Closed vildead closed 3 years ago
see report_01.Rmd
Investigate not just r but also significance of correlations
@vildead this link could be helpful for understading concurvity: https://stat.ethz.ch/R-manual/R-devel/library/mgcv/html/concurvity.html
Note that in our case, as we have discarded smooth surfaces of two time variables (ruled out seasonal features and keep long-term trend) concurvity should be come from high correlation between non-time covariates (i.e. co-linearity in the generalized linear approximation to gam models). Indeed, I've just checked the concurvity values for pike-day model and they are only below 0.5 for the variable "det_therm_deviation_center".
This paper of variable selection in GAMs is worth reading: https://www.sciencedirect.com/science/article/pii/S0167947311000491?via%3Dihub
@rubenrabb01 these are very nice resources! Thank you a lot :)
using caret::findCorrelation() was decided that the thickness and strength are going to be removed from model and replaced by mean temperature gradient
Consider using another variable in models 1) replace thermocline depth by measure of general lake stratification (standard deviation of all temperatures) 2) replace thickness/strength by mean temperature gradient (slope of thermocline curve)