Multiple Pressures nutrient boundary setting Geoff Phillips & (Sebastian Birk FP7 MARS) Slides 2 – 3 were kindly provided by Sebastian Birk from the EU MARS project
Multiple stressors: management implications What is the (combined) effect of stressors? Dominance (1 + 0 = 1) or Additive (1 + 1 = 2) Interactions “Ecological surprises“ Synergistic (1 + 1 = 3) (e.g. Nutrients & Temperature) Requires, for instance, more protective nutrient standards. Antagonistic (1 + 1 = 1) (e.g. Nutrients & Hydropeaking) Requires combined stressor mitigation to avoid worsening.
Paired-stressor effects: interactions LAKES (n= 58 cases) RIVERS (n = 122 cases) * Birk et al., in prep. Share of interactions across lakes and rivers
Dealing with multiple pressures remains a challenge Consider what variables might influence biological status other than nutrients Turbidity suspended sediment Humic substances Shade (rivers) Grazing Toxic substances Sediment deposition Hydromorphological change Boosted regression trees (BRT) Allow us to identify important variables and rank them
Boosted regression to predict Macrophyte EQR Sediment accumulation < -0.19 High Low Combination of 3500 individual regression trees, using boosting to minimise the loss of model performance by adding new tree models that best reduces overall model deviance. Final model is a linear combination of all trees Alkalinity >63 mgCaCO3/l Nitrate <1.7 mg/l Alkalinity >75 mgCaCO3/l Nitrate <0.8 mg/l Optimum model P EQR <0.69 0.77 0.60 0.77 0.78 0.47 0.57 Increase plant deviation from reference Overall model explains 63% of total deviance Sol P EQR is the most important variable (22%) Alkalinity (19)% Sediment accumulation (13%) Nitrate(10%) Indicators of a degraded lowland stream
Partial dependence plots showing predicted Macrophyte EQR for each environmental variable Data England & Wales GAM added to indicate trajectory of change
How does this influence boundary setting? Simplify by looking at 2 pressure combinations For river phytobenthos nutrients (Soluble P & nitrate or total oxidised N) explain most of the variation, but BRT analysis shows fine sediment load is also important How does fine sediment load interact with nutrients? Investigate this using multiple regression models We can specify additive models EQR = c + ax1 + bx2 + error or a model that includes an interaction EQR = c + ax1 + bx2 + d(x1x2) + error
Plot the effect of one stressor while keeping the other stressor constant Here the effects of sediment deposition and nitrate on status of phytobenthos is Antagonistic The response of EQR to nitrate is lower when sediment deposition is high The response of EQR to sediment deposition is lower when nitrate is high Without interaction the slope of the lines would be identical, but the level of the lines would be different Red – 2nd stressor high, green – 2nd stressor medium, blue – 2nd stressor low
River phytobenthos EQR The effect can be visualised by plotting the 2 pressure with the EQR as a response surface (identified by contours (blue line high/good boundary, green line good/moderate, yellow line moderate poor) Solid lines – interaction model Dotted line – no interaction River phytobenthos EQR For some sites reducing either pressure in isolation could reach the good/mod boundary if there was no interaction, but with interaction a much bigger reduction would be required Reducing both provides the most likely chance of reaching the boundary.
Summary Toolkit provides relatively simple approaches to setting boundaries In some situations simple models may not work Combining and dividing data sets may help (e.g. harmonise other pressure levels) Value judgements are needed to make use of other approaches – quantile modelling More use may need to be made of advanced techniques BRTs, multiple regression models more results needed from 2 pressure combinations Should be a general comparability of boundary values across similar EU water bodies Uncertainty and limitations of modelling allow for relatively broad range of boundaries Compare results with other MS Greater emphasis on ecological understanding, general rules, less emphasis on analysis of specific (small) data sets