Multivariate analyses of phytoplankton pigment fluorescence from a

Subscriber access provided by CORNELL UNIVERSITY LIBRARY

Article

Multivariate analyses of phytoplankton pigment fluorescence from a freshwater river network Ruchi Bhattacharya, and Christopher L. Osburn Environ. Sci. Technol., Just Accepted Manuscript • Publication Date (Web): 16 May 2017 Downloaded from http://pubs.acs.org on May 20, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 28

Environmental Science & Technology

1

Multivariate analyses of phytoplankton pigment fluorescence from a freshwater river

2

network

3

Ruchi Bhattacharya*, Christopher L. Osburn

4

Department of Marine, Earth, and Atmospheric Sciences, North Carolina State University,

5

Raleigh, NC, USA 27606

6

*Corresponding author info: [email protected]

7 8 9 10 11 12 13 14 15 16 17

1

ACS Paragon Plus Environment


18

Abstract

19

Monitoring phytoplankton classes in river networks is critical to understand phytoplankton

20

dynamics, and to predict the ecosystem response to changing land-use and seasons. Applicability

21

of phytoplankton fluorescence as a quick and effective ecological monitoring approach is

22

relatively unexplored in freshwater ecosystems. We used multivariate analyses of fluorescence

23

from pigment extracted in 90% acetone to assess the variability in phytoplankton classes,

24

herbivory and organic matter quality in a freshwater river network. Four models developed by

25

Parallel Factor Analysis (PARAFAC) of fluorescence excitation and emission matrices identified

26

6 components: Model 1 (Pheophytin-A, Chlorophyll-A), Model 2 (Chl-B, Chl-C), Model 3 (Phe-

27

B) and Model 4 (Phe-C). Redundancy analyses revealed that in summer, urban and agricultural

28

streams were abundant in chlorophylls, fresh organic matter and organic nitrogen; whereas in

29

winter, streams were high in phaeopigments. A slow moving, light-limited wetland stream was

30

an exception as high phaeopigment abundance was observed in both seasons. The PARAFAC

31

components were used to develop a partial least square regression-based model (r2=0.53;

32

NSE=0.5; n=147) that successfully predicted Chl-A concentrations from an external sub-set of

33

river water samples (r2=0.41; p

165

95% of the variation in the data. An efficient alternating least squares approach is used to

166

minimize residuals between measured and modeled EEMs and usually produces robust models

167

[45; 46]

168

. The PARAFAC Models were validated by first dividing the data array into two halves,

169

and then using Tucker’s congruence coefficient (TCC, set to 95% similarity) to test separate

170

PARAFAC models fit to each half (split-half validation)[24]. Random initialization was next used

171

on 10 models to find the best-fit model having the lowest sum of squared error [46]. Residuals

172

were determined as the difference between the measured and the modeled EEMs [47]. In case

173

distinct residual peaks with intensities of few orders of magnitude lower than the measured

174

PARAFAC components were observed at a specific excitation/emission pair, then residual EEMs

175

were modeled using in-house MATLAB functions built around the DOMFluor toolbox [24]. The

176

modeled residuals were also split-half validated using TCC and random initialization methods

177

described above.

178

The principle component analysis (PCA) - a linear response ordination method [37] was

179

used to explore the spatio-temporal variability in PARAFAC components. The PCA was

180

conducted on the maximum fluorescence intensity (FMax) of each PARAFAC component, 9



181

typically used as a qualitative measure of the strength/presence of each component [44; 45]. The

182

PCA was conducted on the water quality parameters to explore their spatio-temporal variability

183

(Table S2). Prior to conducting PCA the distribution of PARAFAC component Fmax and water

184

quality parameters were checked by constructing q-q plots and histograms; parameters not-

185

normally distributed were centered and log transformed log10 (n+10). Linear regressions were

186

conducted for outlier identification (high leverage) and removal. The PCA was also used to

187

explore the inter-variable relationships by calculating correlation coefficients between the

188

variables [48]. Major gradients in the dataset were identified by calculating correlation

189

coefficients for the variables and each of the ordination axes [48]. Next, the relation between the

190

water quality parameters and the PARAFAC component FMax were determined by redundancy

191

analysis (RDA). In order to identify a minimum subset of variables that significantly explain

192

variation in the PARAFAC components, we first assessed the multi-collinearity in the data set by

193

checking for high variance inflation factor (VIF > 10). Second, the redundant variables were

194

removed through step-wise regression with Monte Carlo permutation tests [49]. Other than the

195

statistical significance of each variable, their relative ecological importance in explaining

196

phytoplankton abundance was also considered prior to variable removal. We further conducted

197

variance partitioning to explore the percent variance uniquely explained by each variable. The

198

PCA and RDA were conducted using “vegan 2.3-1” package in R statistical software [50].

199

Partial least squares regression (PLS) [26] was conducted on PARAFAC component Fmax

200

from using “rioja” package in R statistical software [51]. The PLS model was calibrated with the

201

sample Subset 1 (n = 147), that was analyzed for pigment fluorescence, Chl-A concentrations

202

and other water quality parameters. The water quality parameter that individually explained the 10


Page 10 of 28

Page 11 of 28


203

most variance was used for the development of the calibration model. Leave-one-out cross

204

validation procedure (Jack-knife) was used to validate the error predictions [52, 53]. The model

205

efficiency was evaluated by r2 and Nash-Sutcliffe Efficiency value (NSE). The PLS based

206

calibration model was externally validated using the PARAFAC component Fmax values from a

207

second sample subset (Subset 2, n = 75). The sample subset 2 was analyzed for pigment

208

fluorescence, and Chl-A, but the associated water quality parameters were not available.

209

However, we ensured that both subset 1 and 2 were representative of the environmental and

210

seasonal gradients.

211

11



212

Results

213

PARAFAC Modeling

214 215 216 217 218

Figure 1 PARAFAC model component excitation and emission spectral loading (shown by the solid line). The excitation and emission loadings of a split-half validated component (shown by the dashed line) is also plotted with each PARAFAC component. An overlap between the PARAFAC and split-half validated component indicates good model performance.

219

A total of four PARAFAC models (see Figures S2 and S3) were developed using two sets of

220

data: 1) the chlorophyll pigment EEMs and 2) the phaeopigment EEMs. The first PARAFAC

221

model (Model1) was developed from the chlorophyll pigment EEMs and resulted in two

222

components that were validated by split-half analysis. The excitation (Ex) and emission (Em)

223

spectral loadings of the two-modeled components resembled Phe-A (Model1-Phe-A) and Chl-A 12


Page 12 of 28

Page 13 of 28


224

(M1-Chl-A), respectively (Figure 1A-B, Table 1, Figure S2). Examination of the residuals EEMs

225

from Model 1 revealed two low intensity, distinct fluorescence peaks. The second PARAFAC

226

model (Model 2) was developed using these residual EEMs and two validsated components were

227

identified as Chl-B (M2-Chl-B) and Chl-C (M2-Chl-C) (Figure 1C-D, Table 1). Then, the

228

phaeopigment EEMs were used to develop a two-component PARAFAC model (Model 3), and

229

the validated components corresponded with Phe-A (M3-Phe-A; not shown in Figure 1) and Phe-

230

B (M3-Phe-B) (Figure 1E; Table 1). The model 3 residuals also revealed a distinct low intensity

231

fluorescence peak. Thus, the final, and fourth model (Model 4) was developed from the residual

232

EEMs from Model 3, and the validated component was identified as Phe-C (M4-Phe-C) (Figure

233

1F, Table 1, Figure S2).

234

Spatial and temporal variability in PARAFAC components

235

PCA conducted on the PARAFAC components Fmax (log10 (n+10)) showed that PCA

236

axis 1 and 2 explained 47.7% and 32.9% of the variance, respectively (Figure 2A). The M3-Phe-

237

B and M4-Phe-C vectors were significantly and positively correlated to each other (r2 = 0.4, p

Multivariate analyses of phytoplankton pigment fluorescence from a

Recommend Documents