8 Discards

Information on discarded weight from commercial fishing trips in the Greater Atlantic comes from 3 main sources (in decreasing order of availability):

1. Vessel trip reports (VTRs)
2. Human at-sea observers
3. Electronic monitoring (EM) systems

Despite reporting requirements that should make discarded weight available for any subtrip with a VTR, concerns with accuracy prevent these data from contributing to discard estimation in most circumstances. Additionally, some fisheries do not require VTR submission.

The majority of discarded weight (referred to hereafter as discards), for most species and subtrips, is estimated following the Standardized Bycatch Reporting Methodology (SBRM) originally described by Wigley et al. (2007) and modified by Nitschke et al. (2010) to support quota monitoring. The process involves using a ratio estimator (Cochran 1977) for trips carrying a human at-sea observer (i.e., “observed” trips) to estimate the rate (or ratio) of discarded weight for a given species per total kept weight of all species (i.e., a proxy for effort) for each subtrip of a trip. Unobserved trips are assumed to discard a given species at similar rates to observed trips, depending on a collection of stratification variables describing subtrip attributes (e.g., gear type, mesh, and statistical area fished) and potentially other relevant attributes such as regulation and fishing year. Discard estimation within a fishing year includes additional methods to accommodate sparse data in the early part of a fishing year for species with quota management needs (GARFO 2016).

Here, we describe the approaches used in CAMS to estimate discards, and document the  business rules for determining how an “official” estimate of discards is determined for a given species and subtrip. The latter is particularly important for subtrips that have multiple sources of information on discards and are subject to management rules in one or more fisheries.

8.1 Species of interest defined

A list of species including those that are currently quota monitored, and others identified in the National Bycatch Report, is needed to define the range of discard estimates required for the Greater Atlantic region. The species of interest are restricted to fish/invertebrates, as other taxa (marine mammals, sea turtles, sea birds) involve a variety of analyses for estimating bycatch that have been determined to be outside the scope of CAMS. CAMS provides a structure such that additional species could be included in the future. 

There are currently 52 managed species included in CAMS discard estimation 8.1. CAMS discard processes will also be applied to a greater set of species associated with SBRM analyses and to support the National Bycatch Report (NBR).

Table 8.1: Fishing years (RUN_ID) for discard species in CAMS. The fishing year is a determination of how the data must be structuted for a discard run.

ITIS_TSN	NESPP3	ITIS_NAME	RUN_ID
160845	365	SKATE,UNCLASSIFIED	MAY
564037	373	SKATE,LITTLE/WINTER MIX	MAY
079718	800	SCALLOP,SEA	APRIL
080944	769	CLAM,SURF	CALENDAR
081343	754	CLAM,OCEAN QUAHOG	CALENDAR
082372	801	SQUID,LONG FINNED (LOLIGO)	CALENDAR
082521	802	SQUID,SHORT FINNED (ILLEX)	CALENDAR
097314	727	LOBSTER,AMERICAN	CALENDAR
098678	711	CRAB,JONAH	CALENDAR
160230	351	DOGFISH,SMOOTH	CALENDAR
160617	352	DOGFISH,SPINY	MAY
160855	372	SKATE,CLEARNOSE	MAY
161706	001	ALEWIFE	CALENDAR
161722	168	HERRING,ATLANTIC,SEA	CALENDAR
164499	012	GOOSEFISH	MAY
164712	081	COD,ATLANTIC	GROUNDFISH
164727	269	POLLOCK,ATLANTIC	GROUNDFISH
164730	152	HAKE,ATLANTIC,RED	MAY
164732	153	HAKE,ATLANTIC,WHITE	GROUNDFISH
164740	096	CUSK	CALENDAR
164744	147	HADDOCK	GROUNDFISH
164791	509	HAKE,SILVER (WHITING)	MAY
164793	508	HAKE,OFFSHORE UNC (WHITING,BLACK)	MAY
166774	240	REDFISH,ACADIAN	GROUNDFISH
167687	335	SEA BASS,BLACK	CALENDAR
168543	444	TILEFISH,BLUELINE	CALENDAR
168546	446	TILEFISH (GOLDEN TILEFISH)	NOVEMBER
168559	023	BLUEFISH	CALENDAR
169182	329	SCUP	CALENDAR
171341	512	WOLFFISH,ATLANTIC	GROUNDFISH
172413	215	MACKEREL,CHUB	CALENDAR
172414	212	MACKEREL,ATLANTIC	CALENDAR
172567	051	BUTTERFISH	CALENDAR
172735	121	FLOUNDER,SUMMER (FLUKE)	CALENDAR
172746	125	FLOUNDER,SAND DAB (WINDOWPANE)	GROUNDFISH
172873	122	FLOUNDER,WITCH (GRAY SOLE)	GROUNDFISH
172877	124	FLOUNDER,PLAICE,AMERICAN (DAB)	GROUNDFISH
172905	120	FLOUNDER,WINTER	GROUNDFISH
172909	123	FLOUNDER,YELLOWTAIL	GROUNDFISH
172933	159	HALIBUT,ATLANTIC	GROUNDFISH
564130	366	SKATE,LITTLE	MAY
564136	364	SKATE,ROSETTE	MAY
564139	368	SKATE,BARNDOOR	MAY
564145	367	SKATE,WINTER	MAY
564149	370	SKATE,THORNY	MAY
564151	369	SKATE,SMOOTH	MAY
620992	710	CRAB,RED DEEPSEA	MARCH
630979	250	OCEAN POUT	GROUNDFISH
161702	374	SHAD,AMERICAN	CALENDAR
161703	112	HERRING,BLUEBACK	CALENDAR
161704	173	SHAD,HICKORY	CALENDAR
161701	170	RIVER,HERRING	CALENDAR

8.2 Stratification or “fleets” for discard sampling/estimation defined

For each species, the CAMS discard estimation process utilizes three tables of stratification variables including area definitions (for stocks or management units or estimation areas), gear and mesh, and other relevant attributes (e.g., regulatory groups). Some of the attributes are species specific while others apply similarly to all species. For many species that are quota monitored, stratification schemes are defined by regulations from a fishery management plan (FMP) and have agreed upon attributes. Other species may have stratification similar, but not identical, to the stratification used by SBRM. A species-centric stratification means that any given subtrip falls into one “fleet” depending on the attributes of the subtrip (e.g., gear or FMP) and the species discards being estimated. For example, a “groundfish” subtrip would have estimates of discards for groundfish species that involve sector-specific discard rates, such that the sector identification is used to define the fleet in which that given subtrip is a member for those groundfish species. However, for non-groundfish species, this subtrip would fall into a different fleet and have discard estimates generated by rates that are not sector specific. An important business rule is that any given sub-trip can only belong to one fleet (i.e., stratum) for a given species, and fleets are non-overlapping.

Stratification variables are largely derived from information on VTRs, including the gear, mesh, and statistical area fished. Other regulatory information may come from Allocation Management System (AMS) declarations, permit information, or other sources, existing or emerging, that need to be defined and reliably available for the relevant trips. For example, vessels participating in the sea scallop FMP declare their intention to harvest with either a limited access or a general category permit, a dichotomy that is used in the stratification.

An important requirement for CAMS stratification was the reconciliation of differing categories or groups within a stratification attribute between GARFO and NEFSC. Spatial and temporal stratifications required particular attention. Some species are managed as multiple stocks while others species are unit stocks, but use multiple areas for discard estimation. In-season quota monitoring requires a 12 month interval that matches a given FMP, while stock assessments require a calendar year and SBRM uses a time period that straddles 2 calendar years.

An important feature of discard estimation in CAMS is the departure from stratifying observer and catch information separately. In the past, analysts produced discard rates, by strata, from observer data and applied this to a similarly stratified set of catch data. In CAMS, a single, matched table CAMS_OBS_CATCH, is produced before any stratification or estimation occurs. This has the distinct benefit of having a sole source table for all estimation needs, as well as removing the possibility of mis-matched strata between distinct sources.

8.2.1 Area strata

Area is a primary stratification attribute and different species have different groups of statistical areas 8.1.

Figure 8.1: Map of federal statistical areas.

Stratification by federal statistical areas follows stock definitions for many species. Some species have within stock stratification for discard estimation purposes. These definitions are stored in a support table CFG_STATAREA_STOCK in Oracle that lists the relevant area stratification for each CAMS species and statistical areas within the US Exclusive Economic Zone in the Greater Atlantic region.

8.2.2 Gear/mesh category strata

Gear type is a primary stratification attribute and different species have different groupings of gear types into gear strata, sometimes with specific estimation area and discard mortality rates.

The base gear coding system for use in discards is the northeast gear code (NEGEAR) coming out of the landings process in CAMS 5.3 and from the observers or EM as the independent discard source. However, the NEGEAR mapped from VTR gear codes and the NEGEAR reported by observers do not always match perfectly, sepecially given slight differences in the exact gear definitions. The CFG_NEGEAR support table contains a column SECGEAR_MAPPED to facilitate gear matching between observed hauls and CAMS subtrips. The gear stratification for each species used in CAMS is defined in the CFG_GEARCODE_STRATA table on Oracle. Gear stratification is species specific since different species have different selectivity to various gear types. A final support table of species-gear group-estimation area stratification was created as a comprehensive source of discard mortalities (CFG_DISCARD_MORTALITY_STOCK in Oracle).

Mesh categories apply to certain gear types. Thus far, no support tables have been built to accommodate mesh categories. The CAMS project convened a mesh subgroup which determined the most pragmatic definition for mesh category (MESH_CAT) based on clustering of recorded mesh sizes in both observer and catch records.

Non-gill nets:

• Missing (NULL)

• Small (SM) : < 3.99 (inches) 

• Large (LM) : >= 4.00 (inches)

Gill Nets:

• Missing (NULL)

• Small (SM) : < 3.99 (inches)

• Large  (LM): >= 4.00 (inches) and < 7.99 (inches) 

• Extra large (XL):  >= 8.00 (inches)

Gear other than net and gillnet:

• Missing (NULL)

These definitions are used for both observer and catch records during the matching process.

8.2.3 Regulatory and other strata

Similar to the SBRM stratification (Wigley et al. 2007) , CAMS includes regulatory attributes such as scallop access area (e.g. Closed Area II, Nantucket Lightship) and trip category distinctions (e.g. , LIM for limited access category , GEN for limited access general category) for scallop trawl and scallop dredge subtrips. To support quota monitoring, CAMS includes a sector distinction (sector identification number and sector type) for groundfish trips. This information is included in CAMS_SUBTRIP and is used in stratification for estimation of groundfish on groundfish trips.

Fishing year (FY) of the species being estimated is used explicitly as a stratification variable. There are five distinct 12 month intervals used in the CAMS discard estimation process that correspond to the 14 FMP fishing years in the Greater Atlantic region. These are: January to December, March to February, April to March, May to April, and November to October. This information is stored in the data element FY_TYPE in CAMS_DISCARD_STRATA. Groundfish follow a May start fishing year but have their own FY_TYPE given the unique stratification needs.

One exception to the FY by species stratification are yellowtail and windowpane flounder. These species are groundfish, but have a subACL specific to the federal scallop fishery. Scallop trips within the groundfish time window utilize stratification and receive rates based on the scallop fishing year in which they fall. For these trips, the FY in the final table (CAMS_DISCARD_ALL_YEARS) follow the groundfish year but the FY_TYPE follows the scallop fishing year. The STRATA_USED_DESC column will contain values for FY and FY_TYPE that follow the scallop fishing year and may not match the singular FY column, which always matches the species being estimated. More detail can be found in the discaRd package.

8.3 Subtrip-level discard estimates, variance and CV

Each sub-trip has one or more estimates of dead discards (discards with discard mortality rates applied) depending on the species and the monitoring program in which the vessel participated. All CAMS species and subtrips can receive a discard estimate based on the quantity of kept weight on the subtrip and the estimated discard rate and discard mortality rate for the relevant stratum. The discard estimate that is used for a given objective may be context dependent, though catch accounting requires a single estimate to serve as an official estimate. Direct observations of species discards by qualified observers take precedent when available.

8.3.1 Direct observations

Trips that carry a human at-sea observer, or have EM video that has been reviewed, can provide official estimates of observed discards. These observations comprise separate data streams that require integration in CAMS, including matching to effort/landings records (e.g., VTRs) so that stratification attributes can be identified. Some monitoring programs (e.g., groundfish EM2) operate under the premise that vessel crew members can estimate discards for certain species; these records appear in the VTR and are not a separate data stream. Species and subtrips with discards determined from direct human observation have a null CV and have the DISCARD_SOURCE set to “O”.

8.3.1.1 Prorated discards

Some trips will have direct observations that are incomplete due to missed or unobserved hauls. For these trips, the total discards for a species must be prorated from those hauls that were observed. This calculation is done by applying a ratio of the kept weight to the kept weight on only the observed portion:

\[d_{spp,i} = \left(\sum_{h\in obs_{h}=\text{T}}d_{spp,i,h}\right)\left(\frac{\sum{k}_{i,h}}{\sum_{h\in obs_{h}=\text{T}}k_{i,h}}\right)\]

Here, the estimated discards (\(d\)) for a species (\(spp\)) on trip \(i\) equates to the recorded discards on hauls (\(h\)) that were observed (\(\left(\sum_{h\in obs_{h}=\text{T}}d_{spp,i,h}\right)\)) multiplied by the ratio of the total kept catch across all hauls (\(\sum{k}\)) to the kept catch (\(k\)) for observed hauls (\(obs_{h}=\text{T}\)).

8.3.1.2 Offwatch Hauls

CAMS imputes the statistical area and KALL from offwatch hauls so they can be treated the same as other observed trips where discards are imputed from unobserved hauls with proration as described above. The observer program uses offwatch haul logs in the scallop dredge, scallop trawl, and clam dredge fisheries to record the dates, times, locations and the amount of kept scallop or clam bushels during periods when the observer is “offwatch” from their normal duties. These hauls are recorded as unobserved. The offwatch period can span 1 or more hauls. Typically the first and last haul of the offwatch period have statistical area populated by the observer and all hauls in between are assigned an unknown statistical area. The captain also provides an estimate for the average number of bushels kept during the offwatch period.

Offwatch hauls increase the number of trips in CAMS that have discards estimated from a discard rate because they are unobserved. On observed trips without offwatch hauls, discards from unobserved hauls are imputed to construct a complete discard estimate for the entire trip, which is then used as the actual discard estimate for the trip. Unfortunately, offwatch hauls do not contain statistical area or kept landings (KALL), which are both needed to impute discards on these hauls and stratify by the appropriate area. In a given year, offwatch hauls make about 25-30% of total hauls on observed trips and are concentrated in the scallop and clam fisheries. To mitigate these impacts, the CAMS Team developed methods to impute statistical area and KALL based on previous analyses to support the Standardized Bycatch Reporting Methodology:

Statistical Area Imputation: The approach joins vessel permit number and observer reported haul times to VMS lat/lon positions from standard positional polls (TPREP). Offwatch haul lat/lon locations are interpolated from the sequence of VMS positions and assigned a statistical area using a points-in-polygon operation between the assigned VMS lat/lon and a spatial enabled Oracle table of statistical areas. The offwatch period is determined by the haul start and haul end for hauls identified as offwatch by the observer. In less common situations where the first haul of the trip is an offwatch haul the trip sail date + steam time is used as the offwatch period start. Similarly, if the last haul of the trip is an offwatch haul the trip land date is used for the offwatch period end. In a minority of cases this can result in position assignments to steaming periods that are likely a less accurate estimate of fishing location. Position interpolation is accomplished by counting the number of positional polls per haul within the offwatch haul period, and determining a fractional position (or ping), which is assigned to each haul by multiplying the ordinal haul number and the pings/haul for the offwatch period.

KALL Imputation: For trips with offwatch hauls, imputed kept weights for all species are based on the total kept (includes observed and unobserved hauls) for the trip that are reported on the observer catch log which is then divided by the total number of non-offwatch hauls to derive an average pounds per haul. This average is the imputed value for each offwatch haul. The approach includes ALL hauls and does not filter out hauls that only caught non-living material (rocks, debris, etc.).

8.3.1.3 EM discards

Vessels enrolled in the EM2 program have direct estimates of allocated groundfish species (DISCARD_SOURCE = EM) for those trips where their video was reviewed (50% review rate in FY21 and for new vessels in FY22; 35% in FY22 for vessels that were enrolled in the program in FY21). Note, on EM trips that were reviewed, ratio estimation of discards was still required for non-groundfish species.

8.3.2 Ratio estimation

The discaRd R package was designed for quota monitoring (GARFO 2016) and implements the “separate” ratio estimator (Cochran 1977), consistent with approaches used by SBRM (Wigley et al. 2007). As such, observed trips are used to estimate the rate of species discarding per unit effort, with effort approximated by the total kept catch (i.e., pounds). Given that the discards (\(d\)) and total kept catch (\(k\)) are both measured in pounds, technically \(d/k\) is a ratio. We use ratio and rate interchangeably here, recognizing that units of effort could be (and have been) measured in other ways (e.g., days at sea).

Here, the ratio or rate of species discards to total kept catch within stratum \(j\) is defined as:

\[ r_{spp,j} = \frac{\sum_i^{n_j} d_{spp,i,j}}{\sum_i^{n_j} k_{i,j}} \] with total pounds of discard:

\[ \hat{D}_{spp} = \sum_{j=1}^{J}{K_j r_{spp,j}} \]

and with analytic variance calculated as:

\[ V(\hat{D}_{spp}) = \sum_{j=1}^{J} \left(\frac{N_j-n_j}{n_j N_j}\right) \frac{1}{\left(\frac{\sum_{i}^{n_j}k_{i,j}}{n_j} \right)^2} \left[\frac{\sum_{i}^{n_j}(d_{spp,i,j}^2+r_{spp,j}^2 k_{i,j}^2 - 2r_{spp,j}d_{spp,i,j}k_{i,j})}{n_j - 1}\right] \]

Finally, the preceding terms are defined as follows:

• \(r_{spp,j}\) is the separate ratio for species \(spp\) in stratum \(j\)
• \(d_{spp,i,j}\) is discards of species \(spp\) from observed trip \(i\) in stratum \(j\);
• \(k_{i,j}\) is the kept pounds of all species on observed trip \(i\)
• \(n_j\) is the number of observed trips in stratum \(j\)
• \(N_j\) is the number of total trips in stratum \(j\)
• \(J\) is the number of strata
• \(\hat{D}_{spp}\) is the total estimated discarded pounds for species \(spp\)
• \(K_j\) is the total kept pounds of all species in stratum \(j\)

Note, the analytic variance contains a finite population correction, recognizing that most strata will have sampling rates over 10%.

Stratum-specific estimates of the total discards, \(\hat{D}_{spp,j}\), and associated variance, \(V(\hat{D}_{spp,j})\), can be calculated as needed. Additionally, the variance or error associated with trip-level predictions can also be derived and is included.

Species and subtrips with discards estimated from a discard ratio have an associated variance and CV based on the pool of data in the stratum and DISCARD_SOURCE based on the stratum used (e.g. I = In season, GM = Gear/Mesh).

8.3.2.1 Transition of discard rates between seasons

In-season monitoring of the cumulative discard for a given species and fishery requires the application of a discard rate to all completed trips up to the day of calculation. During the early part of the fishing season, or for low-effort strata, the sample size of observed trips may be too small for a stable and reasonable estimate of the discard rate. To improve in-season estimates of discard, information from the previous fishing year is used in varying degrees as an adjustment.

As such, we calculate a weighted mean between the estimated discard rate of the previous year and that for the current year when the number of observed trips, \(n_j\), is >0 and <5. The weighted mean is calculated as follows:

\[r_{spp,j}^* = \left(\frac{0.7}{n_j}\right)r_{spp,j}^{t-1} + \left(1-\frac{0.7}{n_j}\right)r_{spp,j}^{t}\] Here, \(r_{spp,j}^*\) is the transition discard rate, \(r_{spp,j}^{t-1}\) is the “assumed” discard rate from the previous year (\(t-1\)), and \(r_{spp,j}^{t}\) is the in-season rate from the current year (\(t\)).

8.3.3 Electronic Monitoring

8.3.3.1 Delta Model

For EM subtrips that were not selected for review, not yet reviewed, or failed review due to poor quality (e.g., >10% of fish not able to be identified to species on any haul(s), or a haul(s) were not observed), a generalized mixed model (delta model) was fit to predict discards based on the relationship between self reported VTR discard estimates and EM discard estimates on trips where the EM video was reviewed (Linden et al. 2021). EM data used to fit the model includes data from FY20 forward.

8.3.3.2 Maximum Retention

For maximum retention EM (MREM) subtrips, sublegal allocated groundfish are not discarded and instead kept. Discard of these species on unobserved trips are assumed to be zero. Observed MREM trips will still get the observed value.

For maximum retention EM (MREM) subtrips, sublegal allocated groundfish are not discarded and instead kept. Discards of these species, on unobserved trips, are assumed to be zero. Observed MREM trips will still get the observed value. Other species may be discarded, including un-allocated groundfish (e.g. halibut, windowpane, ocean pout, wolffish). In order to make MREM trips on par with non-MREM trips, adjustments in subtrip kept all (from the VTR) have to be made. This is done by removing sublegal, allocated groundfish pounds, from the kept all (adjusted kept all). The adjusted KALL may then have a discard rate applied from non-MREM trips without inducing bias.

8.4 Support Tables

Figure 8.2: Base flowchart of the CAMS discard process and support tables.

Matching data from observer hauls to CAMS subtrips is a key element of the CAMS discard estimation process resulting in the CAMS_OBS_CATCH table in Oracle. The matching occurs in a staged manner. All commercial trips with a non-null LINK1 field get a value for how many unique CAMS subtrips (CAMSID || SUBTRIP) are associated with it. The vast majority of observed trips are associated with only a single subtrip, and only require matching by LINK1. For multiple subtrips, a match as described above is used (gear, meshgroup, statistical area, and LINK1). Statistical area is only used for matching when more than one area is reported for a trip. This reduces data loss from observer recorded area mismatches to VTR recorded area.

Matching observed records to subtrips reduces the possibility of subsequent mismatches in discard stratification. It also allows observed discards to easily be used as the official discard for a particular trip. Furthermore, we recognize that data errors either from the catch, or observer data, may result in a non-match. This likely reduces the total pool of observed trips that are being used, but we feel the benefits of using this approach outweigh a reduced sample size. Quality control of these data are outside the purview of the CAMS project. Last, we only use observed trips (LINK1) where valid LINK3 (hauls) occurred to alleviate issues of multiple LINK1 records for a single subtrip.

IMPORTANT!: The CAMS_OBS_CATCH table has a one to many row-wise structure. For trips that were not observed, there will be a single row, with all trip metrics, and a total KALL per subtrip. When a trip was observed, there are multiple rows, where the trip metric information is repeated and each row shows species, discarded amount, and other observer recorded information for each row. Total KALL CANNOT be calculated without filtering rows by LINK1 to indicate an observed trip or not. These steps are outlined in subsequent R modules used to run discaRd.

Figure 8.3: The process of joining observer haul information to CAMS subtrips.

The CFG_GEARCODE_STRATA discard support table translates species SECGEAR_MAPPED codes, from joining landings and observer data, to the CAMS gear group code used for discard stratification. These stratifications are based on gear stratification used in past stock assessments and quota monitoring along with the expert opinions from analysts at NEFSC and GARFO through discard working groups and review. The ESTIMATE_DISCARDS column in this table defines gear and species combinations that may be purposely not estimated. Primarily, this is for gear designations that do not receive observer coverage (e.g. rakes, dipnets, diving, harpoon) but could potentially cover other situations (e.g. not extrapolating crab pot discards to all crab pot landings). This column is applied as a multiplier for unobserved subtrips at the end of each discard run.

The CFG_STATAREA_STOCK discard support table translates species statistical areas to species stock areas used for discard stratification based on area stratifications used in current stock assessments and quota monitoring. Species specific area stratifications have been agreed upon by a discard working group.

Mortality rates by species, estimation area, and gear group are found in the CFG_DISCARD_MORTALITY_STOCK discard estimation support table are based on values used in current stock assessments and quota monitoring. Discard mortality rates have been agreed upon by a CAMS discard working group.

8.5 Steps for running discard estimation

• refresh/build Oracle tables (SQL)
  • CAMS_OBDBS_ALL_YEARS
  • CAMS_OBS_CATCH
• import two or more years of CAMS_OBS_CATCH to R depending on the run needed. A minimum of two fishing years is necessary.
• For each run section, a group of species is defined corresponding to a fishing year
  • apply CAMS_GEAR_GROUP according to SPECIES. Support table is CFG_GEARCODE_STRATA
  • apply estimation area according to SPECIES and stock (if needed). Support table is : CFG_STATAREA_STOCK
  • join discard mortality by species/stock/CAMS_GEAR_GROUP using CFG_DISCARD_MORTALITY_STOCK
    • STRATA is assigned dynamically using elements of the imported data. There are currently three flavors of stratification: Groundfish trips, Scallop trips for groundfish estimations, and generic (all others)
    • Two time periods are defined according to focal and previous fishing year, by species. This allows a transition rate calculation
    • Assumed rates (e.g. sector rollup, gear/mesh, gear only, broad stock) are defined as a subset of STRATA
• Estimation is done using the discaRd R package built for 2016 Discard Estimation Peer Review. The run_discard function is applied for each time period and species. Two or three passes of this routine are done, each using a less coarse stratification, depending on module specifics.
• Assign CAMS_DISCARD_RATE through business rules for the trip (e.g., observed trips receive the observed value, EM rules)
  • note the DISCARD_SOURCE
• Apply CAMS_DISCARD_RATE to kept all for the trip. This is the estimated discard.
• Apply discard mortality to discard to get finalized discard estimate.
• Apply ESTIMATE_DISCARDS criteria (0/1) to unobserved subtrips. 0 values receive a DISCARD_SOURCE = N, CAMS_DISCARD = 0 and CV = NULL. Trips falling in strata with no coverage in both focal season and previous season also receive DISCARD_SOURCE = N.

8.6 Discard Source

There are several possible sources of discard based on type of trip and stratification used for that trip.

Table 8.2: Discard source codes and descriptions based on trip characteristics and stratification.

DISCARD_SOURCE	DISCARD_SOURCE_NAME	DESCRIPTION
R	RULE_BASED	Rule-based assignment of a set discard value (e.g. 0 for allocated groundfish on unobserved/unreviewed MREM trips because not allowed to discard).
I	IN_SEASON	In season rate; >= 5 trips in Full Stratification
EM	ELECTRONIC_MONITORING	Electronic monitoring (EM) video based estimate. If the trip was reviewed and passed; and there was no human observer on board.
N	NO_COVERAGE	No coverage in focal season and previous season; discard rate is not possible to estimate. 2) Gear and species combinations where ESTIMATE_DISCARD in CFG_GEARCODE_STRATA = 0. These are gear types that tpyically have no coverage (e.g. rakes, dipnets, etc.)
O	OBSERVED_VALUES	Observed values used from observed trips; discard rate is NOT USED.
B	BROAD_STOCK	For groundfish, broad stock rate is used when other criteria are not met.
A	ASSUMED	Assumed rate for groundfish. This is the second pass rate that rolls up all Sectors. Common Pool trips are separated in this strata.
DELTA	DELTA_MODEL	Delta model estimate of discards on an EM trip that was not selected for review or it was reviewed and failed; and there was no human observer on board.
T	TRANSITION	Transition in season rate; < 5 trips in Full Stratification, year t, AND >= 5 trips in year t-1
GM	STOCK_GEAR_MESH	Assignment is from the second pass (e.g. Stock/Gear/Meshgroup). This is the rate when there were <5 trips in season and <5 in previous season at the full stratification level.
G	STOCK_GEAR	Assignment is from the second pass (e.g. Stock/Gear).

Groundfish estimates from groundfish trips are as follows can receive discard sources I, T, A, B, EM, DELTA, N, or O and non-groundfish trips receive one of I, T, GM, G, N, or O.

Figure 8.4: Decision diagram of the process for assigning a discard source.

The coefficient of variance (CV) of the estimated discard rate is derived for all discard sources except, O, EM, DELTA, and N, and is stored in the CAMS_DISCARD_STRATA data table. To accommodate users who require a variance over aggregated subtrips, we back-calculate this value from the CV and the variance of the discard estimate.

\[ COV(D_{spp, j, s}) = (D_{spp, j, s}*CV_{spp, j}) * \sum_{}\sqrt{V_{spp, j, s}} \]

Finally, the preceding terms are defined as follows:

• \(COV(D_{spp, j, s})\) is the covariance of the discard of species \(spp\) of a single subtrip \(s\) in stratum \(j\)
• \(D_{spp, j, s}\) is the estimated discard of species \(spp\) of a single subtrip \(s\) in stratum \(j\)
• \(CV_{spp, j}\) is the coefficient of variation of the discard rate for species \(spp\) in stratum \(j\)
• \(V_{spp, j, s}\) is the variance discard of species \(spp\) of a single subtrip \(s\) in stratum \(j\)

The back-calculated covariance (CAMS_DISCARD_VARIANCE) can be summed for any group of trips. This allows calculation of a CV associated with the aggregated discards on different groups of trips. This is particulalry useful for stock assessments, which operate on calendar year, for species with offset fishing years (e.g. groundfish).

8.7 Discard Runs in CAMS

Standard discard runs based on GARFO Quota Monitoring (species where in-season estimation is required) :

• May Fishing Year
  • Groundfish
  • Monkfish
  • Dogfish
  • Skates
  • Hakes
• April Fishing year
  • Scallops
    
  • Flatfish bycatch
• January Fishing Year (Calendar Year):
  • Squid/Mack/Butterfish
  • Herring{style=“color:orange”}
  • River Herring/Shad
  • Bluefish
  • Summer flounder
  • Black Sea Bass
  • Scup
• November Fishing Year (January Year from 2022 onward)
  • Tilefish

Originally, separate modules, based on fishing year, were coded for these runs. This was simplified to a few key functions within the discaRd package (discard_groundfish.R,discard_generic.R) to streamline the process.

During the discaRd run, temporary tables (.fst) are produced. Each temporary table is specific to a fishing year and species. These are then uploaded to Oracle, through a table append, to CAMS_DISCARD_ALL_YEARS and CAMS_DISCARD_STRATA.

   CAMS_GARFO.CAMS_DISCARD_ALL_YEARS
    
    select d.*
    from CAMS_GARFO.CAMS_DISCARD_ALL_YEARS d
    join
      cams_subtrip s
      on d.camsid = s.camsid
      and d.subtrip = s.subtrip
     where s.GF = 1
    AND s.permit_state_fed = 'FEDERAL'
    AND d.FY = 2019
    AND d.ITIS_TSN in ('172877' ,
    '164712' ,
    '164732' ,
    '166774',
    '171341',
    '172873',
    '172909',
    '164727',
    '172933',
    '172746',
    '172905',
    '164744', 
    '630979')

8.8 Future design

If all strata were observed then the sampling would achieve a full factorial design, whereby all combinations of stratum attributes have an associated discard rate estimate. Some strata will inevitably have no observed trips (\(n_j = 0\)) in a given fishing season, and even across multiple seasons, for combinations of stratification attributes that result in rare (but non-zero: \(N_j>0\)) effort. The choice of a discard rate for an unobserved stratum has historically been somewhat arbitrary and required manual selection of a “similar” stratum to inform the discard estimate.

One proposed strategy is to take a model-based approach to discard estimation and estimate the additive effects of stratification attributes instead of the multiplicative interactions. This reduces the sampling needed and provides a pragmatic replacement estimate for unobserved strata.

It can be shown that a Poisson regression of discards with log-scale kept catch as an offset can approximate the ratio estimator as follows:

\[ d_{spp,i} \sim Poisson(\lambda_{spp,i} k_i)\] \[ \text{log}(\lambda_{spp,i}) = \mathbf{X}_{i} \boldsymbol{\beta}_{spp}\]

where \(\lambda_{spp,i}\) is the estimated species discard rate per pound of kept catch, \(\mathbf{X}_{i}\) is a design matrix of additive effects for relevant stratification attributes on trip \(i\), and \(\boldsymbol{\beta}_{spp}\) are species-specific coefficients for the additive effects.

As an example, if stratification involved 3 gear types (“A”, “B”, “C”) and 2 areas (“North” and “South”), the ratio estimation would typically estimate rates for 2 x 3 = 6 different strata. The equivalent Poisson model above would require 6 coefficients \((\beta_{AN}, \beta_{BN}, \beta_{CN}, \beta_{AS}, \beta_{BS}, \beta_{CS})\), one for each stratum. If the design matrix specified additive effects instead of the full factorial interactions, only 4 coefficients would be required \((\beta_{0}, \beta_{B}, \beta_{C}, \beta_{S})\). Crucially, if no trips were observed for gear = C in area = South, the full factorial model could not be estimated. The missing observations are not a problem for the additive model, and predictions could still be generated for unobserved trips with those attributes based on the overall effects: \(\text{E}(\text{log}(\lambda)) = \beta_0 + \beta_C + \beta_S\).

To implement the proposed approach, an unsampled stratum needs to be identified and have estimates of \(\text{E}(\lambda_{spp,j})\) replace \(r_{spp,j}\) as described to estimate \(\hat{D}_{spp,j}\), the total species discards for the given stratum.

8.9 References

Cochran, W.G. 1977. Sampling techniques (3rd edition). John Wiley and Sons, New York. Available from http://www.wiley.com/WileyCDA/WileyTitle/productCd-047116240X.html.

Linden, D.W. 2021, September 9. A predictive model of discarded catch that leverages self-reporting and electronic monitoring on commercial fishing vessels. CIE Reviewed White paper, NOAA/NMFS/GARFO. Available from https://media.fisheries.noaa.gov/2022-03/LindenEMdeltamodelpaperCIE-GARFO.pdf.

Linden, D.W., Galuardi, B., and McAfee, B.M. 2016. Methods for examining in-season behavior of the cumulative discard estimation in the Greater Atlantic Region. Working Paper #1. NOAA/NMFS/GARFO. Available from https://www.fisheries.noaa.gov/new-england-mid-atlantic/science-data/discard-methodology.

Nitschke, P. 2010. Estimating in-season discards from the Northeast United States groundfish fishery: Discard Estimator Performance Simulation Study (Part I). Working Paper No. In Discard Estimation Methodology Review. Northeast Fisheries Science Center National Marine Fisheries Service, Woods Hole, MA. Available from https://www.researchgate.net/publication/268257483_Estimating_in-season_discards_from_the_Northeast_United_States_groundfish_fishery_Discard_Estimator_Performance_Simulation_Study_Part_I_Working_Paper_No.

Wigley, S.E., Rago, P.J., Sosebee, K.A., and Palka, D.L. 2007. The Analytic Component to the Standardized Bycatch Reporting Methodology Omnibus Amendment: Sampling Design and Estimation of Precision and Accuracy (2nd edition). U.S. Department of Commerce, Northeast Fisheries Science Center Reference Document 07-09. Available from https://repository.library.noaa.gov/view/noaa/5261.

Last modified: 2025-07-07