View of /trunk/gadgetMerluza2015/updates.txt

######  May 2014  ########  Changes suggested by WGSOUTH 2014 (Daniel Howell)
params file 
	constrained with narow bounds to avoid risk of stupid solutions. DONE!!!
optinfo.long
	set convergence criteria back to 1e-04 for both SA and H&J (instead of 1e-06). DONE!!!
	increase the number if iterations to 40000
longrun.sh
	reduce the number of runs to 8 (instead of 16)

simplify code in files avoiding unnecesary operations like multipliers, etc...  DONE!!!

ToDo: add explanatory labels to length agregation files e.g. "len4-5" instead of "len1" for the first length class in data files
ToDo: add a new diagnostic plots for length distribution: mean of all data together (obs and modeled)


#######   MAY 2012   ########
Updated with 2011 (except Pt CPUE)
2010 was not reviewed.


######   APRIL 2011 ########
Data review for discards time series. Sp and Pt. Main change is in 2009 2nd quarter from Sp (1.3 kt to 0.25 kt)
Updated with 2010 data. NO sp SOP available. Length distributions are provisional. Raised to total catch.


######  NOVEMBER 2010 #######
WGHMM 2010 - Southern hake review of assessment and projections after errors finding.

Previous to ICES WKSHAKE2 (Lisbon, november 2010) a projection model in R was set to evaluate different hake HCR. The model follows GADGET dynamics. In order to check their performance, ICES WGHMM 2010 results were compared with R projections finding some erros in WGHMM 2010 hake assessment and the corresponding advice.

1- Wrong data red for recruitment series. Lengths 1-3 cm were excluded in the reading procedure driving to underestimation of the recruitment figures. This affects plots and tables that were reviewed and updated.

2- Projections. Short and long term projections were set wrong because the same multiplier was used for landings and discards in every quarter. GADGET internaly saves the explotation pattern (suitability in GADGET) scaling maximum to 1. This means that F for landings and discards in every quarter, in projections, were scaled to have the same maximum F when in fact they have not the same. In order to allow GADGET to scale properly these suitabilities a raising have been done before project the population in the future.

The escaling factor (E), for every quarter and fleet (landings and discards) was set, acording with the GADGET userguide (7.3-Linear fleet), from equation E=C/(Δt*S*N)
where:
< C > is the catches in number at length
< E > is the scaling factor for the fleet and quarter
< Δt > is the length of the timestep in years (0.25 for a quarterly based model)
< N > is the number of stock at length
< S > is the suitability function for a given length
E is equal for all length clases.

Escaled factor performance was checked comparing the resulting F at length in the projection years with F at length in the last year (2009) under Fsq. The figures obtained (F at length) were the same.

This scale factors were used for short and long term projections and new results presented for advice. 

Settings for projections.
Recruitment in 2009 was considered unrealistic and substituted with a geometric mean. In order to kept the 2009 catches, GADGET recalculate 2009 mortality (ages 1-3) changing from 0.74 to 0.77. Since F 2009 was estimated as 0.74, that was used as the best proxi for Fsq, escaled factors were raised accordingly.