NEWS

---

cpueTargeting 0.4.0

Selection tools, predictive checks and logbook guides

• select_tactics(): remove (or keep) fishing tactics by cluster number or by species content (e.g. clusters with at least 50% demersal catch) and recompute the PCA and clusters on the remaining operations.
• select_vessels() gains rules on a target species: target, min_years_target, min_share_target, min_catch_target; the vessel table is attached as attribute vessels. New vessel_summary() and plot_vessel_profile() (each vessel's catch by species or group).
• ppc_cpue(): predictive check of the error distribution. Replicate data sets are simulated from the fitted Tweedie or delta models (optionally with parameter uncertainty) and compared with the data: share of zeros, upper quantiles of positive CPUE and the distribution of log CPUE, overall and by year (or any grouping).
• New vignette Setting up the logbook simulator: a user guide; the logbook vignette is now Standardising a realistic logbook: two worked examples (generic preparation, then blackspot seabream and dolphinfish); the real-data guide covers the new tools.

Logbook simulator: specialist loyalty and mixed trips

• simulate_logbook(): specialist vessels are no longer locked to their tactics. home (replacing allowed) gives each vessel type's home tactics and loyalty the share of its trips that stay on them (default 0.8 for demersal and pelagic specialists; the "flexible" type is now mixed, loyalty 0, choosing purely by expected value). The other trips choose among all tactics by expected value. sim$trips$loyal records which trips stayed home. Old settings with allowed still work (loyalty 1).
• logbook$switch_tactic default raised from 0.2 to 0.7: with the default fleet, about 18% of trips now combine two tactics in one logbook record.

Toolbox for real data (steps 1-6)

• cpue_data() brings long or wide logbook data into one row per operation with species columns, merges species synonyms, removes missing effort, and keeps a processing log; subset() method and cpue_matrix().
• Exploration: plot_fleet() (operations by year, vessel x year tiles, fleet turnover, covariate shares), select_vessels() (core fleet), nominal_cpue(), plot_nominal(), plot_composition(), plot_cpue_map().
• Species selection: species_curve(), plot_species_curve() and select_species() (cumulative share of occurrence or weight, optional Other group).
• Habitat-based subsetting of Stephens and MacCall (2004): stephens_maccall() and plot_stephens_maccall().
• Targeting covariates: composition_pca(), compute_oc() (Optimal Coordinates test of the CPUE guidelines, robust when all components pass), cluster_tactics() (CLARA, k-means or Ward), add_targeting(), tactic_table(), plot_oc(), plot_biplot(), plot_targeting().
• Models for real data (steps 7-9): fit_cpue() fits Base, Covariates, DPC and Cluster GAMs (optionally +S spatial and random vessel effects) with mgcv, Tweedie (power estimated jointly, no two-step gamm()) or delta-lognormal / delta-gamma families, and derives indices averaged over a reference set with delta-method log SEs. cpue_index(), export_index().
• Diagnostics: cpue_diagnostics(), plot_index(), plot_step(), influence_cpue() and plot_influence() (Bentley et al. 2012), plot_residuals() (randomised quantile residuals for Tweedie), plot_effects(), plot_vessel_effects().
• New vignette Analysing real catch and effort data: a step-by-step guide.
• simulate_logbook(), logbook_settings(), logbook_truth(), plot_logbook_om(): a ten-species demersal-pelagic fishery with reef/sediment x shallow/deep habitats, seasonal pelagic migrants, discrete tactic choice (Reef, Sediment, Pelagic), specialist and mixed vessels, fleet turnover with efficiency creep, price periods and a closure, reported as a messy trip-level logbook. New vignette A realistic logbook: reef, sediment and pelagic tactics.
• GitHub Actions: R CMD check on every push; generated documentation is committed automatically.
• cluster added to Suggests.

cpueTargeting 0.3.0

• standardize_tuna_indices() is renamed standardize_cpue(), as it works for both simulators. The old name is kept as an alias.
• New plot_clusters(): diagnostics for the Cluster models (composition of each cluster, cluster shares over time, most common cluster in space, hook depth by cluster, clusters in PC space), as in the CPUE guidelines.
• run_replicates() gains simulator, so it also runs simulate_bathymetric(); new read_replicates() combines results saved one file per seed (e.g. from parallel runs).
• New vignettes Targeting along a continuous gradient: the bathymetric example and Tuna longline targeting: flat versus contrasting true trends. They load pre-run results; the simulation scripts are in inst/scripts/ (run_bathymetric.R, run_tuna_flat.R, run_tuna_contrast.R, with shared settings in talk_setup.R), and render_docs.R renders the vignettes to docs/. Results are saved to rdata/ (ignored by git); options(cpueTargeting.reps = FALSE) runs only the single runs used by the vignettes.
• The 1 x 1 spatial control is no longer part of the standard comparisons: positions are reported as 5 x 5 blocks, the realistic, data-limited case.
• Single-run illustrations use seed 123.
• New setting q_index (default 0.2) separates the catchability at which CPUE is recorded from the fishing catchability q. Set-level Tweedie noise is drawn at q_index, and the catch removed scales the same draws by q / q_index. Fishing harder (larger q) therefore no longer removes the zeros and reduces the noise of the CPUE data, and avoids saturating 1 - exp(-q A), which damped the targeting signal. With q_index = q results are as before.

cpueTargeting 0.2.0

Tuna operating model rebuilt

• simulate_tuna() now has four species (YFT, BET, BFT, ALB) in a conceptual Southern Hemisphere ocean. Horizontal habitat is a preference for SST; vertical habitat is a normal depth profile shifted by a thermocline anomaly field, so the effect of HBF varies in space.
• Each HBF level has a typical hook depth (cf. Bigelow et al. 2002, Fig. 2) with lognormal set-level error (set_cv) and lognormal within-set spread (hook_cv). A set's catchability is vessel q times the overlap between hook and fish depth distributions (effective effort).
• Skippers choose cells and HBF from expected catch value.
• Data keeps the latent truth (hook_depth, vertical_avail) and the overlap a habitat model would predict from HBF and the average thermocline (hbs_avail).
• standardize_tuna_indices() follows the guideline workflow (Tweedie GAMs with estimated power). The default comparison covers the three practical approaches to targeting (Cluster, DPC, HBF) and their combinations (Cluster_HBF, DPC_HBF); habitat-based standardisation (HBS) is available on request. Optional random vessel effect. It also works for simulate_bathymetric() output.

Faster and fewer dependencies

• The bathymetric and tuna models share one set of building blocks (R/om_utils.R). Movement matrices are never formed explicitly, the stable initial distribution uses its closed form, and all sets of a year are simulated at once. A 30-year run takes seconds.
• Removed dependencies on fields, reshape2 and tweedie: the Gaussian random field (same covariance as RandomFields::RMgauss) and the Tweedie generator are now internal.
• attach() of the settings is no longer used. Missing settings are filled from the defaults.

Corrections relative to the original Run 22 code

• Species growth rates r were recycled over cells instead of species (r * B with a cells x species matrix); r is now applied by species.
• The initial biomass B0 * 0.5 * rlnorm(4, ...) assumed four species; it now uses the number of species.
• The initial distribution is the exact stable distribution of the movement process (the original iterated 1000 times, which may not converge for short movement distances).
• Reported coarse positions are the centres of 5 x 5 blocks aligned with the grid (obs_res); the original rounding gave 11 uneven blocks.
• A closure set catchability to 1 in open cells; open cells now keep qq_original.
• Because random numbers are drawn in a different order, results do not match the original runs number by number.

Moving thermocline

• New settings thermocline_annual_sd, thermocline_annual_rho and thermocline_annual_field_sd let the thermocline move between years (default 0: fixed). Data$thermocline is the anomaly of that year and the full field is returned in Thermocline.
• standardize_tuna_indices() gains user-defined models via add_models, and engine = "bam" for speed. Appending +S to any model name (e.g. "DPC+S") adds the spatial smoother s(Xs, Ys); spatial_res sets the reporting block size and spatial_k the basis size. The Spatial model is now Base+S.
• New vignette Targeting in a longline tuna fishery: fixed and moving thermocline scenarios, Cluster vs DPC vs HBF, what the PCs capture, and optional spatial-resolution and year-interaction steps.

Standardisation helpers

• n_pc sets the number of principal components used by the DPC models (default 3; the guidelines used 2); all PCs and the PCA loadings are returned.
• n_clusters defaults to n_pc + 1 (Winker et al. 2014), i.e. 4 clusters.
• label, bind_comparisons() and select_models() to combine and reuse fits (e.g. a 1 x 1 spatial control).
• Deep species are deeper and narrower by default (BET 260 m, SD 50; BFT 240 m, SD 55), so deep gear clearly targets them while most sets still catch all four species.
• Default scenario: prices favour ALB (years 1-10), BET (11-20), then BET and BFT (21-30), with strong targeting (target species worth 4 times the others, preference = 1.5, HBF_choice_temp = 0.15).
• Stocks are fished harder (q = 3) and start partly fished: new per-species setting B_start (initial biomass / carrying capacity; default 0.6, 0.7, 0.2, 0.9). With r = 0.30, 0.18, 0.25, 0.25 the true trends differ: ALB is fished down, BET declines, BFT recovers and declines again, YFT dips and recovers.
• effort can be one value per year. By default it increases towards 800 sets per year and is cut by 25% in years 21-30.
• Presets: tuna_settings("mild") (moderate targeting) and tuna_settings("stress_test") (strong targeting of lightly fished, unfished stocks, as in the first runs).

New functions

• plot_om(), plot_indices(), plot_diagnostics(), plot_thermocline(), plot_thermocline_effect(), plot_availability(), plot_pcs(), plot_sections(), bind_comparisons(), select_models(), run_replicates(), plot_replicates() and as_set_data().