jeremieboudreault/fish_habitat_modelling_functional_regression

Scientific research modelling the fish physico-thermal habitat with a functional regression model.

Fish habitat modelling using functional regression 🐟

• A scientific research by Jeremie Boudreault, Fateh Chebana, André St-Hilaire and Normand Bergeron
• This project was part of my master degree in water sciences at Institut National de la Recherche Scientifique.
• All codes and data are made available here under the Creative Common License .
• The research article was published in Journal of Ecohydraulics on 09/02/2021.
• Questions regarding the code or the data should be sent to [email protected].

Data

Data are from field survey that have been conducted during summer 2017 on the Sainte-Marguerite river (SMR) :

• data/field/* : contains the raw .xlsx file filled after each day of field work
• data/* : contains the cleaned and transformed datasets

R scripts

Scripts are all from Jeremie Boudreault. They use the R package mgcv to fit generalized additive models (GAM) and of FDboost to fit functional regression models (FRM) :

• R/Data_initial_cleaning.R : code to clean the field data spreadsheets and produce more adapted datasets
• R/Data_salmons_lengths.R: code to convert the salmon lengths to number of fry and parr
• R/Data_per_site.R : code to produce the observations at each site (mean value or functional observations)
• R/GAMs_all.R : code to fit several types of GLM/GAM on the data using the mgcv package
• R/GAMs_best.R : among all models, do variable selection to find the best GAM models and save them to out/models
• R/GAMs_predictions : code to calculate the leave-one-out predictions for the GAMs
• R/FRMs_all.R : code to fit several types of FRM on the data using the FDboost package
• R/FRMs_best.R : among all models, select the best FRM models and save them to out/models
• R/FRMs_predictions : code to calculate the leave-one-out predictions for the FRMs
• R/Models_coefficients.R : code to extract the coefficients of the best models
• R/Models_performance.R : compare the results between GAMs and FRMs

Results

A folder for the results at each part of the coding process :

• out/data visualisation/* : raw data visualisation and tables
• out/models/* : fitted final models
• out/coefficients/* : coefficients of the models
• out/predictions/* : leave-one-out predictions and goodness-of-fit