Observation error model

The Observation Error Model (OEM) defines the methods for the observation of the system. This includes observations of commercial catches catches and scientific surveys, but more fundamentally also defines the overall structure of the system as required by the Management Procedure (MP). If management resolves fleets, then the OEM returns stock biological information (in the form of FLBiols) and fleet catch information (in the form of FLFisheries) separately. If fleets are not resolved (as is the case for 99% of MPs), then the biological and catch information are combined in to a single structure (FLStocks).

Inputs

We've already introduced the general structure of the OEM input object here, but we now want to break down what each of these inputs and arguments mean. The input OEM object is an FLR object FLoem that defined by the mse package, but has a neat layout that satisfies the requirements for MixME - so we use it here!

MixME makes use of three key slots in the FLoem object:

• observations
• deviances
• args

It's worth taking some time to take a closer look at each of these slots. We need to explore both the required structure for the inputs to each slot and the effect that each argument has on the behaviour of the OEM.

Observation structure (observations)

The observations slot of the OEM contains elements for stocks (stk) and indices (idx).

Scientific survey indices may be age-structured or a lump of biomass.

Observation error (deviances)

The deviances slot of the OEM contains the pre-sampled observation error for stock catch and survey observations. This pre-sampling of error is convenient because it means that stochastic simulations are repeatable for the same set of inputs.

In a typical Management Strategy Evaluation, catch and survey observations will be with observation error. However, you might want to carry out more exploratory simulations that do not have this layer of complexity.

Arguments (args)

use_stk_oem

This is a logical argument. If TRUE, then a user-supplied OEM observation object is populated with observed historical landings and discards (and hence overall catches). If FALSE, then the observation structure is directly derived from the Operating Model (OM).

There are two possible inputs to use_stk_oem. The input is either an unnamed logical element or a vector where there is a named logical element for each observed stock. The value of specifying the input for each stock individually is that some stocks can take a user-suppled OEM observation object whilst others are derived from the OM. If an unnamed logical element is supplied, MixME applies this selection to each stock.

The value of supplying an OEM observation structure is that biological parameters (individual stock, landings and discards mean weights-at-age, proportion mature-at-age, and instantaneous natural mortality-at-age), which are not dynamically observed within the simulation, can differ from the true parameter values.

Note

Supplying an OEM observation structure is currently the only way of specifying an FLStock observation structure. If the observation structure is derived from the OM, then an FLBiol and FLFisheries is created.

use_catch_residuals

This is a logical argument. If TRUE, then user-supplied catch observation error is applied to Operating Model catches. If FALSE, then no observation error is applied. There are two possible inputs: either an unnamed logical element or a vector where there is a named logical element for each observed stock. If an unnamed logical element is supplied, MixME applies this selection to each stock.

use_idx_residuals

This is a logical argument. If TRUE, then user-supplied survey index observation error is applied to simulated survey indices. If FALSE, then no observation error is applied. There are two possible inputs: either an unnamed logical element or a vector where there is a named logical element for each observed stock. If an unnamed logical element is supplied, MixME applies this selection to each stock.

use_om_weights

This is a logical argument. If TRUE, then catch weights are updated from the Operating Model. This is important if changes in fleet activity or overquota discarding leads to changes in the individual mean landings and discards weights at age, which are not reflected in the user-supplied OEM observation object.

There are two possible inputs: either an unnamed logical element or a vector where there is a named logical element for each observed stock. If an unnamed logical element is supplied, MixME applies this selection to each stock.

Note

The value of use_om_weights is if you want to use an FLStock observation structure and perfect observations of landings and discards weights, but potentially observations with error of other stock properties. If the observation structure is taken from the OM, catch weights are automatically up-to-date.

catch_timing

This is a named list with an element for each observed stock. Each element is numeric and describes the observation lag for catches. This is given as the index of the final data year relative to assessment year (e.g. assessment year is 0, assessment year - 1 = -1, assessment year - 2 = -2, ...).

idx_timing

This is a named list with an element for each observed stock. Each element is a numeric vector with elements for each survey index. The order of survey indices in this vector follows the order of surveys given in the survey observations slot for each stock. Like catch_timing, this is an index of the final data year relative to the assessment year (e.g. assessment year is 0, assessment year - 1 = -1, assessment year - 2 = -2, ...).

Methods

The OEM is applied iteratively over each stock. This imposes the constraint that observations of one stock cannot depend on the observations of another stock. This assumption is probably fine for the vast, vast majority of mixed fisheries MSE.

Depending on the input arguments, stock observations are generated and coerced into either a user-supplied OEM observation structure or an observation structure that is derived from the operating model. The specific methods used to generate stock and survey observations are largely adapted from WKNSMSE (2019).

Catch observations

User-supplied FLStock

The processing steps are:

1. extract stock landings and discards numbers for each fleet from the OM
2. (optional) update discard and catch individual mean weights at age from the OM
3. (optional) apply fleet-specific catch observation error and calculate total landings and discards numbers
4. trim the FLStock to the period for which we have catch or survey data (defined by catch_timing and idx_timing)

Note

Catch observation uncertainty impacts both landings and discards numbers, but the fraction of catch that is landed is assumed to unbiased (unaffected by observation error). This is a fairly strong assumption.

User-supplied FLBiol and FLFisheries

Caution

This is not currently implemented in MixME.

Operating Model FLBiol and FLFisheries

The processing steps are:

1. extract stock and fleets data from the OM
2. trim the FLBiol and FLFisheries to the period for which we have catch or survey data (defined by catch_timing and idx_timing)

Important

Catch observation error is not currently applied.

Survey index observations

The process steps are:

1. calculate index observations based on updated values in the OM
2. (optional) apply survey-specific index observation error
3. trim the FLIndices to the period for which we have survey data (defined by idx_timing)

Outputs

The output from the observation error model is a named list with the elements:

• stk: a named list with an element for each stock. FLStock or FLBiol object
• flt: (optional) a named list with an element for each stock. Each element is an FLFisheries object. Alternatively NULL
• idx: (optional) a named list with an element for each stock. Each element is an FLIndices object. Alternatively NULL