There two ways to conduct a hindcast based either on observations, i.e. crossvalidation, or model estimates, e.g. as a backtest. While there are three reasons for doing so, namely to find the "best assessment", select and weight models in an ensemble, or condition Operating and Observation Error Models when conducting Management Strategy Evaluation. We review how stock assessment models are currently validated, summarise the use of the hindcast by the RMFOs, and propose how to adopt hindcasting as an objective approach for selecting, screening and weighting hypotheses.
Impact of Allee effects on reference points
Simulation testing of length-based methods, using Operating Models based on life history relationships
Evaluation of LBIs for detecting overfishing and recovery.
Examples of processing MSE outputs
Beverton and Holt Stock Recruitment Relationships with two regimes, variable $SPR_0$ and autocorrelation.
Selected tuna RFMO stock trends relative to BMSY
The RAM DB is used to generate data poor time series and then JABBA configured as a catch-only model is used to estimate final depletion. Results are then compared using Receiver Operator Characteristics
Derivation of priors for population growth rate, based on the Leslie matrix using life history parameters
Summary of the data used to evaluate the catch-only models
A comparison of reference points from the Operating Model and eqsim.
A comparison of reference points estimated by SS3, for different values of steepness and natural mortality, with those estimated by msy-tools
results from MSE
Full set of plots
Summary of data in RAM Legacy database
Example of an empirical control rule developed using Machine Learning
Examples of screening indicators, and using machine learning to conduct MSE
Validation and Screening of Indicators: Sargasso Sea Case Study
2020 ICCAT SUB-COMMITTEE ON ECOSYSTEM MEETING 4-6 May 2020
Use Pareto Fronts to evaluate trade-offs
Comparison of the dynamics of a low k stock (turbot) with a high k stock (brill)
Pareto Plots to show trade-offs
Update on work conducted under MyDas
The cross correlations are plotted for negative lags between recruitment and catch, i.e. for lobster high catches occur 6 years after a strong recruitment, while for sprat they occur after 2 years. The hypothesis is that the performance of a harvest control is determined by the variability in recruitment and the lag between catch and recruitment, rather than the life history parameters and the production function. This can be tested by simulating stocks with different life history parameters and natural mortality vectors and various recruitment scenarios, then comparing the performance of the control rules
Mydas Presentation at WKLife
Summary of the MyDas project. MyDas is funded by the Irish exchequer and EMFF 2014-2020, and the overall aim is to develop and test a range of assessment models and methods to establish Maximum Sustainable Yield (MSY), or proxy MSY reference points across the spectrum of data-limited stocks.
Simulation of long-term constant Fishing Mortality to compare dynamic projections with equilibrium. Fs are for a range of per recruit reference points.
Proportion of stock>Lmega, i.e. save the BOFFs
Indicator based on proportion of stock below Length at Maturity
A large pelagic that was originally at virgin, then expoited increasingly heavily over 100 years
The Kobe Advice Framework
tentative course outline
dl/dt by days at liberty