First check the histograms of the transformed predictor variables and pair plots for co-linearity among predictors.
Density plots of original and transformed preditors
Pairs plot of transformed predictor variables
Phormidium cover through time
Relationship between Phormidium and predictors by river
Explore the relationship of Phormidium cover with all predictor variables by river.
PCA of environmental variables on each river
Boxplot of environmental variables per river
Fit a GAM model using cubic spline for each predictor variable and using Gaussian (normal) errors with a log link. The model also has an auto-correlation structure of order 1.
The formula for the model is:
## cyanocover + 0.01 ~ cs(week) + cs(sqDIN) + cs(logDRP) + cs(spotTemp) +
## cs(logFlow) + cs(Cond) + cs(logSilt) + SiteName
Estimate | Std. Error | t value | Pr(>|t|) | |
---|---|---|---|---|
(Intercept) | -4.68 | 1.09 | -4.29 | 0.00002 |
cs(week) | -0.04 | 0.01 | -7.14 | 0.00000 |
cs(sqDIN) | 2.20 | 0.50 | 4.38 | 0.00001 |
cs(logDRP) | -0.31 | 0.17 | -1.85 | 0.06537 |
cs(spotTemp) | 0.10 | 0.03 | 3.84 | 0.00014 |
cs(logFlow) | -0.44 | 0.09 | -4.83 | 0.00000 |
cs(Cond) | 0.01 | 0.00 | 3.27 | 0.00115 |
cs(logSilt) | 0.25 | 0.05 | 4.84 | 0.00000 |
Partial effects GAM plots of the effect predictor variables on Phormidium cover. The solid line is a cubic spline fitted (+/- S.E.) and the points a re the model residuals.
## This is mgcv 1.8-11. For overview type 'help("mgcv-package")'.
Two-way interaction surface plots
Standarised residuals for the fitted GAM model using log normal errors
R-squared for the model
## [1] 0.5617614
## Loading required package: survival
## Loading required package: lattice
## Loaded gbm 2.1.1
## var rel.inf
## SiteName SiteName 26.239420
## logFlow logFlow 16.168612
## sqDIN sqDIN 15.669461
## Cond Cond 10.126061
## spotTemp spotTemp 8.970681
## logDRP logDRP 7.985903
## logSilt logSilt 7.697782
## week week 7.142080
Partial plots boosted tree
2-way plots
2-way plots
2-way plots
2-way plots