Phytoplankton Classification Tool (Phase 2)
WFD80
June 2007

EXECUTIVE SUMMARY

Background to research

The Environment Agency and SNIFFER have commissioned Phase 2 of this R & D project to develop the method to classify the ecological status of lakes on the basis of phytoplankton. As part of this assessment, metrics need to be developed for phytoplankton community composition.

Objectives of research
Specific objectives for the project were to develop a robust classification, incorporating:
  1. Prediction of reference scores for UK lakes based on phytoplankton composition
  2. Developing criteria for defining the good/moderate boundary
  3. Classifying the ecological status of a water body in to one of five status classes (High/Good/Moderate/Poor/Bad), based on the calculation of an Ecological Quality Ratio (EQR). An EQR being calculated from the relationship between current observed and reference phytoplankton community composition for a site
  4. Determining uncertainty associated with the classification result, based on statistical confidence or probability of class
Key findings and recommendations
Following collation of a dataset of matching phytoplankton and environmental data from 300 lake samples, a multivariate approach to metric development was adopted. CCA was used to develop a species-environment model for phytoplankton, with the main typology variables (alkalinity, altitude, mean depth) included as significant explanatory variables in the model alongside two variables indicative of eutrophication pressure (chlorophyll and TP concentrations). The CCA model indicated strong correlations between the eutrophication pressure gradients (Chlorophyll and TP) and alkalinity. This highlighted the potential problem of developing simple univariate optima of phytoplankton taxa against pressure gradients.

Optima were derived for 112 of the most common phytoplankton taxa (mixed Genus and species level) along both eutrophication gradients (chlorophyll and TP) using reciprocal averaging. Although this was still a univariate approach, the correlative effect with alkalinity is removed later through the calculation of an EQR by taking account of a site’s alkalinity in the reference score. A metric, the Phytoplankton Index of Eutrophication (PIE) was developed which averages the taxa optima at a site, weighted by their abundance (log10 biovolume), to give an Observed PIE Score. Comparisons of various weightings of the community data clearly showed that this metric showed the strongest relationships with both log10TP (r2 = 0.60) and log10Chlorophyll (r2 = 0.64). The metric showed significant relationships with the two pressure gradients for all lake alkalinity types, although was weakest for low alkalinity lakes.

Three approaches were explored for establishing Expected, or Reference, PIE Scores for a site using data from 50 reference lake samples. Firstly, stepwise regression was carried out examining which typology variables explained significant variance in reference site PIE scores. Alkalinity was the only typology variable selected with a significant positive relationship between PIE score and alkalinity. Using this regression model, site-specific reference PIE scores can be predicted for any UK or Irish lake of known alkalinity. Type- specific reference PIE scores were also established with the regression model on the basis of the median alkalinity measures in the UK and Irish lake dataset. It is recommended that, if possible, modelled site-specific reference conditions are adopted as they are ecologically more appropriate and correlate better with observed PIE scores at reference lakes than type-specific reference scores.

EQRs were calculated from the ratio of Observed to Expected PIE scores, which were then transformed, to produce an EQR ranging from 0 to 1. The High/Good (H/G) boundary was determined from the lower 25% of reference site EQRs. To derive the remaining boundaries, phytoplankton taxa were classified as “positive” (low eutrophication pressure) or “negative” (high eutrophication pressure) indicators of eutrophication pressure, by examining their optima and tolerances in a constrained CCA model. The % biovolume of positive and negative taxa was calculated for each sample and polynomial regression analysis was carried out to examine their relationships with EQR. The crossover point in these two relationships was chosen to represent the Good/Moderate (G/M) status class boundary. The 75% of residuals in the two equations were then used to identify the Moderate/Poor (M/P) status class boundary and the remaining Poor/Bad (P/B) boundary was derived from a division of the remaining EQR scale.

The PIE metric was applied to all UK and Irish lake phytoplankton samples. Observed and Expected PIE scores and resultant EQRs are given in Appendix 2. The mean variance in EQR scores between samples from different months for the same lake was relatively low. The relationship between observed EQR and variance in EQR was used to estimate the confidence in classification of results and mis-classification rate for a given EQR.

A number of sources of uncertainty, or error, in EQRs and associated constituent measures (observed PIE scores, sample biovolume, number of taxa) were examined, focusing on sample processing errors due to the combination of sub-sample and counter analytical error. The resultant estimates of the average within-site sampling/processing variability can also be used, in software such as STARBUGS, to derive estimates of uncertainty in assigning water bodies to a WFD ecological status class. These results are, however, preliminary, based on limited data and unstructured sampling. The effects of larger scale within-lake spatial variability on EQRs also needs to be examined, with estimates in more lakes over a wide range of ecological qualities using nested replicate samples and replicate sub-samples before these results can be considered reliable.

Sources of variation in chlorophyll data due to within-lake spatial variation and laboratory analytical variability were examined and compared with previous work on temporal variability. The largest source of variability in chlorophyll concentrations within a lake is temporal, but if sampling is carried out monthly then much of the seasonal variability is eliminated and becomes smaller than the estimated variance between replicate samples taken on the same day. If sampling is restricted to the outflow of lakes, the least variable location, then the estimate of replicate sampling variance becomes less than laboratory variance.

In terms of minimising uncertainty in both chlorophyll and composition classifications, it is recommended that sampling is carried out at a single specific location within a site, where the location is representative of the lake as a whole (outflow or centre of lake). Samples should be taken from July to September at a monthly frequency.

Key words: phytoplankton, WFD, classification, lake, ecological status

Copies of this report are available from the Foundation, in electronic format on CDRom at £20.00 + VAT or hard copy at £35.00, less 20% to FWR members.

N.B. The report is available for download from the SNIFFER Website