Using maps and models as a tool for conservation and management in the age of the anthropocene pieces of evidence from indigenous protists and a local landscape of the philippine archipelago

The first study reported potential suitable geographical distributions of three different bright-spored myxomycetes namely, Arcyria cinerea, Perichaena depressa, and Hemitrichia serpula.

THAI NGUYEN UNIVERSITY

UNIVERSITY OF AGRICULTURAL AND FORESTRY

JAMES EDUARD L DIZON

USING MAPS AND MODELS AS A TOOL FOR CONSERVATION AND

MANAGEMENT IN THE AGE OF THE ANTHROPOCENE:

PIECES OF EVIDENCE FROM INDIGENOUS PROTISTS AND A LOCAL

LANDSCAPE OF THE PHILIPPINE ARCHIPELAGO

BACHELOR THESIS

Study Mode: Full-time

Major: Environmental Science and Management

Faculty: International Programs Office

Batch: K49 – AEP

Thai Nguyen, 10/22/2021

DOCUMENTATION PAGE WITH ABSTRACT

Thai Nguyen University of Agriculture and Forestry

Degree Program Bachelor of Environmental Science and Management

James Eduard L DizonDTN1754290033Using maps and models as a tool for conservation and management in the age of the Anthropocene: Pieces of evidencefrom indigenous protists and a local landscape of the Philippine archipelago

Supervisor (s)

Abstract: Three independent yet cohesive topics that utilize maps and models to addressthe gaps in major Anthropocene issues related to environmental management in thePhilippines is employed for this thesis The first study reported potential suitable

geographical distributions of three different bright-spored myxomycetes namely, Arcyria cinerea, Perichaena depressa, and Hemitrichia serpula Three different modeling

approaches employing MaxEnt were performed in this study points this: (i) expansion ofthe localized fundamental niches of the three myxomycetes species, (ii) isothermality(BIO3) is the most influential bioclimatic predictor, and (iii) models developed in thisstudy can serve as a useful baseline to enhance the conservation efforts for most habitats

in the country that are directly affecting microbial communities due to rampant habitatloss and rapid urbanization The second study of this thesis performed simple bioclimaticmodeling to update the anecdotal reports of the disease-causing pathogen on our common

maize plants, Peronosclerosopora philippinensis.

Thesis Title

Student name

Student ID

The correlative modeling also performed in this study showed the following: (i) meandiurnal temperature (BIO2) affects the ecological distribution of the disease, (ii) rangeexpansion on other plantations of the country, and (iii) suggest potentialities on placeswhere the species is most likely to infect The last component of this thesis utilizesremote sensing technology to cover the urban coastline of Metro Manila.Interestingly, this component has yielded the following results: (i) between 1992 and

2020, shoreline changes have been detected within approximately 1.5 km decreased,(ii) The northern part of the study area, which shifted from being composed of treesand grasslands to now enormous fishponds, and (iii) the critically important Ramsarsite, LPPCHEA, have maintained the preservation of its natural mangrove forest.Overall, this Bachelor’s thesis has shown how maps and models can be used increating narratives that can address interconnected environmental issues However,despite these advantages, this new mode of visuals should always be treated withcaution and utmost critical interpretations Nevertheless, in silico/computer-assistedstudies is the modern approach that can be used by future environmental scientistsand managers to address pressing issues in this era of the Anthropocene

Keywords: conservation, machine learning, maximum entropy, niches,

urbanizationNumber of pages 78

Submission:

iii

● Firstly, I would like to thank MY FAMILY (Papa, Mama, Kuya, and Miggy)

for all the support they have given me throughout my thesis and my journey in myacademic life I wouldn’t accomplish all of this without them

● To my thesis supervisors, Dr Nikki Heherson A Dagamac and Dr Duong

Van Thao, a big thanks for helping and guiding me in conducting my thesis.

● To Dr Sittie Aisha B Macabago of the University of Arkansas, Fayetteville,

USA, thank you for the help that you gave during my thesis especially on MaxEntmodeling of the bright-spored myxomycetes

● To Dr Reuel M Bennett of the University of Santo Tomas, Manila,

Philippines thank you for sharing your knowledge on the oomycete pathogens,

Peronosclerospora philippinensis.

● To the AEP Family, thank you for the help, support, understanding, updates,

and for answering all the questions about the thesis

● My Vietnam family/friends, Henry, Raphael, Isaiah, JC, Ella, Angel, Elisha,

Jemimah, Ronnieca, Hanna for your continuous love and support.

● To my friends, Dale, Elmo, Austin, Marc, Francis, Noehl for your

understanding and support

● To King for being there when I needed his help and guidance.

● To my mentor/life coach/adviser/brother, thank you for all the lessons that

you have taught me and all the advice that you gave me that helped me inaccomplishing the things that I never thought I would be able to do Thank you forbelieving in me and trusting my abilities, and for seeing the best in me even when Idon't believe it myself

● To all who helped during the process of my thesis from the planning,brainstorming, and up until the very last step, Thank you! To all of those whosupported and believed in me, all the stress, the hard work, the headache paid off.Thank you very much, I appreciate it all

This Bachelor’s Thesis is dedicated to my family for their

never-ending love and support.

My Father, Eric M Dizon

My Mother, Marilou L Dizon And my two brothers, Eric Jason L Dizon & Jericho Miguel L Dizon

You have been my source of inspiration throughout my academic life Your love and support have been my strength during the hard

times and because of all of you, I made it.

v

TABLE OF CONTENTS

List of Figures 1

List of Tables 2

List of Abbreviations 3

CHAPTER I INTRODUCTION 4

1.1 Research rationale 4

1.2 Research questions and hypotheses 5

1.2.1 Maxent modeling of three bright-spored species 5

1.2.2 Peronosclerospora philippinensis (downy mildew) in the Philippines 6

1.2.3 LULC of urban coastline of Metro Manila 7

1.3 Research objectives 8

1.3.1 Maxent modeling of three bright-spored species 8

1.3.2 Peronosclerospora philippinensis (downy mildew) in the Philippines 8

1.3.3 LULC of urban coastline of Metro Manila 9

1.4 Scope and limitations 9

1.5 Definition of terms 10

CHAPTER II LITERATURE REVIEW 11

2.1 Myxomycetes 11

2.2 Species Distribution Modeling (SDM) 13

2.3 Land use/ Land cover classification using remotes sensing and its application to coastline studies 14

CHAPTER III MATERIALS AND METHODS 16

3.1 Maxent modeling for the prediction of the suitable local geographical distribution of selected bright spored myxomycetes in the Philippine archipelago 16

3.1.1 Occurrence data and environmental layers 16

3.1.2 Modeling procedure 17

3.2 Updating the potential Philippine distribution of the maize pathogen, Peronosclerospora philippinensis (downy mildew), using predictive machine learning approach 19

3.2.1 Data Gathering 19

3.2.2 Model performance and calibration 21

3.3 Land use land cover change and coastline change detection of the urban coastline in Metro Manila, Philippines 22

3.3.1 Study Area 22

3.3.2 Gathering of maps and data 24

3.3.3 Processing of images 24

3.3.4 Classifying the data 25

3.3.5 Accuracy Assessment 26

CHAPTER IV RESULTS AND DISCUSSION 29

4.1 Maxent modeling for the prediction of the suitable local geographical distribution of selected bright spored myxomycetes in the Philippine archipelago 29

4.1.1 Results 29

4.1.2 Discussion 36

4.2 Updating the potential Philippine distribution of the maize pathogen, Peronosclerospora philippinensis (downy mildew), using predictive machine learning approach 41

4.2.1 Results 41

4.2.2 Discussion 42

4.3 Land use land cover change and coastline change detection of the urban coastline in Metro Manila, Philippines 44

4.3.1 Results 44

4.3.2 Discussion 50

CHAPTER V SUMMARY AND CONCLUSION 53

REFERENCES 56

APPENDICES 68

vii

List of Figures

Figure 1. A) The map of the Philippines shows the location of Metro Manila B) Metro Manila and the provinces surrounding it C) Landsat Map showing Metro Manila and the chosen study

area 23

Figure 2. Occurrence points of three bright-spored species in the Philippines based on the

published geographic coordinates of species occurrences where each of the three bright-spored

species was recorded 31

Figure 3. Results area under the curve (AUC) analysis, including mean AUC values for eachbright-spored species obtained using the three model approaches 33

Figure 4. Species distribution models for the three bright-spored species of myxomycetes

showing a map of the Philippines and the predictive suitable habitat areas under the three model

approach generated by maximum entropy algorithm The maps were presented on a heat mapbased on the calculated probability of occurrence for the three bright-spored species 35

Figure 5. Species distribution models for the localized distribution of Peronosclerospora philippinenses and the predictive suitable habitat areas under the current and two climate

storylines (A2 and B1 scenarios) generated by maximum entropy algorithm The maps werepresented on a heat map based on the calculated probability of occurrence 41

Figure 6. A) Map of Metro Manila showing the location of LPPCHEA in a thick red box B)

An enlarged map that shows the location of LPPCHEA inside a thick red box 46

Figure 7. Land Use Land Cover change map from 1992-2020 of the Urban coastlines of Metro

Manila 47

Figure 8. Overview of the major changes that happened in the urban coastline of Metro Manila.

A) Map of Metro Manila that shows the part of the coastline that has been changed (Source:Google Earth Pro) In thick black boxes are the highlighted areas that emphasized B & D)Coastline of the year 1992 (marked as the blue thin line) C & E) Coastline of the year 2020(marked as the green thin line) 49

List of Tables

Table 2. Land cover classes used in the study and its definition 25

Table 3. List of environmental variables in the Philippines used for the three-model approachperformed for this study and its percent contribution and Mean AUC values Model approach

1 included all 19 bioclimatic variables with default regularization setting; model approach 2

increased the regularization multiplier suggested after ENMeval calculations; Model approach

3 includes the selected 9 bioclimatic variables after autocorrelation 32

Table 4. Percentage and size of area of each class for the classified image of the study area 45

Table 5. Length of the Urban coastline from 1992-2020 48

Table 6. Overall Accuracy and Kappa Coefficient of the classified datasets 48

2

List of Abbreviation AUC

Diurnal Temperature RangeFeature Type

Global Climate ModelInternational Union for Conservation of NatureLas Piñas – Parañaque Critical Habitat and Ecotourism AreaLand Use Land Cover

Mindoro, Marinduque, Romblon, PalawanOperational Land Imager

Regularization MultiplierReceiver Operating CharacteristicThematic Mapper

United States Geological Survey

CHAPTER I INTRODUCTION

1.1 RESEARCH RATIONALE

The age of the Anthropocene is raised with enormous environmental threats.Besides the obvious problem brought by the changing climate on major naturalresources of the world, anthropocentric activities such as urbanization,industrialization, etc have certainly ameliorated many global pressing environmentalissues including developing third world countries like the Philippines

The Philippines is an archipelago known to have the richest biodiversity in theSoutheast Asian region Despite the known distribution of much indigenous flora andfauna that have been reported for the last decades, major microbial communities thatplay a vital role in agricultural or environmental processes have remainedcircumstantial Moreover, the coastline of the country is exposed to variouscomplicated natural processes that always result in long and short-term changes.Littoral transport is responsible for carrying eroded materials along the beaches bywaves and currents in the near-shore zone, which results in shoreline alteration Thesechanges in the coastal ecosystem all directly affect humankind, infrastructures, land,coastal natural ecosystems, and coastal socio-economic value (Misra and Balaji 2015).For instance, human activities during the Anthropocene caused many coastal habitats

to be severely impacted by eutrophication and chemical pollution in many coastlines

of the Southeast Asian region

Given the same rate of effect of many anthropocentric activities both at thelevel of wetland landscape and indigenous microbial flora in the country, policyrecommendations for sustainable environmental management that utilizes science-

4

based evidence using the aforementioned subjects are still considered in their infancy.

In fact, mapping and modeling techniques promise visuals that can be used to predictpotentialities on the distribution of biological resources, update the expansive nature

of plant pathogens, and explain changes in land use Such visuals are apparentlyadvantageous in creating possible management strategies that can be employed at theconservation of forest ecosystems, agricultural management, and urban development

of the Philippines

These are the main themes that this Bachelor’s thesis wishes to address Hence,this thesis is subdivided into three independent yet cohesive studies that utilize eithermaps or models to address specific questions and hypotheses that are considered to be

a major research gap in terms of environmental management at the age of theAnthropocene

1.2 RESEARCH QUESTIONS AND HYPOTHESES

Background: Among the countries in Southeast Asia, the Philippines have been able to

document the greatest number of records of plasmodial slime molds (myxomycetes) forthe region, currently having a total of 162 species (Macabago et al 2020) Over the lastdecades, since the myxomycetes surveys have been conducted in the Philippines, most ofthe assessments were able to show the following: (1) major terrestrial ecosystems harbor

a diverse myxoflora for the country (Bernardo et al 2018), (2) variation on the diversity

of most myxomycetes species randomly collected on different substrates collected onpriority areas for conservation in the Philippines occurs

(Macabago et al 2017; Pecundo et al 2017), and (3) clear differences on theoccurrences of myxomycete assemblages exist (Dagamac et al 2017) So far, there isonly a single study that used species distribution model (SDM) for Philippinesmyxomycetes In the paper of Almadrones-Reyes and Dagamac (2018), the suitable

habitat for the common dark-spored myxomycetes in the tropics, Diderma

hemisphaericum, was determined It also predicted the range expansion of the species

in other islands of the Philippines in response to two climate change scenarios (A2 andB1) With many reported myxomycetes species in the country, none have tried topredict the geographical niches suitable for bright-spored myxomycetes

Question: Using maxent modeling, what are the probable suitable habitats for the

three selected bright-spored myxomycetes species?

Hypothesis: The bioclimatic factor influences the determination of the expanding

range shifts of putative suitable habitats where the three selected bright-sporedmyxomycetes will thrive

1.2.2 Peronosclerospora philippinensis (downy mildew) in the Philippines

Background: The Philippines' maize-growing agricultural industry has been plagued for

a long time by downy mildews, more specifically by its causal pathogen,

Peronosclerospora philippinensis The earliest record according to Exconde (1982) was

first conducted by Baker (1916), but the most definitive and comprehensivedocumentation of the disease in the Philippines was done only by Weston (1920) For thelast decades, high disease occurrence has been reported in many parts of the Philippinesespecially in Northern Luzon and in many areas of Mindanao despite many

6

major crop techniques that have been utilized to mitigate the spread of the pathogenic

disease Being the most virulent of the downy mildew family, P philippinensis

severely causes economic loss to corn production with ca 40-60% decrease in cropyield observed every time there is an incidence of the pathogen Despite this, therecords for the disease are merely anecdotal In addition, the distribution of the disease

and risk maps for many plant pathogens in the country is still a missing piece.

Question: What is the possible distribution of the pathogen, Peronosclerospora

philippinensis, under different climate change scenarios?

Hypothesis: Similar to other fungal-like protist allies, these pathogens will have an

expanding range shift under different climate change storylines

1.2.3 LULC of urban coastline of Metro Manila

Background: The urban coastline of Metro Manila is a prime example of a polluted

environment (Chang et al., 2009) It is one of the country’s most important bodies ofwater because it is home to an international port, a large fishing area, and an oyster andmussel aquaculture site (Prudente et al 1994) Since pre-Hispanic times, the bay has beenthe center of socio-economic growth, with both local and international ports It also hasextensive natural resources, which have historically been the principal source of incomefor communities in the bay’s coastal section Due to the rapid rise in population andindustrialization in the watershed due to the growing human population, the bay’s waterquality has decreased substantially (Jacinto et al 2006) Increased incidences of hypoxiaand anoxia, regular blooms of toxic microalgae, and chronic red tides generated bydinoflagellates are all consequences of increased organic loads

entering the bay from excessive urban emissions of nutrients (nitrogen and phosphorus)and heavy metals (Chang et al 2009) Manila Bay is one of the marine pollution hot spots

in the East Asian Seas, according to the GEF/UNDP/IMO/PEMSEA project (Maria et al.,2009) However, the solution for these environmental issues described herein entailsproper and sustainable management

Question: What are the major changes in the urban coastline of Metro Manila in the

span of 30 years (1992-2020)?

Hypothesis: A clear change in the coastal land use in Metro Manila’s Bay over the

last 30 years is imminent

● To show the possible suitable habitats and distribution of the three spored species in the Philippines

bright-● To create a predictive distribution map of the three bright-spored species using three different model approaches

1.3.2 Peronosclerospora philippinensis (downy mildew) in the Philippines

● To produce maps that visualize the distribution of Peronosclerospora

philippinensis species in the Philippines under different climate change scenarios.

8

1.3.3 LULC of urban coastline of Metro Manila

● To create a map that shows the changes in the land use/land cover around the coastal area of Metro Manila

● To show the major changes that happen in the coastline in the last 40 years

● Three independent Anthropocene issues are addressed in this study: (1)potential habitat suitable for three selected bright-spored myxomycete species in thePhilippines, (2) updating the distribution of a plant pathogen under changing climatescenarios, and (3) detect the land use and land cover change of the coastline in theurban capital of the Philippines

● To address the issues mentioned above, this study will employ two importantvisualization techniques ([1] predictive models generated using MaxEnt algorithm and[2] LULC classification maps performed utilizing the ArcGIS software) to addressenvironmental management issues at the age of the Anthropocene

● The results of this study are strictly performed using computer-generated insilico analysis, hence no field ground-truthing or site validation has been performeddue to the restriction implemented by the Philippine local government during thecourse of the research study

9

1.5 DEFINITION OF TERMS

Bioclimatic variables are commonly used in species distribution modeling and similar

ecological modeling approaches to represent annual trends, seasonality, and extreme

or limiting environmental circumstances

Interactive Supervised Classification is an ArcMap tool that speeds up classification

that includes all the bands available in the image layer selected.

International Union for Conservation of Nature (IUCN) is an internationalorganization dedicated to the conservation of nature and the sustainable management

of natural resources

PhilGIS is a website where you can access and download different Philippine geospatial

data

10

CHAPTER II LITERATURE REVIEW

2.1 Myxomycetes

Myxomycetes are a small group of species, 998 of which are distributedworldwide It is categorized in the kingdoms of Plantae and Animalia sincemyxomycetes are usually found in the same environments as fungi, and are considered

a taxon within the Fungi kingdom (Baba & Sevindik, 2018) Researchers performed aphylogenetic study of highly conserved, 1-alpha (EF-1α) gene sequences of the) gene sequences of theelongation factor and showed that myxomycetes are not fungi (Baldauf & Doolittle,1997)

A myxomycete's life cycle comprises two morphologically distinct trophic stages,one consisting of uninucleate amoebae, and the other consisting of a distinctive network

of multinucleates; the plasmodium (Baba, 2012) Bacteria are consumed by theplasmodium, hyphae fungi, and other micro-organisms A large variety of microbes canthus function as nutrient species Certainly, bacteria are the most important of thosenutrients (Baba & Sevindik, 2018) The food of myxomycetes are bacteria and fungi, butlater in the life cycle of myxomycetes, the engulfed bacteria or fungi develop mutualitywith myxomycetes (Cohen, 1941) Myxomycetes are phagotrophic bacteriovores andfungivores They might also make use of some organic matter (Ergul et al., 2005) Thepresence of myxomycetes is correlated with rotting or living plant material in terrestrialforest habitats Humidity and temperature play a major role in their diversity andabundance, and other physical and biotic factors such as light intensity, pH substrate,environmental degradation, and the existence of bacteria, fungi,

and insects have also been considered Environmental pollution and the rise in toxinsalso reduce the diversity of Mycetozoa (Ko et al., 2011).

Unlike other protists, myxomycetes or commonly known as 'slime molds’ producemacroscopic fruiting species that are fairly easy to capture and classify (Stephenson &Stempen, 1994) Slime molds are important because they accumulate high metals in theircells, similar to fungi (Keller & Everhart, 2010) Slime molds eat bacteria and othermicroorganisms, but they also provide suitable substrates and habitats for different kinds

of fungi and insects, primarily Coleoptera or beetles, Latridiidae, and Diptera or flies Infact, some beetle species use not only the spores but also the plasmodia of slime molds as

a nutritional source (Stephenson & Stempen, 1994) The distribution of myxomycetes iswidespread and has been observed in a wide variety of habitats, including temperateforests (Kazunari, 2010; Takahashi & Hada, 2009), tropical rainforests (Dagamac, 2012),dry land ecosystems, and northern Siberia tundra Myxomycete diversity has also beeninvestigated in soils as part of the greater protozoan community (Feest & Madelin, 1985;Kamono et al., 2009) These studies found that the abundance of myxomycetes washighest in grassland and agricultural soils (Feest & Madelin, 1985) While urbanization isone of the most ubiquitous types of disruption, few studies have been conducted on theurban ecology of myxomycetes (Ing, 1998) In addition, myxomycetes tend to be immune

to a variety of disturbance types For example, forest fragmentation and habitat loss havegenerally reduced the diversity of myxomycetes in the Amazon rainforest (Rojas &Stephenson, 2013)

12

2.2 Species Distribution Modelling (SDM)

Modeling the potential distribution of many macroecological organisms (Cabral et

al 2017; Connolly et al 2017; Guisan and Rahbek 2011) have been widely tackled inmany kinds of literature including those organisms that are ephemeral in nature, such asfungi (see Ocampo-Chavira et al 2020; Yuan et al 2015; Rohr et al 2011) or fungalallies (see Duque-Lazo et al 2016, Aguilar and Lado 2012) that are once classified asspecies belonging to the Kingdom Fungi In fact, species distribution models are anemerging tool in the study of fungi, and their use is expanding across species and researchtopics (Hao et al 2020) However, in spite of the growing interest for this important tool

to be utilized, most of the reported studies concentrated on macrofungi (see Sato et al.2020), lichens (see Dymytrova et al 2016, Braidwood and Ellis 2012), and fungalpathogens (see Bosso et al 2017; Narouei-Khandan et al 2017) Very limited SDMstudies have been reported so far, especially on fungus-like protists that are widely known

to be an important microbial predator on the soil biota Among these protists,myxomycetes are one of the few groups with macroscopically visible fruit bodies that arefound in a wide array of ecological habitats Unlike the true fungi, myxomycetes arepredominantly sexual (Feng et al 2016) and the fructifications of these protists have verylimited diagnostic morphological characters which can easily be withered Moreover,myxomycetes are classified as a monophyletic taxon within the Amoebozoa (Adl et al

2012, 2018; Ruggiero et al 2015) and are classified into two clades: bright-spored anddark-spored, which are now officially called Lucisporomycetidae andColumellomycetidae, respectively Rostafiński (1875) established the first classificationbased on comprehensible criteria, dividing

myxomycetes into two "subdivisions" based on the color of the spore mass: the spored and bright-spored This classification was based on a combination offructification morphological characteristics, although plasmodium appearance andfruiting body growth were also taken into account to some degree (Ross 1973).

dark-2.3 Land use/ Land cover classification using remote sensing and its application

to coastline studies

Several types of thematic data crucial to GIS analysis, such as data on land use andland cover features, are mostly derived via remote sensing Landsat satellite images andaerial photos are commonly used in assessing the land cover distribution (Rwanga

&Ndambuki, 2017) During the late twentieth and early twenty-first centuries, rapidand uncontrolled population growth, combined with industrialization, accelerated therate of land-use/land-cover (LULC) change many times, especially in developingnations (Talukdar et al 2020)

LULC change is important in a variety of sectors that rely on Earth observations,including urban planning (Hashem et al 2015; Rahman et al 2012), environmentalvulnerabilities, and impact assessment (Liou et al 2017; Talukdar et al 2018; Nguyen et

al 2016), natural calamities and hazards observation (Che et al 2014; Dao et al 2015;Zhang et al 2019), and soil erosion and salinity assessment (Chen et al 2019; Braun &Hochschild, 2017) LULC is becoming more widely recognized as a major driver ofchanges in the environment (Lambin et al 2001; Goldewijk & Ramankutty 2004) Thecurrent challenge is to preserve the natural environment while maintaining or improvingthe economic and social benefits derived from their use As a result, it is important to

14

comprehend the pattern and trends of LULC changes Developments in remotesensing and related technologies have allowed for the collection of usefulspatiotemporal data on LULC Within the last two decades, the search for methodsused in obtaining and producing accurate LULC classification and identifying LULCchange over time has been a major focus of remote sensing research (Manandhar et al.

availability and high spatial coverage Research on mapping, monitoring, andpredicting LULC trends have been conducted over the last few decades usingmedium- and low-resolution observations from satellites such as Landsat, ModerateResolution Imaging Spectroradiometer (MODIS) Indian Remote Sensing (IRS)Advanced Spaceborne Thermal Emission, and Reflection Radiometer (ASTER),Satellite for observation of Earth (SPOT) and others (Mas et al 2017; Wentz et al.2008; Toure et al 2018; Usman et al 2020; Stefanov & Netzband, 2005)

Assessing changes in land use/land cover (LULC) remains significant inenvironmental issues and environmental sustainability since it helps to understandbetter and visualize the changes that have occurred in the environment Significantglobal population growth has been followed by economic activity that has resulted inurbanization and subsequent construction land development, resulting in rapid LULCshifts (Guan et al 2011; Halmy et al 2015; Zheng et al 2015) Monitoring the LULCprovides an effective, sustainable plan for the urbanized coastal area that is significantfor improving future urban development and management

CHAPTER III MATERIALS AND METHODS

3.1 MAXENT MODELING FOR THE PREDICTION OF SUITABLE LOCAL GEOGRAPHICAL DISTRIBUTION OF SELECTED BRIGHT SPORED MYXOMYCETES IN THE PHILIPPINE ARCHIPELAGO

3.1.1 Occurrence data and environmental layers

For this study, three bright-spored myxomycete species were selected based on

their known occurrence in the Philippines: Arcyria cinerea representing the abundantly/cosmopolitan occurrence, Perichaena depressa depicting common occurrence, and Hemitrichia serpula as the occasionally occurring slime molds The

distribution of these bright-spored representatives was surveyed using all known localreports, grey publications, and personal records accounted by the last author of thisstudy To verify the accuracy of all the 201 geographic coordinates used for thiscorrelative modeling study, an initial data checking was conducted All thecoordinates were initially transformed into a CSV file that was then overlaid on aPhilippine map using ArcGIS ver 10.3 All points that were eliminated on the basemap were then rechecked and corrected

The environmental layers for this study were the 19 bioclimatic variables in thePhilippines with a raster resolution of 1km obtained from the PhilGIS website(http://philgis.org/) Since the downloaded environmental layers from PhilGIS wereall in GeoTIFF format, the ArcGIS software was utilized to convert all the 19 GRIDfile layers into an ASCII extension (file format compatible with the modeling softwareused for this study)

16

3.1.2 Modeling procedure

MaxEnt software (ver 3.4.4) was downloaded fromhttps://biodiversityinformatics.amnh.org/opensource/maxent/ MaxEnt generalizesindividual observations of species presence using entropy and does not require or eveninclude points where the species is absent within the theoretical context MaxEntranks a species' habitat suitability on a scale of 0 to 1, with 0 being the least suitableand 1 being the most suitable (Kamyo and Asanok,2020) In this study, threeapproaches (Table 3) were used for each bright-spored myxomycetes species Firstly,with the use of MaxEnt's default settings (see Table 3) Secondly, to provide pseudo-absence correction, the input files were subjected to an ENMeval analysis performedusing R Studio Lastly, the autocorrelations among the 19 bioclimatic variables wereanalyzed, reducing now the possible environmental layers that can be used for thecorrelative modeling

For the first model, the transformed CSV file of the occurrence records of eachbright-spored myxomycetes and the converted ASCII format of the 19 bioclimaticenvironmental layers were used as input files in the MaxEnt software The model wasrun using the default regularization settings (regularization multiplier = 0, feature type

= Auto) in Maxent To determine the significance of each biophysical variable, thefollowing settings were chosen: (i) “Create a response curve” and “Do jackknife test

to measure variable importance,” and (ii) the output format was set to “logistic” Inaccordance with Yang et al (2013), the random test percentage was adjusted to 30%and the file format turned into logistic for all models A total of 10 runs were set for

each model approach The algorithm runs either 1000 iterations of these processes orcontinues until it reaches a convergence threshold of 0.00001 (Yuan et al 2015).

In the second model, ENMeval was used to optimize model complexity in order tobalance the goodness of fit and predictive ability In addition, the use of this R-basedmodeling evaluation modifies the models to improve predictive ability and avoids issues

of possible overfitting (Muscarella et al 2014) For this approach, the fine-tuned settinggenerated from the ENMeval analysis (method = randomkfold, kfold=10) suggested theadjustment of a regularization multiplier (RM) and feature type (FT) for each bright

spored myxomycetes as follows: Arcyria cinerea (2.5 [RM] / LQHPT [FT]); Perichaena depressa (1 [RM]/ LQ [FT]) ; Hemitrichia serpula (2.5 [RM] / H [FT]).

For the third approach, the SDMToolbox in ArcMap 10.3 was utilized to checkfor autocorrelations among the environmental variables The ASCII file of 19environmental layers was uploaded in ArcMap 10.3 and the tool “Remove highlycorrelated variables” was used under the SDMToolbox Variables with correlationcoefficients of >0.8 were chosen following the Spearman correlation for a total of 9variables (BIO2, BIO4, BIO7, BIO8, BIO12, BIO16, BIO17, BIO18, and BIO19).These variables were used to produce the ENMeval analysis for each species in the

third model approach In this case ENMeval suggested the following settings: Arcyria

cinerea (2 [RM] / LQH [FT]); Perichaena depressa (0.5 [RM]/ LQ [FT]); Hemitrichia serpula (2.5 [RM] / LQHPT [FT]) The random sampling process was

performed ten times for all the models to make sure that the results were not affected

by the random collection of points and the average of those ten runs was used in thisstudy

18

The resulting models were then evaluated using the receiver operatingcharacteristic (ROC) analysis, which eventually generated the area under the curve(AUC) scores (Philipis and Dudik, 2008) to determine the model’s goodness-of-fit AJacknife procedure was used to calculate the contribution of the variables used for themodel prediction The final output of all models produced a distribution map withpotential incidence values for each grid cell, ranging from 0 to 1 For bettervisualization, the ASCII output file format from MaxEnt was imported into ArcMap10.3 software The map was then divided using a defined interval as a classificationmethod Based on projected habitat suitability, the habitat suitability of each specieswas divided into five categories, which are as follows: Very low suitability (0-<0.2),Low Suitability (0.2-<0.4), Moderate Suitability (0.4-<0.6), High Suitability (0.6-

<0.8), Very High Suitability (0.8-1)

3.2 UPDATING THE POTENTIAL PHILIPPINE DISTRIBUTION OF THE

MAIZE PATHOGEN, PERONOSCLEROSPORA PHILIPPINENSIS (DOWNY

MILDEW), USING A PREDICTIVE MACHINE LEARNING APPROACH

3.2.1 Data gathering

Geographical coordinates of the pathogen occurrence were retrieved on threedifferent platforms First, grey literature includes anecdotal reports and scientificresearches that cited the appearance of the disease on corn For older records with nogeographical information recorded originally, the geographic centroid was estimated bydoing a triangulation of the closest known locality wherein the agricultural land for corn

19

is most likely planted Second, are the electronic database that reports actualcoordinates, and lastly, the personal recording or anecdotes of people that have beenable to survey the occurrence of the disease in the Philippines Initial data verificationwas performed to confirm the accuracy of all of the 12 geographic coordinates used inthis correlative modeling study All of the coordinates were first converted into a CSVfile, which was then overlaid on a Philippine map using ArcGIS 10.3 For theenvironmental layers, the current climatic conditions of 19 ‘bioclimatic’ variables datawere collected from PhilGIS web portal (http://ww.philgis.org) with a raster resolution

of 1km obtained from the PhilGIS website (http://philgis.org/) To predict the future

potential distribution of P philippinensis in the Philippines, the future climate

projection data were downloaded at the GCM downscaled data portal(http://www.ccafs-climate.org/) The global climate model (GCM), CICRO modelrepresenting simulations from the fifth assessment of the Intergovernmental Panel forClimate Change (CMIP5) were patterned from the study of Almadrones-Reyes &Dagamac (2018), hence selecting future climatic conditions of the year 2080 with twodifferent storylines of future climate as representative for this study The A2 storylinefollows an increasingly growing population with a higher carbon emission rate thatleads to a predicted global mean temperature increasing by 3.4° Celsius The B1storyline assumes a modification in the material production that causes technology toadvance more efficiently, hence, predicting an increase by only 1.8° Celsius to theglobal mean These two future scenarios were downloaded as an ESRI file and werethen converted to ASCII format to fit the requirements in the MaxEnt software

3.2.2 Model performance and calibration

The software MaxEnt (maximum entropy species modeling), Version 3.4.4 was

used to map the potential geographic distribution of the 12 coordinates identified for P philippinensis incidence in the Philippines The selection of MaxEnt is based on the

known advantages of the software it utilized (i) present-background data, (ii) effective atlow numbers of occurrence data, and (iii) robust production of models due to theestimation of the probability distribution of species occurrence based on the currentpresence points and randomly generated background points of environmental conditions

by finding the maximum entropy distribution of the species To optimize modelcomplexity that will improve the predictive ability of the model runs to avoid issues ofoverfitting, ENMeval was used (Muscarella et al 2014) For this approach, the fine-tunedsetting generated from the ENMeval analysis (method = jacknife, kfold=10) suggestedthe adjustment of a regularization multiplier (RM.) and feature type (FT) for each model

runs as follows: Current Climate (2.5 [RM] / LQH [FT]); A2 storyline (4 [RM]/ LQH [FT]); B1 storyline (4 [RM] / LQHPT [FT]) The settings and the model were employed

at convergence threshold (10–5), maximum iterations (5000), and the maximum number

of background points (10000) to run the model A Jackknife test was also used to estimatethe relative importance of each of the selected variables to the model development Themodel performance was tested using the area under the curve (AUC) of the receiveroperating characteristic (ROC) AUC values in the model output range from 0 to 1(unsuitable to highly suitable) When AUC shows the values below 0.5 then it can beinterpreted as a random prediction Response curves were used to study the relationshipsbetween bioclimatic variables and the predicted probability of

21

the presence of P philippinensis The output file that was generated from MaxEnt was

then exported as ASCII file format To get a detailed visualization, the ASCII outputfile format from MaxEnt was imported into ArcMap 10.4 software In the ArcMapsoftware, the map was then divided using defined intervals as a classification methodinto four different categories

3.3 LAND USE LAND COVER CHANGE AND COASTLINE CHANGE DETECTION OF THE URBAN COASTLINE IN METRO MANILA, PHILIPPINES

3.3.1 Study Area

Metro Manila is the Philippines' primary tourist attraction and is also known as the country's national capital region The city serves as the Philippines' commercial, educational, and entertainment hub Although being the smallest region in the

country, it is the most populated of the country's twelve declared metropolitan areas and the world's 11th most populous region, with a population of 16.3 million people Located on the island of Luzon, bordered by the provinces of Bulacan, Rizal, Laguna, Cavite, and stretches along the eastern shore of Manila Bay Since the study focuses

on the coastline, the eastern side of Metro Manila was excluded and the western side which is closer to the coastline was chosen as the study area to visualize the changes

in the urban coastline (Fig 1)

Figure 1 A) The map of the Philippines shows the location of Metro Manila B) Metro Manila and the provinces surrounding it C) Landsat Map showing Metro Manila

and the chosen study area (Source: https://earthexplorer.usgs.gov/

23

3.3.2 Gathering of maps and data

Four Landsat datasets were acquired from United States Geological Survey(USGS) EarthExplorer (https://earthexplorer.usgs.gov) to track changes along theUrban coastline of Metro Manila To avoid misinterpretation during the classification,the images with the least amount of cloud cover for each year were chosen to be used

in this study This dataset includes two images acquired by the Landsat 5 ThematicMapper (TM) sensor for the years 1992 and 2005 (Table 1), and two images from theLandsat 8 Operational Land Imager (OLI) sensor for the years 2014 and 2020 (Table1) Moreover, the vector data that was used to clip the study area were downloadedfrom DIVA-GIS (https://www.diva-gis.org/)

Table 1 Detailed information of the datasets used in this study.

Satellite Sensor No of Bands Acquisition Date Spatial Resolution (m)

Landsat images of each dataset using the “Extract by Mask” tool from the ArcToolbox

of ArcMap 10.8 software to produce the map of the study area that will be used for theclassification Lastly, all the maps for each year were finalized before going on to theclassification.

3.3.4 Classifying the data

In this study, supervised classification was utilized to classify the dataset for eachyear To designate a specific class in supervised classification, the user selects pixels inthe image that have the same properties The software ArcMap 10.8 will automaticallyselect each pixel that meets the class properties when the user assigns the classes (Masria

et al., 2020) Four classes were chosen to classify our research field as shown in Table 2.The Maximum Likelihood approach was used to apply supervised classification to thefour images Bands 4,3,2 for Landsat 8 and bands 3,2,1 for Landsat 5 were used to showthe natural color of the images for better classification

Table 2 Land cover classes used in the study and their definition (Source: https://www.arcgis.com)

Water Areas where water was predominantly present throughout the year; may

not cover areas with sporadic or ephemeral water; contains little to no

sparse vegetation, no rock outcrop nor built-up features like docks

Grass Open areas covered in homogenous grasses with little to no taller

vegetation; wild cereals and grasses with no obvious human plotting (i.e.,not a plotted field); examples: natural meadows and fields with sparse to notree cover, open savanna with few to no trees, parks/golf courses/lawns,pastures

Trees Any significant clustering of tall (~15-m or higher) dense vegetation,

typically with a closed or dense canopy; examples: wooded vegetation,clusters of dense tall vegetation within savannas, plantations, swamp ormangroves (dense/tall vegetation with ephemeral water or canopy too thick

to detect water underneath)

Built Human-made structures; major road and rail networks; large homogenousArea impervious surfaces including parking structures, office buildings, and

residential housing; examples: houses, dense villages/towns/cities, pavedroads, asphalt

25

After assigning the appropriate classes to the image, Interactive SupervisedClassification was used to classify the entire dataset After that, the output of theclassification was examined to see if the classification was done correctly by looking

at the classes and comparing it to the Landsat data After evaluating and validating the

classification result, it was converted to a polygon using ArcToolbox's "Raster to

Polygon '' tool The area size of each class can be determined by converting it to

polygon or vector data and utilizing the field calculator in the categorized layer'sattribute table After the conversion of the results of classified images from raster tovector as "polygons”, the Polygon features are converted to line features using theArcGIS/ArcToolbox Using the field calculator in the attribute table of each layer thelength of the coastline was calculated Afterward, the same techniques are used toobtain the length of the coastline across all years

3.3.5 Accuracy Assessment

Accuracy assessment is the last phase in the remote sensing data process, thatallows us to determine how well pixels were sampled into the appropriate land coverclasses (Anand, 2017) Furthermore, locations that could be properly seen on both theLandsat high-resolution image, Google Earth, and Google Map were prioritized foraccuracy evaluation pixel selection (Rwanga & Ndambuki, 2017) Without taking thenecessary steps to check the accuracy or validity of a map created with remotely sensed

or other spatial data, it cannot be considered a finished result (Anand, 2017) Toaccomplish an effective accuracy assessment, a comparison of two sources of data must

be conducted: (i) an interpreted map/classified image obtained from remote sensing data

and (ii) a reference map, high-resolution images, or ground truth data After aclassification process, the degree of error in the final result, which includes identifiedcategories on the map, must be determined Errors are the result of inaccurate pixellabeling for a category (Anand, 2017).

The overall accuracy, which is defined as the number of correctly identifiedpixels divided by the total number of pixels tested, is a regularly used metric ofmapping accuracy Despite the fact that overall accuracy is a measure of imageaccuracy across all classes, it ignores off-diagonal elements Furthermore, if variednumbers of accuracy sites were employed, it was impossible to compare different totalaccuracy levels The error matrix is also used to calculate two further accuracies:producer and consumer accuracies The errors of the class producers and users aredepicted in the above section's example All of these simple accuracy measurementscan give results owing to random pixel classification, meaning they do not provide away to compare accuracy statistically Therefore, another popular method known asKappa analysis should be used, which involves incorporating off-diagonal factors as aproduct of the row and column marginal totals It is a discrete multivariate techniquefor evaluating classification accuracy from a matrix of errors Kappa analysis yields akappa coefficient or Khat statistics, with values ranging from 0 to 1

Calculating an accurate measure for the entire image across all classes present inthe categorized image is desirable Overall accuracy, which determines the proportion ofpixels correctly identified, can be used to describe the map's overall accuracy for allclasses The overall accuracy is calculated using the following equation:

Overall Accuracy = ! " # x 100 ! $! %

27

The problem with overall accuracy is that the summary value is the overall average, which hides whether errors were evenly distributed among classes or if someclasses were particularly bad or good As a result, various types of accuracy, such as user and producer accuracy, were included The chance that a pixel classified on the image represents that category on the ground is defined as user accuracy The row reliability values (user accuracy) show the class reliability in the classified image Theprobability that any pixel in that category has been correctly identified is defined as the producer's accuracy The following formulas were used to compute the user and producer accuracy:

Number of Correctly Classified Pixels in each Category

User Accuracy = Total number of Classified Pixels in that category (Row Total) x 100

Number of Correctly Classified Pixels in each Category Producer Accuracy = Total number of Reference Pixels in that category (Row Total) x 100

Kappa analysis is a discrete mathematical procedure that is used to assessaccuracy (Jensen, 1996) The Khat statistic is a measure of agreement or correctnessderived from Kappa analysis (Congalton, 1991) The Khat statistic is calculated asfollows:

(TS x TCS) − Σ(Column Total x Row Total)

TS B − Σ(Column Total x Row Total)

The accuracy assessments for classified images were conducted with 30random points provided per class for each map using simple random sampling, withthe corresponding reference classes of each LULC category obtained fromtopographic maps, Google earth services, and their digital maps

28

CHAPTER IV RESULTS AND DISCUSSION

4.1 MAXENT MODELING FOR THE PREDICTION OF SUITABLE LOCAL GEOGRAPHICAL DISTRIBUTION OF SELECTED BRIGHT SPORED MYXOMYCETES IN THE PHILIPPINE ARCHIPELAGO

4.1.1 Results

A total of 201 occurrence points (98 for Arcyria cinerea, 67 for Perichaena

depressa, 36 for Hemitrichia serpula) found on different geographical ranges in the

Philippines were used for this study (Fig 2) The maximum entropy algorithm, whichwas used for species distribution modeling for three bright-spored myxomycetespecies, suggested several important bioclimatic variables that could affect the species'predictive local distribution in the Philippines

In the first model approach, isothermality (BIO3) has the highest percent

contribution for all three bright spored species with 32.6% for Arcyria cinerea, 32% for Perichaena depressa, and 29.7% for Hemitrichia serpula (Table 3) In terms of

permutation importance, the model calculated that isothermality (BIO3) was the

highest for Hemitrichia serpula with 39.2% and temperature annual change (BIO7) was the highest for the two other bright-spored species with 26.5% for Arcyria

cinerea and 23.7% for Perichaena depressa Furthermore, the Area Under the Curve

(AUC) values derived from maximum entropy-based results for the three bight-spored

species are (i) 0.873 for Arcyria cinerea; (ii) 0.841 for Perichaena depressa; and (iii) 0.824 for Hemitrichia serpula, indicating strong predictive model efficiency (Fig 3).

29

For the second model approach, isothermality has the highest percent

contribution for both Arcyria cinerea and Hemitrichia serpula with 38.9% and 50.7%, respectively But for Perichaena depressa precipitation of driest quarter contributed

the most accounting for 31.3% (Table 3) When it comes to permutation importance,the three species ranked isothermality (BIO3) as the highest For the values of AUC,

Arcyria cinerea has an AUC of 0.850, Perichaena depressa has an AUC of 0.793, and Hemitrichia serpula has an AUC of 0.816 (Fig 3).

Among the nine variables that were used for the third modeling approach,Temperature seasonality (BIO4), has the highest percent contribution in the third model

approach for both Arcyria cinerea and Hemitrichia serpula, with 41.1 and 43.5 percent,

respectively While the precipitation of the driest quarter (BIO14) contributed the most to

Perichaena depressa, accounting for 56.8 percent (Table 3) In terms of permutation

importance, the model calculated that temperature annual change was the highest for

Arcyria cinerea with 55.7% and temperature seasonality was the highest for the two other bright-spored species with 36.4% for Perichaena depressa and 33.5% for Hemitrichia serpula Arcyria cinerea, Perichaena depressa, and Hemitrichia serpula each had AUC

values of 0.802, 0.792, and 0.778, respectively (Fig 3)

Figure 2 Occurrence points of three bright-spored species in the Philippines based on the published geographic coordinates of species occurrences where each of the three

bright-spored species was recorded

31

Table 3 List of environmental variables in the Philippines used for the three model approach performed

for this study and its percent contribution and Mean AUC values Model approach 1 included all 19

bioclimatic variables with default regularization setting; model approach 2 increased the regularization multiplier suggested after ENMeval calculations; Model approach 3 includes the selected 9 bioclimatic variables after autocorrelation.

Model Approach 1 Model Approach 2 Model Approach 3

Arcyria Perichaena Hemitrichia Arcyria Perichaena Hemitrichia Arcyria Perichaena Hemitrichia cinerea depressa serpula cinerea depressa serpula cinerea depressa serpula