CONTENTS Abstract Introduction Data & Methods Results Discussion & Conclusions Acknowledgments References List of Acronyms
Northeast Fisheries Science Center Reference Document 07-22
Michael C. Palmer and Susan E. Wigley
Validating the Stock Apportionment of Commercial Fisheries Landings Using Positional Data from Vessel Monitoring Systems (VMS)
NOAA Northeast Fisheries Science Center, 166 Water St., Woods Hole, MA 02543-1026
Web version posted January 14, 2008Citation: Palmer MC, Wigley SE. 2007. Validating the stock apportionment of commercial fisheries landings using positional data from Vessel Monitoring Systems (VMS). US Dep Commer, Northeast Fish Sci Cent Ref Doc. 07-22; 35 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.
Vessel Monitoring System (VMS) positional data from northeast United States fisheries were used to validate the statistical area fished and stock allocation of commercial landings derived from mandatory Vessel Trip Reports (VTR). A gear-specific speed algorithm was applied to 2004–2006 VMS data from the otter trawl, scallop dredge, sink gillnet, and benthic longline fisheries to estimate the location of fishing activity. Estimated fishing locations were used to allocate the landings of 8 federally managed species to stock areas: Atlantic cod (Gadus morhua), haddock (Melanogrammus aeglefinus), yellowtail flounder (Limanda ferruginea), winter flounder (Pseudopleuronectes americanus), windowpane flounder (Scophthalmus aquosus), goosefish (Lophius americanus), silver hake (Merluccius bilinearis), and red hake (Urophycis chuss). Haul location and catch data from the Northeast Fisheries Observer Program (NEFOP) were used to assess the relative accuracy of both VMS and VTR allocation methods.
Overall, the mean VMS-NEFOP agreement rate was 86.4 ± 7.6% compared to a mean VTR-NEFOP agreement rate of 58.5 ± 4.9%. The VMS algorithm had a tendency (approx. 10% of all trips) to overestimate the number of statistical areas fished; when all fishing activity from a given trip occurred in a single statistical area, VTRs more accurately reflected the true fishing location. However, on trips where fishing activity occurred in multiple statistical area, the VMS algorithm showed pronounced gains (77.2 ± 11.2% NEFOP agreement) relative to VTR reports (12.0 ± 5.9% NEFOP agreement). The VMS method achieved distributions of stock landings closer to NEFOP estimates in 18 out of 24 instances (8 species over 3 years). The stock allocations from both the VMS and VTR-based methods were within ± 5% for all stocks, suggesting that the impacts on total stock allocations are relatively minor. However, these small differences represent major relative differences for less abundant stocks such as southern New England/mid-Atlantic yellowtail flounder. In 2005 the VTR-based method allocated 61.9% more yellowtail flounder landings relative to the VMS-based method. The VMS-based method is not a replacement for the VTR-based method; however, it can, and should, be used as a tool to identify those vessels where targeted outreach activities would improve the accuracy of VTR statistical area reporting.
Among federally managed fish species in the northeast United States, eight species are managed and assessed as two or more discrete stocks. The eight species are: Atlantic cod (Gadus morhua), haddock (Melanogrammus aeglefinus), yellowtail flounder (Limanda ferruginea), winter flounder (Pseudopleuronectes americanus), windowpane flounder (Scophthalmus aquosus), goosefish (Lophius americanus), silver hake (Merluccius bilinearis), and red hake (Urophycis chuss). Stock units are composed of statistical area groupings (Figure 1), with stocks defined by divisions that in most cases relate to oceanographic features (e.g., Gulf of Maine, Georges Bank, etc.; Table 1). All of the species are managed under the Northeast Multispecies Fisheries Management Plan (FMP) (NEFMC 1985) with the exception of goosefish, which is managed under the Monkfish FMP (NEFMC 1998).
In the northeast United States, dealer weighout data are assumed to be a census of commercial landings amounts. Commercial landings are allocated to management stocks using the statistical areas fished reported on mandatory vessel trip reports (VTRs) (Wigley et al. 1998). Current VTR regulations (50 CFR §648.7) require submission of paper logbooks upon completion of each fishing trip documenting the total catch by species for each statistical area in which fishing occurs. Despite regulations, it is known that misreporting of statistical area occurs, most frequently in the form of underreporting the number of statistical areas fished when fishing occurs in more than one area (Palmer et al. in press). While underreporting of statistical areas does not necessarily translate to misclassification of commercial landings to stock areas, the potential exists and the entire magnitude of these effects on the allocation of commercial landings is unknown.
The most reliable fisheries-dependent catch and effort data in the region are available from the Northeast Fisheries Observer Program (NEFOP). However, because these data are limited in their coverage (e.g., <5% of all certain fisheries in a given year; Wigley et al. 2007) they cannot provide the synoptic coverage necessary to allocate commercial landings to stock area with any regularity. Vessel monitoring systems (VMS) in the northeast were first implemented for the limited-access scallop fisheries in 1998 (NEFMC 1993); their use has increased over time (Figure 2) and expanded to cover many fisheries (Table 2). Historically, larger offshore vessels participating in the limited-access scallop and special-access groundfish fisheries were more likely to be equipped with VMS compared to the smaller nearshore vessels. With the passage of Framework 17 to the Atlantic sea scallop FMP (NEFMC 2005) and Framework 42 to the Northeast Multispecies FMP (NEFMC 2006), VMS is now required for a greater proportion of the smaller near-shore scallop and groundfish fleets. While VMS does not provide census coverage of these fleets, it does provide census coverage of trips taken by those vessels equipped with VMS. Given the increasing use of VMS in the region, this represents a potential tool to conduct large-scale validation of the statistical areas reported on VTRs.
Vessel positions obtained from VMS have been used as a proxy for location of fishing effort in prior work (Deng et al. 2005, Murawski et al. 2005, Mills et al. 2007). Many VMS programs do not require the transmission of instantaneous vessel speeds; only a vessel position and a date and time stamp are required. This has recently changed in some fisheries (Mills et al. 2007); however, most users of VMS data must infer vessel speed and course from averages calculated from successive reported positions. Northeast United States VMS regulations only require the transmission of date, time, and position information. In the northeast United States VMS data are typically collected once per 30 min from vessels participating in the limited access scallop fishery and once per 60 min from vessels participating in the groundfish fishery (Table 2).
Past work has characterized all activity falling within specific ranges of average vessels speeds to be indicative of fishing activity (Deng et al. 2005, Murawski et al. 2005). The vessel speed method can achieve accuracy levels as great as 99%; however, it can also result in the incorrect classification of non-trawling activity (Mills et al. 2007) leading to an overestimation of fishing intensity. A more complex method utilizing both vessel speed and directionality has been attempted; however, this method did not improve the detection of fishing activity and reduced the inclusion of false positives only slightly (0.7%; Mills et al. 2007). When using the vessel-speed method, the amount of classification error is sensitive to the VMS polling rate (Palmer 2008), the speed ranges used to define fishing activity, and the practices of the fishery under observation (e.g., amount of overlap between the vessel-speed signals of fishing and nonfishing activity, length of individual hauls, etc.). With the exception of Mills et al. (2007), much of the work so far published in the fisheries literature has utilized VMS data without a quantitative assessment of the classification error of fishing vs. nonfishing activity when the vessel-speed method is used. This paper assesses the ability of the VMS vessel-speed method to detect the statistical area fished and allocate fishery landings to stock area by comparing results to matching NEFOP trips. The method is then applied to assess VTR area reporting compliance and its impacts on the current VTR-based allocation method used in the northeast United States.
Data and Methods
VTR logbook trip, gear, and species catch data were extracted from the VTR database (VESLOG tables) for calendar years 2004–2006; prior to 2004, <500 vessels were equipped with VMS units, thus limiting the scope of a VMS-based allocation (Figure 2). The analytical datasets were post-processed to remove any overlapping trips (i.e., trips taken by the same vessel with a date of sailing occurring before the date of landing of a previous trip). Overlaps are due to VTR reporting and/or data entry errors. This process resulted in the removal of 1.2%, 1.7%, and 1.9% of the total reported VTR trips in 2004, 2005, and 2006 respectively. Of the remaining trips, only those trips where at least one of the eight study species were reported as retained catch were kept in the dataset (Atlantic cod, haddock, yellowtail flounder, winter flounder, windowpane flounder, monkfish, silver hake, and red hake). Because the focus was on assessing the impact of statistical area misreporting on the proration of commercial landings, discards were not included in these analyses. All species weights were converted to live weight in kilograms (kg) using standard NEFSC conversion factors. The VTR dataset was further restricted to include only the four major gear types which catch these demersal species in the northeast United States: fish bottom otter trawl (OTF), scallop dredge (DRS), sink gillnet (GNS) and benthic longline (LLB). The VTR database field CAREA (calculated area) was used as the basis for allocating VTR reported retained catch. On each logbook sheet, vessel operators must report both the average fishing location (latitude x longitude or loran bearings) and the statistical area fished (Figure 1). If the statistical area corresponding to the point location is not in agreement, or not adjacent to the reported statistical area, the reported statistical area is used to populate CAREA, otherwise CAREA is populated using the statistical area corresponding to the fishing location. VTR species landings were then assigned to a stock area based on the statistical area fished (Table 1). The final VTR subsets used in this analysis contained approximately 32,000–33,000 trips in a given year (Table 3).
All available VMS data were extracted from the VMS database for each vessel and assigned to the appropriate VTR reported trips by matching on vessel and assigning all VMS point locations with dates between the date of sailing and date landed reported on the VTR to the respective trip. The average vessel speed was calculated by dividing the haversine distance (Sinnott 1984) by the time difference between consecutive fixes. All positions were assigned to a National Marine Fisheries Service (NMFS) statistical area (Figure 1). Summaries of the number of matched trips by year are included in Table 3.
All NEFOP trips which could be matched to the list of VMS-VTR matched trips were extracted from the Observer Data Base System (OBDBS) database. Matches were established on the vessel, date of sailing, and date of landing as reported on the VTR; trips with multiple matches were removed from the analyses. For all matched trips the associated haul duration, statistical area fished, species and retained catch weights were also extracted; retained catch weights were converted to live weight in kilograms (kg) using standard NEFSC conversion factors. Summaries of the number of matches by year are included in Table 3.
Method development and application
Some analyses using northeast US VMS data have differentiated fishing activity from nonfishing activity by using only upper-speed bounds: <3.5 knots for bottom trawl vessels (Murawski et al. 2005) and <5.0 knots for scallop dredge vessels (Rago and McSherry 2002). To our knowledge no attempt has been made to identify fishing activity from the VMS signals of fixed-gear vessels (i.e., sink gillnet, benthic longline). We attempted to improve vessel speed classifications and extend the application to fixed-gear vessels through a combination of visual examination of the percent frequency distributions of VMS-derived average speeds, knowledge of fishing operations, and observations from high-frequency polled Global Positioning System (GPS) data.
Percent frequency distributions of VMS average vessel speed were plotted for all gear types (Figure 3). These distributions were then compared to percent frequency distributions of activity-specific (fishing vs. nonfishing) instantaneous vessel speeds from high-frequency polled GPS data (1 fix/10 seconds) collected from vessels involved in NMFS cooperative research projects (Figure 4). These data sets included precise observations of the dates and times of fishing activity. Four trips taken by four separate vessels were analyzed; two groundfish bottom trawl trips and two scallop dredge trips. Individual vessel speed observations from all trips were combined by gear type, and activity was classified activity as either ‘fishing’ or ‘other’. ‘Fishing’ was defined as the period from winch brake lock to winch brake release (presumably, the period during which the gear is actually in contact with the bottom). Unfortunately, these data were not available for fixed-gear vessels. It is assumed that fixed gears such as sink gillnet and benthic longline gear are likely to be fished in very specific and limited geographic areas on a given trip; thus it is unlikely fishing is occurring on multiple fish stocks on a single trip. If this assumption is true, these analyses will not be as sensitive to misclassification of fixed gear activity compared to mobile gear activity.
VMS-based bottom otter trawl activity exhibits a very pronounced bimodal distribution of vessel speeds. It was assumed that the first mode (2.8 knots) represented fishing activity and the second mode (8.0 knots) was indicative of steaming activity. Fishing activity falls within a very narrow range from approximately 2.0–5.0 knots as evidenced by the distributions observed from the high-frequency GPS data. A fishing speed window of 2.0 knots < fishing activity < 4.0 knots was used. This window fits the high-frequency polled GPS well, correctly classifying 99.2% of fishing activity. However, it also incorrectly categorizes 31.8% of nonfishing activity as fishing activity (Figure 4). It is expected that a portion of the nonfishing activity falling inside the window of fishing speed represents activity associated with the hauling and setting of the gear, which suggests that the impact of false-positives may not be as great as the 31.8% figure implies.
The VMS-based average vessel speed distribution of scallop dredge activity has a nearly trimodal distribution (Figure 3). Unlike bottom otter trawl speed distributions, scallop dredge has a high percentage of activity close to 0.0 knots. This may be indicative of shucking activity when vessels drift, allowing the crew to shuck scallops and clear the deck. The primary mode (4.2 knots) was assumed to represent fishing activity and the 8.2-knot mode was assumed to represent steaming activity. Scallop dredge fishing activity occurs over a broader range than trawl activity, falling between approximately 2–7 knots as evidenced by the distributions observed from the high-frequency GPS data (Figure 4). A fishing speed window of 2.5 knots < fishing activity < 6.0 knots was used. This window fit the high-frequency polled GPS well, correctly classifying 98.3% of fishing activity; however, it incorrectly categorized 69.3% of nonfishing activity.
Like scallop dredge activity, VMS-observed sink gillnet average speed distributions have a trimodal distribution (Figure 3). Based on knowledge of gillnet operations, the first mode (0.6 knots) was interpreted as representing the hauling of gillnet gear, the second mode (3.0 knots) as re-setting the nets, and the third mode (8.2 knots) as steaming activity. Benthic longline average speed distributions have a bimodal distribution (Figure 3). The first mode (0.8 knots) was interpreted as representing the hauling and setting of the longline gear and the second mode (10.0 knots) as steaming to and from the fishing grounds. For both sink gillnet and benthic longline gear, speed bounds of 0.1 < fishing activity < 1.3 were used.
Those VMS locations identified as representative of fishing activity were then used to determine the statistical areas in which fishing occurred. Statistical areas fished were compared across data sources to assess whether the statistical areas derived from VMS-defined fishing activity represented an improvement over VTR-reported statistical areas relative to NEFOP data. Trips were broken into two categories: single subtrip trips (fishing occurs in only one statistical area per trip) and multi-subtrip trips (fishing occurs in more than one statistical area per trip). Because all stock boundaries are divided along statistical area boundaries, correct reporting of multi-subtrip trips are of the greatest concern. These trips have the potential to fish on multiple stocks of fish in a single trip, and misreporting of statistical area(s) may lead to incorrect estimates of stock removals. For each trip, the levels of agreement between the NEFOP, VMS, and VTR statistical areas were categorized as in agreement (‘complete’), not in agreement (‘none’) or in partial agreement (‘partial;’ at least one statistical area was in agreement, but not all). Agreement levels were contingent on agreement between the number of statistical areas reported and the identity of those statistical areas. For example, if a VTR reports that fishing occurred in statistical areas 515 and 521 and VMS positions indicate that fishing occurred in 515 and 521, then the trip would be considered to be in agreement (‘complete’). If the VTR reported fishing in 515 and the VMS data suggests fishing occurred in 515 and 521, then the trip would be considered to be in partial agreement (‘partial’). If the VTR reported fishing in 515 and the VMS data suggested fishing occurred only in 521, then the trip would not be considered to be in agreement (‘none’). The same analysis was also performed on the larger set of VMS and VTR matched trips.
A VMS-based allocation algorithm was devised using the statistical areas fished from the VMS data to reallocate VTR-reported landings to stock area. Fishing activity was assigned to stock area based on the species landed and statistical area in which the fishing activity was occurring. The time spent fishing in each stock area was estimated as the sum of fishing activity blocks occurring in each stock area. (The duration of one activity block is contingent on the VMS polling frequency which is variable, but generally once per 30 minutes for scallop vessels and once per hour for groundfish vessels.) Total VTR trip landings for each species (s) were allocated to stock area (k) based on the ratio of time spent fishing in each stock area as determined from VMS locations (Equation 1).
= VMS prorated trip landings for species s, stock k (kg)
ls = trip landings for species s in stock area, k, as derived from VTR reports (kg)
li = trip landings for species s in stock areas i, where i ≠ k, as derived from VTR reports (kg)
tk = time spent fishing in stock area, k, as derived from VMS positional data (days)
ti = time spent fishing in stock area i, where i ≠ k, as derived form VMS positional data (days)
The results of the VMS-based allocation were compared to landings allocation derived from both NEFOP and VTR data sources to assess the relative accuracy of the VTR-based allocation and determine if the VMS-based algorithm resulted in improved estimates of landings by stock area. VTR and NEFOP species landings were prorated by assigning landings to stock area based on the reported statistical area. All comparisons were performed through examination of percent allocation to stock area as opposed to absolute landings, because percent allocations derived from the traditional VTR source are used to allocate the amounts of commercial landings as determined through dealer weighout data (Wigley et al. 1998). The same analysis was performed on the larger VMS-VTR matched data set.
The VMS-based allocation method assumes a constant species catch-per-unit-effort (CPUE) at all fishing locations (i.e., species catch is distributed only as a function of the time spent fishing in each stock area). This assumption neglects species habitat preferences (e.g., sediment composition, water depth and temperature, etc.) which would result in species being more likely to be caught in some locales and not others. To assess the degree to which this assumption was violated, individual species trip allocations from the VMS method were compared to the same allocations as determined from NEFOP observations using linear regression.
Method validation using NEFOP data
Statistical area agreement between NEFOP and VTR was >94% for single-subtrip trips across all years, but <17% for multi-subtrip trips (Table 4). Nearly all disagreements among the ‘partial’ multi-subtrip trips matches (>98%) are due to underreporting of statistical areas (fewer statistical areas reported on the VTR compared to NEFOP: 105 trips in 2004, 337 in 2005, and 166 in 2006). There was a general trend towards improved VTR reporting of multi-subtrip trips over time; however, given the small sample size and potential for observer-type effects on VTR reporting, such a conclusion may be premature. The statistical area agreement between NEFOP and VMS-based statistical areas was lower (≥88.0%) for single-subtrip trips compared to the NEFOP-VTR comparisons (Table 5). The cause of disagreement among single-subtrip trips is the VMS-based method's overestimation of statistical areas fished. This overestimation results from the VMS-based method misclassifying nonfishing activity as fishing activity. Agreement among multi-subtrip trips is greater (>67%) when using the VMS method compared to the VTR-reported statistical area trips, with no complete disagreement among any of the trips. Among statistical areas in partial agreement there was a tendency for the VMS method to overestimate the number of statistical areas fished (59.5% of partial matches in 2004, 53.3% in 2005, and 50.8% in 2006). The performance of the VMS-based method in detecting statistical areas fished is not equivalent for all gear types; a closer examination of the VMS-NEFOP statistical area comparison in 2005 showed that 80.3% (535 of 666) of trawl trips, 65.4% (17 of 26) of dredge trips, 83.8% (88 of 105) of gillnet trips, and 97.1% (101 of 104) of longline trips have agreement levels of ‘complete.’ This finding supports the assumption that the misclassification of the location of fixed gear fishing activity is less likely compared to mobile gear activity.
The VMS-based allocation method arrived at annual stock allocations closer to NEFOP allocations relative to VTR allocations for 18 of the 24 comparisons examined (eight species over three years; Tables 6–8). There were no species allocations for which the VMS-based allocation underperformed the VTR allocation in all three years; haddock was the only species for which the VMS-based allocation underperformed in 2 of the 3 years. There was general improvement in the VMS-based allocation over time, with the number of species for which it underperformed the VTR allocation decreasing from 3 in 2004 to only one in 2006. Of all species, goosefish, silver hake, and red hake had the greatest percent difference relative to the NEFOP allocation in all 3 years, with the single exception of windowpane flounder in 2004. It is important to consider the implications of the matched trip set composition in the interpretation of these results, since the performance of the VMS-based method is contingent on the number of multi-subtrip trips and the gear composition of the matched data set. For example, a higher proportion of multi-subtrip trips in the examined dataset would appear to improve the performance of the method, and a higher proportion of dredge trips in the matched set would appear to decrease performance. Comparisons of the individual trip stock allocations between the VMS-based method and NEFOP allocation showed strong agreement between VMS and NEFOP stock allocations (r=0.823, p <0.001, n=514; Figure 5); however, there was considerable spread in residuals.
There are large differences in the NEFOP landings compared to VTR landings shown in Tables 6–8 for some species, most notably monkfish (e.g., in 2004 NEFOP estimated 380 mt compared to the VTR estimate of 71 mt). The exact reasons for these discrepancies are unknown; however, there is a tendency for self-reported hail weights to be biased low (Palmer et al. in press). Additionally, monkfish tails constitute a large proportion of monkfish landings and these are often incorrectly reported on VTRs as whole monkfish (Palmer et al. in press). A Commercial Fisheries Database System (CFDBS) conversion factor of 3.32 is applied to monkfish tail landings to convert these to whole weights. Incorrect reporting of monkfish tails as whole monkfish will result in the underestimation of VTR monkfish landings by approximately a factor of 3.
Extrapolation to larger VMS-VTR matched dataset
The NEFOP-VMS-VTR subset of data used to validate the VMS-based method is relatively small compared to the total population of VTR-recorded trips (Table 3). The validation results suggest that for some trips monitored through VMS, the VMS-based allocation method can be used to gauge the accuracy of the stock allocations as determined through VTR reports. The VMS-VTR matched set is a much larger dataset. The subset of VTR reports examined (eight species caught using the four gear types) account for only approximately a quarter of the total VTR reports in a given year (Table 3); however, this dataset accounts for >96% of the landings of all the study species across the time series (Table 9). Similarly, VMS coverage is available for only 5,892 to 19,165 of the VTR trips in a given year (Table 3), but these trips account for 17.6 to 92.0% of the total landings of individual species (Table 9). By 2006, VMS data were available for trips responsible for landing >70% of all species but goosefish; coverage of goosefish landings is low because there are no specific VMS requirements for the goosefish fishery (Table 2). All demersal species examined are primarily caught by the otter trawl fishery except goosefish, for which gillnet gear is responsible for the majority of the landings. Gillnet is the secondary gear type for all species with the exception of haddock and silver hake, which are secondarily targeted by benthic longline (Tables 10–12). VMS coverage of the landings by most gear types is highly variable, though generally increasing with time; there is a general pattern of low gillnet coverage for landings of most species across time.
Examination of the VTR statistical area reporting using VMS-based statistical areas fished showed similar patterns to those observed in the NEFOP-VMS-VTR comparisons. Agreement levels of single-subtrip trips exceeded 92% in all years and was always <6.5% for multi-subtrip trips (Table 13). This level of agreement is less than that observed in the NEFOP-VTR comparison. It is unclear whether these lower rates of agreement are due to the overestimation of the number of statistical areas fished by the VMS method, an observer effect, or some other factor. Closer examination of the partial matches revealed that the number of vessels apparently under-reporting the number of statistical areas fished was 397 in 2004, 477 in 2005, and 629 in 2006. Those vessels that likely frequently under-report trips (>5 trips in a year) are responsible for the majority of the potentially underreported trips. In 2004 there were 179 vessels that appeared to frequently under-report. These vessels accounted for 1,876 of 2,797 of partial agreement trips (67.1%). In 2005, there were 221 vessels in this category; they accounted for 2,787 of the 3,837 partial agreement trips (72.6%) and in 2006 there were 268 vessels which potentially submitted >5 underreported trips, accounting for 3,815 of the 5,251 partial agreement trips (72.7%).
Because the performance of the VMS algorithm is sensitive to the number of multi-stock trips taken in a given year, it is important to understand the types of trips recorded in the VMS dataset and how that composition varies over time. The percentage of multi-stock trips recorded by VMS increased in 2005, followed by a decline in 2006 to levels below 2004 values for all but windowpane, silver hake, and red hake trips (Table 14). Those trips fishing on multiple stocks are predominantly (≥ 99.0%) mobile-gear vessels (Table 15), implying that fixed-gear fishing effort occurs primarily in localized geographic areas; therefore, landings from fixed-gear trips are unlikely to have come from multiple stocks. This supports the prior assumption that the misinterpretation of the VMS speed signals from fixed-gear trips is unlikely to result in the misallocation of landings.
The perceived underreporting of statistical areas in the VTR data led to minor (<5%) differences in the overall stock allocations; only two stocks in the three year time-series exhibited differences in stock allocations exceeding 2.0% (2004 silver hake, ±3.0%; and 2006 windowpane flounder, ±4.7%; Tables 16–18). These figures are similar to the total proportion of species landings potentially misallocated, which was <5% for all species-years examined, again with the exception of 2004 silver hake and 2006 windowpane flounder. However, these small differences in percent allocation have a disproportionate effect on the less abundant stock such as such as Gulf of Maine haddock, southern New England yellowtail, southern windowpane, and northern silver hake. For these stocks, minor differences can be large (≥5.0%) relative to the percent of the total species landings allocated to that stock (Tables 16–18). These impacts are most notable in the stock allocations of the southern New England/mid-Atlantic yellowtail flounder. Stock allocation differences between the VTR and VMS methods were ≤1.6% for all years; however, commercial landings of this stock were ≤6.4% of the total stock landings as estimated from the VTR reports, resulting in relative differences of 53.8, 61.9, and 25.0% for the years 2004, 2005, and 2006, respectively. Of the 54 comparisons analyzed (8 species, 18 stocks, 3 years), the VMS-based method stock allocations had ≥5.0% relative difference compared to the VTR-based allocations for 17 of the comparisons. Only southern New England/mid-Atlantic yellowtail, southern windowpane, and northern silver hake exceeded the ≥5.0% difference in all three years examined.
There was a tendency for the VTR method to over-allocate the predominant Atlantic cod and haddock stocks (i.e., Georges Bank), with the exception of 2004 haddock. For yellowtail and winter flounder there was a tendency for the VTR-method to under allocate the predominant Georges Bank stock and over-allocate the Gulf of Maine and southern New England stocks. The only exception to this was 2005 winter flounder, for which there was a perceived under-allocation of VMS-based landings estimate of the southern New England stock. For all years, there was an over-allocation of landings to the southern goosefish stock using the VTR-method relative to the VMS method. The direction of stock allocation differences for windowpane flounder, silver hake, and red hake was variable from year to year.
Discussion and Conclusions
The underreporting of statistical areas on VTR logbooks is a significant problem affecting >80% multi-subtrip trips. The VTR underreporting rates from this study agree closely with past studies that have used both NEFOP and haul-by-haul self-reported data (Palmer et al. in press). While the impacts of this underreporting are relatively small in regard to overall stock allocation percentages, the relative impacts on less abundant stocks such as southern New England/mid-Atlantic yellowtail can be significant. This is in agreement with the findings of other studies that have examined this issue using smaller data sets which utilized NEFOP-VTR comparisons. These discrepancies have implications on the estimation of fishery removals and the assessment of these stocks. While the impacts are minimal for the majority of stocks examined, the extent of the impacts on those few stocks that are significantly affected suggests a problem that deserves attention.
Many of the stock assessments of these eight species use finer stratification of commercial landings (e.g., quarter, market category, and gear groups) to construct the age-length keys used in virtual population analysis (VPA) or similar assessment models (Mayo and Terceiro 2005). This paper does not consider the impacts of statistical area reporting patterns on these finer scale stratifications of commercial landings; however, the accuracy of finer-scale allocations would be sensitive to the number of multi-subtrip trips included in each strata. It is possible that the effects of statistical area misreporting on stock allocations are reduced due to offsetting errors (i.e., a trip that misallocates 1100 kg to the Georges Bank cod stock could be largely offset by a trip that misallocates 1200 kg to the Gulf of Maine cod stock). However, the spatial accuracy of VTR reports is critical not only for the assessment of fish species, but also of protected species such as sea turtles (e.g., Murray 2004, 2005, 2006; Orphanides and Bisack 2006) and marine mammals (Belden et al. 2006). When these data are used at finer spatial scales the accuracy of VTR reports becomes increasingly important.
It is important to consider that the results of this study apply only to the trips monitored by VMS; however, by 2006 trips responsible for >70% of multispecies landings were monitored by VMS (Table 9). VMS coverage of some fisheries such as the Northeast multispecies is nearing complete coverage, with all vessels required to have a VMS unit installed when fishing under the days-at-sea program (NEFMC 2006). The increased coverage improves the utility of VMS data as a validation tool for managers and data set of spatial fishing patterns for analysts. The number of vessels responsible for the landings of the eight species examined has remained constant at slightly less than 1200 (Table 3); however, the number of these vessels monitored by VMS has increased from 38.5% (453 of 1176) to 76.7% (886 of 1155). The increase in VMS usage appears to have occurred primarily among the smaller nearshore fleet in response to VMS requirements to participate in the general category scallop fishery (NEFMC 2005) and the Northeast multispecies fishery (NEFMC 2006) as indicated by the drop in percentage of multi-stock area trips recorded by VMS from 2004–2006 (Table 11). There was a decrease in the number of multiple stock area trips from 2005–2006 which may explain the greater degree of agreement between the VMS and VTR proration in 2006 for Gulf of Maine cod, haddock, and winter flounder.
The study results are sensitive to the use of average VMS vessel speeds to differentiate fishing activity from nonfishing activity and to the validity of the VMS-based allocation. This study defines fishing activity using narrower speed ranges than have been used in past studies, which should lead to more conservative estimates of fishing effort. The speed range used for the mobile gears agree closely with the speeds obtained from high-frequency polling of vessels GPS units suggesting that these ranges are reasonable. However, instantaneous vessel speeds are not collected by NMFS Northeast Region VMS Program, so this study relied on average vessel speeds. The averaging process blurs activity from observation to observation and results in speeds slower than actual speeds due to a corner-cutting effect (Deng et al. 2005, Palmer 2008). These impacts were not considered in this study and represent an area of uncertainty. The speed ranges adequately classify fishing activity (>98% success for mobile gear), but tend to overestimate the amount of fishing by incorrectly classifying nonfishing effort as fishing (69.3% misclassification of nonfishing scallop activity). The overestimation was apparent in the comparisons of statistical areas fished between VMS and NEFOP data (Table 5). VMS data indicate where it is likely that fishing effort is occurring, but provide no information on catch composition. A critical assumption of the VMS-based allocation is that the proportion of species caught across multiple stock areas on a fishing trip is only a function of the time spent fishing in each stock area. While the relationship between VMS and NEFOP allocations was significant, there was a considerable amount of variability (Figure 5). This assumption is not independent of overestimation errors; disproportionate overestimation of time spent fishing in a particular stock area will have a direct effect on the VMS-based allocation.
The various uncertainties and shortcomings of the VMS allocation method point out that this is not a replacement for a VTR-based allocation. Furthermore, the low vessel coverage of historical VMS data (Figure 2) limits its use as a tool to correct historical misreporting. However, the results do show that VMS data can be used as a tool to monitor the accuracy and completeness of VTRs and guide efforts to improve VTR compliance. The number of vessels which are potentially underreporting statistical areas on a frequent basis is small (<250 vessels) relative to the total number of vessels submitting VTRs (>2,400; Table 3). Improvements are needed in the compliance of VTR reporting regulations, particularly among those vessels likely to be fishing multiple stocks. Given the manageable size of the problem and availability of tools to monitor these data, the quality of self-reported data should be monitored and improved through targeted outreach and education activities.
We thank those vessel captains that allowed us to capture high-frequency GPS polling observations of their fishing operations. Thanks also to Douglas Christel, Lou Goodreau, and Deirdre Boelke for their assistance with assembling the list of management measures affecting VMS use. The quality and scope of this paper benefited greatly from discussions with Thomas Nies, Andrew Applegate, and Christopher Legault.
Belden DL, Orphanides CD, Rossman MC, Palka DL. 2006. Estimates of cetacean and seal bycatch in the 2004 Northeast sink gillnet and mid-Atlantic coastal gillnet fisheries. US Dep Commer, Northeast Fish Sci Cent Ref Doc. 06-13; 24 p.
Deng R, Dichmont C, Milton D, Haywood M, Vance D, Hall N, Die D. 2005. Can vessel monitoring system data also be used to study trawling intensity and population depletion? The example of Australia’s northern prawn fishery. Can J Aquat Fish Sci. 62:611-622.
Mayo RK, Terceiro M, editors. 2005. Assessment of 19 Northeast groundfish stocks through 2004. 2005 Groundfish Assessment Review Meeting (2005 GARM), Northeast Fisheries Science Center, Woods Hole, Massachusetts, 15-19 August 2005. US Dep Commer, Northeast Fish Sci Cent Ref Doc. 05-13; 499 p.
Mills CM, Townsend SE, Jennings S, Eastwood PD, Houghton CA. 2007. Estimating high resolution trawl fishing effort from satellite-based vessel monitoring system data. ICES J Mar Sci 64:248-255.
Murawski SA, Wigley SE, Fogarty MJ, Rago PJ, Mountain DG. 2005. Effort distribution and catch patterns adjacent to temperate MPAs. ICES J Mar Sci. 62:1150-1167.
Murray KT. 2004. Bycatch of sea turtles in the Mid-Atlantic sea scallop (Placopecten magellanicus) dredge fishery during 2003. 2nd edition. US Dep Commer, Northeast Fish Sci Cent Ref Doc. 04-11; 25 p.
Murray KT. 2005. Total bycatch estimate of loggerhead turtles (Caretta caretta) in the 2004 Atlantic sea scallop (Placopecten magellanicus) dredge fishery. US Dep Commer, Northeast Fish Sci Cent Ref Doc. 05-12; 22 p.
Murray KT. 2006. Estimated average annual bycatch of loggerhead sea turtles (Caretta caretta) in U.S. Mid-Atlantic bottom otter trawl gear, 1996-2004. US Dep Commer, Northeast Fish Sci Cent Ref Doc. 06-19; 26 p.
New England Fisheries Management Council (NEFMC). 1985. Fishery Management Plan environmental impact statement regulatory impact review and initial regulatory flexibility analysis for the Northeast Multispecies Fishery. Newburyport, MA: NEFMC.
NEFMC. 1993. Amendment 4 and supplemental environmental impact statement to the Atlantic Sea Scallop Fishery Management Plan: Volume I. Newburyport, MA: NEFMC.
NEFMC. 1998. Monkfish Fishery Management Plan: Supplement 1. Newburyport, MA: NEFMC.
NEFMC. 1999. Atlantic Herring Fishery Management Plan incorporating the environmental impact statement and regulatory impact review (including the regulatory flexibility analysis): Volume I. Newburyport, MA: NEFMC.
NEFMC. 2001. Supplemental environmental impact statement and preliminary regulatory economic evaluation for Framework Adjustment 14 to the Atlantic Sea Scallop Fishery Management Plan. Newburyport, MA: NEFMC.
NEFMC. 2003. Amendment 13 to the Northeast Multispecies Fishery Management Plan including a final supplemental environmental impact statement and a initial regulatory flexibility analysis. Newburyport, MA: NEFMC.
NEFMC. 2004a. Framework Adjustment 16 to the Atlantic Sea Scallop Fishery Management Plan and Framework Adjustment 39 to the Northeast Multispecies Fishery Management Plan with an environmental assessment, regulatory impact review, and regulatory flexibility analysis. Newburyport, MA: NEFMC.
NEFMC. 2004b. Framework Adjustment 40A to the Multispecies Fishery Management Plan including an environmental assessment, regulatory impact review, and regulatory flexibility analysis. Newburyport, MA: NEFMC.
NEFMC. 2005. Framework Adjustment 17 to the Atlantic Sea Scallop Fishery Management Plan and environmental assessment, regulatory impact review, and regulatory flexibility analysis. Newburyport, MA: NEFMC.
NEFMC. 2006. Framework Adjustment 42 to the Multispecies Fishery Management Plan and Framework Adjustment 3 to the Monkfish Fishery Management Plan including an environmental assessment, regulatory impact review and initial regulatory flexibility analysis. Newburyport, MA: NEFMC.
Orphanides CD, Bisack KD. 2006. Analysis of Virginia fisheries effort as a component in the development of a fisheries sampling plan to investigate the causes of sea turtle strandings. US Dep Commer, Northeast Fish Sci Cent Ref Doc. 06-24; 49 p.
Palmer MC, Wigley SE, Hoey JJ, Palmer J. In press. An evaluation of the Northeast Region’s Study Fleet Pilot Program and electronic logbook (ELB) system: Phases I and II. NOAA Tech Memo NMFS NE; in press.
Palmer, MC. 2008. Calculation of distance traveled by fishing vessels using GPS positional data: A theoretical evaluation of the sources of error. Fish Res. 89:57-64.
Rago P, McSherry M. 2002. Spatial distribution of fishing effort for sea scallops: 1998-2000. In: Northeast Region Essential Fish Habitat Steering Committee, Editors. Workshop on the Effects of Fishing Gear on Marine Habitats off the Northeastern United States, October 23-25, 2001, Boston, Massachusetts. Northeast Fish Sci Cent Ref Doc. 02-01; p. 80-86.
Sinnott RW. 1984. Virtues of the Haversine. Sky and Telescope. 68:159.
Wigley S, Terceiro M, DeLong A., Sosebee, K. 1998. Proration of 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.
Wigley SE, Rago PJ, Sosebee KA, Palka DL. 2007. The analytic component to the Standardized Bycatch Reporting Methodology Omnibus Amendment: sampling designend estimation of precision and accuracy (2nd edition). US Dep Commer, Northeast Fish Sci Cent Ref Doc. 07-09; 156 p.
 A. Applegate and T. Nies, NEFMC,
August 17, 2007, pers. comm.
 A. Applegate and T. Nies, NEFMC,
August 17, 2007, pers. comm.
List of Acronyms
CAREA = calculated area
CFDBS = Commercial Fisheries Database System
CPUE = catch per unit effort
DAS = days at sea
DRS = scallop dredge
FMP = fishery management plan
GNS = sink gillnet
GPS = Global Positioning System
LLB = benthic longline
MAFMC = Mid-Atlantic Fisheries Management Council
NEFMC = Northeast Fisheries Management Council
NEFOP = Northeast Fisheries Observer Program
NMFS = National Marine Fisheries Service
OBDBS = Observer Data Base System
OTF = fish bottom otter trawl
VMS = vessel monitoring system
VTR = vessel trip report