the bug appears on CRAN when updating dependencies

a first model is calibrated over the entire dataset using only geographic distance, then residuals
are predicted locally

It can now use .Call instead of .Fortran. The computation time of the Rn similarity matrix is
greatly reduced by calling a Fortran subroutine and parallelising with OpenMP.

difftime does not return a simple numeric in r-devel

a linear model can be built between a list of predictors (provided by the package user) and the
observed similarity of Rn in order to predict the similarity between a gauged catchment and an
ungauged catchment

Ghosh distance, where many independent distances must be computed, is now parallelised directly in
the FORTRAN code using OpenMP

allow to distinguish a matrix for donor selection (mdist) and a matrix for donor weighting
(similarity)

thanks to a recent update of the elevatr package

to meet CRAN's requirements

fix #1

the sampling of each catchment is now paralellised before computing ghosh distance

add shapefile of the river network to the Blavet dataset