Understanding Stream Water Quality and Fish Diversity in the

Great Lake Surrounding Areas: A Path Modeling Analysis

by

Yuhan Zhou

A thesis submitted

in partial fulfillment of the requirements

for the degree of

Master of Landscape Architecture

(Environment and Sustainability)

in the University of Michigan

04 2023

Thesis committee:

Assistant Professor Runzi Wang, Chair

Associate Professor Stan Jones

 Abstract

Fish richness and diversity serve as important indicators of a healthy stream ecosystem,

which are influenced by a complex web of ecological factors, including regional climate,

watershed characteristics, riparian zone quality, and water quality. Investigating how these

factors interconnect and impact fish community is crucial for developing effective

management strategies to safeguard freshwater ecosystems. In this study, we used partial least

squares regression to develop a causal understanding of how watershed development and

climate factors affect fish richness and diversity by altering water temperature, pH,

conductivity, total nitrogen, and total phosphorous (TP) in 277 watersheds in the Great Lake

surrounding areas. After identifying TP as a potential threat to fish biodiversity, we examined

how watershed land use, slope, and soil interacted to drive changes in TP concentration with

multiple linear regression. Results suggested that moderate watershed development (average

5% developed percentage in the study site) can enhance fish diversity by increasing pH,

temperature, and conductivity. However, watershed development and riparian degradation

diminished fish diversity by increasing nutrient concentrations. We also found that land

cover-slope interaction was an important factor affecting TP concentration, while land

cover-soil interaction was not significant. Future ecosystem management in the study area

should therefore emphasize a dual focus on watershed management approaches and riparian

zone preservation to improve fish diversity and stream ecosystem health.

1
Table of Content

Chapter 1: Introduction ....................................................................................................3

Chapter 2: Materials and Methods ...................................................................................8
2.1 Study Site ......................................................................................................................9
2.2 Data and Variables ...................................................................................................... 10
2.3 Analytical Methods ..................................................................................................... 14
Chapter 3: Results ....................................................................................................... 19
3.1 Spatial Distribution of Key Variables ......................................................................... 20
3.2 Variable Correlation .................................................................................................... 25
3.3 Pathways of Fish Richness and Diversity ...................................................................26
3.4 How Watershed Characteristics Interact to Influence Total Phosphorous .................. 30
Chapter 4: Discussion ..................................................................................................34
4.1 The Mechanisms of Fish Richness and Diversity Pathways ...................................... 35
4.2 The Mechanisms and Management Implication of How Watershed Characteristics
Influenced TP ....................................................................................................................38
4.3 Research Limitations and Outlook ............................................................................. 40
Chapter 5: Conclusion .................................................................................................44
References ....................................................................................................................47

2
Chapter 1: Introduction

3
Fish diversity and richness are vital indicators of biodiversity and ecosystem health, where

species with greater diversity and richness adapt to a wider variety of conditions such as

stream disturbance, disease, and climate change (Hiddink et al., 2018; Messemer et al., 2011).

Fish diversity and richness also influence various ecosystem services including aquatic

habitat, recreational opportunities, and fisheries (Keeler et al., 2012; Prudencio and Null,

2018). Today, freshwater fish population are increasingly jeopardized by human activities,

primarily due to factors such as river fragmentation, habitat loss, eutrophication and pollution,

and the introduction of non-native species (Su et al., 2021). These threats manifest at

different spatial scales, ranging from global climate change, watershed-scale land use change,

and riparian scale stream degradation, to in stream-scale change in water physiochemistry.

Such threats induce alterations in fish diversity and richness predominantly by modifying the

freshwater environment, including temperature, nutrients, sediment, and toxins (Keeler et al.,

2012). To the best of our knowledge, little research has been done to investigate this complex

system at cross scales to identify the pathways of fish diversity alteration.

Threats to freshwater fish diversity and richness could be causally and structurally linked. For

example, the combined effect of watershed development and climate change causes flow

regimes alteration and generates greater sediment and nutrients under heavy rainfall events,

thereby driving changes in habitat suitability and fish community composition (Ferreira et al.,

2019; Pörtner and Peck, 2010; Turunen et al., 2021). Another possible pathway is that

damage to riparian vegetation affects fish biodiversity by altering sedimentation, nutrients,

and water temperature (Wilkinson, 1999). Two major gaps are associated in quantifying these

4
pathways and their effects on fish richness and diversity. Firstly, there is limited

understanding of the relative importance of the threats to fish richness and diversity in a

specific ecological context. For example, it remains unknown whether watershed land use

change or riparian zone alteration are more influential, since both alter runoff patterns,

sedimentation, pollution, and habitat suitability; and the relative importance may vary

depending on the characteristics of the freshwater system (Meador and Goldstein, 2003).

Secondly, the effect of a specific pathway changes according to environmental conditions.

For instance, the disturbance of riparian zone is typically considered harmful to fish diversity

as it undermines a diverse range of habitats, elevates soil erosion, and increases water

temperature (Reid et al., 2019). While the intermediate disturbance hypothesis suggests local

species diversity is maximized in the intermediate disturbance condition, moderate disturbed

riparian zone has high biomass due to the abundance of terrestrial and aquatic food

availability (Albertson et al. 2017). Therefore, a causal model is needed to quantify the

strength and effect of different pathways on how watershed characteristics, riparian quality,

as well as water physical, chemical, and biological indicators are intertwined in their effect on

fish richness and diversity. Wang and colleagues (2021) have recently applied a structural

equation model to water quality because of its advantage in constructing latent variables

using measured variables and exploring the causal pathways; however, this model has been

rarely used to investigate fish diversity.

Understanding the change of nutrients is fundamental to investigating fish diversity because

high TP concentrations in streams can reduce fish diversity by promoting algae growth,

5
thereby reducing oxygen levels and creating unfavorable conditions for fish (Dala-Corte et al.,

2016). Extensive literature on the drivers of total phosphorus (TP) concentration in surface

water has identified multiple drivers of phosphorus pollution, such as climate change,

geology, soil type, topography, landscape composition, landscape configuration, seasonal

land use change, and catchment hydrology (Lintern et al., 2018). Past research selected

appropriate variables from these drivers to model in-stream TP concentration; however,

conclusions regarding the effect of each driver were inconsistent. Some researchers proposed

that flat areas may discharge more nutrients when compared to steep areas (Yu et al., 2016),

while others suggested that a decreased deviation in slope is associated with a decrease in

pollutant concentration (Wang et al., 1997; Wissmar et al., 1990). These inconsistent

conclusions may be due to interaction effects between variables having been often

overlooked, especially how land cover interacts with other geophysical factors in influencing

TP (Kaushai et al., 2008).

Slope and soil are key factors that interact with land cover in affecting water quality (Yu et al.,

2016). Slope is an important factor affecting nutrient concentration in the water body because

it is related to surface runoff volume and velocities. Slope was also identified as a key

parameter in predicting rates of water flow across surfaces (Richards et al., 1996). With

increasing slope, greater rates of water flow contribute to soil erosion and the rates of

particulates picking up pollutants, which has the potential to further deteriorate water quality

(Yu et al., 2016). The chemical characteristics of soil (e.g., phosphorus, nitrogen, and salt

content in soil) directly affect the types and concentration of pollutants in water (Dillon et al.,

6
1975). Soil erodibility and sorption capacity also influence constituent mobilization in

catchments (Lintern et al., 2018). The mobilization of sediments is positively correlated to the

susceptibility of the geological deposit and the soil within the catchment to erosion and

weather. Moreover, soil hydrological property affects water quality, with a lower hydraulic

conductivity leading to more residence time of subsurface flow and facilitating pollutants to

be removed by vegetative uptake or biogeochemical processing (Lintern et al., 2018). As a

result, research has found a positive correlation between pollutants in water and soil drainage

capacity (Arheimer et al., 2000; Franklin et al., 2013).

In this study, we hypothesized a causal connection between watershed development, climate,

and fish community, with riparian quality and water quality as mediating factors. Stream

nutrients are affected by watershed development through a combination of land use,

topography, soil, and their interaction effects. Our specific research goals are as follows: (1)

to explore the spatial distribution of important land cover, climate, riparian quality, water

quality, and fish diversity variables and their associations; (2) to investigate the mechanisms

by which watershed development and climate impact fish richness and diversity; and (3) to

assess how land cover-slope interaction and land cover-soil interaction affect TP

concentration.

7
Chapter 2: Materials and Methods

8
2.1 Study Site

We obtained our sampling stations from the EPA’s National River and Stream Assessment

(NRSA) between 2018-2019 as part of the National Aquatic Resource Surveys. We defined

the site as all states surrounding the Great Lakes (i.e., Illinois, Indiana, Michigan, Minnesota,

New York, Ohio, Pennsylvania, and Wisconsin), resulting in the sample size of 277 stations.

We delineated subwatershed boundaries with the location and elevation of these stations as

shown in Fig. 1 with ArcGIS (version 10.8). The major land use types of our study area are

forest and cropland. All samples were collected in summer with an average daily temperature

of 19.9℃ in the area.

Fig. 1. Study site.

9
2.2 Data and Variables

Fish community data were obtained from NRSA 2018-2019; and we calculated the richness

and Shannon diversity index based on the number of each fish species, which represented fish

community structure (J Zhao et al., 2014). Here, H is the Shannon diversity index, and P(i) is

the proportion of each fish species (Eq. 1):

�
H=- �=1
[�(�) ∗ ���(�)] (1)

Water quality data included TP, total nitrogen (TN), pH, conductivity (COND), and water

temperature (WATER_TEM). Phosphorus and nitrogen are the main nutrients associated

with the intensification of cropland in the catchment area, which could have significant

negative impact on aquatic ecosystems function (Owens et al., 2005; Bierschenk et al., 2019).

pH, COND, and WATER_TEM are water quality parameters highly relevant to fish diversity,

as changes in these indicators may influence the transmission of acoustic, visual, and

chemical signals, thereby affecting species interactions and reproductive success (Barrella et

al., 2003; Leuven et al., 1987).

Climate data were obtained from the parameter-elevation relationships on independent slopes

model dataset, which provides gridded climate datasets for the United States (Daly et al.,

2008). Among climate variables, we calculated the daily precipitation amount and daily

average temperature at the subwatershed scale to match the TP concentration collected from

the outlet of the subwatershed (see Table 1). We also calculated the monthly maximum 1-day

10
precipitation (Rx1day), warmest monthly temperature, and coldest monthly temperature to

represent the extreme climates effect on water quality and fish. Rx1day was calculated using

Eq. 2, where Rx1dayj is the maximum 1-day precipitation value for period j, and RRij is the

daily precipitation amount on day i in period j:

Rx1dayj = max (RRij) (2)

Land cover percentages were obtained from the National Land Cover Database (NLCD;

2019), a 30m Landsat-based land cover database. These images rely on the imperviousness

data layer for urban classes and on a decision-tree classification for all other classes (Dewitz

et al., 2021). Among the various land cover types, we selected the percentage of developed,

agricultural, and forest land as variables to represent watershed development conditions.

NLCD divides the developed area into four categories: (1) developed open space (impervious

surfaces <20%); (2) developed low intensity (impervious surfaces 20%-49%); (3) developed

medium intensity (impervious surfaces 50%-79%), and (4) developed high intensity

(impervious surfaces >80%). It divides forest land into three categories: (1) deciduous forest;

(2) evergreen forest; and (3) mixed forest. To simplify the model, we merged the developed

low, medium, and high intensity into developed (“DEV”) land and combined all types of

forest into a single “FOREST” category. The developed open space remained a separate

category because the impervious percentage was lower than 20% and not likely to cause

stream syndrome (Walsh et al., 2015).

11
Soil type were obtained from the OpenLandMap Soil Texture Class (USDA System).

Because the water quality indicators we used were primarily pollutants transported with

surface runoff, we used the 0cm soil depth layer as the soil data source. We then reclassified

the 12 types of soil into three different types (clay, silt, and sand) according to the soil

textural triangle. Silt and sand had a strong correlation, while sand and clay had a weak

correlation. Therefore, we chose sand and clay as variables to represent soil texture in our

study and excluded silt. Slope variable were calculated based on the 30m-resolution NASA

digital elevation model.

Part of the riparian zone variables were from NRSA, including riparian quality index, riparian

vegetation quality index, and riparian disturbance index. These represented on a scale from 0

to 1 the quality and disturbance of the riparian buffer. Other riparian zone variables including

proportion of deciduous forest (R_DECI), evergreen forest (R_EVER), and mixed forest

(R_MIXED) in riparian zone were calculated from NLCD 2019. Specifically, we calculated

the fraction of different forests at the riparian buffer scale defined as 500m surrounding the

water quality stations. Here, we separated forest categories because different forest species

and defoliation along the riparian buffer could have varying impacts on fish diversity (Allan

et al., 2004).

12
Table 1. Variables used.

Variable Description Units

Fish community variables

Richness Abundance of fish species around each water /

quality station

Diversity Shannon diversity index (H) of fish /

community

Water quality variables

TP Total phosphorus ug / L
TN Total nitrogen mg / L
PH Potential of hydrogen /
COND Conductivity uS / cm
WATER_TEM Water temperature °C
Climate variables
TMEAN Daily average temperature °C
PPT Daily total precipitation mm
W_MTEM Warmest monthly temperature °C
C_MTEM Coldest monthly temperature °C
Rx1day Monthly maximum 1-day precipitation mm

Land cover variables

OPEN Proportion developed open land (impervious %

surface <20%)

DEVELOPMENT Proportion developed land (impervious %

surface >20%)

FOREST Proportion forest land %

CROP Proportion agricultural land %
Slope variables

13
SLOPE Average slope %

Soil variables

CLAY Proportion clay in soil %

SAND Proportion sand in soil %
Riparian zone variables

QR Riparian quality index /

QRVeg Riparian veg quality index /

RDIST Riparian disturbance index /

R_DECI Proportion deciduous forest in riparian zone %

R_EVER Proportion evergreen forest in riparian zone %

R_MIXED Proportion mixed forest in riparian zone %

2.3 Analytical Methods

2.3.1 Partial Least Squares Path Modeling

PLS path modeling is a sophisticated technique for estimating intricate causal relationships

within path models that incorporate latent variables (LV; Wold, 1966, 1980). In a PLS model,

the inner model consists of latent variables and their connecting paths, while the outer model

comprises measured variables (MV; Henseler et al., 2016; see Eq. 3). Within the inner model,

path coefficients (pc) quantify the relationships between LVs, whereas in the outer model, the

associations between LVs and MVs are represented by weights (W; Hair et al., 2014). Each

path coefficient represents the impact of independent variables (i.e., exogenous or ‘start of

path’) on dependent latent variables (i.e., endogenous or ‘end of path’). The estimated score

of an endogenous latent variable (e.g., LVp,c in Fig. 2; see Eq. 4) is calculated as the

weighted sum of all its connected exogenous latent variables, where the Ws are represented

by path coefficients:
14
 �
��� = �=1
(��� ∗ �� ) (3)

���,� = ���,� ∗ ��� + ���,� ∗ ��� (4)

Fig. 2. Partial least squares path model used.

The PLS algorithm can produce either formative of reflexive models. Reflexive models

feature paths that originate from LV and terminate at MV, while the arrow directions are

reversed in formative models. In reflexive models, measured variables are considered as

effects of latent variables, while they are viewed as causes in formative models (Boccuzzo

and Fordellone, 2015). In our research, PLS was used as formative model because we

assumed all the latent variables determined the measured variables.

15
We used the coefficient of determination (R2 ) for LVs and the Goodness of Fit (GoF) index

of the whole model to evaluate the model performance, which serves as a pseudo GoF

measure to account for the quality of both the measurement and structural models in

evaluating the overall model performance. In a PLS model, R2 represents the percentage of

variance in the dependent variable that is explained by the predicting constructs (Oliveira et

al., 2019). Generally, the model performance can be classified into three levels: R2 <= 0.2 -

low, 0.2 < R2 < 0.6 - moderate, R2 >= 0.6 - high.

2.3.2 Multiple Linear Regression Models and Post-Hoc Analysis of the Interaction Effect

We applied multiple linear regression to explore the relationship between land cover, soil,

slope, daily climate, and TP. The TP value was log-transformed to approximate a normal

distribution. We added interaction terms by multiplying two independent variables

(Ponce-Palafox et al., 2019): land cover-slope interaction and land cover-soil interaction. To

control the number of independent variables and address the multicollinearity issues, we

performed least absolute shrinkage and selection operator (LASSO) regression for each

model to select the important main effects variables. This is a regression analysis method that

performs both variable selection and regularization, which can enhance the prediction

accuracy and interpretability of the regression models. LASSO model selection resulted in

the following formula for the linear regression model, where Y is the dependent variable (TP),

and bi represents the constant and the coefficient of different independent variables. T, P, O,

D, F, C, S, CL and SA represent the TMEAN, PPT, DEV_OPEN, DEV, FOREST, CROP,

16
SLOPE, CLAY and SAND, respectively. Table 1 describes the variable in detail. The

significant level of regression was set as 0.05:

Y = b0 + b1T + b2P + b3O + b4D + b5F + b6C + b7S + b8CL + b9SA + b10(D*S) + b11(C*S) +

b12(O*CL) + b13(O*SA) + b14(C*CL) + b15(C*SA) + � (5)

To evaluate whether adding interaction variables can significantly improve model

performance, we used R2 and Akaike information criterion (AIC) to compare Eq. 5 and

model without interaction variables. Because the number of independent variables in these

two models differ, it is not appropriate to use only R2 to compare model performance. AIC

is a proven method for model evaluation with the criteria of the number of independent

variables and the maximum likelihood estimate of the model. The best-fit model according to

AIC is the one that explains the largest variance with the least independent variables. When

models are nested and the sample sizes of models are consistent, the model with the smallest

AIC value is preferred.

In post-hoc analysis, we used perspective plots to investigate the significant interaction

effects in the regression models. Perspective plots are particularly useful for exploring

complex interactions, finding optimal solutions, and identifying trends or patterns in data.

The perspective plot can draw a contour graph that explores the relationship between several

independent variables and the dependent variable. According to the contour graph of two

17
independent variables, we can then identify the joint effect of them and how the effect of one

variable changes with different values of the other variable.

18
Chapter 3: Results

19
3.1 Spatial Distribution of Key Variables

The study site and season were characterized by mild climate and favorable ecological

conditions. The site experienced relatively cool temperatures, with 22℃ as the average

highest monthly temperature. Precipitation distribution varied among different sampling sites

and days, ranging from 0 to 46.20mm of daily precipitation. The maximum Rx1day was

120.82mm, with an average value of 32.82mm, indicating the relative wet condition of the

sampling season. The average forest land cover percentage was 37.8%, with notably high

forest coverage in Southeast Lake Ontario, Susquehanna, Allegheny subregions (HUC 4), and

areas around Lake Michigan (see Fig. 3). The major human disturbance was agriculture

development, with a 26.7% average crop coverage. The average developed/urban area was

only 5%, which would not likely cause stream degradation. Riparian quality was generally

good in the study area, with an average QR of 0.59 and an average 0.47 QRVeg. The riparian

forest predominantly consisted of deciduous trees (25.6% average coverage). We also found

the spatially clustering effect of riparian quality was weak and not correlated with watershed

forest coverage. For example, sites in proximity to Lake Michigan and Lake Huron had high

forest percentages but low riparian quality.

Overall, the study area exhibited satisfactory conditions for both water quality and fish

diversity. Fish diversity was higher near Lake Superior and Lake Michigan, including

Northeastern Lake Huron, Northeastern Lake Michigan, and Northeastern Lake Michigan

subregions (HUC4), where forest dominated the watershed landscape. However, some

watersheds in east Lake Ontario and Lake Erie had low fish diversity, such as the

20
Southeastern Lake Ontario and the Allegheny subregion (see Fig. 3). The study area

exhibited low nutrient concentration, with an average TP 0.08 mg/l and an average TN of

1.98 mg/l. The highest TP concentrations were observed in the Great Miami subregions and

Mississippi Headwaters. The conductivity variation was relatively large, with elevated values

in the Southeastern Lake Huron, Southeastern Lake Michigan, and Wabash subregions (see

Fig. 3).

21
Table 2. Summarized statistics for variables used.

Variable M SD Max Min

Fish community variables

Richness 12.87 7.53 38.00 1.00

Diversity 1.67 0.68 3.12 0.00

Water quality variables

TP 89.21 95.19 659.00 6.47

TN 1.98 2.54 16.21 0.11

PH 7.96 0.49 8.67 5.43

COND 445.71 339.87 2773.60 12.20

WATER_TEM 18.25 5.17 27.90 0.001

Climate variables

TMEAN 19.90 4.14 28.27 7.37

PPT 2.59 6.12 46.20 0.00

W_MTEM 22.02 1.76 26.62 18.54

C_MTEM -8.13 4.38 1.10 -18.17

Rx1day 32.82 17.31 120.82 8.18

Land cover variables

DEV_OPEN 4.56 4.16 34.62 0.19

DEV 5.12 8.09 58.60 0.00

FOREST 37.80 27.44 93.33 0.30

CROP 26.72 29.00 94.51 0.00

22
Slope variables

SLOPE 4.11 2.55 15.81 0.48

Soil variables

CLAY 15.91 4.78 39.42 10.00

SAND 40.38 14.29 70.74 16.42

Riparian zone variables

QR 0.59 0.16 0.99 0.18

QRVeg 0.47 0.20 1.00 0.10

RDIST 0.35 0.23 0.86 0.00

R_DECI 25.60 23.26 88.54 0.02

R_EVER 3.25 7.52 44.93 0.02

R_MIXED 7.28 11.23 65.40 0.05

23
Fig. 3. Spatial distribution of the key variables (a) fish diversity, (b) warmest monthly

temperature, (c) forest land cover, (d) riparian quality index, (e) total phosphorus, and (f)

conductivity.

24
3.2 Variable Correlation

The correlation heatmap in Fig. 4 indicates that the variable with the highest correlation to

fish diversity was pH (r=0.44), followed by warmest monthly temperature (r=0.32) and water

temperature (r=0.27). At the riparian zone scale, fish diversity was surprisingly positively

correlated with riparian disturbance score (r=0.26). This might be due to a riparian

disturbance leading to high pH and conductivity, which are both increasing fish diversity in

the study region. Also, more riparian disturbance was found in warm areas (as indicated by

the correlation of 0.43 between riparian disturbance and the warmest monthly temperature)

with high fish diversity. Among all watershed characteristics, the percentages of forest, crop,

and sand in the soil were the most correlated variable with fish diversity (r=0.22, -0.22, and

0.22, respectively). Fish richness was highly correlated with fish diversity, with the effect of

water temperature higher (r=0.38).

Nutrient concentration was associated with both watershed land cover and riparian quality,

while watershed land cover’s association was stronger. Forest was significantly negatively

correlated with TN and TP concentration, while the crop’s correlation was significantly

positive. Riparian forest coverage was negatively correlated with TN and TP concentration,

with deciduous and mixed forests’ correlation higher than evergreen forests. In general, the

riparian forest coverage was stronger related to TN and TP than the riparian quality and

vegetation quality score. Higher forest in the upstream and riparian zone were also associated

with lower pH and conductivity. We also found that watersheds with high cropland had lower

riparian quality and higher riparian disturbance. The warmest monthly temperature had a

25
significant association with many variables, where warm areas had more disturbed riparian

zone, more sandy soil, higher pH, conductivity, and higher nutrient concentration.

Fig. 4. Correlation map of all variables.

3.3 Pathways of Fish Richness and Diversity

In the PLS model, we linked all the measured variables to six latent variables: watershed

development, warm climate, riparian quality, water physiochemistry, eutrophication, and fish

diversity. Watershed development was assessed using land use (developed, developed open,

forest, crop), slope, soil, and their interactions. According to the PLS results, forest, crop, and

26
their interaction with soil were the most influential factors on the latent variable of watershed

development. The warm climate latent variable was quantified using daily temperature,

warmest and coldest monthly temperatures, daily precipitation, and Rx1day. The most

significant factor for this latent variable was the warmest monthly temperature. Additionally,

extreme precipitation as indicated by Rx1day weighed more heavily than daily precipitation.

Watershed development and warm climate were exogenous variables that accounted for the

variations of other latent variables in the PLS model.

The endogenous latent variables were riparian quality, water physiochemistry, nutrients, and

fish diversity. Riparian quality was assessed using different types of riparian forest

percentages (i.e., riparian evergreen forest, mixed forest, and deciduous forest percentages),

riparian quality index, riparian vegetation index, and riparian disturbance index. The

percentages of mixed and deciduous forests in the riparian zone predominantly influenced

this latent variable. Water physiochemistry was measured using three critical water quality

indicators affecting fish diversity: water temperature, pH, and conductivity, with conductivity

and pH weighing more heavily than water temperature. Eutrophication was represented by

total nitrogen and total phosphorus concentrations, which had nearly equal weightings. Lastly,

fish diversity was evaluated using the fish diversity index and richness index.

In terms of model performance, riparian quality, water physiochemistry, eutrophication, and

fish diversity had R2 values of 0.44, 0.46, 0.31, and 0.25, respectively. As a result,

watershed landscape, climate factors, and riparian quality demonstrated moderate predictive

power for water quality (Sanchez, 2013), while fish community characteristics were less

27
effectively captured by the measured variables. The relationships between latent variables can

be expressed as Equations 6-9.

riparian quality=-0.58*watershed development-0.14*warm climate (6)

eutrophication=0.49*watershed development+0.07*warm climate-0.05*riparian

quality (7)

water physiochemistry=0.40*watershed development+0.16*warm

climate-0.24*riparian quality (8)

fish diversity=0.41*water physiochemistry-0.11*eutrophication+0.17*warm

climate-0.004*watershed development-0.04*riparian quality (9)

The above equations indicate that higher fish diversity is associated with higher pH (within

the range of 5.43 to 8.67, with an average of 7.96), higher conductivity (within the range of

12.2 to 2773.60, with an average of 445.71), and warmer water temperatures (within the

range of 0 to 27.90, with an average of 18.25). Eutrophication contributes to a decrease in

fish diversity and richness. Warm and humid climate conditions lead to increased fish

diversity and richness. Furthermore, both eutrophication and water physiochemistry are

strongly affected by watershed development. Specifically, a higher percentage of cropland

and a lower percentage of forest result in higher pH, conductivity, and water temperature.

The effect of urban development on nutrients, pH, conductivity, and water temperature is also

positive, but not as pronounced as cropland. Riparian quality can reduce eutrophication levels

and decrease water temperature, conductivity, and pH. Warm and humid climate conditions

increase eutrophication, water temperature, pH, and conductivity.

28
After substituting the endogenous latent variables (eutrophication, water environment, and

riparian quality) in Equations 6-9 on the right-hand side with water physiochemistry and

warm climate, we derived a model for fish diversity, eutrophication, and water environment

based solely on watershed development and warm climate (Equations 10-12).

eutrophication=0.52 watershed development+0.08 warm climate (10)

water environment=0.54 watershed development+0.19 warm climate (11)

fish diversity=0.18 watershed development+0.07 warm climate (12)

We found watershed development strongly influenced nutrients and water environment by

increasing TP, TN, pH, conductivity, and water temperature. The effect of warm climate on

pH, conductivity and water temperature was more pronounced than that on TP and TN.

Watershed development caused increasing fish diversity because the disturbance in our study

site was only moderate (5% average percentage of development, see Table 2). In general, we

concluded that moderate development led to low degree of eutrophication, thus not

necessarily decreasing fish diversity.

29
Fig. 5. The outcomes of PLS model. (a) path coefficients between LVs within the inner model, (b)
weights of all factors for each LV.

3.4 How Watershed Characteristics Interact to Influence Total Phosphorous

We found adding interaction variables to regression models leads to an increase in R2 and a

decrease in the AIC value, indicating that interaction variables work efficiently with

individual variables in explaining TP concentration. Specifically, regression models with only

main effects had R2 of 0.368 and AIC of 734.15; while the R2 and AIC of regression

30
models with interaction effects were 0.420 and 714.14, respectively (see Table 3).

The effect of crop on TP depended on soil types, while the effect of developed area on TP

varied with slope. Specifically, the main effect of crop was negative but not significant.

However, CROP-CLAY interaction was positive, indicating the influence of crop on TP gets

positive and larger with the increase of clay content in the soil. Also, the CROP-CLAY

interaction graphs (see Fig. 6 (a)) indicates that when the proportion of clay in the soil is less

than 20%, TP increased slowly with the increase of cropland percentage. However, when the

percentage of clay in the soil is larger than 40%, TP increased much more rapidly as the

percentage of cropland increased. Developed area had a significantly negative main effect on

TP, while the effect changed to positive when the slope gets steeper indicated by the

significantly positive DEV-SLOPE interaction (see Table 3). From the DEV-SLOPE

interaction graphs (see Fig. 6 (b)), when average slope WAS less than 5%, the increase in the

percentage of developed land almost had no impact on TP. However, when average slope was

greater than 15%, the percentage of developed land has a strong positive association with TP.

Overall, the individual influence of land cover, slope, and soil on TP presented a potentially

nonlinear effect due to the interaction with each other. In addition, daily precipitation showed

a positive influence on TP, while the effect of forest was negative.

31
Table 3. Multiple linear regression results with interaction variables.

Predictors Estimates CI p

(Intercept) 5.76 4.01 — 7.52 <0.001*

TEMPERATURE 0.01 -0.01 — 0.04 0.244

PRECIPITATION 0.02 0.01 — 0.04 0.002*

DEV_OPEN -18.99 -44.64 — 6.67 0.146

DEV -6.69 -10.52 — -2.87 0.001*

FOREST -1.16 -1.94 — -0.37 0.004*

CROP -1.28 -2.86 — 0.30 0.112

SLOPE -0.09 -0.16 — -0.03 0.003*

CLAY 2.27 0.22 — 4.32 0.141

SAND -1.61 -3.19 — -0.03 0.046*

DEVELOPMENT 2.35 1.06 — 3.63 <0.001*

* SLOPE

CROP * SLOPE 0.29 -0.02 — 0.60 0.069

OPEN * CLAY 103.33 -16.90 — 223.56 0.092

OPEN * SAND 7.97 -14.04 — 29.99 0.477

CROP * CLAY 6.26 0.59 — 11.93 0.031*

CROP * SAND -0.23 -2.94 — 2.48 0.866

Observations 277

R2/ R2 adjusted 0.449 / 0.420

AIC 714.135

32
Fig. 6. Contour graph of interaction variables. (a) CROP-CLAY interaction graphs, and (b)

DEV-SLOPE interaction.

33
Chapter 4: Discussion

34
4.1 The Mechanisms of Fish Richness and Diversity Pathways

We contributed to the existing research on the relationship between watershed development

and fish diversity by identifying two distinct pathways– through changing eutrophication

levels and through altering other physicochemical indicators such as pH, conductivity, and

water temperature. Previous research on the effects of watershed development on fish

biodiversity has yielded inconsistent results. For example, many studies found row-crop

agricultural had deleterious effects on fish communities primarily because higher loads of

fine sediment reduced hatching success (Gido et al. 2010; Sternecker and Geist, 2010).

However, other studies have observed relatively high fish community conditions in heavily

agricultural areas (comprising over 50% of the basin), which may be attributable to the

influence of factors such as nutrient and sediment regulation (Meador and Goldstein, 2003).

Our findings indicated that agricultural land use augments nutrient levels, subsequently

reducing fish diversity. Concurrently, agriculture also increased conductivity, water

temperature, and pH, thereby promoting fish diversity. The combined effects of the two

pathways showed an overall positive effect of watershed development on fish diversity. This

result supported the previous finding that freshwater fish community was only sensitive to

some higher-level threshold of agricultural development (2%-37%, Chen and Older, 2020).

The average agricultural land cover of the study site was 26.72%, which might not have

caused fish diversity degradation. That said, in our study site, the positive effect of

agriculture on fish diversity through changing water physiochemistry outweighed its negative

effect through increasing eutrophication, given the relatively low in-stream TP concentration

(average 0.08 mg/l). However, it is plausible that with the progression of urbanization, this

35
relationship could be inverted. When watershed development reached certain threshold, the

effect of agriculture on eutrophication might surpass the advantageous effects on fish

diversity. The other possible reason could be the site climate was cool and humid, where the

slightly increasing water temperature caused enhanced biodiversity.

The observed positive correlation between watershed development and fish diversity, as well

as the association between riparian disturbance and fish diversity, may be attributed to the

relatively low extent of development. This evidence lends support to the well-established

"intermediate disturbance hypothesis" (Huston, 2014), which posits that sites with

intermediate levels of disturbance may exhibit increased biomass, consequently promoting

biodiversity. Specifically, it was found that food availability was highest in streams of open

meadow habitat, which represented an intermediate level of disturbance compared to a

completed forested stream (Albertson et al. 2017). However, this “intermediate disturbance

hypothesis” was ambiguous due to the imprecise definition of "intermediate." In this study,

we contribute to the substantiation of this theory by offering empirical evidence suggesting

that approximately 5% of urban development may be considered an "intermediate

disturbance" for watersheds surrounding the Great Lakes. Also, in addition to food

availability, we identified the other possible mechanism of how intermediate disturbed site

had high biodiversity—through slightly increasing water temperature, pH, and conductivity.

Existing research has yielded inconsistent conclusions regarding the relative importance of

riparian quality and catchment land use in relation to their effects on water quality and fish

36
diversity, where our research supported the significance of catchment land use. Some

researchers found measures of water physiochemistry and riparian condition might be better

indicators of fish diversity compared to basin wide land use because broader scale land use

might not adequately capture local activities affecting fish community (Meador and Goldstein,

2003; Lammert and Allan, 1999). However, some researches highlighted the terrestrial land

use and urbanization effects on fish community composition, especially the effects of

erosion-prone land use types such as root crop and maize (Bierschenk et al. 2019; Sternecker

and Geist, 2010). Our research indicated riparian quality mainly influenced fish diversity by

changing water physiochemistry, but this effect was not as strong as the effect of watershed

development. Moreover, our findings indicate that an intermediate level of riparian zone

disturbance enhances fish diversity, which contradicts numerous studies that have asserted

that riparian zones promote fish diversity by reducing nutrient and sediment input into

streams (Bierschenk et al. 2019). Riparian quality appeared less critical than watershed land

cover and did not exhibit a marked effect on eutrophication, possibly due to the relatively

favorable riparian quality in the study area, characterized by an average Riparian Quality

Index (QR) of 0.59 and an average deciduous tree cover of 25.6%. Consequently, urban

stream syndrome was not evident in the study area.

The warm climate in the study area exhibited a positive impact on fish diversity, which may

be attributed to the mild average summer temperature of 18.25℃ during the sampling season.

The daily temperature throughout this period did not surpass the optimal growth temperature

for the majority of species (Jobling, 1981; Tsuchida, 1995). Other research also showed fish

37
migration and diversity were positively related to the water temperature because warmer

temperature increased biomass by raising metabolism (Brodersen et al. 2011; Duffy et al.

2016). However, if water temperature keeps rising under climate change, it would decrease

fish diversity when it exceeds fish’s preferred temperature range and when it starts to cause

eutrophication. In addition, extreme precipitation (Rx1day) also had a positive effect on fish

diversity in our study. This association may stem from increased streamflow resulting from

high precipitation, which in turn generates new habitats for fish and supplies additional food

resources (Cheung, 2018; Hollowed et al. 2013). Compared to Rx1day, daily precipitation

was a negligible factor for fish diversity. Overall, we identified the pathways that warm

weather and high precipitation influenced fish diversity directly and indirectly through

changing eutrophication and water physiochemistry. Any future projection of climate change

effects on fish community should jointly consider the direct effects of climate change and the

indirect consequences mediated by changes in water quality.

4.2 The Mechanisms and Management Implication of How Watershed Characteristics

Influenced TP

Both the developed area and slope negatively affected TP when the value of them were small,

but the effect changed to positive with the larger values of developed and slope due to the

significant positive interaction effects. Generally, steep slopes tended to increase surface

runoff volume (Wang et al. 2002; Sueker et al. 2001) and velocities (Lintern et al. 2018; Sliva

et al. 2001), which led to a greater chance of mobilization of sediments and pollutions by

mass failure or landslides (Bednarik et al. 2010). The observed negative effect of slope in our

38
study may be attributed to the prevalence of cropland and scarcity of forest in flat areas.

Previous studies also found that in forest area, the slope had significantly negative

correlations with water quality variables, which was because flat areas were cropland and

urban land (Yu et al. 2016). Therefore, land cover types and land use activities may outweigh

topographic features in their influence on TP concentration (Anbumozhi et al. 2005).

Regarding the main effect of developed area, our results were also not consistent with most

existing research stating the positive influence of urban area on TP. In the process of

urbanization, large vegetative surfaces have been converted into built-up areas, leading to an

increase of impervious surfaces (Kim et al. 2017; Li et al. 2018) and therefore increased

surface runoff (Li et al. 2019) and deteriorated surface water quality (Owens et al. 2002). The

observed inconsistency may be due to the fact that when the percentage of developed area is

minimal, the impact of extensive cropland on water quality surpasses that of developed land,

thereby exerting a significant positive influence on TP (Yu et al. 2016).

The positive interaction of DEV-SLOPE could be explained by the fact that steeper slopes

reduced infiltration and increased surface runoff velocities. According to the results of

contour plot of DEV-SLOPE (Fig. 4), we suggested land use planning consider reducing the

development intensity in areas exhibiting slopes greater than 15%. Furthermore, we

recommend that regulations governing various land uses for TP pollution management be

tailored to specific topographical conditions (Anbumozhi et al. 2005).

39
Among the three types of soil, silt and clay both have high sorption capacity, resulting in their

positive association with TP (Lintern et al. 2018). Other research also found a positive

correlation between stream TP concentration and the silt and clay content of catchment soils

in Finland (Varanka et al. 1989--1999). Different from silt and clay, sand has a higher rate of

porosity and holds less water, which leads to a lower nutrient absorption. Consequently,

sandy soil improves water quality by reducing nutrient content (Andry et al. 2009), which

agreed with our regression model results. The reason why clay soils exacerbated the impact

of cropland on TP may be due to the high sorption capacity of clay, causing an increased

uptake of phosphorus from fertilizers into the soil and subsequent discharge into adjacent

rivers (Johnson et al. 1997). Therefore, we suggested agricultural cultivation should avoid

clay-soil areas in land use planning, especially on areas with clay percent larger than 40%.

Moreover, although sand had positive influence on water quality, a high percentage of sand

could lead to low water-holding capacity, excessive drainage of irrigation, and poor fertilizer

use efficiency when planting (Andry et al. 2009). Therefore, the most suitable soil type for

planting should comprise a balanced composition (e.g., sandy loam), which accommodates

both water quality management and other planting considerations.

4.3 Research Limitations and Outlook

The major limitation of this study is that the quantification of the fish community did not

include species-related information. Given the relatively large study area, the same diversity

index could represent distinct species distributions across different regions, potentially

compromising the internal validity of the PLS model. Environmental factors influence fish

40
species in various ways. For example, research has demonstrated that climate change,

characterized by rising and variable temperatures, impacts diverse fish communities more

significantly than species-poor communities (Duffy et al. 2016). Also, different fish species

occurred at different pH conditions, with the lowest pH ranging from 3.1 to 7.0 (Leuven and

Oyen, 1987). Moreover, the species richness was inherently different in regions with varying

temperature conditions (Dala-Cort, 2016). Therefore, without accounting for species, our fish

diversity and richness index may not be comparable between watersheds. Future research

could consider partitioning data according to species distributions and constructing the

structural equation models separately. Another related issue was that the model did not

consider spatial clustering effects, which could be partially attributed to the spatial pattern of

species distribution. More advanced multivariate analysis might be promising to include

mixed effects in the PLS model to deal with this issue, such as treating different regions as

random effects (Bry et al. 2019).

Due to the data availability issue, the omission of some important variables affecting water

quality and fish was another limitation, which could cause incomplete pathways in the PLS

model. Stream discharge, for instance, can alter fish diversity by affecting habitat quality and

restricting fish movement. Groundwater and baseflow were related to water temperature,

dissolved oxygen levels, and nutrient availability (Franssen et al. 2016; Poff et al. 2010; Rolls

et al. 2018). However, these hydrological regime-related variables were not available to

include in the PLS model. Certain chemical indicators, such as dissolved oxygen and metals,

were also absent from the NRSA dataset. Because the study sites had low urban and industrial

41
land uses, we expected metals indicating toxicity conditions to fish would be very low and

not cause much bias in the results. Dissolved oxygen was highly correlated with water

temperature which was already included in the model. Another omitted variable was the

location of dams relative to the sampling sites. Dams could cause habitat fragmentation and

prevent fish from accessing important spawning and feeding areas (Liermann et al. 2012).

However, the initial quantification of upstream dams was very weakly associated with fish

diversity and therefore we removed it from the PLS model. For the next step, we could try

quantifying the number of dams within different location buffers relative to the sampling sites

to investigate their influence on fish. Furthermore, we also missed some watershed

characteristics affecting TP concentration, such as urban growth form, landscape

configuration, and land management (Liu et al. 2012; Brabec et al. 2009; Yu et al. 2016).

Here we argued these variables were not very important for TP considering the general low

urban development intensity in the study area.

Our models were also subject to some multicollinearity and generalizability issues. Despite

efforts to mitigate multicollinearity among independent variables through the application of

LASSO, the final regression models still exhibited moderate correlations among certain

independent variables (e.g., CROP and FOREST, DEV_OPEN and DEV). A similar situation

arose when we added interaction variables to the regression models which could led to some

biased coefficients in the MLR results. Moreover, the generalization of this research could be

limited by the specific context of the study site. Our study sites were characterized by low

development intensity and favorable habitat quality. The data collection took place during the

42
summer months in cooler regions, with an average temperature of 19.90 ℃. Consequently,

our findings are biased towards favorable environmental conditions and may not be readily

applicable to highly developed or warmer regions.

43
Chapter 5: Conclusion

44
In this study, we identified pathways through which watershed development, climate, and

riparian quality impacted fish richness and diversity by altering water quality indicators. The

most crucial pathway involved watershed development, characterized by increased

agricultural land use and decreased forest land use, which influenced fish richness and

diversity by raising pH, water temperature, conductivity, and nutrient levels. Moderate

watershed development, as indicated by the average of 5% developed areas in the study site,

resulted in increased fish diversity. Although watershed development negatively affected fish

richness and diversity by increasing nutrient concentrations and eutrophication, this pathway

was outweighed by the positive effects of increasing pH, temperature, and conductivity.

Furthermore, watershed development had a more substantial impact on fish communities than

riparian quality. The warmest monthly temperature and Rx1day both positively influenced

fish richness and diversity.

This study also revealed how stream TP concentration was affected by the interaction

between watershed land cover, slope, and soil. The CROP-CLAY and DEV-SLOPE

interactions were both significant factors affecting TP. Specifically, a soil clay content higher

than 40% significantly increased the positive effect of cropland on TP, and a slope greater

than 15% significantly increased the positive effect of developed areas on TP.

We propose two management suggestions accordingly: (1) Future ecosystem management in

the Great Lakes should focus on both watershed management and riparian zone protection to

mitigate the negative impact of nutrients on fish diversity. (2) Land use regulation and

45
nonpoint source programs for stream nutrient management should be tailored to regions with

different topography and soil textures.

46
References

47
Albertson, L. K., Ouellet, V., & Daniels, M. D. (2018). Impacts of stream riparian buffer land

use on water temperature and food availability for fish. Journal of Freshwater Ecology, 33(1),

195-210.

Anbumozhi, Venkatachalam, Jay Radhakrishnan, and Eiji Yamaji. "Impact of riparian buffer

zones on water quality and associated management considerations." Ecological

Engineering 24.5 (2005): 517-523.

Andry, H., et al. "Water retention, hydraulic conductivity of hydrophilic polymers in sandy

soil as affected by temperature and water quality." Journal of Hydrology 373.1-2 (2009):

177-183.

Bednarik, Martin, et al. "Landslide susceptibility assessment of the Kraľovany–Liptovský

Mikuláš railway case study." Physics and Chemistry of the Earth, Parts A/B/C 35.3-5 (2010):

162-171.

Bierschenk, A. M., Mueller, M., Pander, J., & Geist, J. (2019). Impact of catchment land use

on fish community composition in the headwater areas of Elbe, Danube and Main. Science of

the Total Environment, 652, 66-74.

Boccuzzo, Giovanna, and Mario Fordellone. "Comments about the use of PLS path modeling

in building a job quality composite indicator." Univ. Padua (2015).

48
Brabec, Elizabeth A. "Imperviousness and land-use policy: Toward an effective approach to

watershed planning." Journal of hydrologic engineering 14.4 (2009): 425-433.

Brodersen, J., Nicolle, A., Nilsson, P. A., Skov, C., Brönmark, C., & Hansson, L. A. (2011).

Interplay between temperature, fish partial migration and trophic dynamics. Oikos, 120(12),

1838-1846.

Bry, X., Trottier, C., Mortier, F., & Cornu, G. (2020). Component-based regularization of a

multivariate GLM with a thematic partitioning of the explanatory variables. Statistical

Modelling, 20(1), 96-119.

Chen, K., & Olden, J. D. (2020). Threshold responses of riverine fish communities to land

use conversion across regions of the world. Global Change Biology, 26(9), 4952-4965.

Cheung, W. W. L. (2018). The future of fishes and fisheries in the changing oceans. Journal

of Fish Biology, 92(3), 790-803.

Dala‐Corte, R. B., Giam, X., Olden, J. D., Becker, F. G., Guimarães, T. D. F., & Melo, A. S.

(2016). Revealing the pathways by which agricultural land‐use affects stream fish

communities in South Brazilian grasslands. Freshwater Biology, 61(11), 1921-1934.

Dillon, P. J., and W. B. Kirchner. "The effects of geology and land use on the export of

phosphorus from watersheds." Water research 9.2 (1975): 135-148.

49
Duffy, J. E., Lefcheck, J. S., Stuart-Smith, R. D., Navarrete, S. A., & Edgar, G. J. (2016).

Biodiversity enhances reef fish biomass and resistance to climate change. Proceedings of the

National Academy of Sciences, 113(22), 6230-6235.

Fernandes, A. C. P., Sanches Fernandes, L. F., Cortes, R. M. V., & Leal Pacheco, F. A. (2019).

The role of landscape configuration, season, and distance from contaminant sources on the

degradation of stream water quality in urban catchments. Water, 11(10), 2025.

F. Hair Jr, Joe, et al. "Partial least squares structural equation modeling (PLS-SEM) An

emerging tool in business research." European business review 26.2 (2014): 106-121.

Franssen, N. R., Gido, K. B., Guy, C. S., Tripe, J. A., Shrank, S. J., Strakosh, T. R., ... &

Paukert, C. P. (2006). Effects of floods on fish assemblages in an intermittent prairie

stream. Freshwater Biology, 51(11), 2072-2086.

Gido, K. B., Dodds, W. K., & Eberle, M. E. (2010). Retrospective analysis of fish community

change during a half-century of landuse and streamflow changes. Journal of the North

American Benthological Society, 29(3), 970-987.

Henseler, Jörg, Geoffrey Hubona, and Pauline Ash Ray. "Using PLS path modeling in new

technology research: updated guidelines." Industrial management & data systems (2016).

50
Hiddink, J. G., MacKenzie, B. R., Rijnsdorp, A., Dulvy, N. K., Nielsen, E. E., Bekkevold,

D., ... & Ojaveer, H. (2008). Importance of fish biodiversity for the management of fisheries

and ecosystems. Fisheries Research, 90(1-3), 6-8.

Huston, M. A. (2014). Disturbance, productivity, and species diversity: empiricism vs. logic

in ecological theory. Ecology, 95(9), 2382-2396.

Hollowed, A. B., Barange, M., Beamish, R. J., Brander, K., Cochrane, K., Drinkwater, K., ...

& Yamanaka, Y. (2013). Projected impacts of climate change on marine fish and

fisheries. ICES Journal of Marine Science, 70(5), 1023-1037.

Jobling, M. (1981). Temperature tolerance and the final preferendum—rapid methods for the

assessment of optimum growth temperatures. Journal of fish biology, 19(4), 439-455.

Johnson, Lucinda, et al. "Landscape influences on water chemistry in Midwestern stream

ecosystems." Freshwater biology 37.1 (1997): 193-208.

Kaushal, Sujay S., et al. "Interaction between urbanization and climate variability amplifies

watershed nitrate export in Maryland." Environmental science & technology 42.16 (2008):

5872-5878.

Keeler, B. L., Polasky, S., Brauman, K. A., Johnson, K. A., Finlay, J. C., O’Neill, A., ... &

51
Dalzell, B. (2012). Linking water quality and well-being for improved assessment and

valuation of ecosystem services. Proceedings of the National Academy of Sciences, 109(45),

18619-18624

Goonetilleke, Ashantha, et al. "Understanding the role of land use in urban stormwater

quality management." Journal of environmental management 74.1 (2005): 31-42.

Kim, Hyun Woo, et al. "Exploring the impact of green space health on runoff reduction using

NDVI." Urban Forestry & Urban Greening 28 (2017): 81-87.

Lammert, M., & Allan, J. D. (1999). Assessing biotic integrity of streams: effects of scale in

measuring the influence of land use/cover and habitat structure on fish and

macroinvertebrates. Environmental management, 23(2), 0257-0270.

Leuven, R. S. E. W., & Oyen, F. G. F. (1987). Impact of acidification and eutrophication on

the distribution of fish species in shallow and lentic soft waters of The Netherlands: an

historical perspective. Journal of Fish Biology, 31(6), 753-774.

Li, Chunlin, et al. "Effects of urbanization on direct runoff characteristics in urban functional

zones." Science of the Total Environment 643 (2018): 301-311.

Li, Fazhi, et al. "Green infrastructure practices simulation of the impacts of land use on

52
surface runoff: Case study in Ecorse River watershed, Michigan." Journal of environmental

management 233 (2019): 603-611.

Liermann, C. R., Nilsson, C., Robertson, J., & Ng, R. Y. (2012). Implications of dam

obstruction for global fre

Lintern, A., et al. "Key factors influencing differences in stream water quality across

space." Wiley Interdisciplinary Reviews: Water 5.1 (2018): e1260.

Liu, An, Ashantha Goonetilleke, and Prasanna Egodawatta. "Inadequacy of land use and

impervious area fraction for determining urban stormwater quality." Water Resources

Management 26.8 (2012): 2259-2265.

Meador, M. R., & Goldstein, R. M. (2003). Assessing water quality at large geographic scales:

relations among land use, water physicochemistry, riparian condition, and fish community

structure. Environmental Management, 31, 0504-0517.

Messmer, V., Jones, G. P., Munday, P. L., Holbrook, S. J., Schmitt, R. J., & Brooks, A. J.

(2011). Habitat biodiversity as a determinant of fish community structure on coral

reefs. Ecology, 92(12), 2285-2298.

Oliveira, Caroline Favaro, et al. "The modeling of pasture conservation and of its impact on

53
stream water quality using Partial Least Squares-Path Modeling." Science of The Total

Environment 697 (2019): 134081.

Owens, Philip N., and Desmond E. Walling. "The phosphorus content of fluvial sediment in

rural and industrialized river basins." Water research 36.3 (2002): 685-701.

Poff, N. L., & Zimmerman, J. K. (2010). Ecological responses to altered flow regimes: a

literature review to inform the science and management of environmental flows. Freshwater

biology, 55(1), 194-205.

Ponce-Palafox, Jesus T., et al. "Response surface analysis of temperature-salinity interaction

effects on water quality, growth and survival of shrimp Penaeus vannamei postlarvae raised

in biofloc intensive nursery production." Aquaculture 503 (2019): 312-321.

Prudencio, L., & Null, S. E. (2018). Stormwater management and ecosystem services: a

review. Environmental Research Letters, 13(3), 033002.

Reid, A. J., Carlson, A. K., Creed, I. F., Eliason, E. J., Gell, P. A., Johnson, P. T., ... & Cooke,

S. J. (2019). Emerging threats and persistent conservation challenges for freshwater

biodiversity. Biological Reviews, 94(3), 849-873.

Richards‐Zawacki, Corinne L. "Effects of slope and riparian habitat connectivity on gene

54
flow in an endangered Panamanian frog, Atelopus varius." Diversity and Distributions 15.5

(2009): 796-806.

Rolls, R. J., Heino, J., Ryder, D. S., Chessman, B. C., Growns, I. O., Thompson, R. M., &

Gido, K. B. (2018). Scaling biodiversity responses to hydrological regimes. Biological

Reviews, 93(2), 971-995.

Sanchez, G. (2013). PLS path modeling with R. Berkeley: Trowchez Editions, 383(2013),

551.

Sliva, Lucie, and D. Dudley Williams. "Buffer zone versus whole catchment approaches to

studying land use impact on river water quality." Water research 35.14 (2001): 3462-3472.

Sternecker, K., & Geist, J. (2010). The effects of stream substratum composition on the

emergence of salmonid fry. Ecology of Freshwater Fish, 19(4), 537-544.

Su, G., Logez, M., Xu, J., Tao, S., Villéger, S., & Brosse, S. (2021). Human impacts on global

freshwater fish biodiversity. Science, 371(6531), 835-838.

Sueker, Julie K., et al. "Effect of basin physical characteristics on solute fluxes in nine

alpine/subalpine basins, Colorado, USA." Hydrological Processes 15.14 (2001): 2749-2769.

55
Tsuchida, S. (1995). The relationship between upper temperature tolerance and final

preferendum of Japanese marine fish. Journal of Thermal Biology, 20(1-2), 35-41.

Turunen, J., Elbrecht, V., Steinke, D., & Aroviita, J. (2021). Riparian forests can mitigate

warming and ecological degradation of agricultural headwater streams. Freshwater

Biology, 66(4), 785-798.

Varanka, Sanna, Jan Hjort, and Miska Luoto. "Geomorphological factors predict water

quality in boreal rivers." Earth Surface Processes and Landforms 40.15 (2015): 1989-1999.

Wang, Guangxing, et al. "Spatial and temporal prediction and uncertainty of soil loss using

the revised universal soil loss equation: a case study of the rainfall–runoff erosivity R

factor." Ecological Modelling 153.1-2 (2002): 143-155.

Wang, Y., Liu, X., Wang, T., Zhang, X., Feng, Y., Yang, G., & Zhen, W. (2021). Relating

land-use/land-cover patterns to water quality in watersheds based on the structural equation

modeling. Catena, 206, 105566.

Wilkinson, D. M. (1999). The disturbing history of intermediate disturbance. Oikos, 84(1),

145-147.

Wold, Herman. "Estimation of principal components and related models by iterative least

56
squares." Multivariate analysis (1966): 391-420.

Wold, Herman. "Model construction and evaluation when theoretical knowledge is scarce:

Theory and application of partial least squares." Evaluation of econometric models.

Academic Press, 1980. 47-74.

Yu, Songyan, et al. "Effect of land use types on stream water quality under seasonal variation

and topographic characteristics in the Wei River basin, China." Ecological Indicators 60

(2016): 202-212.

Zhao, Jing, et al. "A comparison between two GAM models in quantifying relationships of

environmental variables with fish richness and diversity indices." Aquatic ecology 48 (2014):

297-312.

57