Northeast Fisheries Science Center Reference Document 08-18
A Description of the Allocation Procedure Applied to the 1994 to 2007 Commercial Landings DataSE Wigley, P Hersey, and JE Palmer
NOAA’s National Marine Fisheries Serv., 166 Water St., Woods Hole MA 02543-1026
Web version posted October 10, 2008Citation: Wigley SE, Hersey P, Palmer JE. 2008. A description of the allocation procedure applied to the 1994 to 2007 commercial landings data. US Dept Commer, Northeast Fish Sci Cent Ref Doc. 08-18; 61 p.
Information Quality Act Compliance: In accordance with section 515 of Public Law 106-554, the Northeast Fisheries Science Center completed both technical and policy reviews for this report. These predissemination reviews are on file at the NEFSC Editorial Office.
The purpose of the allocation is to supplement the mandatory commercial landing data (1994 onward) with area fished and effort information using Vessel Trip Reports (VTR). The goal is to eliminate the need for single species allocation for each analysis conducted and to maintain a consistent, comprehensive commercial landings database from 1963–present containing the information needed to address management questions, conduct stock assessments, and perform ecosystem research.
The multi-tier trip-based allocation is designed to combine each mandatory reporting dealer (Dealer) trip with a VTR trip, or group of VTR trips with similar characteristics, to obtain area fished and effort associated with the Dealer trip. Although the trip-based allocation and the single-species proration yield similar results with regard to stock landings (Wigley et al. 2007), the trip-based allocation is an improvement over the single-species proration because it provides area fished at a fine level of resolution (statistical area rather than stock level) for all species. It also estimates effort associated with these landings. The trip-based allocation represents a comprehensive approach to determining area fished and effort in Northeast region’s commercial landings in order to meet scientific and fishery management needs, as well as commercial data reporting requirements to Northwest Atlantic Fisheries Organization (NAFO).
The multi-tier trip-based allocation has been developed to augment commercial landings data with area fished and effort; however, trip characteristics, species landings, and price information will not change. All species on a given trip/subtrip will be assigned the same area and effort. The multi-tier trip-based allocation utilizes VTR data that has been aggregated into four levels: Level A, Level B, Level C, and Level D. At Level A, Dealer and VTR trips are matched one to one. At Levels B, C, and D, VTR trips are grouped together to form a pool of trips with similar characteristics which define the stratification cell within the level of aggregation.
A Dealer trip seeks an area match at Level A and progresses through the increasing levels of aggregated VTR data until a match occurs. Area is obtained first, then effort.
For each area level and stratification cell, a discrete probability distribution function is formed representing the proportion of trips which fished in a unique statistical area. A discrete cumulative distribution is formed using the statistical area probabilities. Each unique statistical area within the VTR group will have a cumulative probability associated with it. Before the allocation begins, every Dealer trip is assigned a random number between 0 and 1. The random number is compared with the cumulative probability associated with each area. The cumulative probabilities are in ascending order; when the random number is greater than or equal to the cumulative probability value, the statistical area associated with the cumulative probability is assigned to the Dealer trip. Thus, a single area fished is assigned to a Dealer trip on a probabilistic basis by sampling (with replacement) the distribution of VTR areas within the group.
Total effort is not known in the Dealer data; each Dealer trip will be supplemented with effort (number of trips, days fished, and days absent) taken directly from a VTR trip or estimated from the pool of VTR trips with similar characteristics. When a match occurs at Level A, days fished and days absent are transferred to the Dealer trip only when both effort metrics have values (both must be not null). If available, the number of hauls, haul duration, crew size, gear quantity, and gear size are also transferred. If a match occurs at Level B, C or D, then an estimate of days fished (DF) per trip and an estimate of days absent (DA) per trip are assigned to the Dealer trip. Both days fished and days absent are estimated by the median of their distributions, respectively, within the cell. The median was selected as the simplest statistic of central tendency for distributions of various shapes.
The estimates of days fished and days absent are assigned to an entire trip. Therefore, all VTR trips within the group must be converted into 'whole trips.' The days fished and the days absent are each multiplied by the inverse of ntrips, the portion of the trip. This calculation is non-consequential for non-split trips; for split trips, the effort is expanded to represent a whole trip. For example, if a subtrip had an ntrips = 0.3, then, to convert this partial trip to a whole trip, the converted DF = DF * 1/0.3 and the converted DA= DA * 1/0.3.
In addition to estimating the median (second quartile), the first and third quartiles are also derived to provide a measure of dispersion. The quartile deviation can be calculated as (q3-q1)/2. The first and third quartiles of DF and DA will be transferred into the Dealer trip record for analysts to use, if desired.
The allocation assumes the follow: (1) Dealer landings are a census of total landings; (2) vessels land only once per trip; (3) each Dealer trip that enters the allocation represents one trip; and (4) VTR data set is a representative subset of the Dealer set.
The proportion of Dealer landings entering the allocation ranges between 19% and 39%. Between 51% and 74% of the landings that enter the allocation to find area fished match at Level A (a one to one match of Dealer and VTR trips). Total commercial landings changed very slightly (< 1 mt) due to rounding of whole species pounds on split trips. An evaluation of input data for allocation revealed the VTR subset generally reflected Dealer data. An evaluation of the random component of the allocation indicated that the random component did not contribute to a wide spread in stock landings, indicating that the random component is not a large source of stock landings variability. Analysts can estimate the uncertainty associated with the random component of the allocation algorithm using a multinomial probability. Although some statistical areas on the biological samples associated with allocated trips changed, the majority of samples remained unchanged.The trip-based allocation will eliminate the need for the single-species proration.
Commercial landings data are used to address management questions, to conduct stock assessments, and to meet reporting requirements for fishery resources off the east coast of the United States. Beginning in June 1994, the National Marine Fisheries Service (NMFS) Northeast Region’s data collection system was changed from a voluntary to a mandatory reporting system for USA fishermen and dealers who catch and buy/sell groundfish species regulated by the Northeast Multi-species Fishery Management Plan (NMFMP). The mandatory reporting system consists of two components: (1) dealer reporting and (2) vessel trip reporting. Each component contains information needed for fishery management and stock assessment analyses: the dealer reports contain total landings by market category, while the vessel trip reports contain information on area fished, kept and discarded portions of the catch, and fishing effort. There is no unique identifier to link these two components into one database. A multi-tier trip-based allocation scheme has been developed to combine information for these two components into a single database, which is consistent with commercial landings data prior to 1994. This comprehensive trip-based allocation will eliminate the need for single species proration. This project was undertaken by the staffs of the Northeast Fisheries Science Center (NEFSC) and the Northeast Regional Office (NERO).
The purpose of this document is to describe the multi-tier trip-based allocation designed to combine each mandatory reporting dealer (Dealer) trip with a vessel trip report (VTR) trip (or a group of VTR trips with similar characteristics) to obtain area fished and effort associated with the Dealer trip. The document describes: (1) allocation design; (2) the qualitative approach used to evaluate if the VTR data set is a representative subset of the Dealer data set; (3) the results of matching within the allocation; (4) an evaluation of the random component of the allocation; and (5) the changes to statistical area previously assigned to biological samples. A detailed technical in-house manual documenting the computer programs has been created.Background
An evaluation of the 1994 VTR data collected under the mandatory system was undertaken in spring 1996 by the Northern Demersal and Coastal/Pelagic Working Groups of the Stock Assessment Workshop (SAW). Findings were reported to the 22nd SAW (NEFSC 1996). The Stock Assessment Review Committee (SARC) recommended that: (1) the data needed further auditing; (2) use of existing data for provisional assessment calculations should be “performed with extreme caution and full awareness of the problems in the database”; (3) analysis and design of the mandatory data collection system should be completed and implemented with consideration given to the following features: (a) an unambiguous linking criterion for dealer, VTR, sea sampling, and effort monitoring databases; (b) pre-audits of all submitted data to eliminate ambiguities and preserve the original integrity of the VTR information; and (c) create user-friendly data collection forms with clear instructions for recording information; and (4) until long-range problems are resolved, immediate steps should be taken to improve the existing data collection process (NEFSC 1996).
Subsequent to the 1996 VTR data evaluation, further auditing of the VTR data has continued at NERO. Since 1997, single species prorations of landings data have been cautiously performed on an ad-hoc basis to meet stock assessment and management needs. The single species proration is narrow in scope, determining stock areas landings (comprising several statistical areas) by calendar quarter (Wigley et al. 1998) and does not estimate effort. The multi-tier trip-based allocation expands upon these limitations and derives area fished for landings of all species caught on a trip to statistical area (Figure 1) and estimates effort, while maintaining the Dealer’s original temporal resolution of month and day. Similar to the single species proration, the multi-tier trip-based allocation does not alter species landings, but augments the Dealer landings data record with area fished and effort.
Other NEFSC analysts have also prorated dealer data by time and area for protected species by port groups (Bisack 2003).
The Dealer data used in the allocation originated from the Commercial Fisheries Database System (CFDBS) Oracle tables maintained by NEFSC. Landings data that were not part of the mandatory reporting system did not enter the allocation; accordingly, these landings were held aside until the allocation was complete, then re-joined for a complete commercial landings data set (Figure 2). The mandatory Dealer data contain species landed and live pounds by market category, date landed, vessel permit, gear type, ton class, port landed, and price. The mandatory Dealer data do not include area fished, gear characteristics (mesh size, gear quantity and gear size), or effort (crew size, number of hauls, haul duration, days fished or days absent); this information will be supplied from the VTR data during the allocation. The steps taken to identify which Dealer trips will enter the allocation process and the procedures developed to prepare the Dealer data are described in a subsequent section.
In May 1, 2004, Dealer Electronic Reporting (DER) was implemented as part of Amendment 13 of the Northeast Multi-species FMP. There are no requirements for Dealers to submit gear information through DER; however, many Dealers do so.Vessel Trip Report data
Northeast multi-species VTR data used in the allocation originated from Oracle tables (DOCUMENT, CATCH and IMAGES) maintained by NERO. These data were used to populate Oracle tables (VESLOGyyyyT, VESLOGyyyyG, VESLOGyyyyS, where yyyy is 4-digit year) created by NEFSC (Appendix Figure 1). The VTR data tables used to determine year were based on the date landed or date sold as these dates would most closely correspond to the date in the Dealer data. These data contain logbooks from charter, party and commercial trips, as well as logbooks which document that no fishing took place during a given month. Only commercial trips which fished and had kept catch were used in the allocation. The VTR data contain information on area fished, kept and discarded species pounds, gear type (gear size, gear quantity, mesh size), and effort (number of hauls, haul duration and crew size). Extensive data summaries and analyses revealed the VTR data were in ‘raw’ form and that procedures were needed to further audit the 1994-2001 VTR data before the data could be used in the allocation scheme. VTRs which did not contain fishing area location data (e.g., statistical area, latitude/ longitude or loran) were eliminated for the data set. The VTR data used in the allocation procedure are described in a subsequent section.
Southeast pelagic vessel trip reports contain area and effort data needed to supplement the large pelagic landings data in the NERO. The Southeast pelagic VTRs will not be incorporated into the allocation at this time; however, the allocation scheme could accommodate multiple but separate sources of VTR data.Data constraints
Several data constraints precluded a simple direct match of each Dealer trip with a VTR trip. These data constraints include: (1) the lack of a unique identifier between the two data sets; (2) not every Dealer trip has a corresponding VTR trip due to the lack of 100% compliance by fishermen (i.e. fishermen are not submitting a logbook for each and every fishing trip); (3) incomplete logbooks (missing data within logbook); and (4) data collection inconsistencies between the two data sets.
In lieu of a unique identifier, data elements common to each data set are essential in establishing an indirect link to match trips between the two data sets. Common elements that uniquely describe a trip or characterize a trip’s fleet include: vessel permit, ton class, month, day, gear type, and port. The use of day, port, and gear as common elements have some associated caveats which are described later in this section.
Although mesh size is a key factor identifying sub-fleets, mesh size is not a common element in the two data sets and therefore could not be used. However, species caught during a trip is an indirect indicator of the mesh size used. As a surrogate for mesh size, VTR and Dealer trips were categorized into 12 main species groups based on the species kept (VTR) or species landed (Dealer) for a given trip. Main species groups are useful in differentiating sub-fleet sectors that are spatially distinct such as the long-line monkfish trips and long-line tilefish trips as well as the large-mesh and small-mesh otter trawl fisheries. Details on these main species groups are discussed later in this section.
There is not a one-to-one correspondence of trips between the Dealer and VTR data sets due to less than 100% compliance for VTR submission and incomplete or unusable VTRs; hence, the VTR data are a subset of Dealer landings. Without a one-to-one correspondence between data sets, it was necessary to develop a multi-tier allocation scheme that would allow both one-to-one matches as well as one-to-many VTR trips of similar characteristics in order to determine area fished and effort associated with the Dealer trip. The VTR data are examined to determine if these data are a representative sample of the Dealer data. A qualitative evaluation to identify potential bias is described in a subsequent section.
During the 1994–2001 period, incomplete VTR logbooks were submitted. In this case, incomplete logbooks consisted of VTRs that did not contain chart (statistical) area fished and/or VTRs which did not report the number of hauls and/or haul duration, the information needed to derive effort in terms of days fished. For VTRs that did not report a statistical area fished but did report a latitude/longitude or Loran, those data are used to derive statistical area, thus increasing the number of VTRs with statistical area. Any VTR trip that did not report statistical area fished or for which an area could not be derived from the point location was eliminated from the VTR data set. Intermittently since 2001, incomplete VTRs have been returned to the fisherman for re-submission of a complete VTR.
Regarding area and effort information, all combinations of logbook completeness existed: area and effort reported; area and no effort reported; effort and no area reported; and no area or effort reported. Given the number of logbooks with incomplete effort (missing number of hauls and/or haul duration) in 1994, the allocation scheme separated the determination of area from the determination of effort in order to maximize the number of VTRs used in the allocation. The allocation consisted of two phases: in the first phase, area fished was determined for each Dealer trip using VTRs with area fished regardless of missing effort. However, if a VTR trip reported subtrips (i.e., a split trip), and effort is missing on one or more of the subtrips, then the entire trip is eliminated from VTR set because subtrip effort is used to partition the proportion of ntrips (number or fraction of a trip) to each subtrip. In the second phase, effort (days fished and days absent) is determined using VTR trips that report both area fished and effort (number of hauls and haul duration).
Data collection and coding inconsistencies between the Dealer and VTR data collection systems necessitated the grouping of similar gear types and similar ports. These data groups are use in the characterization and stratification of trips for the allocation matching; however, the original Dealer data are not replaced with these data groups. The data groups are created as additional fields on the data record.
Gear groups: The formation of gear groups is necessary because not all gear codes in the Dealer data have a corresponding VTR gear code. The gear groups are generally based upon the CFDBS negear2 code, with some exceptions such as negear2 = 05 where otter trawl gear types are separated out into three unique gear group codes (i.e., scallop trawl, shrimp trawl, and otter trawl). On the other hand, where distinct negear2 codes represent similar gear types such as hoes and shovels and rakes, these gear types are coded into one group. The gear groups are formed in an ad-hoc fashion based on general gear knowledge as well as assistance from NERO staff. VTR trips are assigned a gear group for each gear used on a trip. For a Dealer trip, if multiple gear types are reported, then the gear type associated with the plurality of the catch is used.
Port groups: Port groups are formed to facilitate the aggregation of VTR data to capture fleet behavior patterns. Port groups are defined by concatenating the state and county codes (the first two and the last two digits of the 6 digit CFDBS port code), with a fifth column appended at the end; i.e., statecd||countycd||’0'. Qualitative analyses of the gillnet fishery revealed that some ports within a county should not be grouped together due to different spatial fishing patterns (i.e., fishing in different statistical areas) by each port within the county. To capture these port specific spatial patterns within a state/county group, a county is subdivided when a statistical area boundary bisected the county. The fifth column of the port group code is utilized to indicate which counties had been sub-divided and which ports within the split county are grouped together. A zero in the fifth column indicates the county was not split; a value greater than 0 indicate the county was split, and indicate which ports belonged within the sub-county. Those port codes used to represent ‘other county’ ports (e.g., ‘Other Barnstable’) have been re-assigned a port code representing ‘other state’ (e.g. ‘Other MA’) because it is not known which sub-division of the county ‘Other Barnstable’ should be assigned. Thus, the port is ‘bumped up’ to the state level. Each of the five counties listed below have their ‘other county’ port codes re-assigned to the corresponding ‘other state’ port code.
Five counties along the east coast identified as containing ports which may have different spatial patterns are:
Barnstable County in MA where areas 514, 521, and 538 trisect;
Suffolk County in NY ( Long Island) where areas 611,612 and 613 trisect;
Washington County in RI where areas 538 and 539 bisect;
Ocean County in NJ where areas 612 and 614 bisect; and
Cape May County in NJ where area 614 and 621 bisect.
Barnstable County has two sub-divisions representing north and south of Cape Cod. Suffolk County has two sub-divisions representing east and west on Long Island’s south coast (there was insufficient gillnet data to separate Suffolk County trips fishing in Long Island sound from those fishing south of Long Island; the lobster fishery was not examined).
Not all ports reported in the VTR data have a corresponding CFDBS port code, thus some of the detailed port information obtained in the VTR data can not be fully utilized. For example, Moriches is a port whose fleet fishes primarily in statistical area 613, however, this port does not have a unique CFDBS port code, therefore this port is assigned ‘Other Suffolk’ port, but is re-assigned again to the state level because the county is a sub-divided county. Consequently, the trips from this port were grouped with trips from ‘Other NY’, which may have a broad range of statistical areas than the single
port of Moriches. Expanding the list of CFDBS port codes could be useful to capture the spatial patterns of small ports within the allocation.
Throughout the time series, there are some Dealer trips reporting port as ‘other state’ indicating that the state is known but the specific port is not known. To accommodate these Dealer trips within the allocation, an additional VTR port group was formed by combining all VTR data for a given state into one port group for that state.
Another issue relating to port groups involved the distinction between ‘port landed’ and ‘port sold’. In the VTR data, the port landed is the location where the fish were taken off the vessel. In the Dealer data, the port may or may not be the port where the fish are landed since fish product can be trucked and sold in other locations. The Dealer ports most affected are Portland, Gloucester, and New Bedford, where auction houses exist and attract fish product from surrounding ports. Port agents often know when Dealer transactions involve an ‘out-of-town’ vessel where the fish have been trucked and the port agent will either send the weighout slip to that port, or assign the appropriate port landed to that transaction (pers. comm. Scott McNamara, NER port agent, Portland, ME). Of course, it is unrealistic to expect the port agents to track all vessel transactions.
It may be possible to ascertain port sold from the VTR using the dealer permit number; however, limitations occur because not all VTRs report a dealer permit number, and some VTRs report two different dealers in two or more cities or states. Given these limitations and the fact that the dealer permit numbers in the VTR data are not audited, this aspect is not incorporated into the allocation scheme at this time.
For simplicity, the allocation scheme used port landed in the VTR as a corresponding element to port in the Dealer data. Recognition of a potential mismatch is acknowledged for two of the four matching allocation levels (Level C and D utilize port group in the stratification, as described in a subsequent section). The potential mismatch is expected to be minor as preliminary analyses indicated only a small portion of Dealer trips enter these levels. Also, it is expected that the number of Dealer trips affected by this potential mismatch will decline over the time series as more and more Dealer trips have a direct match with a VTR trip, reducing the need for fleet characteristics such as port group.
Main species groups: Dealer and VTR trips are assigned a main species group for each trip and gear group used on that trip. Main species groups are used as a surrogate for mesh size to further sub-divide the fleet definition of ton class, port, gear, and month. Twelve main species groups were formed with the intent to generally capture major sub-fisheries within a fleet. The species groups were defined in an ad-hoc fashion. An exploratory analysis of data grouped by the 12 species groups detected differences in spatial distribution patterns by main species. These analyses were conducted on longline, gillnet and otter trawl gears. There were four gear types (scallop dredge, lobster pot, shrimp trawl, and scallop trawl) identified as having a single primary species associated with the gear; hence, these four gear types are assigned a single main species group. The species groups were formed based on the reported kept quantity in the VTR and the landed pounds in the Dealer data. To derive main species group, each species is assigned to a main species group, then the species weights are summed by main species group for each trip. The main species group is the group with the maximum species weight.
The main species group allows the category defined by ton class, gear group and port to be further subdivided to capture species-specific spatial patterns. For example: Ton class 3 longline boats from Montauk were fishing in spatially distinct areas depending on which species they were targeting (Statistical Area 537 for monkfish, Statistical Area 539 for tilefish).
Main species group and species group are synonymous.
Other caveats: Another issue related to the input data involves the VTR date fields. During the 1994–1996 data entry, date sailed was a required field for data entry and had to be reported at the time the VTR was submitted; however, date landed and date sold were not required fields for submission (i.e., logbooks were accepted with this information missing). At data entry, if logbooks were missing date sold, then date landed was used. If date landed was missing, date sold was used and if date sold was missing then date sailed was used. No indicator or flag was used to identify these trips where a ‘substitute’ date was supplied at data entry; hence, there is no way to identify actual days absent from estimated days absent [days absent is calculated in hundredths of days as (date landed - date sailed + 1); no time component was used in the 1994 allocation]. For those trips that do not report date landed or date sold, days absent will be underestimated; these trips will be incorrectly categorized as day trips. Given the uncertainty in the date landed and date sold, and the lax submission requirements for these fields, it was decided to use the date landed (the field most often reported) as the date to derive year, month and day for the VTR data. More importantly, however, the distinction between date sold and date landed is that date sold represents a transaction at the trip-species level (there may be multiple sold dates) while date landed is a trip-level variable. Date landed was selected because this is a trip-based allocation. It is recognized that fish from one trip may be sold on multiple days or fish from one trip sold on a day that is different from the date of landing. To account for these situations and to bridge this apparent disparity in date landed vs. date sold, the VTR sold date was utilized to the extent possible given the aforementioned issues. The steps taken to utilize date sold and date landed are further described in a subsequent section.
Another data issue in the VTR was the unit of measure of the species quantity kept. Because of the uncertainty of the unit of measure of the species quantity kept, it was presumed that most species weights were in pounds, live weight; however, landed pounds may be have reported for some species. Species codes were reviewed and using Northeast Conversion Factors (established for CFDBS), species kept quantities that were reported in bags, trays, bushels, etc were converted to live pounds. Additionally, some of the reported quantities were questionable (i.e., amounts large enough to sink a vessel); given this, species proportions are used in allocation. The VTR species weights are not used to replace the Dealer landings.
Given the data constraints and the goal of creating a consistent, comprehensive and compatible data set with commercial landings data prior to 1994, a multi-tier trip-based allocation scheme has been developed to match a Dealer trip to a corresponding VTR trip or a group of VTRs based upon fields which are in common to both data sets.
The multi-tier trip-based allocation scheme is designed to resemble, as well as possible, the methods utilized during the voluntary data collection system where port agents collected and/or estimated area fished and effort based upon knowledge of individual vessel and fleet behaviors. Prior to 1994, NMFS port agents would interview vessel captains to obtain area fished (a 10-minute square location within a statistical area) and effort information such as days absent from port, number of hauls, haul duration, crew size, and the quantity and size of the gear used on a given trip. However, not every fishing trip was interviewed. For non-interviewed trips, the port agent used knowledge gained through prior interviews of the vessel and the fleet to assign a statistical area and estimate days fished (time the gear was actively fishing). For non-interviewed trips, the resolution of area fished was not as fine as for interviewed trips, and similarly, detailed effort information was not obtained; however, days fished and days absent were estimated. In the multi-tier trip-based allocation, the VTR trips are considered a sample of the commercial trips under the mandatory system, and thus provide the information previously collected during a port agent’s ‘interview’; accordingly, these VTR trips pooled into vessel and fleet groups form the informational base for the ‘non-interviewed’ trips.
Total commercial landings in the Dealer data are assumed to be known, but the spatial pattern of these landings is not known. The allocation determines an area fished for the landings based upon the spatial patterns observed in the VTR data. Dealer landings (pounds and value) are not altered during the allocation; area fished (statistical area) is added to the Dealer data record. Total effort in the Dealer data is not known; the allocation determines effort based upon the effort reported in the VTR data. The allocation is trip-based; hence, a trip’s area fished and effort will be associated with all the species landings from that trip. The allocation determines area fished first, then effort. The word ‘determine’ is used because in some cases area fished and effort information come directly from a VTR and in other cases, area fished and effort have been estimated based on a group for VTRs.
In this allocation, a trip is defined as a group of data records with the same year, month, day, and vessel permit in both the Dealer and VTR data sets. A split trip is defined as a trip which used either multiple gear types, multiple mesh sizes or fished in multiple statistical areas.Allocation Levels
The VTR data are aggregated into groups containing VTR trips of similar characteristics. Four groups (Levels A, B, C and D) of increasingly aggregated VTR data are created, stored as Oracle tables, and used in the allocation. Two levels (A and B) represent vessel-oriented data and two levels (C and D) represent fleet-oriented data (Figure 3). Level A comprises audited VTR trips that have not been grouped. Level B comprises VTR trips from Level A that have been pooled by vessel permit, gear group, main species group, and month. Level C comprises VTR trips from Level A that have been pooled by ton class, port group, gear group, main species group, and calendar quarter. Level D comprises VTR trips from Level A that have been grouped by port group.
Every attempt was made to keep the allocation scheme at a monthly resolution; however, for the fleet-oriented data (Level C), it was necessary to use quarter-year to ensure sufficient sample sizes within each stratification cell. Thus, Level A uses the month and day, Level B combines trips of the same vessel over the month, Level C combines trips with similar fleet characteristics over the quarter, and Level D combines trips with similar port characteristics. Level D is intentionally a broad group to capture all trips that did not find a match at a previous level. A total of seven data sets were formed: one table for Level A containing area and effort information; and an area determination table and an effort determination table for Levels B, C and D (Figure 3).
If a Dealer trip has a corresponding VTR (i.e., a one-to-one match on vessel permit, month, and day between the two data sets), then the area fished and the effort information, if present, is transferred directly onto the Dealer trip and the Dealer trip data record is complete. The Level A match corresponds to a pre-1994 port agent’s interview. However, if a Dealer trip does not have a corresponding a VTR trip, then the Dealer trip is matched to a group of VTR trips that have similar trip characteristics. If a match occurs with a group of VTR trips (from Levels B, C, or D), then a single area will be assigned to the Dealer trip on a probabilistic basis by sampling (with replacement) the distribution of VTR trips within the group. Days fished and days absent will be assigned to the Dealer trip based upon the median days fished per trip and median days absent, respectively, from trips within the pooled VTR data for that given area fished. The increasing levels (Levels B, C, D) of pooled VTR trips form an information base similar to the pre-1994 data collection system, where the port agent estimated area fished and effort for a non-interviewed trip based upon either previous vessel interviews and/or based on fleet patterns.
The allocation sequentially searches each of the four VTR data levels until a matching VTR trip (or group of trips) has been obtained for a Dealer trip (Figure 2). The first objective is to find an area fished; once area has been determined, then effort can be determined. Since area and effort are acquired sequentially in the allocation, area may be determined at one level, and effort may be determined at the same or higher level. For Dealer trips which do not find a match in one of the four levels, the area and/or effort fields are assigned the CFDBS default values (days fished and days absent are set to null).Meta Fields
Two meta fields have been created and appended to the Dealer data to document which VTR data aggregation level was used to obtain the area and effort information. The area and effort meta fields are independent of each other. The meta fields can guide users to which Dealer data may be appropriate for certain analyses and which may not be appropriate. For example, catch per unit effort analyses would utilize data from Level A only (for which effort has not been estimated). The user will employ these meta fields in a similar fashion as they used the interview_indicator in pre-1994 data to discriminate between actual versus estimated area and effort data.
The meta field Alevel will have character values A, B, C, D indicating that area fished was obtained from either Level A, B, C, or D, respectively. A value of X indicates that the Dealer trip entered the allocation, but did not find a match at any of the four levels. Null indicates that the Dealer trip did not enter the allocation. All trips that enter the allocation are expected to find a match; Level D has been designed to capture all trips.
The meta field Elevel will have character values A, B, C, D indicating that effort was obtained from either Level A, B, C, or D, respectively. A value of X indicates the Dealer trip entered the allocation, but did not find a match at any of the four levels. Null indicates that the Dealer trip did not enter the allocation.Level A: Dealer Trip Matches a VTR Trip
Dealer trips that matched at Level A are augmented with as much VTR information as available including: mesh size, depth, latitude, longitude, ten minute square, quarter degree square, statistical area, crew size, gear quantity, gear size, number of hauls, haul duration, days fished, days absent. For a Level A area match to occur, statistical area must be present on the VTR; all other area fields may or may not be present. If the VTR trip does not contain information on the aforementioned area fields, these fields will be assigned the CFDBS default value. For a Level A effort match to occur, days fished and days absent must be present in the VTR; all other effort fields may be null and will be assigned the CFDBS default value. At Level A, area fished and effort, if available, are assumed known from the VTR; no estimation is performed in the allocation for Level A matched trips. The VTR tripid and gearid  are added to the Dealer data record for documentation purposes. The meta fields for area and effort are set to ‘A’.
If the Dealer trip matched a VTR trip which does not contain days fished and day absent, then the VTR area information is used to augment the Dealer trip at Level A. The search for an effort match continues through Level B, C, and D effort tables to determine days fished and days absent (Figure 2). The VTR tripid is added to the Dealer data record, the area meta field is set to ‘A’, and the effort meta field is set to the level where effort was determined.
A Dealer trip matched at Level A can result in a split trip when the matching VTR indicates a split trip (i.e., the VTR trip fished in multiple areas, or used multiple gears or mesh sizes). The Dealer trip landings (species by market category) and the value (dollar amount) associated with species landings are partitioned into subtrip components. The process of partitioning landings and price among subtrips is described in a subsequent section.Levels B, C, and D: Dealer Trip Matches a Pool of VTR Trips
Dealer trips that matched a pool of VTR trips at Level B, C or D are augmented with an estimate of area fished and an estimate of days fished and days absent. No VTR tripid and gearid are assigned to Dealer trips which match at Levels B, C or D. Fine scale area and effort information, such as latitude, longitude, quarter degree square, ten minute square, crew size, depth, mesh size, gear quantity, gear size, number of hauls, and haul duration are not estimated; these fields are assigned the CFDBS default value of null. No split trips will result from Levels B, C, or D; a single area is estimated for the entire trip and days fished and days absent are determined on a ‘per-trip’ basis. Estimates of area fished and effort are described below.Area Probability Distribution Functions
At Levels B, C, and D, VTR trips are grouped based upon the stratification criteria for each level. For each level and stratification cell, a discrete probability distribution function is formed representing the proportion of trips which fished in each unique statistical area. A discrete cumulative distribution is then formed using the statistical area probabilities. The number of trips within each cell and the number of trips within each unique statistical area are also stored.
Each Dealer trip is assigned a random number between 0 and 1 that has been generated using a large, odd number as the seed; this seed is stored in the software. When a Dealer trip matches a stratification cell, a single area fished (statistical area) is assigned to the Dealer trip on a probabilistic basis by sampling (with replacement) the distribution of statistical areas with the cell. The random number value is compared with each discrete cumulative probability (in ascending order) associated with an unique statistical area. When the random number is less than or equal to the cumulative probability value, the statistical area associated with the cumulative probability is assigned to the Dealer trip. The probability, prob, associated with the statistical area assigned to the Dealer trip, is stored in CFDETSyyyyAA (Appendix Figure 1).
The following example is given for illustration:
For a given Level and stratification cell, there are 6 VTR trips in the cell which fished in 3 unique statistical areas. Three trips fished in Area 521, two trips fished in Area 522, and 1 trip fished in Area 526. For this cell, the probability of fishing in Area 521 is 0.50 (3 trips / 6 trips), 0.33 (2 trips / 6 trips) for Area 522, and 0.17 (1 trip / 6 trips) for Area 526. When ordered by ascending probability, the cumulative probabilities for the three areas 526, 522, and 521 are 0.17, 0.50, and 1.0, respectively. Each Dealer trip is randomly assigned a number between 0 and 1. This number is compared with the cumulative probabilities to determine a single area fished. In this example, if a given Dealer trip with a randomly assigned value of 0.75 matched this cell, the Dealer trip would be assigned Area 521. On average, for Dealer trips which match this Level and cell, 50% of matches would be assigned Area 521, 33% of the trips would be assigned Area 522, and 17% of the trips would be assigned Area 526. This example illustrates whole trips in each statistical area; however, this algorithm also works with a mix of whole and partial trips.Effort Estimation
Effort information is collected only in the VTR component of the mandatory data collection system; therefore, total effort is not known in the Dealer data. Effort can not be distributed in a similar fashion as the landings; each Dealer trip’s effort must be estimated from the VTR data. Dealer trips acquire effort directly from a corresponding VTR trip at Level A or effort is estimated from a group of VTR possessing similar trip characteristics at Level B, C, or D. Exploratory analyses indicate that days fished (DF) and days absent (DA) can be a function of statistical area (i.e., longer trips were made to statistical areas farther from home port). At Levels B, C, and D, DF and DA are estimated using the median days fished and median days absent from the VTR trips within the stratification cell for a given statistical area. Also, the first and third quartiles are derived so the semi-interquartile range [(Q3 - Q1 ) / 2] may be used as a measure of dispersion. In addition to the median, another measure of central tendency, (Q1 + Q3)/2), may also be computed from the quartiles. If needed, the interquartile range could be used as a diagnostic for homogeneity of effort within the cell during the input data set evaluation.
Days fished and days absent, although correlated, are independent of each other. Days fished measure the time (in tenths of days) the gear was actively fishing while days absent measure the time (in hundredths of days) the vessel was away from port. Only VTR trips that contain both DF and DA are used to create the data tables at Level B, C, and D.
To estimate the median value of a distribution of days fished per trip and days absent per trip, effort associated with split trips is multiplied by the inverse of ntrips (or the fraction of the trip which is associated to the subtrip) to convert effort from a partial trip basis to a ‘per-trip’ basis. This allowed all VTR trips and subtrips within a stratification cell to be combined into one distribution and the median value would represent effort on a ‘per-trip’ basis.
Probability density functions (PDF) to determine DF and DA (as used for area fished) were not appropriate because DF and DA may be correlated on a given trip (especially for mobile gear types) and using a separate PDF for each effort measure may result in a mis-match of the two due to the random nature. A joint PDF could have been used, but the objective was to keep the estimation of effort as simple as possible given all the data constraints. The median DF and median DA are selected as the simplest statistic of central tendency for various shaped distributions.
When a match occurs at Levels B, C, or D, the meta fields for area (Alevel) and effort (ELevel) are set to the corresponding letter representing the Level. When no match occurs for effort, effort fields will be assigned their CFDBS default values (null). In addition to meta field Elevel, the effort indicator field used in pre-1994 data, effind, is assigned as given below:
Effort indicator Criteria 4 Alevel = A and Elevel = A 3 Alevel = A, B, C or D and Elevel = B, C, or D 2 Alevel = A, B, C or D and Elevel = X
Two diagnostic fields were created to monitor the matching of Dealer and VTR trips for area and effort. Each time the VTR data are used in a match, a counter is incremented. There is a counter for area and a counter for effort. The counters may be used to evaluate the frequency of cell usage in estimating area and effort. Evaluating how many times a given cell was used provides feedback on allocation, the Levels, and the stratification. The area and effort counters at Level A are careful reviewed to ensure that only a VTR trip was used only once at this level.
- Assume Dealer landings is a census of total landings;
- Vessels land only once per day;
- Each trip (permit-month-day) in the Dealer data set represents only one trip
(consolidated trips are special cases and handled according);
- VTR data are representative subset of the Dealer data.
It is recognized that a trip may sell its catch over several days; if this is not accounted for, the number of trips will be over-estimated. We have addressed this issue to the extent possible when identifying unique Dealer trips. We have established a unique Dealer trip identifier, dlrtrpid. This trip identifier links together all transactions that are associated with a trip.
Although landings data are collected in both dealer and vessel components of the mandatory reporting system, ‘kept’ pounds are recorded in the VTR and ‘landed’ pounds are recorded in the dealer report. It is assumed for the purposes of these analyses that the dealer data contain the most complete record of total landings, and that the VTR data are an unbiased subset of the commercial data set.
In 1994, an exploratory analysis revealed that there were potentially 74 trips that reported the same permit, month, day and had more than one time sailed. On 29 of the 74 trips, one of the multiple time sailed was ‘0000’; on another 17 trips, the two time sailed values were within one hour of each other; thus, over half the apparent two-trips-per-day trips had data errors. Thus, a potential 36 trips out of ~50,000 plus trips were incorrectly combined with another trip made by the same vessel.
We have decided to ignore multiple trips per day because we can not distinguish the trips that land multiple times per day from those that have misreported, i.e., do not fill out the logbook correctly. We recognize that day boats may make multiple trips per day; in this case, the number of trips will be an underestimate. In the near future, when electronic VTRs are implemented and/ or the VTR unique trip identifier is fully in place, this issue will diminish.
PREPARATION OF DEALER INPUT DATA
The Dealer data contains consolidated trips; consolidated trips have landings from multiple trips for the same vessel. Consolidated Dealer trips have day = ‘00’. These trips do not enter the area matching phase at Level A; they are only matched at Level B, C, or D to estimate a single area. To assign an area based on vessel or fleet characteristics seemed more appropriate than to assign a single VTR to multiple trips. Consolidated trips do not enter the effort matching phase; effort (days fished and days absent) are assigned the CFDBS default value. Since a consolidated trip represents an unknown number of trips, it is inappropriate to apply the estimated effort that represents one trip. The effort fields of consolidated Dealer trips are assigned the CFDBS default value (null). To estimate the number of trips the consolidated Dealer trip represents, the number of unique month||docn is used as a surrogate for the number of unique trips.
It should be noted that records within the Dealer data represent ‘transactions’ for which single or multiple transactions may comprise a trip. Multiple transactions may occur on the same day or over several days. Since the allocation is ‘trip-based’, it is necessary to identify all transactions that are associated with a given trip. All transactions for a given trip are assigned a unique Dealer trip identifier.Dealer Trip Identifier
Three types of multiple Dealer transactions exist: 1) multiple transactions on the same day from one trip; 2) multiple transactions on different days from one trip; and 3) multiple transactions on the same day and multiple transactions on different days from one trip.
Type 1 transactions can be identified using only the Dealer data based on the permit, month and day; all transactions for a trip will be tagged with the same dlrtrpid.
Type 2 and Type 3 transactions can not be identified solely with Dealer data, VTR data are needed to identify these transactions.
min sold = max sold = date landed
Trips that sold on a single date, the landed date
(min sold = max sold) <> date landed
Trips that sold on a single day but not the landed date
(min sold <> max sold) = or <> date landed
Trips that sold on multiple days
In the allocation, the Dealer uses month and day (representing sold date) while the VTR uses month and day based on date of landing because this is a trip-level data element and this is a trip-based allocation. It is recognized that fish from one trip may be sold on multiple days or fish from one trip sold on a day that is different from the date of landing. To account for these situations and to bridge this apparent disparity in date landed vs. date sold, the following steps have been developed to: (1) maintain the original date in the Dealer data, and (2) use the VTR landing date for matching purposes only for multi-day (trip-boats) trips which sold on multiple days or the sold date is different from the landed date.
A set of VTR trips are identified that have days absent greater than 1 day (trip-boat trips) and sold on multiple days or sold on a day different from date landed. This set does not include the following: (1) trips with erroneous maximum sold dates (if maxsold - datelnd1 > 10 then delete); (2) trips with more than 3 different sold dates (there are many VTRs where fishermen had reported many trips on one log sheet); (3) trips where minimum date sold is less than the date landed; and (4) overlapping VTR trips with overlapping dates for the same permit. This set also excludes day trips as it is unlikely that they would sell over multiple days if they were a day-boat, with the exception of fishermen who pound/carr (i.e., hold in a cage).
Dealer transactions and the subset of VTR trips (Type 2 and 3 above) are merged based on permit and where the Dealer date landed is between the VTR minimum date sold and maximum date sold. All Dealer transactions associated with the VTR trip from the subset will be assigned the same dlrtrpid and the month and day based on the VTR date land. The original Dealer month and day values are stored and will be used to re-populate the month and day fields after the allocation.
A Dealer trip identifier, dlrtrpid , will be assigned to all transactions in the Dealer data and is defined as the concatenation of month and document number, month||docn. A trip is defined as a unique permit-month-day. Single transaction trips will have the dlrtrpid equal to month||docn. A trip with multiple transactions will have one dlrtrpid for all transactions, and the dlrtrpid will be based on the month||docn of the transaction with the most landed pounds.Non-allocated and Allocated Dealer Data
As noted earlier, not all Dealer trips will enter the allocation. The Dealer data are partitioned into two data sets: allocated and non-allocated.
Dealer data entering the allocation have the following criteria: (1) source = ‘07’ (i.e. mandatory reporting); (2) month between 01 and 12 (except for 1994, where only data between April – December only); and (3) vessel with a unique permit represent a single vessel.
Dealer data not entering the allocation have the following criteria: (1) Dealer data with source != ‘07’ including non-mandatory reporting data such as state data, bluefin tuna and other highly migratory species data, surfclam and ocean quahog fishery trips, landings data from Foreign Sea Sampling and the Potomac River Fish Commission (as identified by the letters E, Q to Z in the document number); (2) Dealer data between January to March 1994 (data prior to mandatory reporting) or data with month = ‘00’; (3) vessels with non-unique permits such as permits in (000000, 190998, 390998); and (4) data where gear is in (400, 040,115,999) or negear2 in = (03, 17), as these gear type have other data reporting systems other than the VTR system.
Trips with transactions that fall in both the non-allocated and allocated sets are identified and all transactions associated with these trips are relocated to the non-allocated set. All transactions associated with a trip are either in the allocated set or the non-allocated set.
PREPARATION OF VESSEL TRIP REPORT DATA
Not all VTR data are used in the allocation. Vessel trip reports from the party and charter industry will not be used to allocate commercial trips. Only commercial trips (tripcatg = 1) that fished and reported an area fished (either a statistical area or a latitude and longitude or Loran from which a statistical area could be derived) will enter the allocation. Although there were several different audit procedures conducted on the 1994 data (NERO audits and NEFSC side-by-side audits) the data needed more auditing before the data could be used in the allocation. Thus, a suite of programs were developed to further audit the VTR data. The allocation audits do not screen every field in the VTR data, only a limited number of fields pertaining to the allocation itself (especially the fields used to match the dealer data with the VTR data) were considered. Other fields were corrected on an ‘ad-hoc basis’; i.e., if an error was discovered, it was corrected. However, a thorough screening of all fields (such as crew size and depth) was not undertaken at this time. Improperly submitted VTRs or trips with no catch were excluded from the allocation.
All records with area recorded as 551, 552 (Canadian waters) were changed to 561 or 562 after verifying the ten minute square was on the Hague Line between Divisions 55 and 56. All records with area recorded as 523 & 524 were re-assigned to 561 & 562 respectively (Areas 561 and 562 were formerly 523 and 524 before the
USAand Canadian boundary line was established). Area is a required field for a VTR to be used in the allocation scheme. Any records for which area was unresolved were not used.
VTR trips that reported the following statistical areas were not used in the allocation: 110, 100, 500, 510, 520, 528, 530, 540, 550, 551, 552, 560, 600, 610, 620, 630, 799, 800, and 899. This list includes statistical areas that (1) are beyond the range of the fishing activity in the Northeast region and/or (2) represent a ‘generic’ statistical area that represents a group of statistical areas. For example, Area 510 represents the collection of statistical areas from 511 to 515, a group of 5 statistical areas.
A trip should be split into subtrips if area, gear, or mesh changed during the trip. A trip is defined a group of VTR records with the same year, month, day, and permit. Year, month, and day are based on date landed. A subtrip is an integer assigned to a record or group of records which make up part of a trip. This number starts with one and is incremented when the gear, area, or mesh size changes within a given permit, year, month, day. The number of subtrips, nsubtrip, is an integer value indicating the number of subtrips for a given trip, nsubtrips= max(subtrip).
To identify trips which may be artificially split when mesh is either not reported on one of the subtrips, reported incorrectly, or entered incorrectly at data entry, additional screening was conducted for mesh.
Other data preparation of the VTR data included: (1) converting species weights, based on species codes, to pounds when quantity kept was reported in bushels, trays, gallons, barrels, etc.; and (2) limited auditing of days fished and days absent was performed to identify and remove outliers (unrealistic values).VTR Data Sets
VTR trips with the following criteria are used in the allocation: (1) statistical area (derived from cnemarea) is not null or 0; (2) tripcatg = 1 (omit charter and party trips); (3) not_fished = 0 or is null (omit trips which did not fish); and (4) vessels that landed in ME, NH, MA, RI, CT, NY, NJ, DE, MD, VA, NC (omit NC for 1994-1996 because NC landings were not included in the Dealer landings for these years).
VTR trips excluded from the allocation include: (1) trips where statistical area could not be derived; (2) trips for which date landed was less than date sailed; (3) trips from vessels with permits in the 800 series representing NY state vessels that are not federally permitted vessels; (4) trips that used the following gear codes (drc, llp, hrp, ptm, gnd, mix, oth, null) and lobster pot gear fishing in zones 0 ,1, or null; (5) trips with more than one subtrip that have any excluded subtrip; and (6) trips with more than one subtrip for which one or more of the subtrips has no effort.
Seven VTR data sets are created. A base data set is created containing all useable individual VTRs trips, this forms Level A. From this data set, six additional data sets are created, an area data set and effort data set for Levels B, C, and D described below.Level B, C and D Area Data Sets
Level B area data set contains VTR trips from Level A grouped into cells stratified by permit, month, gear group, and species group. Trips for which the species group is null are excluded.
Level C area data set contains VTR trips from Level A grouped into cells stratified by ton class, quarter, port group, gear group, species group. Trips for which species group is null are excluded as well as trips that fished in the Grand Banks statistical areas in (330, 340, and 350).
For Level D area data set contains VTR trips from Level A grouped into cells stratified by port group. Trips fishing on the Grand Banks (area 330, 340, and 350) were excluded.
The unique statistical areas fished by these trips are determined within each Level and cell, and their associated probability and cumulative probability are calculated. These data sets contains the following variables for each cell: (a) probability; (b) cumulative probability; (c) number of trips (ntrips) in the cell; (d) given an area, the number of trips in the cell; (e) count of the number of trips or subtrips which formed this cell; (f) area: an unique statistical area within the cell; and (g) a counter of dealer matches within the cell.
This information can later be used to calculate multinomial probability to capture the uncertainty associated with statistical area landings at Levels B, C, and D.Level B, C and D Effort Data Sets
For Level B effort data, the VTR trips in Level A are grouped into cells stratified by permit, month, gear group, species group, and area. Trips for which species group is null are excluded as well as trips for which days fished or days absent or null.
For Level C effort data, the VTR trips are grouped into cells stratified by ton class, quarter, port group, gear group, species group, and area. Trips for which species group is null are excluded as well as trips that fished in the
Grand Banks statistical areas in (330, 340, and 350) and trips with no effort.
For Level D Effort data set, the VTR trips from Level A are grouped into cells stratified by port group and area. Trips fishing on the Grand Banks (statistical areas 330, 340, and 350) were excluded as well as trips with no effort.
MATCHING: Creation of Header records and Species detailed records
The Dealer data comprise (1) ‘header’ records that contain trip landings, trip value and effort for each trip/subtrip, and (2) detailed species records that contain species, market category (grade), weight and value for each species-market category and trip/subtrip.Header Records
When a Dealer trip matches a VTR trip that has subtrips (Level A only), multiple headers will be created. The match will return the number of headers records equivalent to the number of subtrips. The additional header records will have the same Dealer trip identifier, and the subtrips will be sequential; this information will come directly from the VTR trip. The landings and value for the trip are partitioned among the additional headers based on species area proportions observed in the VTR data if available or based on the effort (ntrip).
If a Dealer trip has multiple transactions and the Dealer trip matches a VTR (split or non-split), then effort (ntrips, days fished, and days absent) must be partitioned evenly among the subtrips. In all other circumstances, ntrip is used as the basis to partition effort from the VTR. For non-split trips, landings and value will remain the same; for split trips, the landings and value will be partitioned among the subtrips based on ntrip. Details on the additional headers created and how effort (ntrips, df and da) is partitioned among the subtrips are given in Appendix Tables 1 and 2 for single and multi-transaction trips, respectively.Species Detailed Records
For non-split trips, statistical area will be transferred onto each Dealer species records for a given matched trip. For split trips, Dealer species-market category landings and value will be distributed to the subtrips based on a species-specific area proportion derived from the VTR, if species information is available; otherwise, effort (ntrips) from the VTR trip is used to partition landings and value when species information is not present.Combined Dealer Data Sets
The non-mandatory data and allocated data are combined into one data set. An audit is run of the entire data set using a modified version of a master audit.
Due to poor VTR logbook instructions, gear size and gear quantity for some fixed gear (lobster pots and crab pots) must be nulled out when Alevel = ‘A’ and source = ‘07’. Also, gear size and gear quantity are set to null when negear in (200, 210, 300) and Alevel = ‘A’ and source = ‘07'.
Master Oracle tables CFDETSyyyyAA and CFDETTyyyyAA are created for use (Appendix Figure 1). Biological samples stored in CFLENyyyy and CFAGEyyyy were updated with the allocated area assigned to the trips from which the sample was taken. If a biological sample was taken from a split-trip, the sample was assigned area = ‘000’. This prevents the misuse of these samples for multi-stock species while allowing their use for single stock species.
COMPARISON OF ALLOCATION INPUT DATA SETS
The allocation assumes the VTR data are a representative subset of the mandatory reporting Dealer data. The VTR and Dealer data are compared to identify any potential bias in the VTR data which may exist due to reporting compliance. The comparisons were performed at the same level of resolution at which the allocation would be conducted, i.e., month, quarter, port group, gear group, ton class, and species group. Annual comparisons were qualitatively evaluated based upon the percent distribution of trips in the VTR and Dealer sets by the stratification variables: month, quarter, state, port group, ton class, gear group, and species group. An illustrative example using data from 2000 was selected to display the percent distributions for each stratification variable (Figures 4a-4e). To summarize the percent distributions for all years, and categories within each stratification variable, the differences between Dealer percentage and VTR percentage were calculated and plotted (Figures 4a-4e).
The differences in percentages between the Dealer and VTR for month, ton class, and port group generally ranged between +/- 5% for most years except 2004. For gear groups and species group, the percentage differences generally range between +/- 10% except in 2004–2006. Further examination into the large percentage differences between VTR and DEALER data for the scallop dredge gear group revealed that Dealers are reporting scallop landings from negear 381 (‘dredge, other’), while fishermen are reporting using negear 132 (‘sea scallop dredge’). In general, there is close agreement between the overall percent distributions of the VTR and Dealer data, indicating that the VTR input data general reflects the Dealer data.
Summary of the number of VTR trips and trips with subtrips are given in Table 1. As shown in Palmer and Wigley (2007), some VTR multiple-subtrip trips underreport the number of statistical area fished, resulting in fewer split trips than expected. However, as Palmer and Wigley (2007) show, this does not have serious implications for the overall use of the allocation procedure.
Summary statistics of the number of metric tons landed in the Dealer data, the proportion of landings that entered the allocation and the proportion of landings that matched at each allocation level for area and effort are given in Table 2 for 1994–2007. In the allocation, there is a very small amount (< 1 mt) of increased landings that result from the rounding of species pounds in trips that have subtrips. The proportion of Dealer landings entering the allocation ranges between 19% and 39%. Between 51% and 74% of the landings that enter the allocation to find area fished match at Level A. The percent of total landings subject to the random component of the allocation ranged between 7% and 14%.
Annual species landings and percent landings, by non-allocated and allocation level for area fished, are given in Table 3, Table 4, and Figure 5. The species summarized here are the 8 species with multi-stock components as well as all species combined. Except for 1994, generally more than 90% of each of the 8 species landings entered the allocation for all but 2 multi-stock species (Table 4). Since first quarter of 1994 did not enter the allocation, it is expected to have a lower percentage of total landings entering the allocation. Over all years, both red hake and silver hake have a higher percent on non-allocated (State) landings than the other multi-stock species.
In 2004, Dealer Electronic Reporting (DER) was implemented. As with many new data collection systems, including self-reporting data collections, there may be start-up issues that must be resolved through outreach and education of the data collection participants. The increased percentage of species landings not entering the allocation in 2004-2006 is attributed to the start-up of this new data collection system. By 2007, the percentage of species landings entering the allocation returned to values observed prior to 2004 (Table 4).
As mentioned above, not all trips enter the allocation due to non-unique permit numbers, unknown gear type, etc. Thus there still remain a small percentage of landings where statistical area remains unknown. Generally, the percentage of landings with unknown area is relatively small (less than 3%) across multi-stock species (Figure 6). However, at the beginning of DER, during 2004-2006, there is an increase in landings with unknown statistical area. As mentioned above, this is attributed to the start-up of DER and the percentage diminishes in 2007 for six of the eight species (Figure 6). Red hake and silver hake continue to have trips associated with under tonnage class vessels without unique permits and/or state landings. A standardized species-basis procedure for assigning stock area to landings without statistical area has been developed by Palmer (2008).Biological Samples
Biological samples (lengths and age structures) taken for species landed by trips that entered the allocation will acquire the allocated trip area. For samples taken from split trips, the area is assigned ‘000’. If a trip did not enter the allocation, the original area, if present, remained on the sample.
There are some samples for which statistical areas changed. There are various reasons for the changes, including the sample not having an area and acquired an area via the allocation, or the statistical area changing due to internal consistencies checks performed on the VTR after the VTR was used to assign statistical area to the sample (this represents about 2% of the samples with areas that changed); the allocated Dealer trip matched at Level B, C, or D and an estimated area was obtained for the sample.
A summary comparison of the original area and the allocated area for the samples collected during 1994–2003 is given in Table 5 and Table 6 for lengths and ages, respectively. Comparison of original area and allocated area for individual species and stocks are given in Wigley et al. (2007).
Evaluation of Random Component (1,000 realizations)
To evaluate the random component of the allocation, the 1994 Dealer data was run through the allocation procedure 1,000 times, each time using a different seed to generate a series of random numbers that were assigned to each Dealer trip. The 1994 Dealer data were selected because it was expected to have the largest proportion (49% of allocated data) of Dealer landings that matched at Levels B, C, and D where area was assigned on a probabilistic basis.
There are 8 species in the Northeast with multiple stock components: cod, haddock, yellowtail flounder, winter flounder, windowpane, monkfish, red hake, and silver hake (Table 7; Figure 1). Stock area landings were summed for each species and stock from each of the 1,000 runs. A frequency distribution and the 80% confidence interval were calculated (Table 8; Figure 7). The range of stock landings by species varied; haddock had the smallest range (5 mt) of stock landings while silver hake had the largest range (325 mt). The percent spread, calculated as the range / mean, varied between 1.2% (GB COD) and 32.1% (Multinomial Probability
SNE YT). For most species, the 1994 point estimate from the base run was within the 80% confidence interval of the 1,000 realizations, with the exception of silver hake. For both windowpane flounder and winter flounder, the 1994 point estimate was at the boundary of the confidence interval. For most species, the random component did not contribute to wide spread in stock landings, indicating that the random component is not a large source of stock landings variability.
The probability, prob, associated with each allocated trip that matched at ALevel = B, C, D has been stored in the Oracle table CFDETSyyyyAA . This probability can be used to approximate the uncertainty associated with the random component of the allocation. The variance and coefficient of variance of a multinomial distribution (Equations 1 and 2) can be used to approximate the uncertainty if we assume the coefficient of variance of landings to be equivalent to the coefficient of variance of the trip associated with the landings (Equation 3). Recall that the allocation is a trip-based algorithm and all landings associated with a trip are assigned a single area at Level B, C, or D.
The variance (V) and coefficient of variance (CV) of an allocated trip (and associated landings) at Alevel = B, C, or D, using the multinomial distribution, are given below:
(1) (2) (3) (4)
where p is the probability (prob) of the trip (stored in the Oracle table CFDETSyyyyAA),
T is a given allocated trip at Alevel = B, C or D
L are the landings associated with an allocated trip at Alevel = B, C or D.
This approximation method was found to produce confidence intervals similar to those from the 1,000 realizations for 1994. Legault et al. (2008) applied this multinomial approach to 3 yellowtail flounder stocks and 2 haddock stocks for 1995–2006, and confirmed that the landings associated with the random component of the allocation do not contribute a significant source of uncertainty to the stock assessments, even for small stock components (Legault et al. 2008). The use of the multinomial approach will allow analysts to compute confidence intervals about stock landings for all allocated data sets to assess the impact of the random component of the allocation.
- There is a high percent of landings that match at the vessel level (Levels A and B) to obtain area fished, and a small percentage of landings where area fished is estimated from fleet patterns. This is an improvement over the pre-1994 landings data, where approximately less than 10% of the trips were interviewed by port agents.
- There are trade-offs between using a trip-based allocation procedure and other methods; however, the need to link biological samples to individual trips to obtain area for the sample was a necessary element.
- An evaluation of the random component of the allocation indicated that the random component did not contribute to a wide spread in stock landings, indicating that the random component is not a large source of stock landings variability.
- The allocation is predicated upon using clean VTR data. To the extent possible, VTR data was audited for use in the allocation. Continued efforts to expand the routine auditing of VTR as soon as logbooks are submit are encouraged to improve data quality and accuracy.
- In the future, the allocation can be expanded to include vessels that make multiple trips per day, as well as incorporating the unique trip identifier established in 2004 to link the Dealer and VTR databases. It is premature to use this identifier until further data quality and accuracy procedures are established.
- Due to implementation of Dealer Electronic Reporting in May 2004, further evaluation of 2004 to 2007 allocated data may be needed.
- Examination of effort (number of trips, days fished, day absents) in the 1994 allocated data compared favorably with 1993; however, examination of effort over the entire allocated time series is needed.
- It is anticipated that work will continue to fine-tune the allocation algorithm to support current and future data needs. Additionally, the development of routines to periodically update the allocated tables will be needed as Dealer and VTR databases are revised.
We wish to thank R. Mayo for his guidance throughout this project; M. Rossman and P. Nitschke for their analyses of VTR data in support of allocation stratification; M. Palmer for his analysis of the impacts of the assumptions; Data Management Systems staff, including Barbara North, Heidi Marrota and David Hiltz for their technical assistance; the Northeast Regional Office Fishery Statistics Office staff, including Greg Power and Barry Clifford for their assistance; Chris Legault for his assistance with the multinomial probability, and we thank our reviewers for their constructive comments.
Bisack KD. 2003. Estimates of marine mammal bycatch in the Northeast (
New England) multispecies sink gillnet fishery in 1996. Northeast Fish Sci Cent Ref Doc. 03-18; 21 p. Available at: http://nefsc.noaa.gov/nefsc/publications/crd/crd0318/crd0318.pdf
Legault CM, Palmer MC, Wigley SE. 2008. Uncertainty in Landings Allocation Algorithm at Stock Level is Insignificant. Groundfish Assessment Review Meeting (GARM) Working Paper 4.6; GARM Biological Reference Point Meeting; April 28 – 2 May 2008, Woods Hole, MA 02543; 5 p. Available at: http://nefsc.noaa.gov/GARM-Public/3.%20BRP%20Meeting/TOR%204%20BRPs/WP%204.6%20Uncertainty%20in%20Landings.pdf
Northeast Fisheries Science Center (NEFSC). 1996. Report of the 22nd Northeast Regional Stock Assessment Workshop (22nd SAW): Public Review Workshop. Northeast Fish Sci Cent Ref Doc. 96-16; 45 p.
Palmer MC, Wigley SE. 2007. Validating the stock apportionment of commercial fisheries landings using positional data from Vessel Monitoring System (VMS). US Dept Commer, Northeast Fish Sci Cent Ref Doc. 07-22; 35 p. Available at: http://nefsc.noaa.gov/publications/crd/crd0722/crd0722.pdf
Palmer MC. 2008. A method to apportion landings with unknown area, month, and unspecified market categories among landings with similar region and fleet characteristics. Groundfish Assessment Review Meeting (GARM) Working Paper 4.4; GARM Biological Reference Point Meeting; April 28 – 2 May 2008, Woods
Hole, MA 02543; 10 p. Available at: http://nefsc.noaa.gov/GARM-Public/3.%20BRP%20Meeting/TOR%204%20BRPs/WP%204.4%20Landings.pdf
Wigley SE, Legault CM, Brooks E, Cadrin SX, Col L, Hendrickson LC, Mayo R, Nitschke P, Palmer M, Sosebee K, Terceiro M. 2007. Annual comparisons of the trip-based allocated and single-species prorated commercial landings, biological samples and numbers of landed fish at age. Groundfish Assessment Review Meeting (GARM) Working Paper A.2, GARM Data Meeting; October 29 – November 2, 2007, Woods Hole, MA; 67 p.
Wigley SE, Terceiro M, DeLong A, Sosebee K. 1998. Proration of the 1994-96 USA commercial landings of Atlantic cod, haddock and yellowtail flounder to unit stock areas. Northeast Fish Sci Cent Ref Doc. 98-02; 32 p.
 In May 2004, a unique trip identifier was established; due to limited QA/QC procedures, it was not possible to utilize the unique trip identifier in the allocation for 2004 to 2007. It will be possible to incorporate the trip identifier into the allocation when this field is fully audited.
 We thank M. Rossman for the plots depicting the gillnet fishery spatial fishing patterns by port which revealed port-specific fishing areas.
 We thank Paul Nitschke for the tilefish fishery analyses.
 VTR tripid and gearid are computer-generated numbers that uniquely identify each logbook sheet based on permit, and date/time sailed.
 This information can later be used to calculate the multinomial probability to capture the uncertainty associated with statistical area landings determined at Levels B, C, and D.
 dlrtrpid is similar to link in CFDETT/S (pre-1994 data), however, differs when multiple transactions for a given trip occur.
 Inshore lobster pot gear will not be included in the allocation because (a) it is expected that inshore fisherman would not have a federal permit to fish in federal waters, and (b) gear code ‘PTL’ does not distinguish between offshore and inshore pots (200 and 210)
 If the trip is split, days fished must be present on all subtrips of the trip to calculate the ntrip for each subtrip; if ntrip can not be determined, then the trip can not be used, not even for area, because the ntrip on the subtrip is utilized in the area probability density function.
 It is assumed that the VTR species pounds are reported by subtrip accurately. Using proportions guards against some of the reporting difficulties encountered.
 This assumes that the probability of catching this species is based on the amount of time the vessel fished in an area (and not based on other species catch amounts). It assumes that some of the catch may be mis-assigned to an area, however this is less ‘evil’ than picking a single area and wrongly assigning the species catch to one incorrect area.