Projection settings and scenarios

[grantdadams]

Currently, Rceattle runs projections assuming the terminal year of weight-at-age, selectivity, catchability, and consumption for all projection years. Only recruitment can change in the settings.

For recruitment for the forecast and BRPs, there are two options:

1. Default: Users can project via mean recruitment between styr and endyr by setting proj_mean_rec = TRUE in the build_srr function fed into the recFun argument of fit_mod. Alternatively, mean recruitment between styr and srr_mse_switchyr can be used (default is srr_mse_switchyr = NULL and set to endyr), but this was designed for dealing with MSEs and Jim's stock-recruit curve as a penalty approach as the model is extended beyond the historical period. Future rec_devs, if different from default 0, will only be integrated into dynamic reference points.
2. Users can project using the stock-recruit function (including environmental linkages and rec_devs) by setting proj_mean_rec = FALSE.

HCRs are specified via build_HCR and a 2-stage estimation approach is used. Stage-1 = estimate the hindcast, Stage-2 = rebuild, fixing the hindcast parameters and estimate the BRPs.

For uncertainty under HCRs, projection_uncertainty in fit_mod will do sdreport on all ADREPORTed derived quantities leaving both hindcast and projection (BRPs) parameters turned on AFTER estimation of both. This was designed to account for the the 2-stage estimation. If no HCR is used, no uncertainty in the BRPs will be integrated into the projections and things will be equivalent under projection_uncertainty = TRUE and projection_uncertainty = FALSE.

@Cole-Monnahan-NOAA

For expansion, I think allowing weight-at-age, selectivity, catchability, and consumption to either use a mean between a specified range of years OR projected forward if time-varying would be important.

For DSEM integration, I think allowing both the projection under mean recruitment and DSEM would be valuable for comparing the value of DSEM to improving projection performance AND having backwards compatibility for the current AFSC approach. Turning off estimation of x_tj in the projection and the projection's log-likelihood components would be the key for projecting under mean recruitment.