Limits...
Laurentian Great Lakes phytoplankton and their water quality characteristics, including a diatom-based model for paleoreconstruction of phosphorus.

Reavie ED, Heathcote AJ, Shaw Chraïbi VL - PLoS ONE (2014)

Bottom Line: Further, TP was minimally confounded by other environmental variables, as indicated by the relatively large amount of unique variance in the diatoms explained by TP.We demonstrated the effectiveness of the transfer function by hindcasting TP concentrations using fossil diatom assemblages in a Lake Superior sediment core.The diatom-based transfer function can be used in lake management when retrospective data are needed for tracking long-term degradation, remediation and trajectories.

View Article: PubMed Central - PubMed

Affiliation: Center for Water and the Environment, Natural Resources Research Institute, University of Minnesota Duluth, Duluth, Minnesota, United States of America.

ABSTRACT
Recent shifts in water quality and food web characteristics driven by anthropogenic impacts on the Laurentian Great Lakes warranted an examination of pelagic primary producers as tracers of environmental change. The distributions of the 263 common phytoplankton taxa were related to water quality variables to determine taxon-specific responses that may be useful in indicator models. A detailed checklist of taxa and their environmental optima are provided. Multivariate analyses indicated a strong relationship between total phosphorus (TP) and patterns in the diatom assemblages across the Great Lakes. Of the 118 common diatom taxa, 90 (76%) had a directional response along the TP gradient. We further evaluated a diatom-based transfer function for TP based on the weighted-average abundance of taxa, assuming unimodal distributions along the TP gradient. The r(2) between observed and inferred TP in the training dataset was 0.79. Substantial spatial and environmental autocorrelation within the training set of samples justified the need for further model validation. A randomization procedure indicated that the actual transfer function consistently performed better than functions based on reshuffled environmental data. Further, TP was minimally confounded by other environmental variables, as indicated by the relatively large amount of unique variance in the diatoms explained by TP. We demonstrated the effectiveness of the transfer function by hindcasting TP concentrations using fossil diatom assemblages in a Lake Superior sediment core. Passive, multivariate analysis of the fossil samples against the training set indicated that phosphorus was a strong determinant of historical diatom assemblages, verifying that the transfer function was suited to reconstruct past TP in Lake Superior. Collectively, these results showed that phytoplankton coefficients for water quality can be robust indicators of Great Lakes pelagic condition. The diatom-based transfer function can be used in lake management when retrospective data are needed for tracking long-term degradation, remediation and trajectories.

Show MeSH

Related in: MedlinePlus

Effect of deleting sites on diatom-TP model performance.The effect on transfer function r2 of deleting sites at random (open circles), from the geographical neighborhood of the test site (filled circles), and those that are environmentally similar (crosses) during cross-validation of the Great Lakes diatom training set. The radius of each distance is labelled.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4126776&req=5

pone-0104705-g002: Effect of deleting sites on diatom-TP model performance.The effect on transfer function r2 of deleting sites at random (open circles), from the geographical neighborhood of the test site (filled circles), and those that are environmentally similar (crosses) during cross-validation of the Great Lakes diatom training set. The radius of each distance is labelled.

Mentions: In a check of the effects of sample removal in the training set, random removal of samples had a small effect on apparent model performance (Figure 2). For instance, the observed-inferred r2 decreased from 0.77 for the full model to 0.66 after removal of 80% of samples. However, removal of spatially and environmentally similar sites had a more substantial impact on performance, indicating autocorrelation among sites. Removal of sites that were geographically close to the test site in cross-validation resulted in a sudden drop in the r2, as low as 0.13 after sites within 300 km were removed. This result is not surprising given the known uniqueness of the phytoplankton assemblages in each lake [44], so removal of training set samples in the same lake as the test site undoubtedly had a substantial effect on assemblage analogs. A similarly precipitous drop in performance occurred due to removal of sites that were environmentally similar to the test site, but model degradation was not noticeably different from that due to spatial removal. The relative importance of geographical and environmental neighbors shows how important adjacent sites are for the performance of the transfer function. As explained by Telford and Birks [21], the occurrence of autocorrelation indicates potential problems in a transfer function, and so the modern analog technique (MAT [37]) of inference is not recommended for this model. However, the model may still have predictive power under weighted-averaging scenarios. Based on models derived by random re-mapping of the environmental data over the species map, the actual model performed better 99.8% of the time (Table 3; Figure S1), indicating that the diatom-TP WA transfer function is statistically significant, suggesting that TP can be reconstructed from this training set.


Laurentian Great Lakes phytoplankton and their water quality characteristics, including a diatom-based model for paleoreconstruction of phosphorus.

Reavie ED, Heathcote AJ, Shaw Chraïbi VL - PLoS ONE (2014)

Effect of deleting sites on diatom-TP model performance.The effect on transfer function r2 of deleting sites at random (open circles), from the geographical neighborhood of the test site (filled circles), and those that are environmentally similar (crosses) during cross-validation of the Great Lakes diatom training set. The radius of each distance is labelled.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4126776&req=5

pone-0104705-g002: Effect of deleting sites on diatom-TP model performance.The effect on transfer function r2 of deleting sites at random (open circles), from the geographical neighborhood of the test site (filled circles), and those that are environmentally similar (crosses) during cross-validation of the Great Lakes diatom training set. The radius of each distance is labelled.
Mentions: In a check of the effects of sample removal in the training set, random removal of samples had a small effect on apparent model performance (Figure 2). For instance, the observed-inferred r2 decreased from 0.77 for the full model to 0.66 after removal of 80% of samples. However, removal of spatially and environmentally similar sites had a more substantial impact on performance, indicating autocorrelation among sites. Removal of sites that were geographically close to the test site in cross-validation resulted in a sudden drop in the r2, as low as 0.13 after sites within 300 km were removed. This result is not surprising given the known uniqueness of the phytoplankton assemblages in each lake [44], so removal of training set samples in the same lake as the test site undoubtedly had a substantial effect on assemblage analogs. A similarly precipitous drop in performance occurred due to removal of sites that were environmentally similar to the test site, but model degradation was not noticeably different from that due to spatial removal. The relative importance of geographical and environmental neighbors shows how important adjacent sites are for the performance of the transfer function. As explained by Telford and Birks [21], the occurrence of autocorrelation indicates potential problems in a transfer function, and so the modern analog technique (MAT [37]) of inference is not recommended for this model. However, the model may still have predictive power under weighted-averaging scenarios. Based on models derived by random re-mapping of the environmental data over the species map, the actual model performed better 99.8% of the time (Table 3; Figure S1), indicating that the diatom-TP WA transfer function is statistically significant, suggesting that TP can be reconstructed from this training set.

Bottom Line: Further, TP was minimally confounded by other environmental variables, as indicated by the relatively large amount of unique variance in the diatoms explained by TP.We demonstrated the effectiveness of the transfer function by hindcasting TP concentrations using fossil diatom assemblages in a Lake Superior sediment core.The diatom-based transfer function can be used in lake management when retrospective data are needed for tracking long-term degradation, remediation and trajectories.

View Article: PubMed Central - PubMed

Affiliation: Center for Water and the Environment, Natural Resources Research Institute, University of Minnesota Duluth, Duluth, Minnesota, United States of America.

ABSTRACT
Recent shifts in water quality and food web characteristics driven by anthropogenic impacts on the Laurentian Great Lakes warranted an examination of pelagic primary producers as tracers of environmental change. The distributions of the 263 common phytoplankton taxa were related to water quality variables to determine taxon-specific responses that may be useful in indicator models. A detailed checklist of taxa and their environmental optima are provided. Multivariate analyses indicated a strong relationship between total phosphorus (TP) and patterns in the diatom assemblages across the Great Lakes. Of the 118 common diatom taxa, 90 (76%) had a directional response along the TP gradient. We further evaluated a diatom-based transfer function for TP based on the weighted-average abundance of taxa, assuming unimodal distributions along the TP gradient. The r(2) between observed and inferred TP in the training dataset was 0.79. Substantial spatial and environmental autocorrelation within the training set of samples justified the need for further model validation. A randomization procedure indicated that the actual transfer function consistently performed better than functions based on reshuffled environmental data. Further, TP was minimally confounded by other environmental variables, as indicated by the relatively large amount of unique variance in the diatoms explained by TP. We demonstrated the effectiveness of the transfer function by hindcasting TP concentrations using fossil diatom assemblages in a Lake Superior sediment core. Passive, multivariate analysis of the fossil samples against the training set indicated that phosphorus was a strong determinant of historical diatom assemblages, verifying that the transfer function was suited to reconstruct past TP in Lake Superior. Collectively, these results showed that phytoplankton coefficients for water quality can be robust indicators of Great Lakes pelagic condition. The diatom-based transfer function can be used in lake management when retrospective data are needed for tracking long-term degradation, remediation and trajectories.

Show MeSH
Related in: MedlinePlus