Final Study Plan for
Estimating the Size of the Pacific Walrus Population
Marine Mammals Management, U.S. Fish and Wildlife Service
Alaska Science Center, U.S. Geological Survey
GiproRybFlot, Research and Engineering Institute for the Development and Operation of Fisheries
ChukotTINRO, Pacific Research Institute of Fisheries and Oceanography
A critical requirement for conservation and management of the Pacific walrus (Odobenus rosmarus divergens) is to estimate abundance with sufficient precision to track population trends. The Marine Mammals Management office of the U.S. Fish and Wildlife Service, in cooperation with the Alaska Science Center of the U.S. Geological Survey, the Chukotka branch of TINRO (Pacific Research Institute of Fisheries and Oceanography, Russia), and GiproRybFlot (Research and Engineering Institute for the Development and Operation of the Fisheries, Russia), will estimate the size of the Pacific walrus population with an infrared (thermal) scanner survey and associated aerial photography and satellite telemetry. The survey will be done in U.S. and Russian waters in the spring, when the walrus population is distributed near the southern edge of the Bering Sea ice pack.
The Bering Sea will be partitioned into survey blocks, and a systematic sample of transects within each block will be sampled with airborne thermal scanners using standard strip-transect survey methodology. Thermal signature intensities will be recorded for each detected walrus group, and a sample of the detected groups will be aerially photographed with digital cameras. Counts of walruses in photographed groups will be used to model the relationship between the amount of heat measured by the thermal scanner (thermal signature) and the number of walruses in the same group. Only walruses that are hauled out on the pack ice are detectable in thermal imagery. Therefore, the population estimate derived from thermal scanning will be corrected for walruses that are in the water and unavailable to the thermal scanner. Immediately prior to the aerial survey, satellite transmitters will be deployed on a representative sample of walruses distributed across the Bering Sea pack ice in both the U.S. and Russia. Transmitters will record the proportion of time each tagged walrus is hauled out on the ice or in the water, enabling estimation of an in-water correction factor. The final population estimate will be developed cooperatively by U.S. and Russian scientists.
TABLE OF CONTENTS
List of Figures 3
I. Introduction and Background 4
II. Goal and Objectives 7
III. Overall Approach to Population Estimation 8
A. Study area 8
B. Basic study design 8
IV. Aerial Survey 8
A. U.S. approach to thermal imagery 8
A1. Thermal imagery equipment 9
A2. Details: thermal imagery 9
A3. Calculation of thermal index 9
A4. Limitations or unknowns 10
B. U.S. approach to aerial digital photography 10
B1. Digital photography equipment 10
B2. Photography/thermal scanning coordination 10
B3. Details: counting digital photographs 10
C. Russian approach to thermal imagery 11
C1. Thermal imagery equipment 11
C2. Details: thermal imagery 11
C3. Calculation of thermal index 11
C4. Limitations or unknowns 12
D. Russian approach to aerial digital photography 12
D1. Digital photography equipment 12
D2. Details: counting digital photographs 12
V. Group Size Assessment: correcting for walrus groups too small to be detected 12
VI. Satellite Radio-Tagging: correcting for walruses in the water 13
A. Tagging equipment 13
B. Details: tagging 14
C. Aircraft/ship coordination 15
VII. Timing of Survey Efforts 17
TABLE OF CONTENTS, cont.
VIII. Survey Design and Statistical Analysis 18
A. Survey transect design 18
B. Statistical methods 18
B1. Relating thermal signature intensity to group size 18
B2. Estimating proportion hauled out 19
B3. Estimating totals for surveyed transects 19
B4. Estimating totals for blocks and the population 19
B5. Accounting for walruses on ice in groups too small to be detected 20
B6. Estimating variance 21
IX. Statistical Assumptions 22
Literature Cited 24
Appendix I List of statistical notation 41
Appendix II List of cooperators by agency 43
LIST OF FIGURES
Figure Number Page
1 Approximate ice coverage in the Bering Sea in late spring 27
2 Example of thermal image and matching digital photo 28
3 Example of frequency histogram of AMS thermal imagery 29
4 Example walrus heat/number regression 30
5 Walrus haul-out chronology data from 2004 31
6 Specifications of the icebreaker Magadan 32
7 Specifications of the Stimson 33
8 Specifications of the walrus satellite tag 34
9 The three study areas where tagging of walruses will take place 35
10 Example of aerial reconnaissance track lines 36
11 Example of ice conditions during walrus tagging from skiffs 37
12 Temperature and wind chill analysis 38
13 Chronological order of the tagging and aerial survey missions 39
14 Example of transect and survey block layout 40
I. Introduction and background
The Pacific walrus (Odobenus rosmarus divergens) is a sea ice-dependent pinniped that faces an uncertain future if the arctic pack ice continues to decrease in both extent and duration. Pacific walruses are an important subsistence and economic resource, with total harvests in the U.S. and Russia estimated at about 6,000 animals per year (U.S. Fish and Wildlife Service 2001). The current size and trend of the Pacific walrus population is unknown, and recent changes in the Bering Sea ecosystem (e.g., Hunt and Stabeno 2000) make it more important than ever that we develop a precise population estimate for monitoring and managing this species. A precise population estimate of Pacific walrus is critical for successfully fulfilling the U.S. Fish and Wildlife Service’s mandate under the Marine Mammal Protection Act for conservation and management of this species. The Marine Mammal Protection Act recognizes the important role that marine mammals play in marine ecosystems, and directs that marine mammal species, including Pacific walruses, are maintained at an “optimum sustainable population.”
Pacific walrus inhabit the continental shelf waters of the Bering and Chukchi seas, and occasionally move into the eastern East Siberian Sea and the western Beaufort Sea. In late winter and early spring, they are found in the Bering Sea where open leads, polynyas, or thin ice occur (Fay et al. 1984). There are three regions where walruses are concentrated at this time of year (U.S. Fish and Wildlife Service unpublished data, GiproRybFlot unpublished data). One region ranges from northwestern Bristol Bay to south of Nunivak Island. A second region is found to the southwest of St. Lawrence Island, and a third region is found in Russian waters of the Gulf of Anadyr. Preliminary genetic analyses provide no evidence of separate stocks within the Pacific walrus, but the need for further genetic studies is recognized (Scribner et al. 1997).
Based on large, sustained harvests in the 18th and 19th centuries, Fay (1982) speculated that the pre-exploitation population was represented by a minimum of 200,000 animals. Since that time, population size is believed to have fluctuated markedly in response to varying levels of human exploitation (Fay et al. 1989). Large-scale commercial harvests reduced the population to an estimated 50,000-100,000 animals in the mid-1950s (Fay et al. 1997). The population is believed to have increased rapidly in size during the 1960s and 1970s in response to reductions in hunting pressure (Fay et al. 1989). Between 1975 and 1990, aerial surveys were carried out by the United States and the former Soviet Union at five year intervals, producing minimum population estimates ranging from 201,039 to 234,020 animals. The estimates generated from these surveys are considered conservative population estimates and are not useful for detecting trends (Hills and Gilbert 1994, Gilbert et al. 1992). Efforts to survey the Pacific walrus population were suspended after 1990 due to unresolved problems with survey methods that produced population estimates with unknown bias and unacceptably large confidence intervals for tracking population trends (Gilbert et al. 1992, Gilbert 1999).
In March 2000, the U.S. Fish and Wildlife Service (USFWS) and U.S. Geological Survey (USGS) hosted a workshop on walrus survey methods (Garlich-Miller and Jay 2000). Workshop participants reviewed past efforts to survey the Pacific walrus population and discussed various approaches to estimating population size and trend. The amount of survey effort required to achieve a population estimate with an acceptably small variance to track trends (CV ≤ 0.3) using the existing methodology was prohibitive. Workshop participants recommended investing in research on walrus distribution and haul-out patterns, and exploring new survey tools, including remote sensing systems and development of satellite transmitters, prior to conducting another aerial survey.
A grant from the National Aeronautics and Space Administration (NASA) in 2002 allowed the USFWS Marine Mammals Management office to initiate a study on remote sensing applications in Alaska. This study demonstrated the feasibility of using airborne thermal imagery to detect walrus groups as they rest on the pack ice in the Bering Sea. Over a two-week study period in April 2002, Marine Mammals Management staff collected thermal imagery and matching aerial digital photography of more than 30 walrus groups ranging in size from 1 to 256 animals. In order to determine the limits of this technique, thermal imagery was collected at spatial resolutions ranging from 1-4 meters per pixel. Digital photographs of a subset of walrus groups were taken concurrently. By matching photographs with thermal images, a clear relationship was demonstrated between the number of walruses present in a group and the amount of heat they produced (Burn et al. in press). This relationship existed across all spatial resolutions tested, and demonstrated that the number of walruses present can be estimated using the amount of heat they produce, or their thermal signatures.
Based on the successful results of the 2002 study, a second pilot survey was conducted in the area of St. Lawrence Island in the Bering Sea. In April 2003, nearly 30,000 square kilometers of sea ice habitat were surveyed – an area larger than that covered in any previous visual aerial survey of Pacific walruses. A statistical approach was developed for estimating the number of walruses present on the sea ice within the study area (Udevitz 2005). Results from the pilot survey (Udevitz et al. in prep.) have been used in development of this study plan for a complete survey of the Pacific walrus population.
Also following the 2000 workshop on walrus survey methods (Garlich-Miller and Jay 2000), the USGS Alaska Science Center began developing a system for remotely deploying satellite transmitters on walruses, eliminating the need for animal capture and making it possible to deploy a large number of transmitters quickly and safely. In most previous walrus telemetry studies, walruses were chemically immobilized while telemetry devices were attached to a tusk. This latter method of deploying tags is time-consuming and dangerous to the animal because darted walruses can enter the water and drown while the immobilizing drug takes effect. This is particularly true when walruses are on ice floes where they can escape quickly into the water.
In 2004 and 2005, USGS staff conducted field trials on three remotely deployed tag designs in collaboration with the USFWS and the Greenland Institute of Natural Resources (Jay et al. in press). The satellite tags were deployed with crossbows and air guns, and were fitted with conductivity sensors that continuously recorded wet and dry intervals, which corresponded to periods when the animal was in seawater or hauled out. These data can be used to derive an estimate of the proportion of walruses that are hauled out on sea ice during a given time period. Data from satellite tags ensure that the estimate of walrus numbers derived from thermal scanning of the pack ice can be corrected for walruses that are in the water and unavailable for detection by the scanner. The attachment method for tags and data collection software were specifically developed for use in a spring survey of the Pacific walrus population.
II. Goal and objectives
Goal: Estimate the size of the Pacific walrus population with acceptable precision
(1) Survey pack ice in the Bering Sea in spring with thermal scanners to detect groups of walruses hauled out on ice and measure their thermal signatures
(2) Collect high-resolution digital photographs of a sample of the thermally detected walrus groups to determine the relationship between thermal signatures and the number of walruses in a group
(3) Estimate the number of walruses hauled out on ice
(4) Deploy satellite radio tags on a representative sample of walruses in the Bering Sea to estimate the proportion of time that walruses were in the water and unavailable to be detected by the thermal scanners
(5) Use haulout data from tagged walruses to correct the estimate of the number of walruses hauled out on ice for the proportion of walruses that were in the water and undetected during thermal scanner surveys
(6) Estimate total size of the Pacific walrus population with confidence intervals
III. Overall approach to population estimation
A. Study area
The survey will cover the extent of the pack ice in the Bering Sea, where the sea floor depth is less than 200 m. Walruses usually feed at depths of 100 m or less (Fay 1982, Fay and Burns 1988), so under this criterion all potential spring walrus habitat will be included in the survey. Ice coverage varies among years, but in U.S. waters ice generally extends to the southeast into Bristol Bay and to the south of St. Lawrence Island (Figure 1). In Russia, ice covers the Gulf of Anadyr and may extend along the shore as far south as Karaginskiy Island (Figure 1).
Satellite images of ice coverage in 2005 show that open water existed along the Chukchi Sea coast in the area of Point Hope. If this occurs again in 2006, our study area will be expanded to include areas of open water that could be walrus habitat. The approximate area to be surveyed ranges among years from 171,000 to 237,000 km2 in Russia and from 130,000 to 507,000 km2 in the U.S.
B. Basic study design
The range of the Pacific walrus in spring in both the U.S. and Russia will be surveyed, and American and Russian scientific crews will survey their respective sides. The Bering Sea will be partitioned into survey blocks of a size that can be surveyed within a single day. A systematic sample of transects within each block will be surveyed with airborne thermal scanners using standard strip-transect survey methodology. The amount of heat produced, or thermal signature, will be recorded for each walrus group that is detected by a thermal scanner. A sample of the detected groups will be aerially photographed with digital cameras. Counts of walruses in photographed groups will be used to model a relationship between thermal signatures and the number of walruses in a group. Only walruses that are hauled out on the pack ice are detectable in thermal imagery. Therefore, the population estimate derived from thermal scanning will be corrected for walruses that are in the water and unavailable to the thermal scanners. Immediately prior to the aerial surveys, satellite radio transmitters will be deployed on a representative sample of walruses distributed across the Bering Sea pack ice in both the U.S. and Russia. Transmitters will record the proportion of time each tagged walrus is hauled out on the ice or in the water, allowing estimation of a correction factor for walruses in the water. The final population estimate will be developed cooperatively by U.S. and Russian scientists.
IV. Aerial surveys
For surveys in both the U.S. and Russia, aerial surveys will be comprised of two major components, a thermal scanner survey and digital aerial photography. Differences in available equipment necessitate that data collection methods will differ between the U.S. and Russia.
A. U.S. approach: thermal imagery
Thermal (infrared) imagery will be the primary data collection method. Walruses are generally warmer than the background environment of ice and snow, and groups of walruses hauled out on ice are therefore detectable with thermal imagery (Figure 2). Walruses in the water cannot be detected. A systematic sample of strip transects will be surveyed with the scanner in each survey block. The amount of heat, or thermal signature, of each detected group will be recorded.
A1. Thermal imagery equipment. The remote sensing system is a Daedelus Airborne Multispectral Scanner (AMS), built by Argon ST of Ann Arbor, Michigan. The system has a 0.625 milliradian instantaneous field of view (IFOV), and collects imagery across a sensor array 3,000 pixels wide. The AMS system records a thermal infrared channel (8.5 – 12.5 µm) with 12-bit radiometric resolution. Scanner data will be recorded on removable hard drives and archived on DVDs.
An Applanix unit will be used to correct thermal imagery for distortion resulting from movement (pitch, yaw, and roll) of the aircraft. Movement by the aircraft on one or more of its axes results in duplicate or dropped scan lines, the measurements of scanned surface areas below the aircraft. If not corrected for in the data set, these excess or dropped measurements would distort in two dimensions the shape and size of features such as walrus groups, increasing the variability in the thermal signature/group size relationship and ultimately, in the overall estimate of walrus abundance. Addition of the Applanix unit will increase precision of the abundance estimate. The Applanix unit will also geo-reference the thermal images.
Thermal imagery will be collected from an Aero Commander 690B turbine engine aircraft. The Aero Commander is a twin-engine craft with high wings and bubble windows that allow lateral and downward visibility on both sides. The AMS control panel will be mounted within the cabin of the aircraft, with the scan head in the tail section.
A2. Details: thermal imagery. Thermal imagery data will be collected as the aircraft flies along transect lines within a survey block. Spatial resolution of the AMS system varies linearly with the altitude of the aircraft. Surveys will be conducted at 6,400 m (21,000 ft) Above Ground Level (AGL), yielding a resolution of 4 m and a 12 km wide survey swath (strip width) along each transect. A thermal signature will be recorded for each walrus group that is detected within the 12-km wide scanned strip.
A3. Calculation of thermal index. To determine the threshold temperature value between walrus and the background environment, a frequency histogram of temperature values in the entire image will be examined. Within each image, the point at which the histogram rapidly decreases from thousands of pixels at each value to fewer than ten pixels is identified as the temperature threshold value (Figure 3).
Pixels with temperatures warmer than the threshold value are classified as having some portion of their area covered by walruses. After determining the threshold temperature for each image, an index of the total amount of heat produced by each walrus group will be calculated as:
where hi is the index for group i, a is the pixel area (m2), tij is the temperature for pixel j of group i, Ti is the threshold temperature for group i, and the summation is over all pixels with temperature values above the threshold (i.e., pixels with tij > Ti).
A4. Limitations and unknowns. Surveys must be conducted within weather constraints. Thermal scanning can only take place during periods of good weather. Skies must be clear, the ambient air temperature must be above -12º C (10º F), and the wind chill factor must be above -18º C (0º F).
B. U.S. approach: aerial digital photography
A sample of the thermally detected walrus groups on each transect will also be photographed with low elevation, high resolution digital photography. Counts of aerially photographed walrus groups will be used to model the relationship between the amount of heat measured by the thermal scanner and the number of walruses in a group (Figure 4). A generalized linear model will be used to estimate this relationship (see Section VIII B1, Relating thermal signature intensity to group size). The relationship between walrus thermal signature and group size appears to vary with weather conditions such as wind and temperature (USFWS unpublished data), and methods are being developed to address this (see Section V, Group Size Assessment).
B1. Digital photography equipment. The digital camera will be mounted within the cabin of a second aircraft. The camera is a Nikon D2X 12 megapixel format camera that produces images with dimensions of 4,288 x 2,848 pixels. Digital photographs will be taken using a 200 mm camera lens from an altitude of about 700 m AGL.
B2. Photography/thermal scanning coordination. The photography plane will follow the general course of the thermal scanning plane, with the objective of photographing as many walrus groups as possible within the 12 km swath surveyed by the scanner. Effort will be made to photograph groups on each transect where they occur, so that photographed groups are representative of the environmental conditions that were encountered during the survey.
When a walrus group is detected by the thermal scanning plane, observers will record the group’s position and note its general location in relation to the transect line (e.g., 2 km to the west of the transect). Observers in the thermal scanning plane will communicate by radio to notify the photography plane of the detected group and its coordinates. This protocol was developed and worked well during pilot studies.
B3. Details: counting digital photographs. Prior to counting, photographs of walrus groups will be matched with the corresponding thermal images of the same group. Locations of walrus groups in each photograph and in each thermal image can be identified by their GPS positions and times. Matching of photographs and thermal images will be done using position, time, and visual inspection of the ice field around each walrus group, which tends to have unique features that may be used for orientation.
Thermal images will be used to define walrus groups. A group will be considered distinct when it is separated from nearby groups by 20 m (5 pixels) or more. The number of walruses in a group will be counted on all photographs collected within the thermally scanned area. It is important that groups of walruses in photographs reflect the same groups that were recorded by the thermal scanner (Figure 4). Therefore, only walruses that are hauled out completely on an ice floe will be counted.
Walruses will be counted using the software ERDAS Imagine (Leica Geosystems, Atlanta, Georgia), which summarizes animals that were marked in each photograph by the user. Walrus groups will be counted three times by the same observer, and each count will be temporally separated by at least one day. If the three independent counts differ, a fourth count will be conducted by a different observer. Two observers will then compare counts to reconcile differences and reach a “best count” by consensus.