Biologically based urban response models for the South

Atlantic gulf and Tennessee River basins

T.F. Cuffney, E.M. Giddings, and M.B. Gregory

North Carolina Water Science Center

Objectives:

1. Derive invertebrate response models for Atlanta, GA, Birmingham, AL, and Raleigh, NC.

2. Test and verify urban response models (ATL, BIR, RAL).

a. Comparability among models

b. Applicability of urban models to sites throughout the Southeast

– can models developed in one urban area be applied to other urban areas?

– can urban models be used to predict regional patterns in water-quality?

Denver

Portland

DallasBirmingham

Atlanta

Raleigh

Milwaukee

Boston

Salt Lake City

South Atlantic gulf and Tennessee river basins

Seattle

Sacramento

National Water-Quality Assessment (NAWQA) Program 11 Urban Stream Studies

Biological sampling sites

Birmingham Atlanta

Raleigh

0 90 180Kilometers

360

Raleigh Urban Study AreaPiedmont

Urban models: Site Characteristics

Abbrev Core urban areaDominant ecoregion

Habitat sampled Basin area (km2)

ATL Atlanta, GA Piedmont Woody snags

43-146

RAL Raleigh, NC Piedmont Riffles 5-82

BIR Birmingham, AL Central Appalachian Ridges and Valleys

Riffles 5-66

Design of urban studies

1. Urban intensity gradient defined by amultimetric index based on land-cover, population, infrastructure and socioeconomic factors associated with population density.

2. Uniform environmental setting to keep natural environmental factors as constant as possible.

% Impervious and Urban Intensity index (UII) -- Raleigh

y = -0.0547x2 + 4.1944x + 2.1458

R2 = 0.78

0

10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50

% impervious surface

Urb

an in

ten

sity

(U

II)

Urban models

• Multivariate (ordination) models:– Measure ecological distances among sites

along an environmental gradient (urban) derived from differences in invertebrate assemblages.

• Multiple regression models:– Relate invertebrate assemblage metrics

(EPT) to landscape variables: land cover, land use, census variables, infrastructure.

Multivariate (ordination) models

0

10

20

30

40

50

60

70

80

90

100

-1 -0.5 0 0.5 1 1.5

Ordination site scores (axis 1)

urb

an

in

ten

sit

y

-1 -0.5 0 0.5 1 1.5

Ordination site scores (Axis 1)

0

10

20

30

40

50

60

70

80

90

100

Urb

an

in

ten

sit

y (

UII

)

Ordinations scores & UII

Atlanta

Atlanta urban model (Piedmont)

y = -0.0169x + 1.6651R2 = 0.81

0.0

0.5

1.0

1.5

2.0

2.5

0 20 40 60 80 100

Urban intensity index (UII)

Ord

inat

ion

sco

re (

axis

1)

RTH Inverts: urban model

10% impervious surface

Ordination site scores vs. UII

Urban area a b R2

Atlanta 1.67 -0.017 0.81

Birmingham 1.84 -0.020 0.74

Raleigh 1.75 -0.015 0.74

Y = a + bXY = ordination site scoreX = urban intensity UII

Ordination model: prediction

k k ikkiki yyx /*ATL Site

score

Species abundance from ATL.

Optima for species K: BIR model

Predicting ATL using BIR model

Eigenvalue from BIR model

Atlanta predicted by Birmingham and Raleigh models

BIRy = 0.5376x + 0.1482

R2 = 0.84

RALy = 0.6987x + 0.1365

R2 = 0.91

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

0 0.5 1 1.5 2

Axis 1 site score

Pre

dic

ted

sit

e sc

ore

Observed Axis 1 site score

Woody snags

Riffles

Piedmont Ridge & Valley

Piedmont

Birmingham predicted by Atlanta and Raleigh models

ATLy = 0.6261x - 0.0807

R2 = 0.69

RALy = 0.4056x + 0.0461

R2 = 0.54

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

0 0.5 1 1.5 2

Axis 1 site score

Pre

dic

ted

sit

e sc

ore

Observed Axis 1 site score

Raleigh predicted by Atlanta and Birmingham models

ATLy = 0.6317x - 0.0132

R2 = 0.91

BIRy = 0.3397x + 0.0688

R2 = 0.82

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0 0.5 1 1.5 2

Axis 1 site score

Pre

dic

ted

sit

e sc

ore

Observed Axis 1 site score

Inter-model comparisons: ordination models

1. Ordination models from Atlanta, Birmingham, and Raleigh are interchangeable.

2. Ordination models developed in one urban area can be applied to other closely allied urban areas.

Regression models

Comparison among ATL, BIR, RAL

Multiple Regression Models

• EPTr: richness of pollution intolerant forms (mayflies, stoneflies, caddisflies)

• Xi: – Land-use: (e.g., forest, developed, ag/urban

grass)• % basin area• % stream buffer

– Population density– Road density

Model: EPTr = f(Xi)

EPTr urban model variables

% basin area Census Infrastructure

City Devel Forest Grass Pop den Road den

RAL 0.11 0.01 -1.44

ATL 0.27 0.33 -3.19

BIR -1.13

% stream buffer

City Devel Forest Grass R2 P

RAL 0.13 0.47 0.81 <0.001

ATL 0.45 0.62 <0.001

BIR 0.59 <0.001

Correspondence between predicted and observed EPT richness based

on multiple regression models

ATL modeled using: BIR modeled using:ATL BIR RAL ATL BIR RAL

b 0.67 0.30 0.30 b 0.64 0.61 0.94

r2 0.66 0.51 0.33 r2 0.19 0.59 0.44

p <0.001 <0.001 <0.001 p 0.011 <0.001 <0.001

Expected = a + b * Observed

Predicted EPT richness based on multiple regression models

RAL modeled using:

ATL BIR RAL

b 0.49 0.49 0.47

r2 0.55 0.62 0.61

p <0.001 <0.001 <0.001

Expected = a + b * Observed

Inter-model comparisons: multiple regression models

1. Variability associated with estimating explanatory variables reduces the overall “fit” of the regression models.

2. Multiple regression models are not as interchangeable among urban study areas as are the ordination models.

Extrapolating urban models to South Atlantic gulf and

Tennessee (SAGT) River basins

Can urban models be used to predict regional patterns in water-quality?

Site Characteristics

Small basins (< 160 km2)

Dominant ecoregion Sites UII range

Piedmont 8 3-80

Interior Plateau 35 3-32

Mid Atlantic Coastal Plain 14 12-33

Southeastern Plains 48 4-81

Central Appalachian Ridges and Valleys 1 11

Southwestern Appalachians 2 9-16

Ordination models

Observed vs. expected site scores

Interior Plateau: Ordination

BIR:y = 0.38x + 0.20

R2 = 0.36

RALy = 0.46x + 0.09

R2 = 0.73

ATL:y = 0.48x + 0.39

R2 = 0.49

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

0 0.5 1 1.5 2 2.5

Observed site score

Pre

dic

ted

sit

e s

co

re

ATL

BIR

RAL

MODEL:

UII: 3-32Sites: 35

Southeastern Plains: Ordination

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

0 1 2 3 4 5

Expected site score

Pre

dic

ted

sit

e sc

ore

ATL

BIR

RAL

R2

ATL: 0.002BIR: 0.003RAL: 0.074

UII: 4-81Sites: 48

Multiple regression models

Observed vs. expected EPT richness

Interior Plateau (< 160 km2)

0

5

10

15

20

25

30

0 5 10 15 20 25

EPT richness

Pre

dic

ted

EP

T r

ich

ness

ATL

RAL

BIR

R2

ATL: 0.001BIR: 0.21RAL: 0.20

EPT: multiple regression model

UII: 3-32Sites: 35

Southeastern Plains (< 160 km2)

0

5

10

15

20

25

30

35

40

0 2 4 6 8 10 12 14 16 18 20

EPT richness

Pre

dic

ted

EP

T r

ich

ne

ss

ATL

RAL

BIR R2

ATL: 0.02BIR: 0.04RAL: 0.06

EPT: multiple regression model

UII: 4-81Sites: 48

Tolerances to urbanization differ from tolerances to general pollution

0

1

2

3

4

5

6

7

0 2 4 6 8 10

USEPA and NCDENR tolerance values

Urb

an

to

lera

nce v

alu

es

(U

II)

General pollution tolerance values: USEPA and NCDENR

Urb

an p

oll

uti

on

to

lera

nce

val

ues

: N

AW

QA

(U

II)

Species tolerances

Conclusions1. Invertebrate responses to urbanization provide

very useful predictive models.

2. Ordination models were generally better predictors of responses than were multiple regression models using metrics.

3. Urban models based on ordination were transferable among urban areas -- models developed for one urban area are applicable to other urban areas.

4. Urban models are specific to urbanization and do not provide a general means of modeling regional responses to water quality – network design affects application.

Next steps

• Derive more generalized land-use disturbance models (forest, agriculture, urban).

• Couple biological response models with SPARROW modeling in South Atlantic gulf and Tennessee River basins.