Map of the maximum Phormidium cover for each segment

Correlation among preditor variables and Phormidium cover. Variables are ordered according to average clusters.

Biplot of the pca conducted on predictor variables. Points are coloured relative to three cover categories: absent, <20% and >20%

Predictor variables distribution

Relationship between predictor variables and Phormidium over

Boosted regression tree model

## Loading required package: survival
## Loading required package: lattice
## Loading required package: splines
## Loading required package: parallel
## Loaded gbm 2.1.1

##                                                   var    rel.inf
## log(SegLowFlow + 0.1)           log(SegLowFlow + 0.1) 21.1227541
## ReachHab                                     ReachHab 17.7967354
## USPhosporus                               USPhosporus 11.2944073
## ReachSed                                     ReachSed  8.6985730
## log(SegSlope + 0.1)               log(SegSlope + 0.1)  8.0545036
## sqrt(SegFlowVariability)     sqrt(SegFlowVariability)  5.7194902
## USHardness                                 USHardness  5.3201566
## factor(LANDCOVER)                   factor(LANDCOVER)  4.0878343
## log(SegRipShade + 0.1)         log(SegRipShade + 0.1)  3.4482866
## SegJanAirT                                 SegJanAirT  2.9807707
## USAvgSlope                                 USAvgSlope  2.7372432
## USCalcium                                   USCalcium  2.4666538
## log(T2PastoralHeavy + 0.1) log(T2PastoralHeavy + 0.1)  2.2077788
## log(SegCluesN + 0.1)             log(SegCluesN + 0.1)  1.8500355
## ORDER                                           ORDER  0.7651367
## factor(UpstreamDamEffect)   factor(UpstreamDamEffect)  0.7327511
## sqrt(USDaysRain)                     sqrt(USDaysRain)  0.7168890

GAM model

Fit a log-normal GAM model using a cubic spline for non-collinear predictor variables identified as potential drivers based on data explorartion and GBM model. Then significant variables are selected using a step-wise procedure based on GAIC.

The formula for the model is:

## cyano + 0.01 ~ cs(logFlow) + cs(ReachSed) + cs(ReachHab) + cs(logPastoral) + 
##     cs(USPhosporus) + cs(sqrt(SegFlowVariability)) + cs(SegJanAirT) + 
##     cs(logN)
## cyano + 0.01 ~ cs(logFlow) + cs(ReachSed) + cs(ReachHab) + cs(logPastoral) + 
##     cs(SegJanAirT)
Results of the GAMM for Phormidium percentage cover using a log normal error distributions
Estimate Std. Error t value Pr(>|t|)
(Intercept) -12.43 2.36 -5.26 0.00000
cs(logFlow) 0.29 0.08 3.61 0.00034
cs(ReachSed) 0.53 0.22 2.40 0.01656
cs(ReachHab) 1.99 0.53 3.78 0.00017
cs(logPastoral) 0.34 0.07 4.60 0.00001
cs(SegJanAirT) 0.15 0.09 1.56 0.11886