholm, M. and J. Kahlert. 2005. Avian collision risk at an offshore wind farm. Biological Letters 1:296–298. Desholm, M., A.D. Fox, P.D.L. Beasley, and J. Kahlert. 2006. Remote techniques for counting the number of bird-wind turbine collisions at sea: a review. Ibis 148:76-89. Di Lorenzo, E., N. Schneider, K.M. Cobb, P.J.S. Franks, K. Chhak, A.J. Miller, J.C. McWilliams, S.J. Bograd, H. Arango, E. Curchitser, T.M. Powell, and P. Rivière. 2008. North Pacific Gyre Oscillation links ocean climate and ecosystem change. Geophysical Research Letters 35:1-6. Di Lorenzo, E, G. Liguorio, N. Schneider, J.C. Furtado, B.T. Anderson, and M.A. Alexander. 2015. ENSO and meridional modes: a null hypothesis for Pacific climate variability. Geophysical Research Letters 42:9440–9448. Dusenbery, D.B. 1992. Sensory ecology: How organisms acquire and respond to information. W.H. Freeman and Company, New York, NY. 558 pp. Ecology Consulting. 2012. Hellrigg Wind Farm: Goose refuge monitoring report winter 2011-12. Report to RWE Npower Renewables. Ecology Consulting. 2013. Hellrigg Wind Farm: Goose refuge monitoring report winter 2012-13. Report to RWE Npower Renewables. Eisenhawer, A.E. and T.E. Reimchen. 1990. Inland flight patterns of marbled murrelets, Brachyramhus marmoratus, on the Queen Charlotte Islands British Columbia. Canadian Field-Naturalist 104:439-444. Elliot, K.H., M. Hewett, G.W. Kaiser, and R.W. Blake. 2004. Flight energetics of the marbled murrelet, Brachyramphus marmoratus. Canadian Journal of Zoology 82:644-652. Erickson, W., G. Johnson, D. Young, D. Strickland, R. Good, M. Bourassa, K. Bay, and K. Sernka. 2002. Synthesis and comparison of baseline avian and bat use, raptor nesting and mortality information from proposed and existing wind developments. Report by Western EcoSystems Technology Inc. Erickson, W., J. Jeffrey, K. Kronner, and K. Bay. 2004. Stateline wind project wildlife monitoring final report, July 2001 - December 2003. Technical report by Western EcoSystems Technology Inc. and Northwest Wildlife Consultants Inc. Evans Mack, D., W.P. Ritchie, S.K. Nelson, E. Kuo-Harrison, P. Harrison, and T.E. Hamer. 2003. Methods for surveying marbled murrelets in forests: a revised protocol for land management and research. Marbled Murrelet Technical Committee, Pacific Seabird Group. Everaert, J. and E. Kuijken. 2007. Wind turbines and birds in Flanders (Belgium). Preliminary summary of the mortality research results. Research Institute for Nature and Forest, Brussels. Everaert, J. and E.W.M. Stienen. 2007. Impact of wind turbines on birds in Zeebrugge (Belgium): significant effect on breeding tern colony due to collisions. Biodiversity and Conservation 16: 3345–3359. Falkowski. P.G., D. Ziemann, Z. Kolber, and P.K. Bienfang. 1991. Role of eddy pumping in enhancing primary production in the ocean. Nature 352:55-58.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 46 Humboldt Wind Project September 2019

Fernley, J., S. Lowther, and P. Whitfield. 2006. A review of goose collisions at operating wind farms and estimation of the goose avoidance rate. A report by Natural Research Ltd, West Coast Energy and Hyder Consulting. 18 pp. Flemming, R.A. 2001. The Weibull model and an ecological application: describing the dynamics of foliage biomass on Scots pine. Ecological Modeling 138:1-3. Furnace, R.W. 2013. Review of avoidance rates for seabirds. Appendix 3 in Review of avoidance rates in seabirds at offshore wind farms and applicability of use in the Band collision risk model—December 2013. Prepared by MacArthur Green Ltd., Glasgow, for the Hornsea Offshore Windfarm Project Two, United Kingdom. 16 pp. Furness, R.W., H.M. Wade, and E.A. Masden. 2013. Assessing vulnerability of marine bird populations to offshore wind farms. Journal of Environmental Manag.ment1 19:56–66. Graham, M.H. 2003. Confronting multicollinearity in ecological multiple regression. Ecology 84:2809-2815. García-Reyes, M. and W.J. Sydeman. 2017. California Multivariate Ocean Climate Indicator (MOCI) and marine ecosystem dynamics. Ecological Indicators 72: 521-529. Garrison, T. and R. Ellis. 2016. Oceanography: An invitation to marine science. Cengage Learning, Boston, MA. 605 pp. Gaston, A.J. and S.B. Dechesne. 1996. Rhinoceros auklet (Cerorhinca monocerata), Ver 2.0. In The Birds of North America (A. F. Poole and F. B. Gill, Editors). Cornell Lab of Ornithology, Ithaca, NY, USA. doi: 10.2173/bna.212. Gaston, A. J. and D. N. Nettleship, 1982. Factors determining seasonal changes in attendance at colonies of the thick-billed murre Uria lomvia. Auk 99:468-473. Gelman, A., J.B. Carlin, H.S. Stern, D.B. Dunson, A. Vehtari, and D.B. Rubin. 2013. Bayesian data analysis (3rd ed.). CRC Press, Boca Raton, FL. Gladics, A.J., R.M. Suryan, J.K. Parish, C.A. Horton, E.A. Daly, and W.T. Peterson. 2015. Environmental drivers and reproductive consequences of variation in the diet of a marine predator. Journal of Marine Systems 146:72-81. Golightly, R.T. and S.R. Schneider. 2009. Year 8 of a long-term monitoring effort at a marbled murrelet nest in northern California. Unpublished Report by Humboldt State University Department of Wildlife, Arcata, CA. Golightly, R.T. and S.R. Schneider. 2011. Years 9 and 10 of a long-term monitoring effort at a marbled murrelet nest in northern California. Unpublished Report by Humboldt State University Department of Wildlife, Arcata, CA. Golightly, R.T., C.D. Hamilton, and P.N. Hébert. 2009. Characteristics of marbled murrelet (Brachyramphus marmoratus) habitat in northern California. Unpublished Report by Humboldt State University Department of Wildlife, Arcata, CA. Gordon, C.E. and C.S. Nations. 2016. Collision Risk Model for “rufa” red knots (Calidris canutus rufa) interacting with a proposed offshore wind energy facility in Nantucket Sound, Massachusetts. Department of the Interior, Bureau of Ocean Energy Management, Sterling, Virginia. OCS Study Bureau of Ocean Energy Management 2016-045. 90 pp. Graham, M. H. 2003. Confronting multicollinearity in ecological multiple regression. Ecology 84:2899-2815. Grémillet, D., G. Kuntz , C. Gilbert , A.J. Woakes , P.J. Butler, and Y. le Maho. 2005. Cormorants dive through the Polar night. Biology Letters 1:469-471. Grünkorn, T., J. Blew, O. Krüger, A. Potiek, M. Reichenbach, J. von Rönn, H. Timmermann, S. Weitekamp, and G. Nehls. 2017. A large-scale, multispecies assessment of avian mortality rates at land-based wind turbines in northern Germany. Pages 43-65 in Wind Energy and Wildlife Interactions (Köppel, J., Ed.). Springer International Publishing, Berlin, Germany. Hall, M.I. and C.F. Ross. 2007. Eye shape and activity pattern in birds. Journal of Zoology 271:437-444. Hamer Environmental. 2009. Use of Radar to Determine Passage Rate and Height Distributions of Marbled Murrelets at the Proposed Radar Ridge Wind Resource Area, Pacific County, WA. Prepared by Hamer Environmental, Mt. Vernon, WA, for Energy Northwest, Richland, WA. Hamer, T.E. and S.K. Nelson. 1995. Nesting chronology of the Marbled Murrelet. Pages 49-56 in Ecology and conservation of the Marbled Murrelet, General Technical Report PSW-GTR-152 (C.J. Ralph, G.L.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 47 Humboldt Wind Project September 2019

Hunt, Jr., M.G. Raphael, and J.F. Piatt, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station, Albany, CA. Hamer, T.E., B.A. Cooper, and C.J. Ralph. 1995 Use of Radar to study the movements of marbled murrelets at inland sites. Northwestern Naturalist 76:73-78. Harper, W.L, B.K. Schroder, I.A. Manly, and J.A. Deal. 2004. Direct comparison of tilted and untilted radar for monitoring Marbled Murrelet Brachyramphous marmoratus populations. Marine Ornithology 32:35- 41. Hébert, P.N. and R.T. Golightly. 2006. Movements, nesting, and response to anthropogenic disturbance of marbled murrelets (Brachyramphus marmoratus) in Redwood National and State Parks, California. Report 2006-02 by Humboldt State University Department of Wildlife, Arcata, CA to California Department of Fish and Game, Sacramento, CA. 321 pp. Hébert, P.N. and R.T. Golightly. 2008. At-sea distribution and movements of nesting and non-nesting marbled murrelets Brachyramphus marmoratus in northern California. Marine Ornithology 36:99-105. Holmstrom, L., T. Hamer, E. Colclazier, N. Dennis, J. Verschuyl, D. Ruché. 2011. Assessing avian-wind turbine collision risk: an approach angle dependent model. Wind Engineering 35:289-312. HRC [Humboldt Redwood Company]. 2018. Marbled murrelet inland monitoring annual report 2017. Report by the Humboldt Redwood Company, Scotia, CA to the U.S. Fish & Wildlife Service, Arcata, CA. 38 pp. Jacobs, G.A., H.E. Hurlbert, J.C. Kindle, E.J. Metzger, J.L. Mitchell, W.J. Teague, and A.J, Wallcraft. 1994. Decade-scale trans-Pacific propagation and warming effects of an El Niño anomaly. Nature 370:360- 363. James, P.C. 1986. How do manx shearwaters Puffinus puffinus find their burrows? Ethology 71:287-294 JISAO [Joint Institute for the Study pf the Atmosphere and Ocean]. 2019. PDO Index. URL http://research.jisao.washington.edu/pdo/PDO.latest.txt [Accessed 9 September 2019]. JNCC [Joint Nature Conservation Committee]. 2014. Joint Response from the Statutory Nature Conservation Bodies to the Marie Scotland Science Avoidance Rate Review. 8 pp. Johnson, G.D., W.P. Erickson, M.D. Strickland, M.F. Shepherd, and D.A Shepherd. 2000. Avian monitoring at the Buffalo Ridge, Minnesota wind resource area: results of a 4-year study. A report by Western EcoSystems Technology Inc. 262 pp. Johnson, G.D., W.P. Erickson, and K. Bay. 2002. Baseline ecological studies for the Klondike Wind Project, Sherman County, Oregon. A report by Western EcoSystems Technology Inc. 50 pp. Johnson, G.D., W.P. Erickson, J. White, and R. McKinney. 2003. Avian and bat mortality during the first year of operation at the Klondike Phase 1 Wind Project, Sherman County, Oregon. A report by Western EcoSystems Technology Inc. 17 pp. Johnstone, J.A. and T.E. Dawson. 2010. Climatic context and ecological implications of summer fog decline in the coast redwood region. Proceedings of the National Academy of Sciences of the United States of America 107:4533–4538. Keedwell, R.J. 2004. Use of population viability analysis in conservation management in New Zealand. 1: Review of technique and software. Science for Conservation 243:5‐37. Killworth, P.D., D.B. Chelton, and R.A. De Szoeke. 1997. The speed of observed and theoretical extratropical planetary waves. Journal of Physical Oceanography 27:1946-1966. Kirsch, E.M., M.J. Wellik, M. Suarez, R.H. Diehl, J. Lutes, W. Woyczik, J. Krapfl, and R. Sojda. 2015 Observation of sandhill cranes’ (Grus canadensis) flight behavior in heavy fog. The Wilson Journal of Ornithology 127(2):281–288. Kissling, M.L., S.B. Lewis, P.M. Lukacs, J.N. Waite, and S.M. Gende. 2016. Breeding decisions of Kittlitz’s Murrelets Brachyramphus brevirostris, Icy Bay, Alaska. Marine Ornithology 44:171-182. Krijgsveld, K.L., K. Akershoek, F. Schnenk, F. Dijk, and S. Dirksen. 2009. Collision risk of birds with modern large wind turbines. Ardea 97:357-366. Kruschke, J.K. 2015. Doing Bayesian Data Analysis: A Tutorial with R, JAGS, and Stan. 2nd edition. Elsevier. 759 pp. Lai, C.D, D.N. Pra Murthy, and M. Xie. 2006. Weibull distributions and their applications. Pages 63-78 in The Handbook of Engineering Statistics. Springer, London.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 48 Humboldt Wind Project September 2019

Limpert, E., W.A. Stahel, and M. Abbt. 2001. Log-normal distributions across the sciences: keys and clues. Bioscience 51:341-352. Lorenz, T.J., M.G. Raphael, T.D. Bloxton, and P.G. Cunningham. 2017. Low breeding propensity and wide- ranging movements by marbled murrelets in Washington. Journal of Wildlife Management 81:306-321. Lorenz, T.J., M.G. Raphael, and T.D. Bloxton. 2019. Nesting behavior of Marbled Murrelets Brachyramphus marmoratus in Washington and British Columbia. Marine Ornithology 47: 157-166. Maclean, I.M.D., L.D. Wright, D.A. Showler, and M.M. Rehfisch. A review of assessment methodologies for offshore windfarms. Report COWRIE METH-08-08 to the British Trust for Ornithology. 76 pp. Manley, I.P. 2006. Inventory methods for marbled murrelet radar surveys: Standards for components of British Columbia’s Biodiversity No. 10a. Prepared for Ecosystems Branch of the Ministry of Environment for the Resources Information Stewards Committee. 22 pp. Mantua, N.J. and S.R. Hare. 2002. The Pacific Decadal Oscillation. Journal of Oceanography 58:35-44. Mantua, N.J., S.R. Hare, Y. Zhang, J.M. Wallace, and R.C. Francis. 1997. A Pacific interdecadal climate oscillation with impacts on salmon production. Bulletin of the American Meteorological Society 78:1069-1079. Manuwal, D.A. 1974. The natural history of the Cassin’s auklet (Ptychoramphus aleuticus). Condor 76:421-431. Martin, G.R., S.J. Portugal, and C.P. Martin. 2012. Visual fields, foraging and collision vulnerability in Gyps vultures. Ibis 154:626-631. Masden, E.A. 2015. Developing an avian collision risk model to incorporate variability and uncertainty. Unpublished report from the University of Highlands and Islands Environmental Research Institute, UK, to the Natural Environment Research Council, UK. 26 pp. Masden, E.A. and A.S.C.P. Cook. 2016. Avian collision risk model for wind energy impact assessments. Environmental Impact Assessment Review 56:43-49. Mateos, M., G.M. Arroyo, A. Rodriguez, D. Cuenca, A. De La Cruz. 2010. Calibration of visually estimated distances to migrating seabirds with radar measurements. Journal of Field Ornithology 83:302-309. May, R. 2015. A unifying framework for the underlying mechanisms of avian avoidance of wind turbines. 190:179-187. May, R, A.B. Gill, J. Köppel, R.H.W. Langston, M. Reichenbach, M. Scheidat, S. Smallwood, C.C. Voight, O. Hüppop, and M. Portman. 2017. Future Research Directions to Reconcile Wind Turbine–Wildlife Interactions. Pages 255-276 in Wind Energy and Wildlife Interactions (J. Köppel, Ed.). Springer Nature, Switzerland. McIver, W., J. Baldwin, M.M. Lance, S.F. Pearson, C. Strong, N. Johnson, D. Lynch, M.G. Raphael, R. Young, T. Lorenz and K. Nelson. 2019. Marbled murrelet effectiveness monitoring, Northwest Forest Plan: 2018 summary report. 22 pp. Montgomery, D.C., E.A. Peck and G.G. Vining, 2001. Introduction to Linear Regression Analysis (3rd Edition). John Wiley and Sons, New York, NY. 672 pp. Morris, W. D. Doak, M. Groom, P. Kareiva, J. Fieberg, L. Gerber, P. Murphy, and D. Thomson. A practical handbook for Population Viability Analysis. The Nature Conservancy. Naslund, N.L. 1993. Why do Marbled Murrelets attend old-growth forest nesting areas year-round? Auk 110: 594-602. Naslund, N.L. and B.P. O’Donnell. 1995. Daily activity patterns of Marbled Murrelet activity at inland sites. Pages 129-134 in Ecology and conservation of the Marbled Murrelet, General Technical Report PSW- GTR-152 (C.J. Ralph, G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station, Albany, CA. Nations, C.S. and W.P. Erickson. 2009. Marbled murrelet—Wind turbine collision model for the Radar Ridge Wind Resource Area. Prepared by Western EcoSystems Technology, Inc., Cheyenne, WY for Energy Northwest, Richland, WA. 38 pp. Nelson, S.K. 1997. Marbled murrelet (Brachyramphous marmoratus), version 2.0. In The Birds of North America (A.F. Poole and F.B. Gill, Eds. Cornell Lab of Ornithology, Ithaca, NY. Nelson, S.K. and T.E. Hamer. 1995. Nesting biology and behavior of the Marbled Murrelet. Pages 57-67 in Ecology and conservation of the Marbled Murrelet, General Technical Report PSW-GTR-152 (C.J.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 49 Humboldt Wind Project September 2019

Ralph, G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station, Albany, CA. Nelson, S.K. and R.W. Peck. 1995. Behavior of Marbled Murrelets at nine nest sites in Oregon. Northwestern Naturalist 76:43-53. Nelson, S. K., M. H. Huff, S. L. Miller, and M. G. Raphael. 2006. Marbled Murrelet biology: habitat relations and populations. Pages 9-30 in Northwest Forest Plan – The first 10 years (1994-2003): status and trends of populations and nesting habitat for the Marbled Murrelet, General Technical Report PNW- GTR-650 (M. H. Huff, M. G. Raphael, S. L. Miller, S. K. Nelson, and J. Baldwin, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station, Portland, OR. New, L., E. Bjerre, B. Milsap, M.C. Otto, and M.C. Runge. 2015. A collision risk model to predict avian fatalities: An example using Golden Eagles 10:1-12. Newman, M., M.A. Alexander, T.R. Ault, K.M. Cobb, C. Deser, E. Di Lorenzo, N. Mantua, A.J. Miller, S. Minobe, H. Nakamura, N. Schneider, D.J. Vimont, A.S. Philips, J.D. Scott, and C.A. Smith. 2016. The Pacific Decadal Oscillation, revisited. Journal of Climate 29:4400-4427. Nichols, T.F. and T.R. Powers. 1964. Moonlight and night visibility. Declassified research Memorandum, Task SWINGSHIFT, U.S. Army Training Center Human Research Unit, Monterey, CA. 59 pp . NDBC [National Data Buoy Center]. 2019. Historical and climatic summaries for Station 46022 (LLNR 500)- Eel River-17 NM WSW of Eureka, CA. URL https://www.ndbc.noaa.gov/station_history.php?station =46022 [Accessed 11 April 2019]. NOAA Earth System Research Laboratory. 2019. Multivariate ENSO Index Version 2 (MEI.v2). URL https://www.esrl.noaa.gov/psd/enso/mei/ [Accessed 11 April 2019]. NOAA CPC [National Oceanic and Atmospheric Administration’s Climate Prediction Center]. 2019a. Global ocean monitoring: recent evolution, current status, and predictions: August 12, 2019 (slide 17). URL https://www.cpc.ncep.noaa.gov/products/GODAS/ocean_briefing_gif/global_ocean_monitoring_ 2019_08.pdf [Accessed 9 September 2019]. NOAA CPC [National Oceanic and Atmospheric Administration’s Climate Prediction Center]. 2019b. El Niño Southern Oscillation cold and warm episodes by season (Oceanic Niño Index ERSSTv5). URL https://origin.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ONI_v5.php [Accessed 11 April 2019]. NOAA NWS [National Oceanic and Atmospheric Administration National Weather Service]. 2019. Fog types. URL https://www.weather.gov/source/zhu/ZHU_Training_Page/fog_stuff/fog_definitions/Fog_defini tions.html [Accessed 13 August 2019]. NOAA PFEL [National Oceanic and Atmospheric Administration’s Pacific Fisheries Environmental Laboratory]. 2019. Upwelling Indices. URL https://www.pfeg.noaa.gov/products/PFEL/ modeled/indices/upwelling/NA/data_download.html [Accessed 11 April 2019]. Painter, S., Little, B. & Lawrence, S. 1999. Continuation of Bird Studies at Blyth Harbour Wind Farm and the Implications for Offshore Wind Farms. Harwell Laboratory Energy Technology Support Unit Report W/13/00485/00/00. 52 pp. Panuccio, M., G. Dell’Omo, G. Bogliani, C. Catoni, and N. Sapir. 2018 Migrating birds avoid flying through fog and low clouds International Journal of Biometeorology DOI: https://doi.org/10.1007/s00484- 018-01656-z. Pearson, S.F., B. McIver, D. Lynch, N. Johnson, J. Baldwin, M.M. Lance, M.G. Raphael, C. Strong, and R. Young, T. Lorenz, and K Nelson. 2018. Marbled murrelet effectiveness monitoring, Northwest Forest Plan: 2017 summary report. 19 pp. Peery, M.Z. and R.W. Henry. 2010. Recovering Marbled Murrelets via corvid management: a population viability analysis approach. Biological Conservation 143:2414-2424. Paton, P.W.C. and C.J. Ralph. 1990. Distribution of the marbled murrelet at inland sites in California. Northwestern Naturalist 7172-84. Peery, M.Z., S.R. Beissinger, S.H. Newman, E.B. Burkett, and T.D. Williams. 2004a. Applying the declining population paradigm: diagnosing causes of poor reproduction in the Marbled Murrelet. Conservation Biology 18:1088-1098.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 50 Humboldt Wind Project September 2019

Peery, M.Z., S.R. Beissinger, S.H. Newman, B.H. Becker, E.B. Burkett, and T.D. Williams 2004b. Individual and temporal variation in inland flight behavior of marbled murrelets: implications for population monitoring. Condor 106:344-353. Piatt, J.F. and N.L. Naslund. 1995. Abundance, distribution, and population status of Marbled Murrelets in Alaska. Pages 285-294 in Ecology and conservation of the Marbled Murrelet, General Technical Report PSW-GTR-152 (C.J. Ralph, G.L. Hunt, Jr., M.G. Raphael, and J.F. Piatt, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station, Albany, CA. Plissner, J.H., B.A. Cooper, R.H. Day, P.M. Sanzenbacher, A.E. Burger, and M.G. Raphael. 2015. A review of Marbled Murrelet research related to nesting habitat use and nest success. ABR. Inc., Environmental Research and Services, Forest Grove, OR. Plummer, M. 2018. RJAGS: Bayesian Graphical Models using MCMC. R package version 4-8. URL https://CRAN.R-project.org/package=rjags [Accessed 11 April 2019]. R Core Development Team. 2019. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/ Raphael, M.G., G.A. Falxa, and A.E. Burger. 2016. Marbled Murrelet. Pages 300-350 in Synthesis of science to inform land management within the Northwest Forest Plan area, General Technical Report PNW- GTR-966 (T.A. Spies, P.A. Stine, R. Gravenmier, J.W. Long, M.J. Reilly, Eds.). U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station, Portland, OR. Raphael, M.G., G.A. Falxa,and A.E. Burger. 2018. Marbled murrelet. Pages 301-370 Synthesis of Science to Inform Land Management within the Northwest Forest Plan Area (T.A. Spies, P.A. Spine, R. Gravenmier, J.W. Long, and M.J. Reilly, Tech. Coords.). USDA General Technical Report PNW-GTR- 966. Reed. T.E., M.P. Harris, and S. Wanless. 2015. Skipped breeding in common guillemots in a changing climate: restraint or constraint? Frontiers in Ecology and Evolution 3:1-13. Reed, T.E., P. Warzybok , A.J. Wilson, R.W. Bradley, S. Wanless, and W.J. Sydeman. 2009. Timing is everything: flexible phenology and shifting selection in a colonial seabird. Journal of Animal Ecology 78:376–387. Regular, P.M., A. Hedd, W.A. Montevecchi. 2011. Fishing in the dark: A pursuit-diving seabird modifies foraging behavior in response to nocturnal light levels. PLoS ONE 6:1-6. Rodway, M.S. and H.M. Regehr. 2000. Measuring Marbled Murrelet activity in valley bottom habitat: bias due to station placement. Journal of Field Ornithology 71:415-422. Rodway, M.S., H.M. Regehr, J.P.L. Savard. 1993. Activity patterns of marbled murrelets in old-growth forest in the Queen Charlotte Islands, British Columbia. Condor 95:831-848. Ross-Smith, V.H., C.B Thaxter, E.A. Masden, J. Shamoun-Baranes, N.H.K. Burton, L.J. Wright, M.M. Rehfisch, A. Johnston. 2016. Modelling flight heights of lesser black-backed gulls and great skuas from GPS: a Bayesian approach. Journal of Applied Ecology 53:1676–1685. RPS 2011. Galloper Offshore Wind Farm Ornithological Technical Report, Appendix 3: Avoidance Rates. Unpublished Report for Galloper Wind Farm Ltd. Rydell, J., R. Ottvall, S. Pettersson, and M. Green. 2017. The effects of wind power on birds and bats: An updated synthesis report 2017. Report by the Swedish Environmental Protection Agency. 132 pp. Sæther, B. and Ø. Bakke. 2000. Avian life history variation and the contribution of demographic traits to the population growth rate. Ecology 81:642-653. Sanzenbacher, P.M. and B.A. Cooper. 2010. Marbled murrelet fatality models for the proposed bear river wind project, California: Collision fatality estimates at wind turbines. Prepared by ABR, Inc.— Environmental Research & Services, Forest Grove, OR. Prepared for Shell WindEnergy Inc., Houston, TX. 35 pp. Sanzenbacher, P.M., B.A. Cooper, J.H. Plissner, and J. Bond. 2014. Intra-annual patterns in passage rates and flight altitudes of Marbled Murrelets Brachyramphus marmoratus at inland sites in northern California. Marine Ornithology 42:169-174. Sanzenbacher, P.M., T.J. Mabee, and B.A. Cooper. 2015 A radar and visual study of marbled murrelets at the proposed Skookumchuck Wind Energy Project, Summer 2013 and 2014. Prepared by ABR, Inc.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 51 Humboldt Wind Project September 2019

Environmental Research & Services, Forest Grove, OR for RES America Developments, Broomfield, CO. 38 pp. Schmidt, A.E., L.W. Botsford, J.M. Eadie, R.W. Bradley, E.D. Lorenzo, J. Jahncke. 2014. Non-stationary seabird responses reveal shifting ENSO dynamics in the northeast Pacific. Marine Ecology Progress Series 499:249-258. Schneider, S.R. 2018. Reproductive performance, foraging effort, and diet of an apex predator, the common murre, at one of the largest nesting colonies in the California Current System. MS Thesis, San Jose State University’s Moss Landing Marine Laboratory, Moss Landing, CA. 44 pp. Schreiber, R.W. and E.A. Schreiber. 1984. Central Pacific seabirds and the El Niño Southern Oscillation: 1982 to 1983 perspectives. Science 225:713-716. Schrimpf, M.B., J.K. Parrish, S.F. Pearson. 2012. Trade-offs in prey quality and quantity revealed through the behavioral compensation of breeding seabirds. Marine Ecology Progress Series 460:247-259. Schroeder, I.D., B.A. Black, W.J. Sydeman, S.J. Bograd, E.L. Hazen, J.A. Santora, and B.K. Wells. 2009. The North Pacific High and wintertime pre-conditioning of California current productivity. Geophysical Research Letters 40:1-6. Schwing, F.B., T. Murphree, P.M. Green. 2002. The Northern Oscillation Index (NOI): A new climate index for the northeast Pacific. Progress in Oceanography 53:115-139. Shultz, M.T., J.F. Piatt, A.M.A. Harding, A.B. Kettle, and T.I. Van Pelt. 2009. Timing of breeding and reproductive performance in murres and kittiwakes reflect mismatched seasonal prey dynamics. Marine Ecology Progress Series 393:247-258. Scotland. 1991. Natural Heritage (Scotland) Act 1991. Chapter 28 in United Kingdom Public General Acts. His/Her Majesty’s Stationery Office Publications Center, London. 52 pp. SNH [Scottish Natural Heritage]. 2000. Windfarms and birds - calculating a theoretical collision risk assuming no avoiding action. Scottish Natural Heritage Guidance Note. SNH [Scottish Natural Heritage]. 2010. Use of avoidance rates in the SNH wind farm collision risk model. Scottish Natural Heritage Guidance Note. SNH [Scottish Natural Heritage] Database. 2014. Data used to derive avoidance rates from studies combining collision rates with estimates of the rate at which birds pass through a windfarm. Appendix 7 of the Scottish Marine and Freshwater Science (Volume 5 Number 16): The avoidance rates of collision and offshore wind turbines (Cook, A.S.C.P, E.M. Humphreys, E.A. Masden, and N.H.K. Burton, Eds.). British Trust for Ornithology Research Report 656. 244 pp. URL https://www.gov.scot/binaries/ content/documents/govscot/publications/progress-report/2014/12/scottish-marine-freshwater- science-volume-5-number-16-avoidance-rates/documents/00464980-xls/00464980- xls/govscot%3Adocument [Accessed 11 April 2019]. SNH [Scottish Natural Heritage]. 2017. Avoidance rates for the onshore SNH wind farm collision risk model. Scottish Natural Heritage Guidance Note. 3 pp. Smallwood, K.S. 2017. The challenges of addressing wildlife impacts when repowering wind energy projects. Pages 175-185 in Wind Energy and Wildlife Interactions (Köppel, J., Ed.). Springer International Publishing, Berlin, Germany. Smallwood, K.S., and C.G. Thelander. 2008. Bird mortality in the Altamont Pass Wind Resource Area, California. Journal of Wildlife Management 72:215–223. Smallwood, K.S., L. Rugge, and M.L. Morrision. 2009. Influence of behavior on bird mortality in wind energy developments. Journal of Wildlife Management 73: 1082-1098. Stantec. 2018. Humboldt Wind Energy Project marbled murrelet radar survey report – Year 1. Prepared by Stantec Consulting Services, Inc. for Humboldt Wind, LLC., San Diego, CA.46 pp. Stantec. 2019. Humboldt Wind Energy Project marbled murrelet radar survey report – Year 2. Prepared by Stantec Consulting Services, Inc. for Humboldt Wind, LLC., San Diego, CA. Stumpf, J.P., N. Denis, T.E. Hamer, G. Johnson, and J. Verschuyl. 2011. Flight height distribution and collision risk of the marbled murrelet Brachyramphus marmoratus: Methodology and preliminary results. Marine Ornithology 39:123-128. Swiss Ornithological Institute. 2018. Birds of Switzerland Online Database. Accessed 16-Jan-2019, https://www.vogelwarte.ch/en/birds/birds-of-switzerland/.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 52 Humboldt Wind Project September 2019

Thaxter, C.B., G.M. Buchanan, J. Carr, S.H.M. Butchart, T. Newbold, R.E. Green, J.A. Tobias, W.B. Foden, S. O'Brien, J.W. Pearce-Higgins. 2017. Bird and bat species' global vulnerability to collision mortality at wind farms revealed through a trait-based assessment. Proceedings of Biological Sciences 284:20170829. The Landmark Practice 2010. Birds and wind turbines at Avonmouth Docks Year 2 Monitoring Report for Ecotricity. Landmark Ref: 06/EC/P1931/ECOTR. 20 pp. The Landmark Practice 2011. Birds and wind turbines at Avonmouth Docks Year 3 Monitoring Report for Ecotricity. Landmark Ref: 06/EC/P1931/ECOTR. 31 pp. The Landmark Practice 2013. Birds and Wind Turbines at Avonmouth Docks Year 5 Monitoring Report for Ecotricity. Bristol. Landmark Ref: 06/EC/P1931/ECOTR. Tucker, V.A. 1996. A mathematical model of bird collisions with wind turbine rotors. ASME Journal of Solar Energy Engineering 118:253-262. Upadhyay S.K. and M. Peshwani. 2003. Choice between Weibull and Lognormal model: A simulation based Bayesian study. Communications in Statistics: Theory and Methods 32:381-405. Urmy, S.S. and J.D. Warren. 2016. Quantitative ornithology with a commercial marine radar: standard-target calibration, target detection and tracking, and measurement of echoes from individuals and flocks. Methods in Marine Ecology and Evolution. doi: 10.1111/2041-210X.12699. USFWS [U.S. Fish & Wildlife Service]. 1997. Recovery plan for the Marbled Murrelet (Brachyramphus marmoratus) in Washington, Oregon, and California. U.S. Fish and Wildlife Service, Region 1, Portland, Oregon. USFWS [U.S. Fish & Wildlife Service]. 2013. Eagle conservation plan guidance. Module 1—land-based wind energy. Version 2, U.S. Fish and Wildlife Service, Division of Migratory Bird Management, Falls Church, VA. 103 pp. Uz, B.M., J.A. Yoder, and V. Osychny. et al. 2001. Pumping of nutrients to ocean surface waters by the action of propagating planetary waves. Nature 409:597-600. Weibull, W. and S.M. Sweden. 1951. A statistical distribution function of wide applicability. Journal of Applied Mechanics 18:293-296. West, J.B., R.R. Watson, Z. Fu. 2007. Major differences in the pulmonary circulation between birds and mammals. Journal of Respiratory Physiology and Neurobiology 157:382-390. Whitfield, D.P. and M. Madders. 2006. Upland raptors and the assessment of wind farm impacts. Ibis 148:43- 56. Winkelman, J.E. 1992a. De invloed van de Sep-proefwindcentrale te Oosterbierum (Friesland) op vogels, 1: aanvaringsslachtoffers. [The impact of the Sep wind park near Oosterbierum (Friesland), The Netherlands, on birds, 1: collision victims]. RIN-rapport 92/2. DLO-Institute for Forestry and Nature Research, The Netherlands. 144 pp. Winkelman, J.E. 1992b. De invloed van de Sep-proefwindcentrale te Oosterbierum (Friesland) op vogels, 2: nachtelijke aanvaringskansen [The impact of the Sep wind park near Oosterbierum (Friesland), The Netherlands, on birds, 2: nocturnal collision risks]. RIN-rapport 92/3. DLO-Institute for Forestry and Nature Research, The Netherlands. 206 pp. Winkelman, J.E. 1994. Bird/Wind turbine investigation in Europe. Proceedings of the National Avian Wind Power Planning Meeting, Lakewood, CO. Pages 43-47 and 120-150.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates 53 Humboldt Wind Project September 2019

Appendix A. Avian avoidance of wind turbines

Avoidance probability is the likelihood that a bird approaching a rotating wind turbine will perceive the turbine and take evasive action. This includes avoiding the entire area occupied by turbines (macro-avoidance), flying at a distance from the turbine (meso-avoidance), and maneuvering to avoid collision with the rotor swept area while in striking distance of turbine structures (micro-avoidance). Overall avoidance includes all three. In CRMs, all three forms of avoidance are often lumped together as one value due to the difficulty of separately quantifying each of these distinct forms of avoidance. In fact, most studies have only measured one or two of these constituents of avoidance and total avoidance is greater than what has been reported. This is because avoidance rates are most frequently derived by comparing the number of observed collisions post-construction with the predicted no-avoidance collision estimate and these numbers often exclude macro-avoidance, especially in terrestrial systems where AV surveys with minimal detection range are often used rather than radar which can detect up to 1.5 km from the radar.

For a CRM to be useful, realistic avoidance rates must be used as very small changes in the avoidance factor results in substantial changes in predicted collisions. For all early windfarms, data to inform the appropriate avoidance rate were lacking and the need for a better understanding of avoidance was recognized as essential to improve the utility of CRMs, specifically their ability to accurately predict collisions. This need was recognized in the early 2000’s (Chamberlin et al. 2005). Although the earliest measures of avoidance at operational wind farms were more than 0.99 (0.996-0.999; Winkelman 1992a, Painter et al. 1999) the Scottish Natural Heritage (SNH) also recognized the need to err on the side of sensitive species until more data were available. As such, the SNH recommended a precautionary default avoidance rate of 0.95 for onshore locations in the mid-2000s (Band et al. 2007). This recommendation was based on expert opinion rather than monitoring data (Furness 2013).

Over time, additional empirical evidence from onshore and offshore wind facilities has supported an upward revision of SNH’s recommended default avoidance to 0.98 as a conservative measure applicable to most situations and species, including many seabirds (SHN 2010). This guidance was reaffirmed in 2017 and, for species where avoidance has been well studied, SNH provided species-specific recommendations that were between 0.99 and 0.998 for all species (divers, swans, geese, kites, harriers, golden eagles, and skuas) except the kestrel and white-tailed eagle for which they recommended an avoidance of 0.95.

In subsequent analyses intended to assess potential avoidance of birds that could be expected in offshore situations, Cook et al. (2014) reviewed avoidance by seabirds at 35 onshore coastal windfarms, they concluded that these seabirds (including some sensitive gull species who spend a lot of time near turbines) were likely to exhibit avoidance greater than 0.99. In a review intended to better understand avoidance by seabirds, Furness (2013) also concluded a precautionary 99.5% avoidance was appropriate for seabirds. Maclean et al. (2009) recommended a default avoidance of 99% but noted that 99.5% was more appropriate for seabirds in the offshore environment.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-1 Humboldt Wind Project September 2019

Because efforts to better quantify this avoidance began almost 30 years ago, avoidance rates for species comparable to murrelets (non-raptorial) of less than 0.9 have not been reported, with two exceptions. Both are from gulls at two facilities using older generation wind turbines that are much smaller and can reach rotation rates of 43 rpm, which revolves at a rate 2.6 times greater than the maximum revolutions reached by modern turbines. Also, for all “reliable” (per their definition) records of avoidance in the SNH Database (2014), only 12 of 64 records reported an avoidance between 0.90 and 0.980. The average precision of these 12 estimates was 0.02 which is not very informative in that, if the reported avoidance was of 0.97, then one fewer observed collision would result in an avoidance rate of 0.99 and one additional collision would result in an avoidance rate of 0.95. To be informative, estimates of avoidance would be precise on the order of 0.001 so that, based on the example above, one fewer observed collision would result in an avoidance rate of 0.971 and one additional collision would result in an avoidance rate of 0.969. The remaining 52 estimates were of avoidance measured to be greater than 0.98.

There is little empirically derived, murrelet-specific data to inform the avoidance in this species. This is because wind turbines have not yet been installed in a space frequented by murrelets. Further, murrelets themselves are very difficult to study, with radar or radio telemetry being the most productive methods of study. However, murrelets are known to nest in old-growth forest canopies and are adept at maneuvering around complicated objects in their airspace, even during low-light conditions when accessing their nest at dawn. Although they fly over human modified habitat while transiting from the ocean to nesting habitat further inland, there are no data to suggest they collide with structures like buildings, signs, or other man-made structures. There has been one study of murrelet avoidance, and this was a quantification of avoidance for individuals at risk of collision along a transmission line (Cooper 2010); this study documented an avoidance of 100% based on a total observation of 423 murrelets using the corridor during the early morning activity period, 79 of which were at risk height and 10 of which were predicted to collide based on chance alone (Cooper 2010).

We recognize that, in the absence of murrelet-specific studies, some skepticism may exist regarding the use of an avoidance rate of 0.98 or greater. Specifically, questions have been raised regarding whether this rate is potentially unrealistic for the marbled murrelet. However, based on a comprehensive review of avian avoidance of wind turbines, avoidance rates in excess of 0.98 (and often as great as 0.995) have been repeatedly derived for birds at modern wind energy facilities with very few exceptions. Due to a preponderance of evidence supporting the notion that 0.98 is a conservative estimate of wind turbine avoidance for most bird species, including species behaviorally and biologically comparable to the marbled murrelet, and considering that there is no objective evidence indicating that murrelets should be categorized differently, we conclude that 0.98 is a very conservative estimate of the ability of most birds to detect and avoid wind turbines. Therefore, for the deterministic CRM, we used 0.98 as a conservative avoidance rate for marbled murrelets. For the probabilistic CRM, we considered all values of avoidance to be possible and used mathematical inference to assign credibility to each value of avoidance.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-2 Humboldt Wind Project September 2019

In the absence of species-specific data, the only option available was to use the best science available for related taxa including auks, seabirds, and birds in general to facilitate calculation of a conservative yet realistic estimate of the ability of murrelets to avoid turbines. Based on all available literature and species studied thus far, there is no reason to expect that marbled murrelets exhibit an avoidance lower than 0.98 in terrestrial habitats, and that it is likely higher in offshore habitats.

There is a preponderance of evidence that murrelets would exhibit an avoidance of turbines similar to the majority of other avian species that pass through onshore wind facilities (0.98 as a conservative estimate). Avian species at greatest risk of collision are ones that either spend significant portions of their day near operational turbines and/or are engaged in behaviors that distract them as they fly near turbines, such as foraging and territorial defense activities (Smallwood et al. 2009, Grünkorn et al. 2017). Additionally, some vultures have a blind spot in their visual field such that, given the way they position their head while searching for prey, they are unable to detect objects in their direction of travel despite having comprehensive abilities to visualize the ground and lateral obstacles (Martin et al. 2012).

This is not the case for murrelets that will be transiting across Bear River and Monument Ridges; these murrelets have the sole objective of navigating from point A to point B and there is very little reason for these individuals to become distracted as they fly near turbines installed at this location. Presumably, their attention is dedicated to ensuring successful transit rather than prey acquisition (they forage at-sea) or defending territory (there is no indication of territoriality by murrelets at this scale). Thus, it is justifiable to assume that murrelets would exhibit an avoidance that is greater than what has been observed for birds that are frequently distracted as they fly near wind turbines.

Other CRM’s for murrelets (Nations and Erickson 2009, Sanzenbacher and Cooper 2010, Sanzenbacher et al. 2015) have used a range of values for avoidance that, with time, have begun to approach the general recommendations for seabirds. All authors have argued that values of 0.9 and lower were unrealistic. In the CRM for the proposed Skookumchuck Wind Energy Project, Sanzenbacher et al. (2015) discussed avoidance factors and noted several related situations with other seabird species or anecdotes about murrelet flight that suggest that murrelets were likely quite capable of avoiding human structures. Further, Sanzenbacher et al. (2015) concluded that the avoidance rate for murrelets is likely much greater than 0.95 (one of the three avoidance rates they modeled). In the Bear River Wind Project (Sanzenbacher and Cooper 2010), a range of avoidance rates were used but they noted that murrelets may be as adept at avoiding collisions as what has been observed for other seabirds. Cooper used radar to investigate murrelet interactions with powerlines and concluded an avoidance rate of 1.0.

Effect of darkness and flight timing One anecdotal argument that has been expressed is that murrelets fly in the dark, which justifies the use of a lower avoidance rate for this species. Murrelets are adapted to regularly fly in complex environments at high speed and in low light conditions. Murrelets nest in old-growth coniferous forests and must be able to navigate through the canopy of these forests. Also, these seabirds travel quite fast; based on an evaluation of 3000 flights,

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-3 Humboldt Wind Project September 2019

Elliot et al. (2004) found that murrelets travel at an average speed of 50.55 miles per hour (max = 100 mph). Radar data collected by Stantec confirm these average and maximum speeds; the average speed at which murrelets were observed transiting the ridge was also 50.55 mph (max = 92.6 mph).

Contrary to flying in darkness, murrelets are crepuscular in their inland transit times and generally restrict inland flights to dawn and, to a lesser extent, dusk. The radar data collected on Bear River and Monument Ridges in 2018 provides evidence that murrelets cross the ridge when ambient light is orders of magnitude greater than on a dark night. Twilight, which occurs prior to sunrise, is a period of rapidly changing light intensity that is often divided into three phases: astronomical twilight (the darkest phase of twilight where the sun is between 12-18 degrees below the horizon and the sun begins to illuminate the sky but bright stars are still visible to humans for navigational purposes), nautical twilight (the second phase of twilight where the sun is between 6- 12 degrees below the horizon and object outlines are distinguishable by the human eye), and civil twilight (where the sun is between 0-6 degrees below the horizon and objects can still be distinguished by the human eye; Nichols and Powers 1964, Dusenbery 1992). When present during twilight, the moon can increase ambient light intensity during astronomical and nautical twilight such that they are comparable to civil twilight.

Assuming murrelet vision is as good as human vision (which is not necessarily the case and may be better in terms of detecting moving objects in low light conditions, addressed below) then the only time that structures would not be visible to them would be during astronomical twilight. Based on the timing of 55 ridge crossings documented in 2018 at Bear River and Monument Ridges, only 3.6% flights occurred during astronomical twilight and these flights only occurred when the moon was present and relatively bright (between 75 and 83% illuminated); thus, murrelets did not transit across the ridge during the darkest portion of twilight unless moonlight was sufficient to enable visual navigation across the landscape. An additional 34.5% of flights occurred during nautical twilight, and 58% of these transits occurred when the moon was present. When the moon was not present, flights occurred closer to civil twilight (and thus during times of increased light intensity) relative to moonlit nights (p=0.07, df=14; we note that the p-value for this analysis was not significant but was very close and would likely become significant with additional samples). The remaining 61.8% of ridge crossings occurred during civil twilight and daylight, when light intensity would allow murrelets to easily see turbines regardless of moon presence. It is important to note that light intensity is much less under the forest canopy relative to open habitats such as ridgelines and, for nesting murrelets that fly inland before sunrise, they must (and presumably are adapted to) negotiate flight between trees and other obstacles to successfully land at their nest at these reduced light levels. These recent data are consistent with the timing of 757 morning flights tracked by Hébert and Golightly (2006) using radio telemetry, where the initial time of inland flights began 17.7 ± 1.9 minutes before sunrise consistent with the assertion that murrelets initiate inland flights primarily during civil twilight. These data are also consistent with detailed report of Lorenz et al. (2019). Hébert and Golightly (2006) also report the timing for an additional 71 evening flights, which initiated 23.2 ± 11.5 minutes before sunset when light levels are sufficient for navigation.

There are several examples that illustrate that murrelets are likely adept at visualizing their surroundings even in low-light conditions. Marbled murrelets are a visually-oriented pursuit-diving predator that forages in waters

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-4 Humboldt Wind Project September 2019

up to 30 meters in depth to capture small fish and invertebrate prey (Nelson et al. 2006). In productive coastal zones where murrelets forage, the thin layer of ocean that is shallow enough for sunlight to penetrate (the photic zone) generally only extends to between 40 and 100 meters deep (Garrision and Ellis 2016) implying that murrelets must regularly forage in low-light conditions.

Because there is little information regarding the underwater behavior of murrelets, we discuss the capabilities of a much better studied taxonomic relative to illustrate the capabilities of Alcids to visualize objects even in low-light conditions. Similar to murrelets, common murres (Uria aalge) are visually-oriented pursuit-diving predators; this species dives up to 177 meters and is thought to dive as deep in the water column as they can while still maintaining efficient visual prey detection (Regular et al. 2011). Almost none of the visible light spectrum reaches depths of 150 meters, even in very clear water, and thus light intensities at 177 meters are extremely low (this is much deeper than murrelets are thought to dive). Common murre have also been documented capturing prey underwater during twilight periods and even between astronomical dusk and dawn (i.e., nocturnally) when there is lunar light available to aid in visualizing prey in the water column (Regular et al. 2011), however, they do not attempt to capture prey during nocturnal periods where only starlight is available, presumably because this level of light is insufficient for them to efficiently detect prey underwater. Regular et al. (2011) extended the hypothesis put forth by Hall and Ross (2007) that the low-light conditions imposed by the underwater environment have selected for the evolution of improved visual sensitivity in pursuit-diving seabirds. Although murrelets do not forage as deep as common murre, they do forage in low-light conditions and it is likely that they are better adapted for visual detection of prey and other underwater obstacles (rocks, kelp, predators, etc.) relative to species that do not need to forage in low-light conditions.

Finally, while many members of the auk family, such as murrelets and murres, are not considered nocturnally active there are a few members (e.g. rhinoceros and Cassin’s auklets) that, while nesting, restrict activity to nocturnal periods and are able to forage, navigate, and are able to locate nests in darkness (Manuwal 1974, Gaston and Dechesne 1996). There are many additional examples of good night-vision in various additional seabird families. Great cormorants are also visual, pursuit-diving predators that must hunt during the polar winter when the sun does not rise, and individuals continue to dive in relative darkness (light levels less than 1 lux; Grémillet et al. 2005). There are also many burrowing species of petrel that are nocturnally active and rely on visual cues to find their burrow in complex terrain, including within forest canopies (Brook 1989). Manx shearwaters use vision to find their burrows at night (James 1986). Thus, there is good evidence that many seabirds have evolved to see well in low-light conditions.

Species vulnerable to contact with the turbine In the context of species-specific abilities to avoid collisions with wind turbines, vulnerable species are not necessarily species with a special status (e.g., threatened or endangered). Rather, behavior is a primary determinant of risk of colliding with wind turbines (Smallwood et al. 2009, Smallwood 2017). Specifically, the most susceptible avian species are ones that become distracted as they are transiting in the vicinity of wind turbines.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-5 Humboldt Wind Project September 2019

Species that exhibit a lower avoidance than expected are often raptorial in nature (Barrios and Rodriguez 2004, Smallwood and Thelander 2008), but it is their behavior and not necessarily their taxonomy that causes this to be the case (Thaxter et al. 2017). Primary behaviors that cause some species of raptor to exhibit less avoidance include slow and less maneuverable flight, a reliance on thermals for soaring flight, and frequently becoming distracted by foraging opportunities or territorial disputes near turbines.

For example, one species noted by SNH (2017) as having relatively low avoidance rate was the white-tailed eagles (Haliaeetus albicilla). SNH (2017) recommended using a conservative avoidance of 0.95 for this species based on a study by Bevanger et al. (2008) at the Smøla wind-power plant (68 turbines) located at a small island archipelago in Norway. Bevanger et al. (2008) estimated avoidance by white-tailed eagles to be 95.4% using vantage point surveys and a more recent study at this same location by May et al. (2011) estimated an avoidance of 97.5% using satellite telemetry data. Prior to construction of the Smøla wind-power plant, the breeding density of white-tailed eagles was highest in the exact area where the wind-power plant was situated (Bevanger et al. 2010) and now often overlaps the area where turbines now exist; for example, 45 breeding territories were documented in the Smøla archipelago in 2010 (Dahl et al. 2013). This species has all the ﬂight characteristics of a soaring raptor and its social activities, such as chasing or ﬁghting and spiraling–playing, which are important for mating. Furthermore, these eagles regularly hunt among turbines for ptarmigan (B. Roy, pers. comm.). This combination of factors resulted in elevated exposure to turbines as well as many opportunities for individuals to become distracted while flying near turbines. Similarly, the American kestrel (Falco sparverius) is another species noted by SNH (2017) as having relatively low avoidance and this has been attributed to their tendency to be more focused on hunting while near turbines in open, agricultural landscapes, which results in reduced awareness of turbines. In both cases, these observations of behaviorally compromised avoidance by raptors are not analogous to murrelets on a focused transit.

We do not expect distracted behavior for murrelets flying through the proposed turbine string because, as mentioned earlier, this ridgeline habitat is unsuitable for foraging and there is no need for murrelets to defend territory in this location. Presumably murrelets crossing these ridgelines will be focused solely on flying between the sea and inland nesting habitat to the east of the proposed turbines and will remain in the airspace potentially occupied by turbines for a short duration where they are unlikely to become distracted from navigation.

Seabird exposure and avoidance in onshore versus offshore environments Seabirds typically remain at-sea and generally only encounter offshore wind turbines. Therefore, by necessity, avoidance rates measured for seabirds tend to be derived in the offshore environment. The installation of wind turbines in the offshore environments is a recent phenomenon relative to installation in onshore environments. As a means of forecasting the potential impact of turbines to birds in the offshore environment, the earliest offshore projects generalized what had been learned about avoidance in the onshore environment. Based on studies at onshore wind developments, 0.98 was (and continues to be) considered a conservative estimate of avoidance for most avian species (Cook et al. 2014, SNH 2017, Aschwanden et al. 2018). However, as more studies of avian avoidance at offshore facilities have been conducted, the applicability of 0.98 as a conservative avoidance rate to the offshore environment has come under scrutiny for being too conservative. Many argue,

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-6 Humboldt Wind Project September 2019

including Cook et al. (2018), Cook et al. (2014), and Furness et al. (2013), that 0.98 is “overly precautionary” for seabirds in an offshore environment and they conclude that an avoidance of 0.99 to 0.995 represents a more appropriate conservative estimate of avoidance in an offshore environment. To quote Cook et al. (2018) directly:

“Previous guidance of the use of avoidance rates in CRMs was that 0.98 (derived from onshore studies) should be considered the default value for seabirds (Scottish Natural Heritage 2010). Whilst significant gaps in knowledge remain, this review highlights that, for the species most likely to be affected by collision, avoidance rates are estimated to exceed 0.99. Whilst this may seem a trivial difference, it will result in the predicted collision rate being more than halved.”

Similar to Cook et al. (2018), Furness (2013) concludes:

“Before data on bird behavior at wind farms were available, SNH [Scottish Natural Heritage] recommended use of a precautionary avoidance rate estimate of 95%, but with accumulation of data at terrestrial wind farms this has been revised to 98% for most species (SNH 2010). With development of offshore wind farms, there is a need to consider whether avoidance rates of seabirds, and appropriate avoidance correction factors for application of the Band models, are the same as, or different from those of terrestrial birds…We conclude that it would be appropriate to use a 99.5% total avoidance rate as a precautionary default for seabirds given the data summarized above, but that appropriate avoidance corrections for Band models would have to be computed from this depending on which version of the Band model was being used.”

Some authors have eluded to differences between onshore and offshore exposure, often confusing exposure with avoidance potential. Corman and Garthe (2014) and Ross-Smith et al. (2016) address flight heights of birds and the resulting exposure and collision risk. Differences in flight altitude between onshore and offshore environments are irrelevant to this marbled murrelet collision risk assessment as the flight height of ridge crossings were empirically derived by vertical radar measurements. In support of the statement by Cook et al. (2018), we found that most ridge crossings (94.8%) by murrelets were at risk height; the collision estimated by the marbled murrelet collision risk model accounts for this flight behavior (it is essentially almost at maximum). Furthermore, and in recognition of the inaccuracy (due to variability in the exact tilt of the radar) in vertical radar measurements of up to ± 30 meters, for determining exposure risk, we treated any flight determined to be up to within 30 meters beyond the upper extent of largest turbine rotor being considered for use (all flights between 0 and 210 meters above ground) as at risk, and therefore have assumed maximum exposure and risk. Simply based on the inclusion of birds from the larger effective swept area, the estimates of passage could be almost double and result in a significant overestimation of collision potential.

Regardless, Cook et al. (2018) cites these papers and seems to suggest that differences in flight height between onshore and offshore locations may have implications for avoidance, but the primary intention of citing these two papers was to highlight that in the offshore environment many seabirds often fly much lower (and often below turbine height) which results in a lower exposure of seabirds. The reason that Cook et al. (2018) invoked the concept of differential avoidance in association with differences in the flight behavior of birds in offshore

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-7 Humboldt Wind Project September 2019

and onshore environments was to support their over-arching conclusion that a default of 0.98 for seabird avoidance in offshore environments was overly conservative and should be revised upwards to 0.99.

As studies at offshore wind facilities continue, there is mounting evidence that many seabirds exhibit much greater macro-avoidance (avoiding the entire windfarm; Furness 2013, Furness et al. 2013). Observations of greater macro-avoidance in the offshore environment may be real or may also be an artifact of sampling method; offshore studies of avoidance typically require long-distance radar (3 km) versus onshore studies which use fatality searches and human observers. Consequently, onshore data often lacks measurements of bird activity at the scale necessary to truly quantify a larger-scale macro-avoidance and only provides data on within- windfarm avoidance (meso- and micro-avoidance): thus, terrestrial measures of avoidance are often underestimates of true avoidance. It is possible that studies of avoidance in the onshore environment, due to differences in methodology, underestimate avoidance by failing to account for macro avoidance.

Although the marbled murrelet is a seabird, for purposes of collision risk assessment at an onshore wind facility, for the deterministic CRM we used the avoidance rate regarded as conservative for birds in the onshore environment (0.98) rather than the avoidance rate regarded as conservative for seabirds in the offshore environment (0.99-0.995). Therefore, we are taking into account that avoidance rates may differ in onshore environments by using the more conservative avoidance rate. It is possible that, in terms of micro-avoidance (avoiding collision when within striking distance of a structure), murrelets would be much better than other seabirds in both onshore and offshore environments as they are adapted to navigate through complex terrestrial environments (old-growth forests) at high speeds and in low light conditions (at dawn under the forest canopy) without population-level consequences.

Recent terrestrial investigations In a study of primarily nocturnally migrating songbirds through a mountainous region, Aschwanden et al. (2018) derived an avoidance rate of 97.9% with a 95% confidence interval of 97% to 98.5%. They documented this avoidance using a sample of 1.65 million birds that passed over a turbine array, 390,500 of which were flying at risk-height; the avoidance rate derived by this study included all species that passed over this wind facility. The study area was the Jura Mountains in the Swiss Alps and although only species found in carcass searches were specifically listed in this paper, Switzerland has a diversity of bird species; the Swiss Ornithological Institute’s Birds of Switzerland database contains 422 bird species of which 110 are listed as either regionally extinct, critically endangered, endangered, near threatened, or vulnerable (Swiss Ornithological Institute 2018).

Importantly, the vulnerability of a species due to its population status or other factors does not determine its ability to avoid collision with a wind turbine. In the case of the study by Aschwanden et al. (2018) the most frequently impacted species were also relatively common in Switzerland (Swiss Ornithological Institute 2018). Notably, many of the birds that passed through the wind facility were nocturnal migrants, thus the avoidance rate derived from this study is a good representation what would be expected for an onshore wind facility in the darkest portion of the night; Aschwanden et al. (2018) specifically concludes that their study “clearly shows that collision risk for nocturnally migrating (small) passerines at onshore wind farms is not necessarily lower

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-8 Humboldt Wind Project September 2019

than for other bird groups.” Further, the period of time over which avoidance was calculated included some periods of inclement weather; therefore the calculated avoidance rate was also inclusive of flight during those periods of inclement weather. The findings of this study further justifies the conclusion that an avoidance of 0.98 for marbled murrelets is conservative; marbled murrelets are crepuscular, rather than nocturnal, in terms of their timing of inland flights and murrelets restrict their use of the ridge to times of twilight and daylight, with the majority of flights occurring during periods of relatively high ambient light (e.g., greater moon illuminance, civil twilight, and daylight).

Conclusions For the deterministic CRM we followed the guidance of SNH (2010, 2017) and used an overall avoidance of 0.98. This rate balances the need to be cautionary, but realistic. In our attempt to be cautious, in the month of April, we lowered the rate to 0.95 with the presumption that inbound birds could include first-year birds that may be naïve to turbines or other structures. Although very conservative, this is only speculative logic and neither the naïve nature of first-year birds nor the estimated decreased avoidance probability for naïve individuals is supported by empirical evidence (in other words, 0.98 is likely a conservative estimate of avoidance even in April). Furthermore, based on empirically derived avoidance rates of auks, other seabirds, and many birds in general that come from recent studies of currently installed wind energy facilities, it is likely that the actual avoidance of turbines by murrelets throughout the year will be greater than 0.98. Thus, the avoidance rates used to estimate murrelet collision risk for this Project by the deterministic CRM should be considered conservative and may, along with the previously described project infrastructure and management assumptions, further contribute to a CRM overestimate of collision.

In contrast to the deterministic model, where only two values of avoidance were considered for the CRM, the probabilistic model used a Bayesian approach to approximate the probability of all values of avoidance possibly exhibited by marbled murrelets, based on prior knowledge of avoidance of turbines by ecologically similar birds as they travel through wind energy facilities. This was based on a database of 35 estimates of avoidance selected from an initial 152 reports of avoidance based on criteria outlined in Section 2.2.5.2 (Table A-1. The 35 estimates of avoidance that were selected from a total of 152 estimates because they met criteria detailed in text for use in a Bayesian analysis that generated a model of avoidance to be input into the Probabilistic Collision Risk Model. The types of avoidance that the estimate represents was indicted as 1 for micro-avoidance, 2 for meso-avoidance, and 3 for macro-avoidance. Behaviors were categorized as either local (L), meaning these individuals may have spent time near turbines engaging in various behaviors in addition to navigation, or migratory (M), for individuals whose sole intent was navigational and they quickly passed through areas occupied by turbines.). All 35 estimates were from monitoring reports or studies that appropriately corrected for searcher efficiency or carcass persistence or otherwise had good accuracy in the determination of fatalities resulting in accurate estimation of avoidance (Table A-2). Thus, the avoidance rates used to estimate murrelet collision risk for this Project by the probabilistic CRM fully accounts for uncertainty associated with the true value of avoidance that may be exhibited by murrelets.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-9 Humboldt Wind Project September 2019

To the best of our ability, we strictly adhered to the criteria described above with the exceptions of the last two. Consistent application of Criteria 8 and 9 would have been ideal, but this would have caused many otherwise robust estimates of avoidance to be excluded from our analysis.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates A-10 Humboldt Wind Project September 2019

Table A-1. The 35 estimates of avoidance that were selected from a total of 152 estimates because they met criteria detailed in text for use in a Bayesian analysis that generated a model of avoidance to be input into the Probabilistic Collision Risk Model. The types of avoidance that the estimate represents was indicted as 1 for micro-avoidance, 2 for meso-avoidance, and 3 for macro-avoidance. Behaviors were categorized as either local (L), meaning these individuals may have spent time near turbines engaging in various behaviors in addition to navigation, or migratory (M), for individuals whose sole intent was navigational and they quickly passed through areas occupied by turbines. Avoidance Behavior ID Wind energy facility Habitat Species Value Type Type Time Source 1 Boudewijn canal Rural/industrial black-headed gull 0.9586 1,2 L,M Diurnal 1, 2, 3 a 2 Boudewijn canal Rural/industrial black-headed gull 0.9988 1,2 L,M Diurnal 4, 2, 3 a 3 Boudewijn canal Rural/industrial herring gull 0.9903 1,2 L,M Diurnal 1, 2, 3 a 4 Boudewijn canal Rural/industrial herring gull 0.9987 1,2 L,M Diurnal 4, 2, 3 a 5 Zeebrugge Coastal breakwater black-headed gull 1.0000 1,2 L,M Diurnal 1, 2 6 Zeebrugge Coastal breakwater common tern 0.9703 1,2 L,M Diurnal 1, 2, 3 a 7 Zeebrugge Coastal breakwater common tern 0.9720 1,2 L,M Diurnal 1, 2, 3 a 8 Zeebrugge Coastal breakwater common tern 0.9988 1,2 L,M Diurnal 4, 2, 3 a 9 Zeebrugge Coastal breakwater common tern 0.9989 1,2 L,M Diurnal 4, 2, 3 a 10 Zeebrugge Coastal breakwater herring gull 0.9976 1,2 L,M Diurnal 1, 2 a 11 Zeebrugge Coastal breakwater lesser black-backed gull 0.9990 1,2 L,M Diurnal 4, 2 a 12 Zeebrugge Coastal breakwater little tern 1.0000 1,2 L,M Diurnal 4, 2 13 Zeebrugge Coastal breakwater little tern 1.0000 1,2 L,M Diurnal 4, 2 14 Zeebrugge Coastal breakwater sandwich tern 0.9895 1,2 L,M Diurnal 1, 2, 3 a 15 Zeebrugge Coastal breakwater sandwich tern 0.9940 1,2 L,M Diurnal 1, 2, 3 a 16 Zeebrugge Coastal breakwater sandwich tern 0.9991 1,2 L,M Diurnal 4, 2, 3 a 17 Zeebrugge Coastal breakwater sandwich tern 0.9995 1,2 L,M Diurnal 4, 2, 3 a 18 Nysted Offshore common eider, geese 0.9910 1,2,3 M Nocturnal 5 19 Hellrigg Coastal hills black-headed gull 1.0000 1,2 L,M Diurnal 1, 4, 6, 7 20 Hellrigg Coastal hills common gull 1.0000 1,2 L,M Diurnal 1, 4, 6, 7 21 Hellrigg Coastal hills herring gull 1.0000 1,2 L,M Diurnal 1,4,6,7 22 Bouin Coastal bottoms black-headed gull 0.9520 1,2 L,M Diurnal 1 23 Bouin Coastal bottoms gulls 0.9903 1,2 L,M Diurnal 1 24 Oosterbierum Coastal bottoms gulls 0.9982 1,2 L,M Diurnal 4, 8 25 Various Onshore gulls 0.9893 1,2 L,M Diurnal 9 26 Le Peuchapatte Ridgeline songbirds 0.9790 1,2 M Nocturnal 10 27 Avonmouth Coastal bottoms gulls 1.0000 1,2 L,M Diurnal 4, 11 28 Avonmouth Coastal bottoms gulls 1.0000 1,2 L,M Diurnal 4, 12, 13 29 Avonmouth Coastal bottoms herring gull 1.0000 1,2 L,M Diurnal 1, 4, 11 30 Avonmouth Coastal bottoms herring gull 1.0000 1,2 L,M Diurnal 1, 4, 12, 13 31 Blythe Harbor Coastal harbor gulls 0.9925 1,2 L,M Diurnal 4, 14 32 Buffalo Ridge Rolling hills Canada and snow geese 1.0000 1,2 L,M Diurnal 15, 16

(continued from previous page) Avoidance Behavior ID Wind energy facility Habitat Species Value Type Type Time Source 33 Klondike Bluff/ridgeline Canada geese 0.9980 1,2 L,M Diurnal 15, 17, 18 34 Nine Canyons Rolling hills Canada geese 1.0000 1,2 L,M Diurnal 15, 19 35 Stateline Wind Farm Rolling hills Canada geese 0.9990 1,2 L,M Diurnal 15, 20 a Avoidance estimated by Cook and Furness for these records were regarded as independent estimates as they made different assumptions regarding inputs to their collision risk models 1 Cook et al. 2014 2 Everaert and Kuijken 2007 3 Everaert and Steinen 2007 4 Furness 2013 5 Desholm and Kalert 2005 6 Ecology Consulting 2012 7 Ecology Consulting 2013 8 Winkelman 1992a 9 Cook et al. 2018 10 Aschwanden et al. 2018 11 The Landmark Practice 2010 12 The Landmark Practice 2011 13 The Landmark Practice 2013 14 RPS 2011 15 Fernley et al. 2006 16 Johnson et al. 2000 17 Johnson et al. 2002 18 Johnson et al. 2003 19 Erickson et al. 2002 20 Erickson et al. 2004

Table A-2. The name, location, and general year of installation for each wind energy facility from which avoidance estimates input into the Probabilistic CRM were derived. Avoidance was estimated in either Cook et al (2014), Cook et al. (2018), or Furness (2013) and was calculated by dividing the number of collisions observed by the number of collisions expected in the absence of avoidance (see Table A-1). The expected number of collisions in the absence of avoidance were calculated using the Band (2012) model and were based on the species-specific passage rate individuals through each facility. The observed number of collisions were derived from carcass searches within each facility, where the total number of carcasses observed were corrected to account for missed carcasses (e.g., searcher efficiency and carcass persistence). For each, the duration of monitoring, method used to determine passage rate, and method used to correct carcasses found in fatality searches was indicated. The ID corresponds with the ID in Table A-1.

Wind energy facility Monitoring Method to determine ID Name Location Installed effort passage rate Correction method for missed carcasses Source 1 Boudewijn canal Belgium ? 1 year Visual Derived from previous study 1, 2, 3 a 2 Boudewijn canal Belgium ? 1 year Visual Derived from previous study 4, 2, 3 a 3 Boudewijn canal Belgium ? 1 year Visual Derived from previous study 1, 2, 3 a 4 Boudewijn canal Belgium ? 1 year Visual Derived from previous study 4, 2, 3 a 5 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 1, 2 6 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 1, 2, 3 a 7 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 1, 2, 3 a 8 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2, 3 a 9 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2, 3 a 10 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 1, 2 a 11 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2 a 12 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2 13 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2 14 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 1, 2, 3 a 15 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 1, 2, 3 a 16 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2, 3 a 17 Zeebrugge Belgium <2000 4 years Visual Search efficiency, carcass persist 4, 2, 3 a 18 Nysted Denmark 2003 6 years Aerial transect/radar Radar, cameras, sensitivity index 5 19 Hellrigg UK 2011 2 years Radar, observer Search efficiency, carcass persist 1, 4, 6, 7 20 Hellrigg UK 2011 2 years Radar, observer Search efficiency, carcass persist 1, 4, 6, 7 21 Hellrigg UK 2011 2 years Radar, observer Search efficiency, carcass persist 1,4,6,7 22 Bouin France <2002 4 years Visual Search efficiency, carcass persist (grid) 1 23 Bouin France <2002 4 years Visual Search efficiency, carcass persist (grid) 1 24 Oosterbierum Belgium 1985 2 springs Visual/radar Search efficiency, carcass persist 4, 8 25 Various Europe Various Various Visual Various 9 26 Le Peuchapatte Switzerland ? 265 days Radar Search efficiency, carcass persist(short grass) 10 27 Avonmouth UK 2007 2 years Visual Carcass search (full visibility) 4, 11 28 Avonmouth UK 2007 2 years Visual Carcass search (full visibility) 4, 12, 13

(continued from previous page) Wind energy facility Monitoring Method to determine ID Name Location Installed effort passage rate Correction method for missed carcasses Source 29 Avonmouth UK 2007 2 years Visual Carcass search (full visibility) 1, 4, 11 30 Avonmouth UK 2007 2 years Visual Carcass search (full visibility) 1, 4, 12, 13 31 Blythe Harbor UK 1993 11 years Visual Search efficiency, carcass persist 4, 14 32 Buffalo Ridge US 1994 3 years Visual Search efficiency, carcass persist 15, 16 33 Klondike USA 2002 1 year Visual Search efficiency, carcass persist 15, 17, 18 34 Nine Canyons USA 2002 1 year Visual Search efficiency, carcass persist 15, 19 35 Stateline Wind Farm USA 2001 2 years Visual Search efficiency, carcass persist 15, 20 a Avoidance estimated by Cook and Furness for these records were regarded as independent estimates as they made different assumptions regarding inputs to their collision risk models 1 Cook et al. 2014 2 Everaert and Kuijken 2007 3 Everaert and Steinen 2007 4 Furness 2013 5 Desholm and Kalert 2005 6 Ecology Consulting 2012 7 Ecology Consulting 2013 8 Winkelman 1992a 9 Cook et al. 2018 10 Aschwanden et al. 2018 11 The Landmark Practice 2010 12 The Landmark Practice 2011 13 The Landmark Practice 2013 14 RPS 2011 15 Fernley et al. 2006 16 Johnson et al. 2000 17 Johnson et al. 2002 18 Johnson et al. 2003 19 Erickson et al. 2002 20 Erickson et al. 2004

Appendix B. Relating inland activity of murrelets to ocean conditions

Photosynthetic organisms, which form the foundation of the marine food web and are essential for the survival and reproduction of marine organisms, require nutrient-rich surface waters to become abundant (Beardall et al. 2001). However, nutrient availability in surface waters is limited and requires an input of nutrient-rich deep water. To reach the surface, these deep waters must be forced to upwell by winds (Bakun 1973), meso-scale eddies (Falkowski et al. 1991), and waves resulting from planetary rotation (Uz et al. 2001). These various processes can cause fluctuations in marine productivity over periods as short as hours (for winds) or as long as decades (for planetary waves; Killworth et al. 1997).

For seabirds nesting in the California Current System, any fluctuation in ocean productivity permeates through the food web (Croll et al. 2005, Gladics et al. 2015) and results in interannual variability in breeding effort (Gaston and Nettleship 1982, Reed et al. 2015, Kissling et al. 2016) and, for individuals that nest, the success of their nests (Schultz et al. 2009, Schrimpf et al. 2012, Schmidt et al. 2014). Because inland flights of marbled murrelets result from the need of breeding pairs to regularly access nests, inland activity during the breeding season is likely to reflect breeding effort and relative proportion of the population participating in nesting. Thus, we hypothesized that interannual variability in inland activity by murrelets could be predicted by measurements of marine productivity (see Lorenz et al. 2017).

Our objective was to determine if inland activity of murrelets varied as a function of marine productivity and, if so, use this relationship to estimate the degree to which murrelet activity in the vicinity of the proposed turbine string might vary across a 30-year period. To do so required (1) a multi-year dataset that quantified inland activity of murrelets near the Project, (2) identification of the most relevant indicators of marine productivity, and (3) evaluation of the degree to which marine productivity can serve as a predictor of inland activity. If the predictive relationship is significant and fluctuations in marine productivity explain at least 50% of variability in inland activity, then this relationship can be applied to estimate the potential range of activity expected for Bear River and Monument Ridges over the 30-year period of turbine operation.

So far, there has been a single multi-year effort to quantify inland activity of murrelets near the Project; the Humboldt Redwoods Company (HRC) has measured murrelet activity since 2002 using marine radar at various reserve sites (areas likely to be frequented by nesting murrelets) along the Eel River (HRC 2018). We used only the average count of murrelet-like targets in the reserve sites between 2002 and 2011. The dataset provided by HRC extended through 2018, however after 2011 there was a directed downward decline in the number of

murrelet detections (Linear Regression: F1,5=62.2, r2=0.926, p=0.0005). Changes in interannual variability of activity due to ocean productivity can only be detected if productivity is the primary driver of variation. Directed trends in murrelet activity (either upwards or downwards) over several years may be caused by circumstances in addition to ocean productivity and would confound measurement of the relationship between inland activity

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates B-1 Humboldt Wind Project September 2019

and productivity. Additionally, following the recommendations of HRC in their own reporting, the 2009 data were excluded as counts from that year were based on a single survey during the peak of breeding (June/July) and therefore are biased high (HRC 2018).

Because the physical processes that regulate marine productivity (wind-driven upwelling, eddies, and planetary waves) interact with each other in very complex ways (Newman et al. 2016), methods to separate the effects of each do not currently exist. Instead, indices reflective of marine productivity have been developed and linked to the biological response across trophic levels. We considered six such indices in this analysis: 1. Sea surface temperature (SST), which indicates localized upwelling of deep water that is cooler than surface waters; there is one SST sensor (NDBC 46022) located the 19 km west of the mouth of the Eel River that has collected data since 1982 (NDBC 2019); 1 2. Pacific Decadal Oscillation (PDO), an index of oceanic response atmospheric processes in the northeast Pacific that best accounts for fluctuations of sea surface temperature in the region (Mantua et al. 1997, Mantua and Hare 2002); 2,3 3. North Pacific Gyre Oscillation (NPGO), an index of oceanic response atmospheric processes in the northeast Pacific that best accounts for fluctuations of nutrients in the region (Di Lorenzo et al. 2008, Di Lorenzo et al. 2015); 4 4. Northern Oscillation Index (NOI), an index of oceanic response to interactions between atmospheric processes in the northeast Pacific with atmospheric processes in the tropical Pacific that impact the north Pacific on seasonal, annual, and decadal scales and represents the northern equivalent of the El Niño Southern Oscillation (ESNO; Schwing et al. 2002); 5 5. Multivariate El Niño Southern Oscillation Index (MEI), an index of the oceanic response to atmospheric processes in the tropical Pacific which result in global climate disruptions that can persist for several years (NOAA Earth System Research Laboratory 2019); 6 6. Multivariate Ocean Climate Index (MOCI), a regional indicator of productivity intended to predict biological responses along California’s continental shelf system that incorporates various indices (PDO, NPGO, NOI MEI, the Bakun Upwelling Index, sea level, alongshore wind stress, sea surface temperature, sea air temperature and sea level pressure; García-Reyes and Sydeman 2017). This index is region specific, so we used values reported for northern California. 7

At the time of this study, data for each of these indices except MOCI were regularly updated and available online at the link associated with each footnote. Each of the six indices were summarized at seasonal and annual

1 https://www.ndbc.noaa.gov/station_history.php?station=46022 (Historical Standard Meteorological Data) 2: http://research.jisao.washington.edu/pdo/PDO.latest.txt (Jan 1900 to Oct 2018 available as of 9 Sept 2019) 3 https://www.cpc.ncep.noaa.gov/products/GODAS/ocean_briefing_gif/global_ocean_monitoring_2019_08.pdf (Nov 2019 to August 2019) 4 http://www.o3d.org/npgo/npgo.php 5 https://www.pfeg.noaa.gov/products/PFEL/modeled/indices/NOIx/data/noix.txt 6 https://www.esrl.noaa.gov/psd/enso/mei/ 7 Supplement A to García-Reyes and Sydeman (2017)

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates B-2 Humboldt Wind Project September 2019

time scales for each year between 2002 and 2011. Seasonal assignments were made by dividing the year into 3- month intervals, with winter being January through March.

To evaluate the degree to which marine productivity can predict inland activity of murrelets, we first identified which of the five time periods (winter, spring, summer, fall, or annual) best explained variability in inland activity of murrelets each index using univariate linear regressions. On an index-by-index basis we found that inland variability in murrelets was best explained by annual SST, winter PDO, spring NOI, winter MEI, and winter MOCI (Figure B-1). For NPGO, none of the five time periods were related to murrelet activity (p>0.5, r2<0.05; Figure B-1) so this variable was dropped from the analysis.

After identifying which of the five time periods (winter, spring, summer, fall, or annual) best explained variability in inland activity of murrelets for each index, we intended to use a multivariate general linear model with a stepwise model selection approach to assess which combination of indices best predicted murrelet activity. Before doing this, we assessed if the data met an important assumption of multivariate general linear models, that the explanatory variables be unrelated to each and not colinear. We found that each of the indices exhibited strong collinearity (Figure B-2), with correlation coefficients ranging between 0.64 and 0.92 and variance inflation factors ranging from 4.8 to 13.6 (Montgomery et al. 2001). Even low levels of collinearity bias

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates B-3 Humboldt Wind Project September 2019

analyses in terms of the calculated fit of the model (r2) resulting in inaccurate model parameterization and exclusion of significant predictor variables (Graham 2003).

Figure B-2. Visual representation of the correlations between ocean productivity predictor variables, with the size of dots indicating the strength of the correlation and the color representing the direction of the association.

Since the independent variables violated the assumptions of multivariate regression, the most appropriate approach was to conduct a separate regression for each variable. Model fit was compared using AIC and the best predictor(s) of inland activity were identified. We found that both winter PDO and annual SST were similarly good predictors of inland variability in murrelet activity (∆AIC = 1.44), with PDO explaining 56.9%

(F1,7=9.25, p=0.019) of the variability in inland activity of murrelets in the Eel River Valley and SST explaining

49.5% (F1,7=6.58, p=0.035) of the variability.

For our purposes, winter PDO and specifically the February PDO was selected over SST to predict the range of interannual variability expected for passage rate of murrelets through the proposed turbine string using both a deterministic and a probabilistic (Bayesian) approach. The PDO had slightly greater predictive power (as indicated by the r2 value). Also, the PDO provides a complete dataset that extends back to 1900 (Mantua et al. 1997) whereas the SST from at the Eel River buoy only began in 1982 and the sensor at this station has frequently malfunctioned (e.g.: 2004, 2010, 2011, 2012, 2013 and 2014), and often malfunctions for substantial amounts of time (as long a year; NDBC 2019).

For the deterministic model, we used a simple linear regression to estimate the best-fit equation of the line describing the relationship between the predictor, PDO, and the predicted, inland activity. We assumed that passage rates along the ridge would be proportional to passage rates in the Eel River Valley (where the HRC indexes murrelet activity on an annual basis) so that PDO-based estimates could be proportionally adjusted to reflect passage rates along ridge. Using this logic, the average, minimum, and maximum values of PDO

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates B-4 Humboldt Wind Project September 2019

observed over the previous 35-years (1983-2018) were used to predict the average, minimum, and maximum passage rates expected along the ridge over the life of the Project.

For the probabilistic model, interannual variability in passage rate was assessed using a similar method as described for the deterministic model except instead of generating a single best-fit line for inland activity and PDO the Bayesian approach allowed for uncertainty in the slope and intercept of the regression line. This resulted in many (100000) unique equations that each approximated the relationship between PDO and inland activity. To predict how inland activity would change across many years, a PPD of PDO was generated using all values of winter PDO since 1900 and 100000 values were randomly selected as inputs for the equations predicting inland activity of murrelets. Like the deterministic model, these estimated were proportionally adjusted for the ridges to estimate the interannual variability of passage along the ridge based on a 118-year dataset.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates B-5 Humboldt Wind Project September 2019

Appendix C. Bayesian likelihoods, priors, and posteriors

For the probabilistic CRM, a Bayesian approach was used to generate a posterior predictive distribution (PPD) for three critical inputs for which there was either variability or uncertainty regarding the most appropriate value to use: flight speed, avoidance, and interannual variability in inland activity. Ultimately, PPDs generated by Bayesian inference could be input into the CRM such that the final output of the model was a probability distribution representing the credibility of all possible values of collision risk for murrelets given variability and uncertainty associated with these three inputs. Additional inputs were set at their maximum value (e.g., rotational speed) or did not vary in a substantial way (e.g., murrelet wingspan) and therefore did not add uncertainty to the outcome (see Section 2.1.2.3)

Generating a PPD using Bayesian inference required technical mathematical specifications. For each variable of interest, the following must be specified: 1) a likelihood distribution that will allow the posterior distribution to approximate the distribution of the actual data and 2) prior assumptions about the parameters responsible for defining the location, scale, and shape (for distributions that can assume more than one shape) of that likelihood function. For example, if the underlying data being modeled are normally distributed, a distribution of likelihood that would allow the posterior distribution to approximate the distribution of the actual data would be the normal distribution and a prior assumption about the value of parameters responsible for determining the location (mean) and scale (variance) of that normal distribution would be specified. For this analysis, likelihood distributions were selected to best represent the distribution of the empirical data and prior information about the value of parameters were specified to be generally uninformed so that they did not influence the resulting posterior distribution. It is a minor detail, but many statistical analyses generally require or report variance whereas the RJAGS package (Plummer 2018) used for this analysis required that scale parameters be specified as precision (the inverse of variance). We specified precision in the analysis but report it here as variance.

As described in Sections 2.2.4 2.2.5 2.2.6 comprehensive datasets were generated for each of the three critical inputs to the probabilistic CRM. Flight speeds were based on the flight speed of the 145 total ridge crossings by murrelets detected by the radar surveys associated with this CRM. Avoidance was based on 35 reports that met selection criteria and were derived for other avian species, mostly seabirds, at onshore wind facilities (see Appendix A). The relationship between inland activity and ocean productivity (specifically, winter values of the Pacific Decadal Oscillation; see Appendix B) was based on 10-years (2002-2011) of radar surveys in the Eel River Valley by Humboldt Redwoods Company (HRC 2018) and concurrent measures of ocean condition. Finally, for input into this regression-based process model for which ocean conditions served as a predictor of inland activity, a 120-year (1900-2019) dataset of the Pacific Decadal Oscillation was used.

Because the likelihood distribution should approximate the distribution of the actual data (Kruschke et al. 2015), the appropriate likelihood functions for each dataset was determined first by examining if data were normally

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-1 Humboldt Wind Project September 2019

distributed. Based on normal quantile plots (Figure C-1), murrelet activity and ocean conditions were well approximated by a normal distribution whereas flight speed and avoidance were not.

Figure C-1. Normal quantile plots for marbled murrelet flight speed, avian avoidance of turbines, inland activity of murrelets and ocean productivity measures at 10-year and 120-year scales. Datasets for which most points fall within the 95% confidence interval (represented by the gray shading around the line of fit) were well approximated by the normal distribution.

To assess which distribution best approximated flight speed and avoidance, these data were fitted to the normal, log-normal, gamma, exponential, and Weibull distributions using the Fitdistrplus package in R (Delignette- Muller and Dutang 2019) and the degree of fit achieved was assessed using AIC (Burnham and Anderson 2002). Flight speed data were best approximated by a log-normal distribution (AIC = 880.5, ∆AIC = 3.3 to 318.2; Figure C-2). Avoidance data were best approximated by an exponential distribution (AIC = 47.2, ∆AIC = 42.7 to 69.5; Figure C-2) with the Weibull distribution serving as the next best approximation (AIC = 89.9). As would be expected given the shape of the exponential distribution, using this to model the empirically derived avoidance data caused a majority of the probability distribution at to be concentrated at 100% avoidance which, given the density curve of the underlying data, likely overestimates avoidance (Figure C-2). In contrast, using the Weibull distribution to model the empirically derived avoidance data resulted in an underestimation of avoidance given the density curve of the underlying data (Figure C-2). Given the need to be conservative and err in a precautionary manner, the Weibull distribution was selected over the exponential distribution as the most appropriate distribution for modeling avoidance.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-2 Humboldt Wind Project September 2019

Figure C-2. Distributional comparisons for marbled murrelet flight speed and avian avoidance data used to generate posterior predictive distributions using Bayesian inference. For flight speed and avoidance, the density distribution of the actual data (red line) and the various distributions considered to serve as a model for those underlying data were depicted to facilitate visual assessment of fit for each.

A log-normal distribution best described right-skewed data (Limpert et al. 2001); in the case of murrelet flight speeds, individuals are more likely to fly slower than the average speed (mode=19.6 m/s, mean=23.3 m/s based on the 145 crossings but see Elliot et al. 2004 for a more comprehensive dataset). Because flight speed is directly related to collision risk (with slower speeds having greater risk) a log-normal likelihood for flight speed facilitated a more accurate estimation of collision risk by generating a PPD where flight speeds slower than average were more likely to occur. This would result in the distribution of collision risk to peak at a value greater than would be the case if a normal likelihood were assumed because, if flight speed were modeled as normal, the most likely estimate of collision risk would coincide with the smaller average value rather that the larger most-likely value.

The Weibull distribution provides a concise yet flexible framework to describe the likelihood of data (Flemming 2001, Upadhyay and Peshwani 2003). This distribution is versatile and able to approximate various other distributions (including the normal, log-normal, gamma, and exponential distributions as well as variants thereof; Weibull and Stockholm 1951, Lai 2006). Traditionally, the Weibull distribution has been used to model survival, hazard, and time to failure (Abernethy et al. 1983, Lai 2006). Avoidance of wind turbines by non- raptorial birds had not previously been modeled using a Bayesian framework and, therefore, an appropriate distribution to model this phenomenon has not been previously identified. As such, Weibull distribution was the most appropriate choice given that it was the best conservative approximation of the 35 estimates of avoidance relative to the other distributions considered and its shape is flexible such that future inputs of new data will allow its shape to be refined to match the true distribution of avoidance.

As stated already, both murrelet activity and the values of PDO were normally distributed. However, since the intention was to relate these two normally distributed variables to each other, some additional details regarding the nature of their relationship had to also be identified. We assumed that murrelet activity was linearly related to PDO (Appendix B). Further, a description of the distribution of error about this liner relationship had to be made; observed data often contain outliers due to extraneous influences that are not explicitly accounted for.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-3 Humboldt Wind Project September 2019

As such, a Student’s t-distribution was used to specify the distribution of points around a best-fit line (i.e., residuals) following a recommendation by Kruschke (2015). The t-distribution matches a normal distribution when the shape parameter (i.e., degrees of freedom) exceeds 30 and, if less than 30, remains robust against outliers.

The appearance of parametric likelihood distributions is determined by a set of parameters that define where they fall on the x-axis (location parameter), how precise they are about the location parameter (scale parameter), and in some instances the shape of the distribution (shape parameter). For the most familiar distribution, the normal distribution, the location is set by the mean (µ) and the scale is set by the variance about the mean (σ). The log-normal distribution also has two parameters and is very similar to the normal distribution; the location is set by the log of the mean (log µ) and the scale is set by the log of variance about the log of the mean (log σ). The Weibull distribution can have more than two parameters but, in this instance, only the scale (α) and shape (β) parameters were specified; shape parameter is what gives the Weibull distribution its flexibility (e.g., β of 1=exponential distribution, β of 3=normal distribution). A linear relationship is described by where it intercepts the y-axis (β0; location parameter) and the slope of the line (β1; scale parameter). Finally, a Student’s t-distribution has two parameters, a scale parameter (σ) and a shape parameter (ν).

Bayesian analyses require that prior assumptions regarding each parameter associated with likelihood functions to be explicitly stated. This is because Bayesian inference combines prior assumptions regarding these parameters with observed data to estimate a posterior distribution of location, shape, and/or scale parameters associated with the likelihood function associated with each variable of interest. These prior assumptions can exert influence over the appearance of a PPD and must be carefully selected to ensure they only exert a justifiable level of influence. In the case of this analysis, we have chosen priors that were generally uninformed or were justifiably informed.

A completely uninformed prior would be one that assumes a uniform distribution that can range anywhere from 0 to infinity. However, for practical purposes including minimizing computing time, it is best to specify priors that are at least mildly informed to match the scale of the data being modeled; in most instances, this is the approach we took (Table C-1). For priors assumed to be uniform, the range of values specified for each can be compared to the posterior value to ensure that the prior assumption was overly inclusive of the true parameter value, thereby exerting no influence on the posterior value of the parameter.

For flight speed, the prior on the mean was slightly informed by a study reporting the speed of 3000 murrelets flights tracked using radar and accurate to ±1 m/s at 15 study sites (Elliot et al. 2004). The mean of flight speed to be centered around a value of 3.1 m/s on the log scale (or 22.2 m/s on the actual scale, which is 0.4 m/s slower than observations by Elliot et al. 2004; Table C-1). Despite a moderately informed mean value of the prior, the variance about this mean (3.84 on the log scale or 46.6 on the actual scale) was so large that, functionally, this was an uninformed prior (Figure C-3A). Because we wanted to ensure a conservative outcome,

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-4 Humboldt Wind Project September 2019

in terms of final collision estimates, we opted to use this uninformed prior although enough data on murrelet flight speed (Elliot et al. 2004) already exists to justify use of a highly informed prior.

Priors associated with the relationship between ocean productivity and the inland activity of murrelets were also not necessarily assumed to be uniformly distributed; the slope and intercept parameters associated with the line of this relationship were set to follow a normal distribution centered at zero (Table C-1). However, the variance of the distribution about the mean was extremely large (set to 1000; Table C-1) and, therefore this served as an uninformed prior (Figure C-3B). If anything, this prior was more conservative relative to a uniform distribution as it was set to bias values of slope and intercept towards zero rather than considering all values to be equally likely. Furthermore, the prior for the shape parameter of the Students-t distribution (which set the expected distribution of residuals about the best-fit regression line) was specified to be exponentially distributed with a mean of 30 (or lambda of 1/30; Table C-1, Figure C-3C) based on the recommendations and logic presented by Kruschke (2015) for estimation of regression lines that are robust to the influence of outliers.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-5 Humboldt Wind Project September 2019

Likelihood Associated Prior distribution for Exact value of priors for Posterior for Input PPD Figure function parameter parameters parameters parameters (95% HDI) (1) Flight speed (m/s) Log-normal log(µ) normal log(µ) = 3.1; log(σ) = 3.84 a 3.09-3.16 C-4 A log(σ) uniform min = 0.0002 ; max = 196 0.20-0.25 C-4 A (2) Avoidance (%) Weibull α uniform min = 0 ; max = 200 111.95-153.67 C-4 B β uniform min = 0 ; max = 200 99.48-99.97 C-4 B (3) PDO and inland activity: relationship Linear β0 normal µ = 0; σ = 1000 35.12-42.61 C-4 C β1 normal µ = 0; σ = 1000 -10.94-2.38 C-4 C (4) PDO and inland activity: residuals Student's t ν exponential µ= 30 (min = 1, max = ∞) b 110.95-153.67 C-4 C σ uniform min = 0.0001 ; max = 1000 2.53-9.31 C-4 C (5) Long-term PDO Normal µ uniform min = 0.00005 ; max = 53.6 0.0047-0.026 C-4 D σ uniform min = 0.0009 ; max = 959.9 0.96-1.24 C-4 D a To directly compare these numbers to flight speeds, they must be back transformed from the logarithmic form b The more familiar form of the parameter of the exponential distribution is lambda (λ) which is equivalent to 1/µ

Figure C-3. Probability distributions specified for the priors of (A) the mean of flight speed [log(µ)], (B) the slope [β1] and intercept [β0] of the relationship between ocean productivity and inland activity of murrelets, and (C) the shape parameter [ν] that described the distribution of residuals about the best-fit line relating measures of ocean productivity to inland activity of murrelets. Although the priors for mean of flight speed were specified on a logarithmic scale, for intuitive purposes it is visualized here on the actual scale.

The primary goal of Bayesian analyses presented here was to generate a PPD for each input identified as having variability or uncertainty that could cause collision risk estimates to differ substantially from what had been calculated using a deterministic approach (Figure C-4). To generate the PPD required that posterior estimates of each of the parameters are input into the specified likelihood function for each variable (Table C-1). The resulting PPD for each of the critical inputs of the probabilistic CRM is a likelihood of occurrence assigned to the entire range of flight speed, avoidance, the relationship between measures of ocean productivity (PDO) and inland activity of murrelets, and the PDO itself. As such, basing calculations of the probabilistic model on these PPDs allowed it to quantitatively consider the likelihood of all potential estimates of collision risk.

Figure C-4. Raw data, a representative subset of credible distributions or regression lines resulting from 100000 successive random selections from four Markov Chains (in blue), and the final posterior predictive distribution (in black) of (A) murrelet flight speed, (B) avoidance, (C) the relationship between ocean productivity and the inland activity of murrelets, and (D) variability in ocean productivity (specifically, the Pacific Decadal Oscillation or PDO; see Appendix B). Only a subset (n=1500) of credible distributions or regression lines were plotted as additional lines would overlay what has already been depicted.

To evaluate the validity of Bayesian analyses, posterior predictive checks are essential. The first posterior predictive check was to confirm that the Markov Chain Monte Carlo (MCMC) values for parameters used to generate each PPD were representative of the posterior distribution for each parameter. This was accomplished by (1) examining a trace plot of the trajectory of each of the chains to confirm they were overlapping (i.e., converged) and did not settle into unusual regions of parameter space (see Figure C-5A for an example), (2) examining a density plot of parameter values sampled by each chain to confirm that they overlapped (see Figure

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-7 Humboldt Wind Project September 2019

C-5B for an example), and (3) confirming that a numerical measure of convergence, the Gelman-Rubin statistic (also called 푅̂; Gelman-Rubin 1992), was between 1.0 and 1.1 (see Figure C-5C for an example). For all parameters presented in Table C-1, quantiles for the posterior distribution have been provided in addition to the exact value of the Gelman-Rubin statistic (Table C- 2).

The next posterior predictive check was to assess if the effective sample size (ESS) for each parameter used to generate the PPDs was enough to produce a stable and accurate estimate of the distribution. ESS is a measure of chain length describing the amount of independent information provided by the chains after accounting for autocorrelation between sequential steps (see Figure C-5D for an example). Larger values of ESS tend to produce a more stable and accurate estimate of the central tendency and HDI for parameters; Kruschke (2015) heuristically recommended an ESS of 10,000 but noted that lower values of ESS could also be sufficient. For all parameters presented in Table C-1, ESS has also been provided (Table C- 2).

Figure C-5. A representative example of the plots used to confirm that the Markov Chain Monte Carlo (MCMC) values were representative of the posterior distribution of a single parameter and that the effective sample size (ESS) was large enough to produce a stable and accurate estimate of the distribution.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-8 Humboldt Wind Project September 2019

The final posterior predictive check was to confirm that the PPDs approximated the underlying data reasonably well and that there were no systematic discrepancies indicating a need to expand or change the model so that it better described the data. For each input subjected to Bayesian inference for the probabilistic CRM, the credible distribution and regression lines generally encompassed the original data (Figure C-4); this was especially true for the relationship between ocean productivity and the inland activity of murrelets as well as the PPD for the long-term PDO. It appeared that modeled flight speed may favor slightly faster flight speeds relative to the underlying data but, on the other hand, the model for flight speed also considered flight speeds slower than the slowest speed observed for murrelets by this study (18 m/s) to be likely. Because this would result in collision probabilities to be greater than and less than what would be expected based on the observed data, the discrepancies between observed and modeled flight speed are likely inconsequential to the final estimate of collision risk. It is also possible that the model for avoidance could be refined; while the peak of the credible distribution lines does coincide with the most frequently observed avoidance, the height (i.e., likelihood) of the probability distribution at the peak for modeled avoidance was lower than would be expected given the relative likelihood of the observed data at its peak. This discrepancy between modeled and observed avoidance resulted in a very conservative model of avoidance as lower values were assigned more credibility than would be expected given the underlying data and what is currently known about avoidance in other species of birds (Appendix A). We conclude that the PPDs generated by Bayesian inference as inputs for variables identified as critical to the estimates produced by the probabilistic collision risk model served as a good approximation of the underlying data.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates C-9 Humboldt Wind Project September 2019

Table C- 2. For all parameters detailed in Table C-1, the quantiles of the posterior distribution and information relevant to posterior predictive checks have been provided, specifically the Gelman-Rubin Statistic (푹̂) and the effective sample size (ESS).

Likelihood Associated Prior for Posterior estimates Quantiles Input function parameters parameters Mean SD 2.50% 50% 97.50% 푹̂ ESS (1) Flight speed (m/s) Log-normal log(µ) normal 3.12 0.02 3.09 3.12 3.16 1.00 101,177 log(σ) uniform 0.22 0.01 0.20 0.22 0.25 1.00 99,567 (2) Avoidance (%) Weibull α uniform 139.03 11.70 110.95 141.55 153.67 1.00 164,149 β uniform 99.72 0.12 99.48 99.72 99.97 1.00 296,828 (3) PDO and inland activity: relationship Linear β0 normal 38.80 1.89 35.12 38.78 42.61 1.00 96,138 β1 normal -6.61 2.14 -10.94 -6.59 -2.38 1.00 98,155 (4) PDO and inland activity: residuals Student's t ν exponential 32.38 29.32 2.65 23.65 111.19 1.00 76,171 σ uniform 4.80 1.81 2.53 4.42 9.31 1.00 61,504 (5) Long-term PDO Normal µ uniform 0.10 0.07 0.00 0.09 0.26 1.00 96,005 σ uniform 1.09 0.07 0.96 1.08 1.24 1.00 99,325

Appendix D. Limited evidence of avian fatalities caused by rotor turbulence at wind energy facilities

For small birds, there is little evidence that encountering rotor turbulence results in a fatal, non-collision outcome. The original source for this issue in the literature appears to be Winkelman (1992b) as summarized in Winkelman (1994b); they provided an anecdotal description of flight behavior in a monitoring report (however, assessment of down-stream turbulence had not been part of the monitoring design). Winkelman (1992) reported that 3 songbirds were fatally impacted by turbulence during nocturnal migration. However, visual observations of bird flight behavior were made using a night scope coupled with a video recording device and the turbulence generated by the turbines was never quantified. Winkelman (1992b) described the impact of these 3 songbirds: “during and after passage, strongly fluttering, just after passing the mast the bird suddenly flutters down with no wing beats anymore” (Appendix 16b of Winkelman 1992b). For these 3 individuals, the presence of external or internal injury was not confirmed by necropsy as none of the bodies were physically recovered. Larger birds (gulls, ducks) were also observed passing through the rotors (n=4), but none were fatally impacted by rotor turbulence.

Other subsequent monitoring efforts have failed to identify turbulence as a cause of mortality (see Desholm 2006, Krijgsveld et al. 2009, May et al. 2017). There are several reasons why this might be the case. For one, formal studies intended to directly link down-stream turbulence to avian fatality have not occurred. Winkelman (1992) monitored an old-style wind farm where turbines had a rotor radius of 15 m and maximum rotation speed of 48 rpm. In contrast, typical modern turbines have a rotor radius near 75 m and maximum rotation speed of 17 rotations per minute. Turbulence associated with the larger turbines used today may exert different influences on small birds transiting through the rotor (Krijgsveld et al. 2009). Another potential reason that avian mortality due to turbulence has not been subsequently observed is that conclusions made by Winkelman (1992b) could be mistaken; although direct collision was not seen using the thermal imaging technology, it is possible that these birds did in fact come into direct contact with a turbine blade which resulted in a fatal outcome. Finally, concern may result from confusion with the impacts of rotor-generated turbulence on bats, specifically the affliction known as “barotrauma” (Rydell et al. 2017); this phenomenon is distinct from the potential impact suggested by Winkelman (1992b) on birds. Barotrauma is unknown in birds (Baerwald et al. 2008) because the respiratory system of birds, specifically their rigid lungs with unidirectional ventilation and cross-current blood-gas relationship, is anatomically distinct from that of bats (West et al. 2007).

Widely applied collision risk models (CRM) for birds, including Band (2012) and Tucker (1996), do not include a specific modifier for turbulence wake behind the rotor of a wind turbine. For marbled murrelets, inclusion of a turbulence factor in a CRM would likely result in overestimate of fatality. Relative to songbirds, murrelets are 2 to 40 times heavier, have different wing loading characteristics, and fly relatively fast as they transit between the ocean and inland nesting habitat (Elliot et al. 2004). To the best of our knowledge, there is only one CRM that has included a factor for turbulence effects in the wake of a rotor: the marbled murrelet

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates D-1 Humboldt Wind Project September 2019

CRM from the Bear River Wind Project (Sanzenbacher and Cooper 2010). This factor was included at the request of the regulatory agencies reviewing this project, and the authors of the CRM specifically noted an absence of evidence to justify this adjustment factor beyond the first and only report by Winkelman (1992b). Other murrelet-specific CRM’s have not included downstream turbulence; Nations and Erickson (2009; known as Radar Ridge project) did not include down-stream effects and noted that any negative bias would probably be negligible. In a CRM for Red Knots (Calidris canutus), Gordon and Nations (2016) could not find any quantitative basis for including turbulence as a factor and suggested that the effects of turbulence were unlikely to be significant enough to include in their CRM. They also indicated that, in their extensive review of the literature at the time, they did not find other CRMs that had incorporated this factor.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates D-2 Humboldt Wind Project September 2019

Appendix E. Comparison of collision risk estimated for oblique approach angles relative to a simplified method for marbled murrelets

Accurate estimation of avian collision risk can be a critical step in the planning and operation of wind energy facilities. To estimate avian collision risk, a kinematic calculation of the probability that a bird flying through a rotor swept area will be struck by a blade is required. Exact calculation of this probability is very complicated and all Collision Risk Models (CRMs) make assumptions and simplifications so that reality can be mathematically approximated. However, over-simplification can result in inaccurate estimates of the true collision risk if they no longer account for principal features of avian-turbine interactions.

One potentially important feature that has been generally overlooked by the most widely used CRMs is oblique approach angles; an assumption that avian flight paths are perpendicular to the turbine orientation is built into the calculations derived by both Tucker (1996) and Band (SNH 2000, Band 2012). This simplification was justified by Band (2012) based on the logic that, as the angle of approach increasingly deviates from 90-degrees, the area occupied by a rotor from the perspective of a bird becomes increasingly smaller until the area occupied by the rotor at 0 and 180 degrees is very small. For example, using turbine dimensions specified in Table 3 of this report, the area occupied by the rotor would be reduced by 99.6%. Probabilistically, it is very unlikely that a bird would enter a rotor swept area at this extreme angle because the total front of interaction is very small; however, the chance of being struck would also become quite great. Band (2012) assumed that these two opposing issues would cancel each other out. Unlike Tucker (1996), Band (2012) recognized that the collision risk would be greater for birds flying against a wind as they pass through a rotor swept area and this model does account for this by calculating collision risk separately for flights that are with and against the wind.

Refinements to calculations made by Tucker (1996) and Band (2012) that enable determination of collision probabilities for birds that approach turbines at oblique angles have been proposed by Holmstrom et al. (2011) and Christie and Urquhart (2015), respectively. Following presentation of the equations needed to assess collision probability at perpendicular angles, both studies present case studies to demonstrate the potential importance of incorporating these refinements. Based on their case studies, both studies concluded that the angle of approach should be considered when estimating avian-turbine collision risk.

Based on recommendations to consider the effect of oblique approach angles in the assessment of collision risk, we compared the collision probability estimated using the Band Model (2012) with collision probabilities estimated by the refinement on the Band Model (2012) developed by Christie and Urquhart (2015). Using the Excel spreadsheet provided as supplementary material by Christie and Urquhart (2015), we generated a collision probability for each of the observed murrelet ridge crossings in 2018. To do this we first determined the angle that each flight would enter a rotor swept area based on the flight direction and wind directions recorded for

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates E-1 Humboldt Wind Project September 2019

each crossing (Stantec 2018). We then input crossing-specific bird speed (speed over ground) and wind speed recorded for each of the crossings.

All estimates were based on passage through the large turbine modeled previously (See Table 4 for specifications). The Band Model (2012) estimated that a murrelet 25 cm in length with a wingspan of 41 cm, and average speed of 22.7 m/s over ground would have a 6.5% chance of colliding with a blade when flying against the wind and a 2.5% chance of colliding with a blade when flying with the wind (Figure E-1); when considering the proportion of 55 crossings from 2018 that were upwind versus downwind, the average probability of collision estimated by the Band Model for this scenario was 4.32%. The probability of collision derived using the Christie and Urquhart Model (2015) for each of the crossings ranged from 1.8% to 8.8% and averaged 4.56% (Figure E-1). Based on the differences in average collision probability derived from these two approaches, the Christie and Urquhart (2015) method would cause fatality estimates to rise by 0.0186 birds per year (0.56 birds over 30 years). Based on the similar outcomes from these two approaches, we conclude that Band’s (2012) assumption that the tradeoff between the area occupied by the rotor swept area and risk of collision for the various angles of approach would cancel.

For the murrelet crossings used to assess collision risk, the assumptions of the Band Model closely approximated the approach recommended for use by Christie and Urquhart (2015). Murrelets detected crossing the ridges associated with the Project were typically confronted with winds that were paralleling their flight path such that they would encounter turbines in a perpendicular orientation (Figure E-2). This similar orientation between the wind direction and bird direction likely facilitated a reliable approximation of collision risk by the Band Model (2012) to in this situation.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates E-2 Humboldt Wind Project September 2019

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates E-3 Humboldt Wind Project September 2019

Figure E-2. Circular histogram generated using a 10-degree bin that depicts the proportion of flight directions (MAMU) and wind directions (WIND) determined for 55 marbled murrelet ridge crossings detected during radar surveys. To facilitate comparison, murrelets and wind both move towards the indicated direction.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates E-4 Humboldt Wind Project September 2019

Appendix F. The influence of fog on navigation and ridge- crossing by murrelets

Coastal marine fog is a global phenomenon that occurs along the east sides of the world’s ocean basins, also called eastern boundary currents; these currents carry water much cooler relative to the air which promotes predictable formation of seasonal “sea fog” along the western coasts of various continents including Europe, Africa, and South America in addition to North America’s west coast. Fog is not only restricted to coastal areas and various physical processes promote the formation of fog. Different fog types include radiation fog, precipitation fog, steam fog, upslope fog, valley fog, freezing fog, and ice fog, as well as “sea fog”, which is a type of advection fog (NOAA NWS 2019).

Despite the ubiquity of fog, very few studies have been conducted with the intent of documenting if, and how, birds navigate through fog (Becciu et al. 2019). The most rigorous assessment of avian navigation in fog linked various widespread meteorological conditions to migration intensity of songbirds and identified fog as an important factor influencing migration intensity, concluding that birds actively avoided flying into fog (Panuccio et al. 2018). The remaining few studies have been opportunistic rather than designed to specifically investigate if and how fog impacts navigation by birds. A study of roosting sandhill cranes (Grus canadensis) found that fog caused individuals to depart night roosts later than usual, upon departing individuals remained relatively close to the roost, and were more likely to return to the roost site rather than navigate to foraging areas (Kirsch et al. 2015). Fog has also been opportunistically linked to an interruption of transits by little penguins (Eudyptula minor) between the ocean and their inland nesting areas; over half of breeding individuals were either unable or unwilling to return to the colony during an unusual fog event and, individuals that chose to navigate inland during fog arrived to nests 2 hours later than typical on fog-free evenings (Chiaradia et al. 2007). From the few studies conducted to date, the common conclusion was that various types of birds—from migrating songbirds, to roosting cranes, to nesting penguins—appear reluctant to transit through fog. It has been speculated that this reluctance is because fog obstructs their ability to navigate using visual cues and, possibly, their ability to detect the presence of predators.

Interestingly, all seabirds nesting in the Pacific northwest frequently experience thick fog and often nest on complicated landscapes (cliffs, offshore rocks). Based on the studies described above, it seems that the ubiquity of fog in coastal areas of the Pacific northwest would noticeably interfere with ability of seabirds to produce young as the presence of fog might inhibit navigation between foraging areas and nesting locations. Marbled murrelets successfully navigate frequently through fog or adjust their flight times and mode of flight to accommodate navigation when encountering fog in the Pacific northwest. This is because they nest in the upper canopy of coastal old-growth forests; these forests depend on thick fog throughout the spring and summer months to provide essential hydration during these rain-free periods (Johnstone and Dawson 2010, DelaSalla et al. 2015). Despite this apparent need to frequently navigate in and around fog while nesting, the willingness and ability of seabirds nesting in the Pacific northwest (including marbled murrelets) to successfully navigate

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates F-1 Humboldt Wind Project September 2019

through fog remains poorly studied. What little information exists comes almost exclusively from studies of murrelets transiting inland along flyways in the presence of fog using audiovisual and radar-based methods.

Similar to what has been found in other bird species (summarized above), murrelets seem to predictably alter their inland transit patterns in the presence of fog. Foggy weather results in lower-light levels relative to clear mornings and murrelets have been noted to delay and prolong their inland flights in response to fog (Hamer and Cummins 1990, Paton and Ralph 1990, Rodway et al. 1993, Naslund and O’Donnel 1995, Cooper and Blaha 2002); this delayed transit time has been corroborated by nest-based observations, where incubation exchanged by nesting pairs are similarly delayed on mornings with thick fog and low cloud ceilings (Naslund and O’Donnel 1995, Nelson and Peck 1995). In addition to delaying inland transit times to navigate in adequate light levels, murrelets also have been documented altering their inland flight patterns during periods of dense fog. Most often murrelets have been observed to respond to these conditions by increasing vocalization rates and flying relatively low in elevation (Naslund and O’Donnell 1995) or restricting movements to just above the valley floor (Hamer et al. 1995). There have also been anecdotal reports of murrelets flying above the fog and waiting until just above the forest canopy to drop below the fog (Naslund and O’Donnell 1995).

Since murrelets primarily fly along waterways (Eisenhawer and Reimchen 1990, Paton and Ralph 1990, Rodway and Regehr 2000), the propensity of murrelets to engage in ridge crossing behavior has only recently become a topic of interest given its relevance to assessing the potential impact of installing turbines along ridgelines on the route to murrelet nesting habitat. It is important to recognize that murrelets transiting inland to nesting habitat east of the Humboldt Wind Project will primarily navigate to these nesting stands along watercourses rather than by ridge crossing. To better assess if, and to what degree, murrelets alter their ridge crossing behaviors in fog, we present a preliminary assessment linking ridge crossings by murrelets along Bear River and Monument Ridges observed during the 2018 nesting season using radar (Stantec 2018). The virtue of using radar to assess murrelet navigation in fog is that radar penetrates through fog and can continue to “see” when other methods do not (Burger 1997, Evans Mack et al. 2003, Stumpf et al. 2011, Sanzenbacher et al. 2015).

This assessment is preliminary because understanding how fog influenced murrelet use of these ridges was not an original goal of this risk assessment. Although the presence and absence of fog was noted on each survey day in 2018, notes regarding fog were recorded with less detail than ideal for a rigorous assessment; for example, specific details regarding the impact of fog on the visibility of airspace to be occupied by turbines was not consistently noted. As such, all surveys in 2018 noted to have fog were treated as if the fog was thick enough to impact visual navigation by murrelets and represents a conservative estimate that could be better refined with more detailed environmental notes. We hypothesized that murrelets would be equally likely to cross the ridge in the presence and absence of fog and this was compared using a chi-squared test of independence.

Fog was present during 26.2% (n=32) surveys and absent during 73.8% (n=90) surveys in 2018. Thus, if murrelet crossings were independent of fog, one would expect that approximately 26% would occur during surveys with fog and 74% would occur during surveys without fog. During these 122 surveys, 55 ridge crossings were observed; 14.5% (n=8) were observed during fog and 85% were observed in the absence of fog. Thus,

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates F-2 Humboldt Wind Project September 2019

this preliminary assessment provides direct evidence that murrelets are less likely to engage in ridge crossing behaviors in the presence of fog (χ2=3.88, df=1, p=0.0489). This observation also corroborates conclusions by Hamer et al. (1995) that murrelets tend to remain at low elevation in valleys as they transit between the ocean and nests in the presence of fog.

Conclusions Marbled murrelets are less likely to make ridge crossings in the presence of fog. This is because they preferentially fly below the fog and tend to remain at low elevation in their primary travel corridors. Anecdotally, there are also observations of murrelets flying above fog and, once over nesting habitat, descending into fog to access nests.

In the event that murrelets do cross ridges in the presence of thick fog, it could reduce their ability to visualize their surroundings. This may be compensated for by later timing allowing for brighter lighting and better visualization. Either way, the avoidance rates used in the Deterministic and Probabilistic CRMs do account for what would be expected for birds crossing Bear River and Monument Ridges in the presence of fog. The full suite of weather conditions experienced by birds that are approaching turbines are included in the calculation of avoidance rates (including fog, darkness, rain, etc.).

Avoidance has been quantified at wind facilities installed onshore along eastern-boundary currents that experience seasonal marine fog analogous to what occurs in Humboldt County. For example, the United Kingdom (UK) is located on the cold North Sea and is frequently blanketed by the marine fog. Importantly, the UK is committed to being coal-free by 2025 and is 6th in the world in terms of installed wind power capacity and produces 7x more megawatts per unit area than the United States; as of July 2019, Scotland generated twice as much energy needed to power each of their 2.6 million homes by wind energy alone. Many studies to better understand bird avoidance of wind turbines at both terrestrial and offshore facilities have occurred in the UK. Because fog is a regular feature, the avoidance rates quantified by these studies account for the effect of fog on collision risk. In reference to the appropriateness of 0.98 as an overall avoidance rate, this value has been recommended by the Scottish version of the U.S. Fish & Wildlife Service since 2010, and is based on all terrestrial studies of avian avoidance that have occurred in and around the UK, including at foggy coastal wind facilities, for use by Scotland, a country that regularly experiences marine fog formed by the same physical processes that cause fog to be present at the Project site. Thus, we conclude that murrelets may cross Bear River and Monument Ridges and that the avoidance rate used to estimate collision risk for those murrelets includes exposure to fog as part of the calculated avoidance.

Marbled Murrelet Collision Risk Assessment H. T. Harvey & Associates F-3 Humboldt Wind Project September 2019