Diatom Intercalibration Workshops

1
Diatom Intercalibration Workshops
4 meetings attended by Annie, Elinor,
Kaarina Dave Ryves guest appearance Issues
covered ID notes for 100 common taxa Clumping
of problem species complexes Excluded
taxa Cross counting exercise
2
Clumping of problem species complexes
Planothidium delicatulum agg. P. delicatulum, P.
haukianum, P. septentrionalis, P.
engelbrechtii Fragilaria elliptica agg.
Staurosira elliptica (12), S. punctiformis,
Staurosirella sopotensis, Pseudostaurosira
perminuta, P. zeillerii, Opephora krumbeinii,
Fragilaria neoelliptica, F. sopotensis, F.
guenter-grassii
3
Complexes continued
Tabularia fasciculata agg. T. affine, T. laevis,
T. fasiculata, T. tabulata, T. waerneii Nitzs
chia frustulum agg. N. amphibia, N. amphibia f.
rostrata, N. frustulum, N. inconspicua, N.
liebtruthii, N. liebtruthii v. major
4
Excluded taxa
Chaetoceros spp. resting spores as identification
to species level not possible. Chaeotoceros spp.
vegetative cells, Skeletonema costatum and
Rhizosolenia spp. spines as preservation in
sediments is uneven.
Cross Counting Exercise All 3 diatomists within
5 of each others counts after the exercise
finished. (We werent after the first attempt!)
Still To Do Authorities and references for MOLTEN
taxa Agreement on data-base images
5
Transfer Function Requirements
Unimodal models The under-lying taxa-response
should predominantly be unimodal Use Detrended
Canonical Correspondence Analysisgradient
lengths greater than 2 have many taxa showing a
unimodal response,less than 1 and no taxa will
have a unimodal reponse Variables to be
reconstructed should explain a significant (and
unique) portion of the variance in the diatom
data CCA (and partial CCA) tested by Monte
Carlo permutation tests are used to determine
this.
6
Dutch training set
30 sites within country Sylt/Seine samples 308
taxa in total 53 taxa with 1 abundance in 2 or
more samples
Thalassiosira pseudonana (42), Delphineis
surirella (29) and Cymatosira belgica (20) were
among the most widespread and abundant taxa.
7
Dutch data set DCA (26 sites 54 taxa)
DCA
Adding Western Danish samples gave only small
improvement DCA gradient 2.3 DCCA on TN gradient
0.8
DCA Gradient length 2.5 (without SOELKK
1.8) DCCA on TN gradient 0.7
8
Dutch diatom data
All sites
SOELKK removed
DCA
DCA
Gradient length 2.5
Gradient length 1.8
9
Dutch data set RDA (SOELKK removed)
Linear version of CCA Interpret in same
way Significance testing with Monte Carlo
permutation tests indicated that only NH4 had a
significant relationship to distribution of
diatom taxa.
10
Dutch training set Performance
PLS on NH4 (DCCA gradient length 1.3)
C1 C2 C3 RMSE 0.08 0.05 0.03 R2 0.67 0.87 0.93 RM
SEP 0.12 0.17 0.18 Boot_R2 0.38 0.15 0.08
11
Danish training set
91 sites 508 taxa in total 152 taxa with 1
abundance or more in 2 or more samples
Opephora mutabilis (22), Fragilaria elliptica
agg. (26) and Thalassiosira sp. 1A (55) were
among the most frequent and abundant taxa.
12
Danish site DCA
Shallow, sandy, saline
Deep
Shallow, brackish
Gradient length 3.1
13
Danish data set CCA (91 sites 152 taxa)
DCA gradient length 3.1 DCCA on TN gradient 1.7
Variables in BLUE are significant in forward
selection
14
Danish Diatom CCA
Nutrient variables (with exception of PO4) are
colinear.
15
Danish predicted vs obs TN
C1 C2 C3 RMSE 0.14 0.11 0.089 R2 0.69 0.81 0.88 R
MSEP 0.18 0.16 0.17 Boot_R2 0.56 0.65 0.66
16
Swedish Data Set (30 sites 75 taxa)
DCA Gradient length 3.7 DCCA on TN gradient 2.5
CCA
17
Swedish predicted vs Obs TN
C1 C2 C3 R2 0.76 0.86 0.94 RMSE 0.085 0.065 0.044
Boot_R2 0.60 0.63 0.65 RMSEP 0.12 0.12 0.12
18
Finnish Data Set (55 samples 73 taxa)
DCA gradient length 2.8 DCCA on TN gradient
2.2 DCCA on TP gradient 1.9
CCA
19
Finnish predicted vs obs TN
Model performance   C1 C2 C3 RMSE 0.094 0.076 0.0
63 R2 0.69 0.79 0.86 RMSEP 0.13 0.13 0.14
Boot_R2 0.48 0.51 0.48
Model performance (fi_46 removed)   C1 C2 C3 RMSE
0.084 0.066 0.055 R2 0.75 0.85 0.89 RMSEP 0.13 0.
12 0.12 Boot_R2 0.56 0.60 0.62
20
Finnish predicted vs obs TP
Model performance   C1 C2 C3 RMSE 0.083 0.073 0.0
66 R2 0.67 0.74 0.79 RMSEP 0.11 0.10 0.11
Boot_R2 0.53 0.59 0.61
21
All Sites DCA
Open deep muddy
Open shallow sandy
Baltic
22
MOLTEN Data Set CCA (202 sites 234 taxa)
DCA gradient length 4.13 DCCA on TN gradient
1.9 DCCA on TP gradient 2.1 DCCA on salinity
gradient 3.0
CCA
CCA with forward selection indicated ALL
variables to be significant
23
Baltic Data Set CCA (75 sites 86 taxa)
East coast Sweden and Finland
DCA gradient length 3.3 DCCA on TN gradient
2.5 DCCA on TP gradient 2.0
CCA
Forward selection indicated ALL variables to be
significant, those in BLUE explained highest
total variance
24
Baltic Predicted vs obs TN
Model performance   C1 C2 C3 RMSE 0.10 0.08 0.07
R2 0.70 0.79 0.84 RMSEP 0.13 0.13 0.14
Boot_R2 0.55 0.57 0.54
25
Baltic predicted vs obs TP
Model performance   C1 C2 C3 RMSE 0.10 0.08 0.07
R2 0.61 0.73 0.77 RMSEP 0.12 0.12 0.13 Boot_R2 0.4
9 0.51 0.52
26
Danish with Swedish west coast (101sites 163
taxa)
DCA gradient length 3.1 DCCA on TN gradient
1.6 DCCA on TP gradient 1.4 DCCA on salinity 2.1
CCA
27
Danish and Swedish West Coast gt 10m CCA (56
sites 121 taxa)
CCA
DCA gradient length 2.9 DCCA on TN gradient
1.7 DCCA on TP gradient 1.6 DCCA on salinity
gradient 2.0