SWEPACS: a running water prediction and classification system using benthic macroinvertebrates

1
SWEPACS a running water prediction and
classification system using benthic
macroinvertebrates

• Leonard Sandin
• Department of Environmental Assessment
• Swedish University of Agricultural Sciences

2
Outline

• RIVPACS background
• General model development
• Swedish data
• Swedish model development
• Results
• Quality bands
• Summary

3
RIVPACS background

• River InVertebrate Prediction And Classification
  System
• Started 1977 in the UK (Institute of Freshwater
  Ecology)
• Biological classification of unpolluted
  running-water sites (macroinvertebrates)
• Assessment of whether the fauna at a site could
  be predicted using physical and chemical features

4
General model development

• 1. classification of reference sites into
  biologically similar groups
• - using some cluster algorithm (e.g., TWINSPAN,
  k-means)

5

• 2. development of a discriminant model with data
  collected from reference sites to estimate the
  probabilities of a new site belonging to each of
  the site groups defined in (1)

Moss (2000)
6

• 3. calculation of the probabilities of all taxa
  in the regional taxa pool occurring within each
  reference site group

7

• 4. calculation of the probabilities that each
  taxon will occur at a new site based on (2) and
  (3)
• - using Discriminant Function Analysis

• using environmental data a "new" stream is
  placed in group I with 80 probability
• in group II with 20 probability

8
- and the weighted probability of taxa occurrence
9

• 5. summation of the estimated probabilities of
  capture of all taxa to estimate the number of
  taxa expected (E) at a new site

10

• 6. calculation of Observed/Expected
• ratio

11
Swedish data

• National stream survey 1995
• 696 streams sampled
• Macroinvertebrates
• gt 100 environmental variables
• Substratum, chemistry, vegetation, riparian zone,
  catchment characteristics, geographical variables

12
Reference or least impaired condition

• 365 samples included in model building
• No acidification (exceedence of S Critical Load)
• No eutrophication (lt 20 arable land in
  catchment)
• No point source pollution
• Sampling error removed
• No liming

13
Illies biogeographical regions

• arctic/alpine ecoregion region 20
• northern and middle boreal ecoregions
  region 22
• southern, boreo-nemoral and nemoral ecoregions
  region 14

14
Building the models

• Classification (TWINSPAN)
• Discriminant Function Analysis - choosing
  explanatory environmental variables
• Predicting taxa occurrence
• Calculating Observed/Expected ratios

15
The three models

• 30 - 186 sites
• 72 - 173 taxa
• 4 - 6 TWINSPAN groups
• 7 - 8 predictor environmental variables
• Latitude, longitude, altitude in all models

16
Region 14 (southern Sweden)

• Five groups
• Seven environmental variables
• 63.5 into correct group
• 54.0 using cross-validation

17
Environmental variables

• Latitude, longitude, altitude in all models
• Stream velocity, depth, substratum
• Vegetation in the stream
• Catchment area, vegetation in catchment
• Above/Below highest coastline of the last
  glaciation

18
Inclusion threshold (probability)

• The probability level where an expected taxon is
  included in calculating the O/E ratio
• RIVPACS 0 probability
• AUSRIVAS 50 probability
• SWEPACS 25 probability
• Very common and very rare are not very good
  indicators - more is not always better!

19
Inclusion threshold - results

• Models for region 14 and 22 had similar taxon
  richness
• Lower taxon richness in region 20 (arctic/alpine)
  complex

20
Inclusion thresholds - results

• Number of observed and expected taxa increased as
  the inclusion threshold decreased

21
Validation of models

• Internal validation - model data set
• External validation - independent reference sites
• Discriminatory power - impacted sites

22
Validation of models

• O/E ratios
• Region 14
• Internal
• External

23
Validation of models

• Internal and external (5 ) reference tests (3 -
  10 sites)
• p gt 0.05 for all pairs (regions) Internal -
  external

24
Test of perturbed sites

• Region 14
• O/E number of taxa
• gt 100 mg tot P/l
• Model 0.981
• External 0.798
• Affected 0.782
• p lt 0.01

25
Test using ASPT

• Region 14
• O/E ASPT
• gt 100 mg tot P/l
• Model 0.945
• External 0.997
• Affected 0.841
• p lt 0.01

26
Summary

• Model is built in four steps
• Reference or least impaired conditions
• Illies biogeographical regions
• 4 - 6 TWINSPAN groups
• 7 - 8 predictor environmental variables
• Latitude, longitude, altitude in all models
• Validation of models (internal, external,
  perturbed)
• O/E ratios e.g., number of taxa, indices