1. Session 2a Working with more difficult data sets: short gradients
Case study: phytobenthos in very large rivers in Romania (Danube and tributaries)
Source:

2. Phytobenthos in very large rivers in Romania
Adopted intercalibration metric (pICM) as national metric
Adopted mean position of H/G and G/M boundaries as national boundaries
But:
Shallow slope
Low r2 (0.060)
Few sites below (intercalibrated) G/M boundary
Extrapolation leads to very high boundaries
Consider using data from other Member States
X-GIG intercalibration dataset, in this case …

3. Phytobenthos in very large rivers in Romania
Understand the reasons for a weak regression
e.g. look at levels of other potential stressors
Here, phosphorus data are plotted against other water chemistry variables
Dotted lines show present Danube Commission thresholds
No obvious candidates here for an alternative stressor
Kelly, Chiriac, Soare-Minea, Hamchevici & Birk (2018) Hydrobiologia (in press)

4. Phytobentos in very large rivers in Romania
Adding X-GIG data extends the gradient and confirms that Romanian stretches of the Danube generally have relatively low concentrations of dissolved phosphorous.
R2 increases to 0.29
Several options now for using toolkit
Similar options may be possible in other cases
Need to allow for ”country” effects (see IC documentation)
May need to be normalized (see TK_Normalise.xlsx)
(Easy for Romania as they use pICM as national metric)

5. Phytobentos in very large rivers in Romania
Check linearity
Blue line: linear regression
Red line: loess regression (via stat_smooth() command in ggplot2)
See also 01_TKit_check_data.R in toolkit
Apply toolkit approaches …

6. Potential boundary values for Romanian very large rivers (summarised as g L-1 P)
Method
Lower
Centre
Upper
OLS1
139
229
RMA
109
162
OLS2 70 97
Average medians 90
Average quartiles 79 47
75th quartile 66
mismatch
107
Logistic regression 91
140
Decision tree 94
Range of national boundary values (Table 4-4 from CIS Guidance Manual)
25th percentile
70 (as TP)
Median
140 (as TP)
75th percentile
200 (as TP)
Current Romanian P good/moderate standards for large rivers:
g L-1 soluble P
280 – 500 g L-1 total P

7. Dealing with short gradients
How to increase gradient length
Data from similar types in neighbouring MS?
Intercalibration datasets?
Data from other types in your own country?
Boundaries in this case were consistent: may need to normalise these in other cases
(normalisation template included)
Need to account for differences between datasets
See IC documentation for guidance on calculating offsets
Linear mixed effect models with MS (or type) as fixed effect