Effects of Nonrandomness on Line Transect Estimates of Dolphin School Abundance

Line transect analysis is a census method that has been used to derive estimates of dolphin school abundance from sightings data collected by observers on tuna purse seine vessels. The method is based 01: the assumption that movements of the sighting platform ( tuna vessel) and sighted objects (dolphin schools) are random with respect to each other. In practice, neither schools nor vessels move randomly. Stratification of sightings data has been used to alleviate partially the effects of this nonrandomness, but the effectiveness of this stratification cannot be tested with data from commercial vessels because the movements of the vessels cannot be controlled. As a n alternative, we have used a relatively simple mathematical simulation model to investigate the severity of bias introduced into school abundance estimates by nonrandom movements of schools and vessels, and by the data stratification procedure. Simulations show that nonrandom movements on a scale of a fen hundred miles, coupled with the data stratification procedure, can lead to overestimates of dolphin school abundance by as much as a factor of two. These results focus attention on the need to understand patterns of dolphin school distribution in smaller scales of space and time than have been studied previously, and to develop data stratification methods more robust against the effects of small-scale nonrandomness. The National Marine Fisheries Service (NMFS) monitors mortality of dolphins involved in fishing operations by the United States purse seine fleet for yellowfin tuna, Thu~i)zzis albacares, in the eastern tropical Pacific Ocean (ETP), to determine whether mortality has exceeded an annual quota implemented by an act of the U.S. Congres s . T h e quo ta leve ls depend upon whether dolphin populations are thought to be increasing or decreasing in number, relative to population levels during previous years. The most effective method currently available for detecting trends in relative abundance is analysis of population abundance estimates collectElizabeth F. Edwards and Pierre M. Kleiber. Southwest Fisher ieb C e n t e r , Kat ional Marlne Fisher ies Servlce. NOAA. P.O. Box 271. La Jolla. CA 9203X. ed over a period of 5-15 years. The most effective method currently available for making these abundance estimates is line transect analysis of dolphin school sightings data (Holt 1987; Buckland and Anganuzzi 1988). Two data sources are available for these line transect estimates of abundance: 1) data collected by observers during research surveys (RSOD-Research Survey Observer Data) and 2) data collected by observe r s du r ing commercial fishing opera t ions (TVOD-Tuna Vessel Observer Data). N M F S has used RSOD because research surveys can be designed specifically to satisfy the assumptions required by line transect analysis (Smith'). However, research surveys are very expensive and are becoming more so. This expense causes RSOD to be sparse relative to TVOD and possibly unavailable in the future. TVOD are a potential solution to these problems, having three significant advantages over RSOD: TVOD are much more abundant, are relatively inexpensive, and are likely to continue being collected as long as fishermen set on and kill dolphins. Observer-days from tuna vessels account for roughly 95% of the annual observer effort in the ETP, while observer-days from research vessels account for only 5%. TVOD are inexpensive relative to RSOD because TVOD are collected by the observers in addition to monitoring dolphin mortality, the latter being the main reason the observer program was initiated. This monitoring program has been in operation for the past 14 years, will continue into the foreseeable future, and monitors about 30% of trips by purse seiners (both U.S. and nonU S . vessels) each year in the ETP'. Ideally, TVOD could be used in place of RSOD to monitor changes in abundance of dolphins. 'Smith, T. D. 19%. Estimates of sizes of two populations of porpoise ( S / e v e / / n ) in the eastern tropical Pacific Ocean. Admin. Rep. No. LJ-75-67, Southwest Fish. Ce;t.. Natl. Mar. Flsh. S e n . , NOAA, La Jolla, CA. -Inter-Amencan Tropical Tuna Commission, Annual Reports 1980-1988, Scripps Institution of Oceanography. La Jolla. CA 9203X. Manuscript accepted April IBRY. Flshery Bulletin. V.S R i 859-876. 859 F I S H E R Y I(I.LI.ETIS YOL. ki. S o 4 l!lh!l methods of line transect analysis to estimate dolphin abundance from TVOD (Buckland and Anganuzzi 1988) raised serious but unanswered questions about the effects of these factors on the abundance estimates derived. The philosophy behind building a relatively simple model was that biases shown to be troublesome and methods shown to be inadequate in a simple computer model are likely to be even more troublesome and inadequate in the real world. I t is both more efficient and more economical t o investigate these biases and methods first with a simple simulation model, prior to developing expensive field experiments. We have specifically applied the t ene t s of Occam’s Razor in developing this model. making it as simple as possible while still incorporating the major processes and features contributing to the TVOD data collection process. In this study, we focused only on estimating abundance of dolphin schools. leaving questions about abundance of i u d i ~ i d t t d dolphins for a later day. We also assumed that data were collected without artifacts, leaving also that problem for a later set of simulations. Both of these omissions are examples of factors that probably have strong effects on analyses of TVOD, but which are at this stage unnecessary refinements to the simulation model. Such refinements could be added later if no problems were identified during simulations with the early, most simplified versions of the model. This paper presents results of testing one hypothesis about one of the most fundamental factors suspected to affect seriously line transect estimates of dolphin abundance derived from TVOD. Specifically. we tested the effect of nonrandom clustering by dolphin schools on abundance estimates. As part of this analysis we tested also the effects of three types of data stratification prior to line transect estimation of school abundance: 1) no stratification. 2) stratification by raw encounter r a t e per 1” square, and 3) stratification by smoothed encounter rate per 1” square, using the smoothing and interpolation algorithm developed by the Inter-American Tropical Tuna Commission for deriving estimates of dolphin abundance from line transect analysis of TVOD (Buckland and Anganuzzi 1988). We were primarily interested in the third type of stratification, because the properties of the smoothing algorithm are poorly understood. The other two stratifications were conducted to provide a basis for comparison with the smoothing procedure. Reluctance to use TVOD to monitor the relative abundance of dolphins stem from concerns that TVOD 1) seriously violate some of the fundamental assumptions of line transect analysis (Polachek 1983), 2) are subject to serious but unquantified and possibly inconsistent biases, and 3) may be plagued with artifacts arising from the data collection process. Artifacts include, for example, differences between RSOD and TVOD in the sighting frequencies of various dolphin species reported by observers on research vessels compared to tuna vessels (Barlow and Holt3), environmental factors affecting sighting ability (e.g., sun glare, sea state, and cloud cover; Holt and Cologne 19871, and shifting areas of concentrated search effort (Buckland and Anganuzzi 1988). However, problems of this type are common to most commercial fisheries data and analyses derived from them. It is important to determine whether, despite these difficulties, useful estimates can be derived from such data sets. Toward t h i s end , we have developed a relatively simple model simulating the TVOD collection process. Our purpose in developing the model was twofold: 1) to test the effect of suspected biasing factors on line t ransect estimates of abundance and 2) to test new methods of abundance estimation prior to conducting expensive field tests. There are two unique advantages of simulation modeling in this context. First, we are simulating dolphin abundances and vessel movements within the model itself; therefore, we have available the “truth” against which to compare our modelgenerated estimates of abundance. Second, we have the capability of investigating effects on estimates that are due to combinations of biasing factors which may not have occurred during the years we happen to have been collecting data, but which can be expected to occur. Biasing factors include, for example, small-scale nonrandomness in school and vessel movements and spatial distributions, choice of data stratification method, changes in fishing objectives, practices, and areas of concentrated search, and changes in sighting protocol and recording procedures. We chose to focus first on the effects of nonrandomness and on t h e method of da t a s t ra t i f icat ion because r ecen t ly developed ‘Barlow. J. .and R. S. Holt. 1986. Geographic distributions of species proportions for dolphins in the eastern tropical Pacific. Admin. Rep. No. LI-84-2i. Southwest Fish. Cent., Natl. Mar. Fish. Sew.. NOAA. La Jolla. CA.