and Determination of Bird Mortality
by
Michael L. Morrison, University of Arizona
Introduction by Holly Davis, National Renewable Energy Laboratory
With the increasing development of wind power in the United States, concern over the number of birds that might be killed by wind turbines has increased. To respond to the increasing concern and to address the many critical questions that still need answering, the National Renewable Energy Laboratory (NREL) is in the process of developing an avian-wind power research program. NREL is interested in the work that the Avian Workgroup of the National Wind Coordinating Committee (NWCC) is doing to develop a research agenda and to understand all parties' needs and priorities. Because of the need to allocate NREL's 1995 dollars before the end of the 1995 fiscal year, NREL gathered together an ad hoc group of researchers versed in avian-wind interactions to develop a list of the most important research topics for future study. The group agreed on the need for two general types of studies: (1) pre/post construction surveys measuring utilization of the area by birds and their mortality both before and after a wind farm is constructed, and (2) studies to determine the effect of various treatments to turbines (such as painting blades or using perch guards) on avian mortality.
The two topics of research were brought before the Avian Workgroup at its June 14, 1995, meeting in Jackson, Wyoming, to generate consensus among all participants regarding proceeding with Requests for Proposals (RFPs) on these topics. Agreement was reached at the meeting and NREL began developing a competitive solicitation. To ensure that the data gathered in these studies meet basic scientific standards and are transferable, NREL included the following two protocols as part of the RFP, and offerors were expected to follow the protocol if awarded a subcontract. These protocols were developed specifically for the RFP, but likely will have a much wider application.
Project Protocol I: Evaluation of Existing Wind DevelopmentsThe following protocol was initially developed at the request of the California Energy
Commission (CEC) as part of their evaluation of the impacts of several existing wind
resource areas (WRAs) on birds. Richard Anderson, CEC, has been involved with the
continued development and field testing of this protocol (see p. 74ff of these
Proceedings).
Objectives.—The protocol will determine the relative abundance and utilization rates of birds in an area, sample for bird mortality, and then determine the bird risk and attributable risk due to the WRA. This approach should allow researchers to focus quickly on key areas for further inquiry and uncover potential relationships that could be verified through follow-up studies, as warranted. The general goals of the protocol are
- Establish a methodology for conducting avian mortality monitoring studies that will set standards for other such studies.
- Determine if differing risk levels of avian mortality are attributable to the WRA, and if so, determine if they represent potentially significant problems for a population.
- Develop research methods and conduct field research on increasingly focused problem areas, and develop recommendations that provide resolution of the problem(s) in order to facilitate siting of future developments.
Expected Research Outcomes.—The protocol is designed to determine the relative level of increased risk to birds that is directly attributable to the development and operation of the WRA. If the attributable risk is determined to be negative, zero, or only slightly increased, then the conclusion would be that the WRA does not pose a significant increased risk to birds relative to non-developed areas.
The protocol is not designed to determine the absolute number of birds dying in the WRA nor the absolute net difference between a WRA and one or more undeveloped comparison area(s). Such a number, in isolation from information about the impact on the actual population, is of little use in evaluating the impact of a WRA on birds. Thus, it is prudent to start with research that can identify the relative risk due to the project, and estimate if this risk is so large that it is likely to be having a negative impact on the population. If so, more intensive studies (e.g., population analysis of selected species) would then be warranted.
Thus, this protocol will allow a conclusion to be made regarding the relative risk that a WRA poses to birds. It also will allow formulation of hypotheses about inferred causal effects based upon any statistically significant correlations that are found.
Research Outline.—The goal is to determine if the development and operation of a Wind Resource Area (WRA) results in an increased risk of bird mortality.
Key Questions: The key question to be addressed is, "What influence does the development of the WRA have on birds?" More specifically,
- Does the WRA development influence the level of bird activity, called utilization rate, compared to that of nearby undeveloped areas?
- Does the WRA development influence the rate of bird mortality, called mortality rate, compared to that of nearby undeveloped areas?
When comparing the utilization and mortality rates in the WRA and undeveloped areas, is there any change in the risk to birds that is attributable to the WRA development?
Does attributable risk, utilization rate, or mortality rate vary by type of technology (e.g., different turbine types, infield power lines) or vegetation types available?
Definitions and Concepts: The following definitions and concepts are central to the proposed approach:
Bird utilization rate: The number of birds detected using the area during set periods of time. Rates can be developed for different species, vegetation types, locations within the WRA, and the like (if adequate sample sizes can be accumulated).
Fatalities: The number of dead birds found during sampling.
Bird mortality rate: The number of dead birds divided by utilization rate; this equates with bird risk in the WRA.
Attributable risk: The risk of death associated with a bird being in the WRA relative to the risk for a bird not in the WRA. This is derived by calculating mortality rate for both the WRA and undeveloped areas.
Sampling Design.—The protocol calls for an initial Phase I study based on a standardized protocol applicable in any WRA. This would be followed, if necessary, by Phase II or III studies focused on specific topics identified as important during earlier phases.
Phase I Studies: Studies to be done during Phase I should include the following:
1. Parallel transects traversing the WRA will be walked from randomly selected, strategic starting points, chosen to include all types of natural communities, developed WRAs, and non-developed comparison (non-WRA) areas. Transects should be 400-600 m apart, and are not placed to follow strings of turbines. The non-WRAs can be areas immediately surrounding the WRA, or areas similar in environmental conditions located nearby. Ideal locations for non-WRA sampling are nearby areas that are suitable for wind development, but have not as yet been developed as such.
2. Ten-minute point counts to determine bird utilization rates will be conducted every 400-600 m along each transect, with the first point randomly established within 300 m of the transect's starting point. The number of points established will be based on the size of the WRA. Data recorded will include species, number, behavior, distance from a turbine if in WRA, or distance from WRA if outside. Sampling can be conducted throughout the day during weather conditions favorable for observing birds. Points need only be sampled once per season, unless the WRA is so small that few total points have been established. In the latter situation, data collected at each point within a single season will be averaged.
3. Phase I sampling will be conducted until an adequate number of samples have been collected over at least four seasons. The anticipated Phase I study period is 1 year.
4. Dead bird searches will be conducted within a circle of 25-m radius around each point-count location. The field observers will start at the point-count location and walk a spiral path outward to 25-m radius, expanding outward so that a complete search of the area is made. The distance between successive coils of the spiral, and the time spent in each 25-m circle, will be based on the density of the vegetation.
5. Data should first be evaluated for adherence to parametric statistical assumptions (normality, equality of variances). Appropriate univariate or nonparametric tests should then be applied to test the hypothesis of no difference in utilization and mortality rates between WRA and non-WRA.
6. Scavenging studies should be conducted during each season to determine if scavenging differs between the WRA and non-WRA. Scavenging studies will be done at point-count sites at varying distances from turbines. A minimum of three general distances categories (near turbine, 500 m, 1 km) will be studied, with marked dead birds being placed and monitored at 10-30 point count sites per distance category. Replicates can be conducted within seasons as time allows. If significant differences in scavenging are found, a correction factor must be applied to the dead bird values.
7. Studies of observer bias will include replicate comparisons of each observer compared to other observers for both bird utilization and dead bird detection efforts.
Phase II and III Studies: The results of Phase I studies will determine if additional work is warranted, and if so, the nature and scope of the work.
1. Areas of high mortality: In areas where mortality rates are substantially higher than elsewhere, additional sampling will be warranted. This sampling is designed to obtain an adequate number of birds for necropsy so that causes and timing of death can be determined adequately.
2. Other structure sampling: Other structures such as power poles and meteorological towers will be sampled to determine their contribution to overall deaths. Specific sampling methods will be determined taking account of local structures and situations.
3. Behavioral documentation: Areas of high utilization and mortality determined during Phase I will be more intensively observed to evaluate the causes of bird use and mortality. Behavioral protocols will be developed according to the individual situation discovered. For example, turbines known to harbor perching birds can be observed according to the time of day, duration of perching activity, and entry and exit direction of the birds. Areas of known high utilization (e.g., for soaring, hunting) can be observed according to time of day, distance from turbine (vertical and horizontal), and outcome of hunting.
4. Prey abundance: Abundances of prey can be determined for areas found to harbor high concentrations of birds or high mortality rates during Phase I studies. Comparison data should be obtained for matched WRA and non-WRA areas of lower bird utilization and mortality. The specific methods used should conform to standard sampling techniques applicable for the species and environmental conditions present. In all cases, however, sampling should include an evaluation of the adequacy of the duration and intensity of trapping (e.g., number of trap nights; number and spacing of traps).
5. Nocturnal bird use: If Phase I studies show high mortality of nocturnally active birds (owls, nocturnally-migrating birds), then the use of night vision equipment or radar might be indicated.
Data Forms.—Data forms being used by the CEC in their ongoing studies in the Tehachapi Pass WRA are included in an Appendix, along with descriptions of the variables being recorded.
Project Protocol II: Determination of Bird Mortality
Problem Statement.—A central issue in wind power development is the mortality of birds within wind farms. Individuals from industry, the scientific community, conservationists, and regulators have postulated that mortality can be reduced by modifying towers to reduce perching, painting disruptive patterns on turbine blades, and other actions. However, the prevailing sentiment is that finding dead birds in wind farms is such a rare event that statistically valid analyses of the effectiveness of treatments designed to reduce mortality are not feasible. Thus, some have suggested that a reduction in bird use on and around towers, and/or marked changes in bird behavior there, would justify concluding that treatments have been effective. The weakness of this argument is that mortality is the issue, and changes in behavior could also cause increases in mortality even if use of turbines has declined. Further, without quantification of dead birds, no statements can be made regarding the influence of turbines on the abundance and dynamics of bird populations. If the risk per visit stays the same for a bird, then by that measure the mortality rate has not been reduced even if fewer birds visit. "Visit" must be carefully defined in all applications. For example, a visit might be defined as an approach within a certain distance (e.g., 100 m) of a turbine, or a bird simply entering a wind farm.
Thus, the goal of this protocol was to determine the best possible study design and testable hypotheses concerning the effect of treatments on bird mortality and/or use in wind farms or around individual turbines. Also of interest were protocols to test the effects of treatments on measurable variables potentially correlated with total mortalities and/or mortality rates. This protocol was developed with the assistance of Drs. Larry Mayer, Lyman McDonald, and Dale Strickland.
Issue Development.—If we test modifications to turbines or wind farms without considering both bird mortality and bird utilization, then the experiment is poorly designed; we will not know whether any decrease in deaths was due to decreased utilization, decreased risk, or both. When we separate utilization from risk, it is clear that a modification reducing utilization of a wind farm could have a devastating effect on the population whether it decreases or increases the risk associated with flight in the wind farm. The farm could actually enhance a population, e.g., by enhancing food supply.
Population Effects: While data on population effects are ultimately desirable, they require an extremely intensive study that is beyond most budgets and may be unnecessary. Thus, we must design studies that address bird behavior and mortality in and around wind farms without directly studying population effects. Such "weight of evidence" results are, at a minimum, a good starting point to determine if one should even worry about more intensive studies. Such studies are of the "intermediate outcome" variety. They function in a stepwise fashion leading toward determination of the influence of a wind farm's impact on populations of birds.
Utilization: Measures of utilization can be based on many different parameters, including the number of birds, number of flights, number of landings, etc. The question is, do the changes on the wind farm effect the risk of death?
Definitions: Suppose we institute an intervention that can be viewed as a preventive intervention or as a factor that removes a risk. The following definitions are provided to help clarify the various types of risk:
Attributable risk: the maximum proportion of risk that would be removed if the risk factor (e.g., all perching) were removed.
Preventable fraction: the proportion of risk that would be removed if all birds got the preventive intervention (e.g., if we removed all perches).
Note that the attributable risk and preventable fraction are the same if we view a preventive intervention as the removal of a risk factor. The prevented fraction is quite different:
Prevented fraction: the actual reduction in mortality resulting from the preventive intervention as implemented (e.g., the proportion of risk actually prevented by removing certain perches).
These measures all assume that the risk factor does not interact with any other factor affecting mortality. For example, removing the perch is assumed not to increase risk of starvation.
Case Study Approach: Case studies have high utility in evaluating mortality. Here, one collects dead birds inside and outside a wind farm, and conducts blind analysis to determine the cause of death. Unfortunately, under most situations very few dead birds could be found outside the farm. However, all dead birds found in a study should be subjected to blind analyses because this information will assist with evaluation of observational data.
The case study approach suggests that epidemiological analysis can often be combined with clinical analysis to extend the inferential power of a study. Here the clinical analysis would be the necropsies of the birds. Suppose that we are successful at finding dead birds inside a wind farm. If we look at proportional mortality—the proportion of the birds killed by blunt trauma, sharp trauma, poisoning, hunting, natural causes, etc.—then the proportions should differ significantly between the wind farm and the control area. It is assumed that the probability of finding a given dead bird is not affected by its cause of death.
Mortality Rates: The ideal denominator in epidemiology is the unit that represents a constant risk to the individual. In the bird/wind farm context, the unit might be miles of flight, hours spent in the farm, or years of life. If the denominator is the total population number then we are assuming that each bird bears the same risk by being alive. In human epidemiological studies, the total population size is usually used because we cannot estimate units of time or units of use. In avian studies, actual population density is extremely difficult to estimate. If the risk is caused by being in the area, then deaths per hour in the area is probably the best epidemiological measure in avian studies. This rate is then extrapolated to the population by estimating the utilization rate of the area for the entire population. Measuring utilization is difficult, however, and must be approached carefully.
Thus, we have two major ways to calculate mortality rate:
(1) = no. dead birds/no. birds in population, versus
(2) = no. dead birds/bird use.
Equation (1) is the ideal, but as discussed above, is usually impractical. Equation (2) is feasible, but results will vary widely depending upon the measure of bird use selected. In addition, for (2), the background (non-wind farm) mortality rate must also be determined for comparative purposes. Thus, equation (2) should be the center of further discussion.
Summary of Study Design:
Primary objective: measure bird use with different treatments (perching, flying, etc.).
Secondary objective: count number of dead birds with different treatments and estimate mortality rates by equation (2).
Analysis:
Test for differences between treatments for primary objective (utilization); must achieve a reasonable level of statistical power.
Test for differences between treatments for secondary objective (mortality rate); power will be poorer than for primary objective.
It is feasible to design treatment vs. control studies for inferences on measures of use. Determination of mortality (using eq. 2) is possible, but statistical power to conclude that treatment and control sites have different mortality rates will be low. For example, in a randomized pairs design, most pairs are expected to result in zero mortalities, with tied values and no mortalities on either member of a pair. The high frequency of zero values effectively reduces the sample size for most analyses.
Study Design.—Experimental units: There are two main options:
Wind-farm based study: In this design, a relatively large portion of the wind farm would serve as an experimental unit. For example, a group of 100 turbines would receive treatment (e.g., perch guards, painted blades), and a similar group of turbines would serve as a control. This basic approach could be applied both to existing farms and in planned farms. Unless preliminary studies are first conducted, an educated guess would be necessary to determine how many turbines to include in an experimental unit. Further, it will usually be difficult to replicate the pairs of experimental units. With a few pairs (1, 2, or 3), this design is most comparable to a series of observational studies even if treatments are randomly assigned to one member of the pair. With this design, however, extrapolation to the entire wind farm is relatively easy.
Small-plot based study: In this design, an individual turbine or a small group of turbines (e.g., a string) serves as the experimental unit. For example, pairs of turbine strings are selected and one string of each pair receives the treatment. This design has the advantage of being centered on discrete units that can be replicated and readily observed; it has more of the features of "classical" experimental design. However, extrapolation to the entire wind farm is relatively difficult.
The latter design is preferred because of the relative ease of gaining an adequate sample size. A relatively large number of pairs of units can be analyzed in the sense of a 'true' experiment. Extrapolation to the entire wind farm is possible in a limited sense if each pair consists of one unit that is randomly sampled and then matched with a second similar unit. The treatment is randomly applied to one member of each pair.
Design Considerations: Treatments and controls can be reversed after the initial experimental period. This would strengthen the test, and would be especially useful in the wind farm based study because of the likely small number of replicates.
Variable Selection: One primary variable will usually drive the study design; thus the initial sample size should be aimed at that variable. However, it is assumed that a reasonable sample size will also be gathered for the other, secondary variables. Sampling can be adjusted as data are collected (i.e., sequential analysis of sample size).
With the small, paired-unit design, 1-2 primary variables on use (e.g., passes through the blade plane, perch attempts) will likely be adequate. The minimum number of pairs to be sampled should be 12. However, a greater number of pairs would be desirable, at least initially. The sampling unit can be either individual turbines, or strings of turbines. It is expected that string length will range from 5 to 10 turbines, depending upon the size and configuration of the wind development; portions of longer strings can be subsampled.
Study Protocol.—The following study protocol concerns the small, paired-unit design, in which one turbine or a small group of turbines (e.g., a string) is the experimental unit.
Objective: To test the null hypothesis of no difference in primary variables and/or mortality rate following treatment.
Definitions: Utilization or use of turbines will be evaluated by measuring two primary variables: perching attempts, and number of passes by distance from the swept blade area. Mortality rate follows equation (2), above.
Secondary variables can include any measurements that do not interfere with the accurate recording of the primary variables.
Experimental Units: The basic sets of experimental units will be pairs (or larger blocks if there are 3 or more treatments) of turbine strings or turbine groups. Within each pair or block, turbines should have
 |
similar environmental conditions and/or |
|
similar breeding (nesting) densities for the species of interest, and |
|
(if possible) a similar history of mortalities. |
The number of turbines/string will be based on the configuration of the wind farm. The researcher should attempt to sample a minimum of 12 pairs or blocks of strings. Treatments are randomly assigned to the members of each pair or block of experimental units.
Sampling Frequency: Sampling should be as frequent as possible initially; it can be scaled back after preliminary data are analyzed. It is recommended that each pair or block be sampled at least weekly. Sampling should be stratified by time so that adequate samples are taken both within and between days. If 12 pairs of strings are under observation, then a minimum of 4 observers would be necessary.
Stratification of sampling by major weather condition (i.e., high or low wind; clear or moderate to heavy fog) can be initiated if funds are available for the additional observers who would be necessary to take advantage of such conditions.
Variables: Two primary variables should be measured:
Number of passes: Record the number of passes by a bird by distance (at closest approach) from the swept blade area of the turbine. Multiple passes by the same bird (if identification known) should be recorded such that repeated observations of the bird can be identified in the data set. A bird flying onto, and then leaving, a turbine to perch is also recorded here. All birds flying by a turbine (regardless of distance) could be recorded. As a recording rule, assign the bird to the nearest turbine it passes in the string you are observing. An alternative sampling protocol would be to develop an activity budget for any bird approaching the string. Here, data (behavior, distance) would be taken at a fixed time interval (e.g., every 1 min).
Perching attempts: Record the number of times that a bird attempts to perch, or does perch, on a turbine. Also record the location of the perch, the perch type, the amount of time spent perching, and the apparent activity of the bird (e.g., preening, scanning the ground, eating). As noted above, identify repeated perches or perching attempts by the same individual.
Dead Bird Surveys: An area within 100 m on either side of the turbine string will be searched for dead birds.
Data Analysis: Apply standard univariate analyses appropriate to the specific experimental design being used. Begin by evaluating equality of variances, normality, and other assumptions.
An attendee enquired about the rationale for recommending a minimum of 12 pairs (or blocks) of turbine units. Dr. Morrison indicated that this recommendation was a general one based mainly on intuition and experience with other types of studies involving paired experimental units: 6 pairs rarely if ever provide enough information for meaningful statistical analysis, 12 pairs is generally a bare minimum, and 24 pairs usually provides good statistical power. Preliminary sampling (a pilot study) is needed in order to get a better sense of the required sample size.
Twelve pairs may be adequate to detect a difference in bird utilization, but not to study mortality. For mortality studies, more pairs likely will be needed because results from some pairs will be zero-zero ties.
The possibility of pooling results of similar tests at different wind plants was also discussed. It was suggested that results from different areas should not be simply pooled and treated as a single overall test, as many aspects of the two (or more) wind plants are likely to differ. However, if the same hypothesis is tested at two or more sites, results could be combined using the techniques of meta-analysis. These methods can derive a single overall hypothesis test from the combined results of separate studies.
Dr. Morrison noted that there are useful design and statistical approaches for studying rare events such as bird deaths at wind turbines [see, for example, R.H. Green and R.C. Young (1993), Ecological Applications 3(2):351-356]. This type of problem is not unique to the bird/wind turbine situation. Also, by obtaining similar types of observations in more than one study area, it may be possible to acquire sufficient data to draw at least a tentative conclusion based on the "weight of evidence".
Appendix: Data Forms and Variables
The following pages provide suggested data forms for point counts of bird utilization and for mortality/injury searches. Also attached are lists explaining the variables to be recorded on the two data forms.
