Finalize model implementation

[cedric]

TODO

general

Fisnish integrating transport

see ticket#124

Integrate sudoe areas (GEREM) in river dataset

see ticket#123

Test parallel processing for caret
problems with transport

see ticket #121

Remaining ticket#120

finish testing model validation

Create a variable mixing distance sea and dam

integrate distance to gibraltar

integrate other metrics

change type of fishing for GT6

ALberto Aguirre Belén Muñoz Carlos Antunes Carlos Fernandez Francisco Hervella Iker Azpiroz Isabel Domingos Josu Iñaki Elso Huarte Lluis Zamora

com 1163 50 30 0 4595 1298 41 11 0
oth 41 891 50 91 651 0 912 46 139

Lucía García Florez Nacho Ferrando Belloch Rafael Miranda Ferreiro

com 200 0 5375
oth 10 182 265

invert distance from gibraltar in the med to avoid independence between alt and mediterranean
silver_eel$dist_from_gibraltar_km[silver_eel$codesea=="M"]<- -silver_eel$dist_from_gibraltar_km[silver_eel$codesea=="M"]

delta

Find best fit => nearly there model structure OK

Try to integrate separate fit for Rhin and Meuse (without excluding them altogether as transported)

Try model using only transported areas.

Check high collinearity for gam using additional height...

Compare results for different height

Run model with altitude or altituderemain

finalize model choice

finalize model graphs => cedric

fit with distance to gibraltar

test cumulated surface downstream

gamma

Find best fit => nearly there model structure OK

Try to integrate separate fit for Rhin and Meuse (without excluding them altogether as transported)

Try model using only transported areas.

finalize model choice

finalize model graphs => Maria

Delta.gamma predictions

Adapt prediction scripts

predict data

write database

integrate new variables in prediction by merging

create maps

caret

Create function for delta model validation

Create function for gamma model validation

[cedric]

best model for delta so far

densCS>0 ~ s(year, by=country, k = 6) +

s(ef_wetted_area, k=3) +

emu +
ef_fishingmethod+
cs_height_10_n.,distanceseakm

# s("cs_height_10_n.,distanceseakm") best model deviance 72.7 AIC 9115

[cedric]

One of the question is how to treat these huge number of data where we expect we have transport. One of the solution is to remove the few points where we have an eel and model with this

# OR (see discussion in ticket 118)
pb_transport <- ddd$transport!="unknown" # 15640
ddd <- ddd[!pb_transport,] #  33658