Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The relationships between rapid urban development and vegetation in the pan Pacific region : spatio-temporal… Lu, Yuhao 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2018_september_lu_yuhao.pdf [ 11.86MB ]
Metadata
JSON: 24-1.0368676.json
JSON-LD: 24-1.0368676-ld.json
RDF/XML (Pretty): 24-1.0368676-rdf.xml
RDF/JSON: 24-1.0368676-rdf.json
Turtle: 24-1.0368676-turtle.txt
N-Triples: 24-1.0368676-rdf-ntriples.txt
Original Record: 24-1.0368676-source.json
Full Text
24-1.0368676-fulltext.txt
Citation
24-1.0368676.ris

Full Text

The	relationships	between	rapid	urban	development	and	vegetation	in	the	pan	Pacific	region:	spatio-temporal	quantification	using	satellite	images		by			Yuhao	Lu			B.Sc.,	The	University	of	British	Columbia,	2014			A	THESIS	SUBMITTED	IN	PARTIAL	FULFILLMENT	OF	THE	REQUIREMENTS	FOR	THE	DEGREE	OF		DOCTOR	OF	PHILOSOPHY			in			The	Faculty	of	Graduate	and	Postdoctoral	Studies	(Forestry)			 		The	University	of	British	Columbia		(Vancouver)			April	2018	©	Yuhao	Lu,	2018	ii		The	following	individuals	certify	that	they	have	read,	and	recommend	to	the	Faculty	of	Graduate	and	Postdoctoral	Studies	for	acceptance,	the	dissertation	entitled:		Relationship	between	rapid	urban	development	and	vegetation	in	pan	Pacific	regions:	spatio-temporal	quantification	using	satellite	images		submitted	by	 Yuhao	Lu	 	 in	partial	fulfilment	of	the	requirements	for	the	degree	of	 Doctor	of	Philosophy	in	 The	Faculty	of	Graduate	and	Postdoctoral	Studies	(Forestry)		Examining	Committee:	Dr.	Nicholas	Coops	Supervisor		Dr.	Tongli	Wang	Supervisory	Committee	Member		Dr.	Jeanine	Rhemtulla	Supervisory	Committee	Member	Dr.	Mark	Johnson	University	Examiner	Dr.	Cecil	Konijnendijk	University	Examiner			Additional	Supervisory	Committee	Members:	NA	Supervisory	Committee	Member	NA	Supervisory	Committee	Member	iii		Abstract		Cities	strive	for	economic	strength	while	recognize	the	necessity	of	being	environmentally	sustainable.	The	balance	between	economic	development	and	the	environment	has	been	challenging	particularly	for	cities	in	the	pan	Pacific	region,	which	is	seeing	some	of	the	most	rapid	urban	growth	rates.	Remotely	sensed	satellite	images	offer	much	larger	and	more	consistent	spatial	and	temporal	coverages	than	conventional	census	data	therefore	are	increasingly	being	utilized	for	regional	and	global	urban	studies.	Two	key	remote	sensing	datasets,	namely	urban	vegetation	cover	derived	from	Landsat	time	series,	and	brightness	generated	from	NOAA’s	nighttime	lights	datasets	to	represent	urban	development	were	the	focus	of	this	dissertation.	I	first	extracted	annual	urban	vegetation	characteristics	using	spectral	indices	(e.g.	EVI)	as	well	as	a	spectral	mixture	analysis	from	1984	to	2012.	Nighttime	lights	brightness	was	used	to	assess	urban	expansion	and	its	relationship	with	census-derived	variables.	Lastly,	I	examined	the	relationships	between	urban	development	and	the	environment	using	Environment	Kuznets	Curve	(EKC)	theory	as	a	lens,	addressing	how	urban	vegetation	responds	to	urban	nighttime	brightness	in	25	cities	across	the	pan	Pacific	region.		I	identified	inter-	and	intra-city	patterns	of	vegetation	and	brightness	changes	that	were	strongly	related	to	social	and	economic	contexts.	Spectral	indices	demonstrated	opposing	trends	between	urban	vegetation	and	built-up	area	both	spatially	and	temporally.	Spectral	mixture	analysis	successfully	extracted	the	urban	vegetation	fraction	at	a	sub-pixel	level,	setting	a	robust	base	for	cross-city	comparisons.		I	found	that	urban	vegetation	changed	linearly	both	positively	and	negatively	with	urban	brightness,	particularly	in	higher	income	cities	in	North	America.	Pixels	with	statistically	strong	quadratic	relationships	between	vegetation	and	brightness	were	less	prevalent	but	more	spatially	clustered	in	comparison	to	those	that	expressed	a	linear	relationship.	Overall,	there	are	three	key	contribution	of	this	dissertation.	Firstly,	the	integration	of	gap-free	satellite	images	and	innovative	processing	techniques	unlocked	new	ways	of	informing	urban	environmental	and	socio-economic	dynamics.	Secondly,	a	classic	econometric	model	(i.e.	Granger	causality	test)	was	used	to	examine	the	casual	relationship	between	census	and	remote	sensing	nighttime	lights	data.	Lastly,	a	pixel-based	model	fitting	was	use	to	confirm	EKC	at	a	sub-city	scale.					iv		Lay	Summary			This	dissertation	investigated	the	spatial	and	temporal	dynamics	between	the	urban	environment	and	economic	development	using	satellite	images	for	25	cities	in	the	pan	Pacific	region	from	1984	to	2012.	This	dissertation	made	key	contributions	to	practical	urban	management.	Gap-free	satellite	time	series	revealed	both	with-	and	across-	vegetation	and	nighttime	brightness	dynamics	at	a	pixel	level.	Comparing	to	conventional	census	data,	satellite-derived	information	is	more	spatially	comparable,	unlocking	new	ways	of	regional	and	global	urbanization	studies.	The	relationship	between	vegetation	and	nighttime	lights	brightness	was	primarily	linear	yet	varying	degrees	of	quadratic	relationship	was	also	present	at	a	pixel	scale,	confirming	the	existence	of	the	Environmental	Kuznets	Curve.																v		Preface		The	research	questions	and	objectives	of	this	dissertation	were	originally	conceived	from	discussions	between	me	and	my	supervisory	committee.	Portions	of	this	dissertation	appear	as	co-authored,	peer-reviewed	journal	articles.	For	these	publications,	I	performed	the	primary	research,	data	analysis	and	interpretation,	and	prepared	the	final	manuscript:			§ Chapter	3:	Lu,	Y.,	Coops,	N.	C.,	&	Hermosilla,	T.	(2016).	Regional	assessment	of	pan	Pacific	urban	environments	over	25	years	using	annual	gap	free	Landsat	data.	International	Journal	of	Applied	Earth	Observation	and	Geoinformation,	50,	198-210.	§ Chapter	4:	Lu,	Y.,	Coops,	N.	C.,	&	Hermosilla,	T.	(2017).	Estimating	urban	vegetation	fraction	across	25	cities	in	pan	Pacific	using	Landsat	time	series	data.	ISPRS	Journal	of	Photogrammetry	and	Remote	Sensing,	126,	11-23.	§ Chapter	4:	Lu,	Y.,	Coops,	N.	C.,	&	Hermosilla,	T.	(2017).	Chronicling	urbanization	and	vegetation	changes	using	annual	gap	free	Landsat	composites	from	1984	to	2012.	In	Urban	Remote	Sensing	Event	(JURSE),	2017	Joint	(pp.	1-4).	IEEE.	§ Chapter	5:	Lu,	Y.,	Coops,	N.C.	Bright	lights,	Big	City:	Causal	effects	of	population	and	GDP	on	urban	brightness	(in	press	Plos	One).		§ Chapter	6:	Lu,	Y.,	Coops,	N.C.,	Wang,	T.	(under	review).	Confirming	the	EKC	theory	at	a	pixel	scale.			 	vi		Table	of	Contents		Abstract	.......................................................................................................................................................	iii	Lay	Summary	..............................................................................................................................................	iv	Preface	.........................................................................................................................................................	v	Table	of	Contents	........................................................................................................................................	vi	List	of	Tables	.............................................................................................................................................	viii	List	of	Figures	..............................................................................................................................................	ix	List	of	Abbreviations	...................................................................................................................................	xi	Acknowledgements	..................................................................................................................................	xiii	Chapter	1	......................................................................................................................................................	1	1.	 Introduction	....................................................................................................................................	1	1.1.	 Urbanization	worldwide	.........................................................................................................	1	1.2.	 Big	city	versus	green	city	........................................................................................................	2	1.3.	 The	opportunity	of	studying	urbanization	using	remote	sensing	...........................................	4	1.4.	 Research	approach	and	objectives	.........................................................................................	8	1.5.	 Dissertation	overview	.............................................................................................................	9	Chapter	2	....................................................................................................................................................	12	2.	 Study	area	and	Data	......................................................................................................................	12	2.1.	 Study	area	.............................................................................................................................	12	2.2.	 Landsat	time	series	...............................................................................................................	14	2.3.	 DMSP-OLS	nighttime	lights	time	series	................................................................................	16	2.4.	 Data	overview	.......................................................................................................................	18	Chapter	3	....................................................................................................................................................	19	3.	 How	can	metrics	derived	from	remotely	sensed	data	inform	environmental	and	socio-economic	dynamics	within	and	across	cities	in	the	pan	Pacific	region?	Introduction	............................................	19	3.1.	 Materials	and	methods	.........................................................................................................	21	3.2.	 Results	..................................................................................................................................	26	3.3.	 Discussion	.............................................................................................................................	34	Chapter	4	....................................................................................................................................................	40	4.	 How	to	can	urban	vegetation	be	mapped	at	the	sub-pixel	scale?	................................................	40	4.1.	 Introduction	..........................................................................................................................	40	4.2.	 Materials	and	methods	.........................................................................................................	42	4.3.	 Results	..................................................................................................................................	46	4.1.	 Discussion	.............................................................................................................................	53	Chapter	5	....................................................................................................................................................	57	vii		5.	 Are	bright	cities	big	cities	?	...........................................................................................................	57	5.1.	 Introduction	..........................................................................................................................	57	5.2.	 Materials	and	methods	.........................................................................................................	58	5.3.	 Results	..................................................................................................................................	62	5.4.	 Discussion	.............................................................................................................................	67	Chapter	6	....................................................................................................................................................	71	6.	 Testing	EKC	theory:	What	is	the	relationship	between	urban	vegetation	and	nighttime	brightness	across	pan	Pacific	cites?	.........................................................................................................................	71	6.1.	 Introduction	..........................................................................................................................	71	6.2.	 Materials	and	methods	.........................................................................................................	74	6.3.	 Results	..................................................................................................................................	77	6.4.	 Discussion	.............................................................................................................................	82	Chapter	7	....................................................................................................................................................	84	7.	Conclusion	..........................................................................................................................................	84	7.1.	 Research	innovation	.............................................................................................................	84	7.2.	 Answers	to	proposed	research	questions	............................................................................	85	7.3.	 Testing	the	EKC	theory	.........................................................................................................	89	7.4.	 Research	challenges	.............................................................................................................	90	Reference	...................................................................................................................................................	93	Appendix	1	...............................................................................................................................................	109	Appendix	2	...............................................................................................................................................	110	Appendix	3	...............................................................................................................................................	111	Appendix	4	...............................................................................................................................................	112				 	viii		List	of	Tables		Table	1.1	Literature	review	on	previous	urban	remote	sensing	studies	........................................................................	6	Table	2.1	Summary	table	of	studied	cities.	..................................................................................................................	13	Table	2.2	NTL	DMSP-OLS	Version	4	average	visible	stable	cloud-free	lights	(Years	highlighted	contain	data	acquired	by	two	sensors.	............................................................................................................................................................	17	Table	3.1	Cluster	analysis	summary	............................................................................................................................	32	Table	4.1	.	Correlations	of	interpreted	versus	estimated	vegetation	fractions	with	median	values.	..........................	47			 	ix		List	of	Figures		Figure	1.1	A	conceptualized	overview	of	the	dissertation	–	An	integration	between	Environmental	Kuznets	Curve	(EKC)	and	4	chapters	of	this	dissertation.	Chapter	2	and	Chapter	3,	addresses	the	challenges	of	measuring	urban	vegetation	consistently	over	time.	Chapter	4	utilizes	the	NTL	image	archive	and	local	census	economic	data	to	examine	the	driving	socio-economic	variables	behind	Nighttime	Lights	(NTL)	changes.	Final	research	Chapter	5	utilizes	results	from	previous	chapters	and	further	investigates	and	confirms	the	EKC	hypothesis	within	25	selected	cities.	.............................................................................................................................................................................	9	Figure	2.1	Geographic	locations	of	studied	cities.	.......................................................................................................	14	Figure	2.2	Time	frame	and	sensors	used	for	each	research	chapter.	..........................................................................	18	Figure	3.1	Examples	of	a	hexagon	based	ring	model;	(a)	Vancouver,	Canada;	(b)	Shanghai,	China;	(c)	Bangkok,	Thailand;	(d)	Melbourne,	Australia.	............................................................................................................................	24	Figure	3.2	Temporal	changes	of	EVI	and	NDBI	of	four	studied	urban	environments.	.................................................	27	Figure	3.3	Example	annual	spectral	trajectories	of	EVI	(left	column)	and	NDBI	(right	column).	See	Appendix	1	for	all	25	cities.	......................................................................................................................................................................	28	Figure	3.4	Separability	metrics	estimated	by	DTW	for	a.	EVI	and	b.	NDBI.	.................................................................	30	Figure	3.5	Average	trajectories	of	a.	EVI	and	b.	NDBI	for	each	cluster.	......................................................................	31	Figure	3.6	Temporal	change	metrics	of	a.	EVI	and	b.	NDBI.	........................................................................................	32	Figure	3.7	Silhouette	width	based	on	a.	temporal	vegetation	changes,	b.	temporal	built-up	changes,	c.	spatial	vegetation	patterns,	and	d.	spatial	built-up	patterns	(groups	are	identified	by	colors).	.............................................	33	Figure	3.8	Number	of	times	cities	were	grouped	together	during	K-means	classification	based	upon	spatial	and	temporal	variables.	Number	of	groupings	ranged	from	2	to	4.	..................................................................................	34	Figure	3.9	A	conceptual	representation	of	urbanization.	............................................................................................	36	Figure	4.1	Image	mosaicking	and	transformation	process	using	a).	Pixel	based	compositing	(PBC)	multispectral	images	(6	spectral	bands);	b).	Mosaicked	PBC	image	(6	spectral	bands)	and	c).	Minimal	Noise	Fraction	transformed	PBC	mega	(3	components	bands).	...............................................................................................................................	43	Figure	4.2	Landsat	proxy	image	(left	panel)	and	unmixed	vegetation	fraction	results	(right	panel)	in	year	2000	of	a).	Tokyo,	b).	Shanghai,	c).	Melbourne,	d).	Mexico	City	e).	Las	Vegas,	and	f).	Vancouver	(scale	1:	800,000).	.................	46	Figure	4.3	Comparison	between	estimated	and	interpreted	vegetation	fraction	of	a)	Shenzhen-Hong	Kong,	b)	Las	Vegas,	c)	Tokyo,	and	d)	Vancouver.	Error	bars	represented	the	standard	deviation	of	all	sample	points	within	each	20%	increment.	............................................................................................................................................................	47	Figure	4.4	Annual	vegetation	fraction	(0	–	100%)	results	of	Las	Vegas.	.....................................................................	49	Figure	4.5	Theil-Sen	estimated	vegetation	trend	slope	(p	<	0.05)	of	a).	Calgary,	b).	Dalian,	c).	Seattle,	d).	Las	Vegas,	e).	Manila,	f).	Shenzhen-Hong	Kong,	g).	Seoul,	and	h).	Vancouver	(scale:	1:700,000).	Water	is	colored	as	grey.	......	50	Figure	4.6	Vegetation	trend	slope	median	per	ring	(600-meter	per	ring).		Green	represents	median	slope	value.	Grey	represents	confidence	interval	of	smoothed	slope	value	(black).	Appendix	2	shows	circular	histogram	of	all	25	cities......................................................................................................................................................................................	51	Figure	4.7	Circular	histogram	of	vegetation	trend	with	slope	median	value	per	bin	(1-degree	per	bin).	The	bar	length	indicated	the	percentage	of	pixels	with	a	decreasing	vegetation	trend.	The	median	slope	value	was	colored	with	a	red-yellow-green	scheme	with	red	representing	the	most	negative	slope,	green	representing	the	most	positive	slope,	and	yellow	represent	stable	vegetation	fraction.	Water	is	colored	as	black.	Appendix	3	shows	circular	histogram	of	all	25	cities.	............................................................................................................................................	52	Figure	5.1	Sum	of	normalized	difference	index	(SNDI)	derived	from	raw	image,	Zhang	et	al.,	(2016),	Elvidge	et	al.,	(2014),	and	Lu	(this	chapter).	......................................................................................................................................	63	Figure	5.2	NTL	change	rate	represented	by	Theil-Sen	slope	values	showing	the	rate	of	change	from	1992	to	2013.	Water	is	colored	as	white.	Cities	were	grouped	based	on	its	growth	intensity.	Left	panel	contains	cities	with	fast	and	more	dynamic	urban	growth	while	the	right	panel	include	cities	with	more	stable	and	less	development.	...............	64	x		Figure	5.3	.	The	year	when	a	given	pixel	within	each	urban	environment	exceeded	the	pre-defined	DN	value.	Dark	grey	pixels	represent	existing	urban	areas	prior	to	1992	while	light	grey	indicating	areas	with	no	sufficient	light	sources	in	2013.	...........................................................................................................................................................	65	Figure	5.4	Urban	land	breakdown	of	changed,	undeveloped,	and	existing	urban	areas,	ranking	the	25	cities	from	the	highest	proportion	of	lit	pixels	(i.e.	TKO)	to	the	least	(i.e.	HAK).	...........................................................................	66	Figure	5.5	Causal	interactions	among	NTL	(nighttime	lights),	POP	(population	size),	and	GDP	(Gross	Domestic	Product)	of	a)	all	cities,	b)	established	cities,	and	c)	dynamic	cities.	A	solid	line	represents	a	statistically	significant	causal	relationship	while	a	dotted	line	indicates	no	significant	causality.		Arrow	head	indicates	the	direction	of	causal	relationship	and	a	double-headed	arrow	represents	a	bi-directional	causal	relationship.	..............................	67	Figure	6.1	An	Illustration	of	previous	studies	attempting	to	quantify	the	relationship	between	economic	indicators	versus	an	environmental	variable.	(Note	that	all	figures	have	been	simplified	to	only	highlight	the	relationship	between	the	corresponding	variables.	........................................................................................................................	75	Figure	6.2	Vegetation	fraction	(VF)	time	series	regresses	against	Nighttime	time	light	(NTL)	time	series	pixel	by	pixel.	Three	candidate	models	(linear,	quadratic,	and	cubic)	I	re	used	to	determine	the	most	appropriate	model	based	on	AIC	score	and	statistics	significance.	The	red	pixel	symbolizes	where	the	city	center	is	while	the	blue	pixel	represents	water	which	was	excluded	from	all	subsequent	model	fitting.	.................................................................	76	Figure	6.3	Histograms	of	r2	values	for	each	selected	model.	.......................................................................................	78	Figure	6.4	Percentage	of	pixels	for	each	selected	best	model	type	at	a	city	level.	Only	pixels	with	significant	changes	were	used	for	calculating	the	percentage	of	each	model.	Pixels	that	did	not	fit	any	of	the	three	functions	were	not	concluded.	...................................................................................................................................................................	79	Figure	6.5	The	top	panel	represents	the	best	model	selected	based	on	AIC	score	and	F-test	at	pixel	level.	Coefficients	of	leading	variables	for	each	selected	model	were	shown	in	the	bottom	panel.	The	sign	of	each	coefficient	determines	the	directionality	of	the	relationship	while	the	absolute	value	of	the	coefficient	indicates	the	magnitude	of	impact	of	NTL	on	VF.	...............................................................................................................................................	80	Figure	6.6	Positions	of	the	examined	cities	on	the	EKC	curve.	It	classifies	the	25	cities	using	the	summed	value	of	NTL	and	VF	pixels	excluding	water.	....................................................................................................................................	81	Figure	6.7	Spatial	autocorrelation	represented	by	the	ratio	between	total	counts	of	joins	and	a	random	spatial	pattern	calculated	by	Join-count	statistics	for	each	model.	Theoretically,	with	a	higher	Rj/r,	I	would	expect	a	stronger	spatial	association.	......................................................................................................................................................	81			 	xi		List	of	Abbreviations			AIC	–	Akaike	information	criterion		BAP	–	Best	available	pixel		BSS/TSS	–	Ratio	of	the	between	to	the	total	sum	of	square	CBD	–	Central	business	district		DN	–	Digital	number		DTW	–	Dynamic	time	warping	EKC—Environmental	Kuznets	Curve		EVI	–	Enhanced	Vegetation	Index		GADM	–	Global	administrative	Data		GDP	–	Gross	domestic	product	NDBI	–	Normalized	Differenced	Built-up	Index		NDI	–	Normalized	Distance	Index		NDVI	–	Normalized	Differenced	Vegetation	Index		NTL	–	Nighttime	lights		PBC	–	Pixel	based	compositing		POP	–	population		SMA	–	Spectral	mixture	analysis	TDN	–	Total	digital	number		TGDP	–	Total	gross	domestic	product		TPOP	–	Total	population		TS	–	Theil	Sen	slope		VF	–	Vegetation	fraction		bak	–	Bangkok	cal	–	Calgary	csx	–	Changsha	dal	–	Dalian	den	–	Denver	xii		edm	–	Edmonton		fuz	–	Fuzhou	hak	–	Haikou	har	–	Harbin	hksz	–	Hong	Kong-Shenzhen		kul	–	Kuala	Lumpur	lav	–	Las	Vegas	man	–	Manila	mel	–	Melbourne	mex	–	Mexico	City	ncx	–	Nanchang	phx	–	Phoenix	sea	–	Seattle	sel	–	Seoul	shh	–	Shanghai	sin	–	Singapore	City	tjn	–	Tianjin	tko	–	Tokyo	van	–	Vancouver			 	xiii		Acknowledgements		This	dissertation	was	supported	by	funding	provided	to	Dr.	Nicholas	Coops	from	Natural	Sciences	and	Engineering	Research	Council	of	Canada	(NSERC,	RGPIN	31	1926-13)	as	well	as	the	Asia-Pacific	Network	for	Sustainable	Forest	Management	and	Rehabilitation	through	Dr.	Guangyu	Wang.	The	images	used	in	this	dissertation	are	archived	and	distributed	by	United	States	Geological	Survey	and	National	Oceanic	and	Atmospheric	Administration.	 My	committee	members,	Dr.	Jeanine	Rhemtulla	and	Dr.	Tongli	Wang,	have	offered	tremendous	support	since	entering	graduate	school.	Dr.	Jeanine	Rhemtulla	has	always	brought	unique	and	fresh	perspectives	to	my	research	and	constantly	reminded	me	to	think	broader.	The	knowledge	and	expertise	on	modelling	and	spatial	data	analysis	from	Dr.	Tongli	Wang	is	more	than	I	could	ever	asked.	 Dr.	Nicholas	Coops	is	more	than	just	a	supervisor.	He	created	and	maintained	a	friendly,	productive,	and	enjoyable	environment	for	all	IRSS	members	and	alumni.	His	passion	and	work	ethic	towards	remote	sensing	is	mind-blowing.	He	is	a	great	mentor	and	a	best	friend.	 The	time	at	the	IRSS	lab	is	a	life	changing	experience.	I	wrote	my	very	first	piece	of	code	with	Dr.	Txomin	Hermosilla.	I	could	not	recall	how	many	times	Dr.	Paul	Pickell	proofread	and	edited	my	broken	manuscripts.	It	was	always	fun	to	discuss	mapping	ideas	and	colour	options	with	Andrew	Plowright.	Dr.	Ryan	Frazier	was	always	happy	and	uplifting	no	matter	how	early	the	hours	were	in.	IRSS	is	also	full	of	some	of	the	greatest	athletes:	Dr.	Ricardo	Tortini,	Chris	Mulverhill,	Ethan	Berman,	Lukas	Jarron	etc.	They	are	humble,	strong,	and	taught	me	a	lot	of	lessons	that	will	be	carried	on	beyond	my	career.	 I	did	not	spend	too	much	time	with	my	parents	during	my	time	in	Canada.	They	are	the	most	incredible	human	beings	I	know.	They	are	my	closest	friends.	I	am	thankful	for	everything	they	have	done	and	sacrificed	for	me.	Yang	is	the	silliest	golden	retriever.	I	left	home	when	he	was	just	5	years	old.	I	hope	he	will	stay	strong	and	healthy.		 I	have	also	met	many	new	friends	from	all	of	the	world.	Some	of	them	I	have	never	had	a	chance	to	meet	in	person	(yet).	A	special	thank-you	to	all	the	places	I	visited	in	Iceland	and	Canada,	and	all	the	members	in	eARTh.	There	are	simply	too	many	of	you.		Y.L.		April	18,	2018 1		Chapter	1		1. Introduction	1.1. Urbanization	worldwide			The	rate	of	human	modification	of	the	terrestrial	biosphere	has	intensified	over	the	past	several	decades	(Hugo,	2017).	The	term	urbanization	describes	not	only	the	physical	expansion	and	densification	of	cities	but	also	implies	the	complex	transformation	of	human	society,	demography,	and,	more	importantly,	the	relationship	among	them	(Boone	&	Fragkias,	2012).	Cities	and	their	surroundings	(synonymous	with	urban	environments	in	this	dissertation)	cause	some	of	the	most	profound	anthropogenic	impacts	on	Earth	in	spite	of	their	disproportionately	small	spatial	footprint.	Covering	approximately	3%	of	the	landscape,	cities	are	responsible	for	three	quarters	of	global	energy	consumption	and	approximately	80%	of	greenhouse	gas	emissions	(Ash,	Jasny,	Roberts,	Stone,	&	Sugden,	2008)	in	accommodating	over	54%	of	the	global	population,	a	number	that	is	expected	to	exceed	70%	by	2050	(UNDESA,	2014,	2017).		Concepts	of	urbanization	have	now	evolved	to	not	only	refer	to	a	collection	of	isolated,	locally	dense,	commercial	districts	of	concrete	and	artificial	lights,	but	extend	to	networks	of	social,	ecological,	and	economic	mosaics	functioning	interdependently	as	a	whole	(Pickett	&	Zhou,	2015).	The	classic	view	of	urbanization	patterns	is	now	being	challenged	by	a	more	connected	and	heterogeneous	decentralizing	trend.	The	localizing	mindset	of	“the	closer	the	cheaper”	that	drove	early	urban	development	is	no	longer	a	primary	objective	espoused	by	city	planners.	Efficient	transportation	and	global	trading	networks	have	nodalized	cities	worldwide	regardless	of	their	physical	distances	and,	as	a	result,	urbanization	has	become	a	global	phenomenon	with	unprecedented	rates	of	growth	(Lee,	2013;	Martinus	&	Tonts,	2015).		Despite	decentralization,	there	remains	recognition	that	cities	are	driven	by	local	geopolitical,	social,	and	ecological	environments	(K.	C.	Y.	Seto,	2014),	making	models	of	urban	development	both	conceptually	and	practically	challenging.	The	scale	and	pace	of	modern	urbanization	also	affect	spatial	urbanization	processes.	In	the	early	20th	century,	cities	were	conceptualized	by	a	series	of	concentric	circles	(Burgess,	1925)	that	expand	radially	from	the	city	centre.	Around	the	central	core	is	a	buffer	area,	also	known	as	a	loop,	which	isolates	the	core	from	the	third,	or	workmen’s,	zone	inhabited	by	industrial	2		workers.	Beyond	that	is	the	fourth,	residential	zone	with	high	class	apartments	and	single	family	houses.	Outside	the	city	limits	lies	the	commuters’	zone,	which	includes	all	suburban	areas	and	other	satellite	cities	within	a	30	to	60	minutes	drive	from	the	city	centre	(Burgess,	1925).		Spatially,	contemporary	urbanization	posits	two	general	models:	fragmentation	and	polycentrism	(Jenks	et	al.,	2008).	Particularly	in	developing	regions,	cities	render	spatial	patterns	that	are	recognized	as	urban	fragmentation	(Balbo	&	Navez-Bouchanine,	1995).	The	concept	of	fragmentation	is	described	and	interpreted	as	the	“dividing”	and	“partitioning”	of	modern	metropolises	(Jenks	et	al.,	2008)	or	is	the	result	of	a	combination	of	urban	dislocation	and	discontinuity	(Mieg	&	Töpfer,	2013).	Burgess	(1925)	defined	urban	fragmentation	as	“a	spatial	phenomenon	that	results	from	the	act	of	breaking	up,	breaking	off	from,	or	disjointing	the	pre-existing	form	and	structure	of	the	city	and	systems	of	cities.”	Although	the	phenomenon	of	urban	fragmentation	is	often	understood	to	have	a	negative	impact	on	urban	demography,	many	cities	have	experienced	decentralization	throughout	their	developing	stages.		In	contrast,	the	polycentrism	model	allows	for	regions	that	contain	separate	and	distinct	cities	or	small	human	settlements	with	substantial	interactions	(Gordon	et	al.,	1986)	while	fragmented	urban	structures	display	disconnection	and	highly	heterogeneous	patterns.	Although	it	would	appear	to	integrate	small	cites,	the	polycentric	urban	form	often	intensifies	fragmentation	instead	of	reversing	it	(Jenks	et	al.,	2008).	Infrastructure	used	to	reduce	urban	fragmentation,	such	as	mass	public	transit	or	freeways,	may	in	fact	exacerbate	it.	Despite	these	two	theoretical	models,	the	scarcity	of	knowledge	about	how	cities	grow	and	interact	spatially	and	temporally	limits	the	advancement	of	healthy	and	efficient	urban	structure	and	organization.			1.2. Big	city	versus	green	city			Urbanization	today	strives	to	grow	not	only	economically	strong	but	also	environmentally	sustainable	during	the	process	of	urbanization.	The	perception	that	“big	cities”	yield	overwhelmingly	positive	outcomes	for	the	population	has	started	to	lose	momentum	over	time.	Urbanization	has	not	only	caused	economic	saturation	(Wheaton	&	Shishido,	1981)	but	also	numerous	social,	health,	and	environmental	challenges	such	as	increased	crime	(Shelley,	1981),	disease	(Nicolaou,	Siddique,	&	Custovic,	2005),	urban	heat	island	effect	(Oke,	1982),	and	water	and	air	pollution	(Beckerman,	1992).		3		As	a	result,	there	has	been	a	recent	increase	in	the	value	placed	on	urban	greenspace,	or	urban	vegetation,	by	both	the	scientific	and	political	communities	(Zhao,	Liu,	&	Zhou,	2016).	Definitions	of	urban	greenspace	and	urban	vegetation	are	just	as	complicated	as	urbanization	itself;	however,	they	are	often	used	interchangeably	to	include	any	functional	vegetated	land-cover	and	land	use	type	within	an	urban	setting,	exclusive	of	agriculture	and	cropland.		The	presence	of	urban	vegetation	is	known	to	be	beneficial	to	the	local	climate,	and	thus	also	to	the	social,	and	physical	environments	through	temperature	control	(Oke,	1982),	air	pollution	reduction	(Nowak,	Crane,	&	Stevens,	2006),	noise	and	storm	water	control	(Glass	&	Singer,	1972),	and	habitat	preservation	(Nowak	&	Dwyer,	2007).	Studies	have	also	indicated	significant	social	(Grahn	&	Stigsdotter,	2003),	economic	(Tyrväinen,	Pauleit,	Seeland,	&	De	Vries,	2005),	and	aesthetic	values	associated	with	urban	vegetation	(Jim	&	Chen,	2006;	Tyrväinen	et	al.,	2005).	Such	greenspace	has	been	utilized	as	an	effective	tool	to	achieve	sustainable	and	functional	urban	environments.	Efforts	towards	preserving	healthy	urban	vegetation	can	be	found	worldwide,	particularly	in	developed	regions,	such	as	in	North	America	and	Europe	(Nowak	et	al.,	2006).		The	question	many	researchers	and	planners	have	raised	however	is	at	which	urbanization	stage	a	city	can	afford	to	be	moving	towards	environmentally	sustainability	or	“being	green”	(Stern,	1998).	In	less	developed	areas,	urbanization	often	receive	higher	prioritization	than	preserving	and	maintaining	urban	vegetation,	despite	the	resulting	benefits	and	services	(Grimm	et	al.,	2008).	Vegetation	in	these	urban	environments	often	grows	in	more	isolated	and	fragmented	patches	compared	to	that	in	novel	and	well	managed	urban	environments,	making	it	more	challenging	to	manage	for	local	urban	planners	in	these	areas.	Another	concern	associated	with	poorly	managed	and	fragmented	urban	vegetation	is	ecological	inequity	(N.	Heynen,	Perkins,	&	Roy,	2006),	which	causes	uneven	access	to	quality	urban	green	space	among	local	residents.		It	is	evident	worldwide	that	growing	or	maintaining	economic	development	in	an	environmentally	sustainable	manner	is	difficult,	particularly	for	less	developed	regions	(Egli	&	Steger,	2007;	Glaeser,	2011;	Stern,	2004).	The	interplay	between	economic	growth	and	urban	greenspace	has	spurred	debate,	forming	various	theories	to	better	understand	the	tensions	that	divide	economic	development	and	environmental	sustainability.	The	Environmental	Kuznets	Curve	(EKC)	theory	(Kuznets,	1955)	hypothesises	a	non-linear,	U-shaped	relationship	between	environmental	quality	and	economic	development	where	environmental	performance	decreases	at	early	stages	of	economic	development	and	recovers	as	the	economy	reaches	a	certain	turning	point.	However,	the	scarcity	of	reliable	and	4		consistent	assessment	of	the	relationship	between	economic	growth	and	environmental	degradation	limits	our	ability	to	understanding	and	testing	theories	such	as	the	EKC.			1.3. The	opportunity	of	studying	urbanization	using	remote	sensing			Remote	sensing	derived	metrics	have	been	widely	used	to	assess	urban	vegetation	and	economic	development	in	a	more	spatially	and	temporally	consistent	manner	than	conventional	census	and	ground	measurements.	New	imaging	and	mapping	technologies	such	as	Geographic	Information	Systems	(GIS)	and	remotely	sensed	imagery	have	simplified	the	geographic	identification	of	cities.	The	age	of	open	access	satellite	images	has	arrived	(Woodcock	et	al.,	2008;	Wulder	et	al.,	2008;	Wulder	&	Coops,	2013).	These	freely	accessible	remotely	sensed	time	series	data	pose	many	strengths	in	urban	contexts,	including:	(i)	the	capability	to	capture	the	full	temporal	profile	of	urbanization	rather	than	snapshots	of	individual	time	periods;	(ii)	pixel-based	compositing	(PBC)	which	unlocks	the	limit	of	the	traditional	scene-based	analysis	approach;	(iii)	continuous	measurement	of	the	spectral	response	in	urbanization	and	urban	greenspace	that	allows	for	the	application	of	various	modelling	options;	and	(iv)	enables	large-scale	systematic	and	compatible	comparisons	across	countries.	As	a	result,	remotely	sensed	data	allow	monitoring,	extracting,	and	estimating	changes	in	three	key	components	of	urban	environments:	namely,	urban	built-up/impervious	area,	urban	vegetation,	and	socio-economic	indicators.	The	literature	on	urban	remote	sensing	indicates	a	substantial	amount	of	diversity	in	terms	of	study	locations,	temporal,	and	spatial	scales	(Table	1.1).		Zha,	Gao,	&	Ni	(2003)	proposed	a	new	spectral	index	–	Normalized	Difference	Built-up	Index	(NDBI)	for	a	fast	urban	impervious	mapping	and	classification.	Bagan	&	Yamagata	(2014)	studied	50	cities	globally	using	Landsat	TM/ETM+	imagery	at	30m	spatial	resolution,	and	used	a	maximum	likelihood	classifier	to	classify	urban	land	use	patterns	between	1985	and	2010.		For	vegetation,	various	spectral	indices	such	as	Normalized	Difference	Vegetation	Index	(NDVI,	e.g.	Boone	&	Fragkias,	2012;	Lin,	Liu,	Li,	&	Li,	2014),	and	Enhanced	Vegetation	Index	(EVI,	e.g.	Zhang	et	al.,	2003)	have	been	used	to	extract	and	monitor	urban	vegetation	dynamics.	Sub-pixel	or	spectral	unmixing	analysis	(SMA)	is	also	a	popular	method	for	investigating	urban	vegetation	trends	(Phinn,	Stanford,	Scarth,	Murray,	&	Shyy,	2002;	Ridd,	1995;	Tooke,	Coops,	Goodwin,	&	Voogt,	2009).	SMA	spectrally	decomposes	a	given	pixel,	allowing	users	to	pre-define	pure	spectra	and	compute	a	fractional	score	that	represents	the	abundance	of	a	given	land	use	or	land-cover	type	(i.e.	vegetation).	Compared	to	5		conventional	spectral	indices	such	as	NDVI,	SMA	derived	vegetation	estimation	is	less	likely	to	be	saturated	(Ridd,	1995).		Historically,	accurate	spatial	representation	of	socio-economic	activities	was	dominated	by	local	census	data	with	little	use	of	remote	sensing.	Welch	(1980)	discovered	the	potential	of	utilizing	nighttime	light	images	(NTL)	to	map	urban	population	and	energy	consumption.	Ever	since,	NTL	data	have	been	used	at	local	(Ma,	Zhou,	Pei,	Haynie,	&	Fan,	2012),	regional	(Klotz,	Kemper,	Geiß,	Esch,	&	Taubenböck,	2016),	and	global	scales	(Small,	Pozzi,	&	Elvidge,	2005),	representing	a	variety	of	urbanization	indicators	such	as	population	density	(P.	Sutton,	Roberts,	Elvidge,	&	Baugh,	2001),	income	level	(Ebener,	Murray,	Tandon,	&	Elvidge,	2005),	GDP	(P.	C.	Sutton,	Elvidge,	Ghosh,	&	others,	2007),		light	pollution	(Bennie,	Davies,	Duffy,	Inger,	&	Gaston,	2015),	and	even	carbon	emissions	(Ghosh	et	al.,	2010).		A	literature	review	(Table	1.1)	reveals	three	key	findings	on	the	use	of	the	remote	sensing	of	urban	environments.	First,	despite	the	increasing	use	of	remote	sensing	in	urban	studies,	it	is	apparent	that	the	value	of	remote	sensing	at	a	global	scale	is	limited	by	the	availability	and	quality	of	continuously	collected	image	data.	Studies	involve	either	monitoring	a	small	number	of	cities	over	a	long	continuous	period,	or	monitoring	a	large	number	of	cities	at	fewer	time	steps.	For	example,	Wang	et	al.	(2014)	covered	1985	to	2013	using	data	acquired	from	alternative	years	for	a	single	city	(Toronto)	while	Bagan	&	Yamagata	(2014)	analysed	50	cities	but	only	using	imagery	acquired	in	1986	and	2010.	In	addition,	recent	urban	expansion	studies	(e.g.	Castrence	et	al.,	2014;	Chen	et	al.,	2014;	T.	Liu	&	Yang,	2015;	Ma	et	al.,	2012),	despite	analysing	a	moderate	number	of	cities	and	data,	tended	to	focus	on	local	urban	dynamics	with	limited	regional-	and	global-scale	synthesis.	Lastly,	literature	reviews	confirm	the	Landsat	data	archive	is	the	dominant	spatial	data	source	in	urban	studies	with	over	80%	of	papers	reviewed	using	data	from	the	Landsat	series	of	satellites.		The	overall	theme	of	previous	literature	focuses	primarily	on	image	classification	using	images	from	a	limited	number	of	cities.		Although	Landsat	is	the	longest	and	most	consistent	Earth	observation	program,	urban	time	series	studies	are	still	scarce	in	the	current	literatures.6			Table	1.1	Literature	review	on	previous	urban	remote	sensing	studies	Article	 Data	 Scale	 Attribute(s)	measured	Tooke	et	al.	2009	 LiDAR	 Metropolitan	 Urban	greenspace	Rottensteiner	&	Briese,	2002	 LiDAR	 Metropolitan	 Building	extraction	Secord	&	Zakhor,	2006	 LiDAR	 City	blocks	 Urban	tree	detection	Singh,	Vogler,	Shoemaker,	&	Meentemeyer,	2012	 LiDAR	and	Landsat	TM	 Metropolitan	 Land-cover	classification	Fauvel	&	Benediktsson,	2008	 Hyperspectral	images	 City	blocks	 Land-cover	classification	Kong,	Yin,	and	Nakagoshi	2007	 SPOT-4	 Province/State	 Green	space	Marconcini	et	al.	2014	 TSX/TDX	 Global	 Urban	extent	Small	2003	 IKONOS	 Global	 Reflectance	properties	Cheng	et	al.	2007	 Landsat	TM	 Metropolitan	 Urban	park	cooling	effect	Wang	et	al.	2014	 Landsat	TM	 Metropolitan	 Urban	extent	Zheng	et	al.	2014	 Landsat-8	OLI	 Rural	 Impervious	surface	Yang	et	al.	2014	 Landsat	ETM+	 Metropolitan	 Green	space	Waqar	et	al.	2012	 Landsat	TM	 Metropolitan	 Built-up	and	bare	soil	Chen	et	al.	2014	 Landsat	TM	 Province/State	 Urban	extent	Schneider	&	Woodcock	2008	 Landsat	TM	 Global	 Urban	fragmentation	Ward,	Phinn,	&	Murray	2000	 Landsat	TM	 Province/State	 Land-cover	classification	Liu	&	Yang	2015	 Landsat	TM	and	WV-2	 Metropolitan	 Land-cover	classification	Handayani	&	Rudiarto	2014	 Landsat	TM/ETM+	 Metropolitan	 Built-up	and	population	Fan	&	Fan	2014	 Landsat	TM/ETM+	 Metropolitan	 Urban	extent	Yang	et	al.	2014	 Landsat	TM/ETM+	 National	 Urban	greenspace	Ji	et	al.	2006	 Landsat	TM/ETM+	 Metropolitan	 Urban	density	Griffiths	et	al.	2010	 Landsat	TM/ETM+	 Metropolitan	 Urban	extent	Michishita,	Jiang,	&	Xu	2012	 Landsat	TM/ETM+	 Metropolitan	 Reflectance	properties	Xu	and	Min	2013	 Landsat	TM/ETM+,	CBERS,	HJ-1	 National	 Urban	extent	Shen	et	al.	2015	 Landsat	TM/ETM+,	MODIS	 Metropolitan	 Urban	temperature	7		Tian	et	al.	2014	 Landsat-MSS	and	TM	 Metropolitan	 Land-cover	and	land	use	Bagan	&	Yamagata	2014	 Lansat	TM/ETM+	 Global	 Land-cover	change	Gonçalves	et	al.	2014	 MODIS,	SPOT,	AVHRR/NOAA	 Metropolitan	 Land	use	temporal	trend	Li	et	al.	2013	 DMSP/OLS	Nighttime	lights	 Metropolitan	 GDP	Ghosh	et	al.,	2010	 DMSP/OLS	Nighttime	lights	 Global	 CO2	emission	Huang	et	al.,	2015	 DMSP/OLS	Nighttime	lights	 National	 Spatial	distribution	Bennie	et	al.,	2015	 DMSP/OLS	Nighttime	lights	 Regional	 Light	pollution	Welch,	1980	 DMSP/OLS	and	Landsat	 National	 Energy	consumption	Bennett	&	Smith,	2017	 DMSP/OLS	and	Suomi	NPP	VIIRS	 Global	 Review	Small	et	al.,	2005	 DMSP/OLS	Nighttime	lights	 Global	 Urban	extent			8		1.4. Research	approach	and	objectives			Given	the	need	for	and	scarcity	of	accurate	spatio-temporally	consistent	records	of	urban	vegetation	and	economic	development,	the	overall	research	objective	of	this	dissertation	is	to	test	what	trends	and	interrelationships	exist	between	remotely	sensed	derived	vegetation	and	economic	indicators	within	and	across	pan	Pacific	urban	centres.	Three	specific	research	questions	are	proposed:		1. How	can	metrics	derived	from	remotely	sensed	data	inform	environmental	and	socio-economic	dynamics	within	and	across	cities	in	the	pan	Pacific	region?		2. What	models	exist	to	examine	the	relationship	between	urban	environmental	and	socio-economic	developments	over	time	and	space?	3. What	similarities	and	differences	exist	across	cities	in	the	pan	Pacific	region	both	spatially	and	temporally?	These	three	questions	are	structured	in	such	a	way	to	allow	examination	of	the	EKC	hypothesis	(Figure	1.1).	Research	question	1	investigates	how	time	series	of	satellite	images	can	be	used	to	spatially	and	temporally	assess	the	dynamics	of	urban	environment	and	development.	Research	question	2	examines	the	EKC	theory	at	a	pixel-level	using	measurements	from	research	question	1.	The	final	research	question	examines	how	individual	cities	conform	to	the	hypothesis	of	EKC	theory.		9				1.5. Dissertation	overview			The	structure	of	this	dissertation	is	as	follows.		Chapter	2	introduces	the	main	study	sites:	25	cities	across	the	pan	Pacific	region,	in	terms	of	their	geographic	location,	climate,	and	economic	status.	Chapter	2	also	provides	an	overview	of	the	two	main	remote	sensing	products	used	in	this	dissertation.			Chapter	3	addresses	the	first	research	question	using	a	pixel-based	image	composite	technique	to	generate	annual	gap-free	surface	reflectance	Landsat	composites	from	1984	to	2012.	Using	time	series	composites,	spectral	indices	were	calculated	and	compared	using	a	hexagonal	grid	ring	model	to	assess	Figure	1.1	A	conceptualized	overview	of	the	dissertation	–	An	integration	between	Environmental	Kuznets	Curve	(EKC)	and	4	chapters	of	this	dissertation.	Chapter	2	and	Chapter	3,	addresses	the	challenges	of	measuring	urban	vegetation	consistently	over	time.	Chapter	4	utilizes	the	NTL	image	archive	and	local	census	economic	data	to	examine	the	driving	socio-economic	variables	behind	Nighttime	Lights	(NTL)	changes.	Final	research	Chapter	5	utilizes	results	from	previous	chapters	and	further	investigates	and	confirms	the	EKC	hypothesis	within	25	selected	cities.			10		changes	in	vegetative	and	urban	built-up	patterns.	Trajectories	are	then	clustered	to	further	investigate	the	spatio-temporal	dynamics	and	relationships	among	the	25	cities.	Outcomes	from	this	chapter	demonstrate	the	value	of	utilising	annual	Landsat	time	series	composites	for	assessing	urban	vegetation	and	urban	dynamics	at	regional	scales	and	potential	use	in	achieving	and	evaluating	sustainable	urban	planning.	Chapter	4	focuses	on	extracting	and	characterizing	urban	vegetation	by	innovatively	applying	sub-pixel,	spectral	unmixing	on	Landsat	time	series	composites	from	Chapter	2.	Vegetation	change	trends	were	then	analyzed	using	Mann-Kendall	statistics	and	Theil-Sen	slope	estimators.	The	outcomes	of	this	chapter	indicate	that	unmixing	approaches	successfully	map	urban	vegetation	for	pixels	located	in	urban	parks,	forested	mountainous	regions,	as	well	as	agricultural	land	(correlation	coefficient	ranging	from	0.66	to	0.77).	Using	temporal	trend	analysis,	our	results	suggest	that	it	is	possible	to	reduce	noise	and	outliers	caused	by	phenological	changes	particularly	in	cropland	using	dense	new	Landsat	time	series	approaches.	Chapter	5	focuses	on	the	socio-economic	aspects	of	urbanization	in	the	pan	Pacific	region.	It	uses	remotely	sensed	nighttime	light	images	(NTL)	as	a	proxy	to	map	urbanization	and	subsequently	examines	the	driving	socio-economic	variables	in	cities.	Using	a	classic	econometric	approach,	panel	causality	tests	are	undertaken	to	analyze	causal	relationships	between	NTL	and	socio-economic	development	across	the	pan	Pacific	region.	Panel	causality	tests	show	a	contrasting	effect	of	population	and	gross	domestic	product	(GDP)	on	NTL	in	fast	and	slowly	changing	cities.	Information	derived	from	this	chapter	quantitatively	chronicles	urban	activities	in	the	pan	Pacific	region	and	offers	data	for	further	studies	on	spatially	tracking	local	policy	progress	on	sustainable	urban	development.	Chapter	6	brings	the	results	from	Chapter	3	and	4	to	spatially	test	the	EKC	hypothesis	within	the	selected	25	cities.	Three	fitted	models	were	developed	(i.e.	linear,	polynomial,	and	cubic)	and	the	best	fit	was	selected	using	AIC	(Akaike	Information	Criteria)	scores.	The	results	of	this	chapter	suggested	that	in	most	cases,	urban	vegetation	varies	linearly	with	NTL	however	the	relationship	and	the	rate	of	change	varies	over	time.	Within	individual	cities	polynomial	models	tend	to	spatially	cluster	together	more	than	linear	and	cubic	models.	Further,	this	chapter	bridges	the	gap	between	the	conventional	econometric	theory	(i.e.	EKC)	and	advanced	earth	observation	satellite	data,	thus	overcoming	some	of	the	difficulties	of	using	census	data	which	are	less	suitable,	updated	infrequently,	and	spatially	incompatible	among	cities.	11		Chapter	7	concludes	with	some	key	research	findings,	limitations,	and	possible	opportunities	for	future	research.			 	12		Chapter	2		2. Study	area	and	Data	2.1. Study	area		Defined	by	UNESCO,	the	pan	Pacific,	or	Pacific	Rim,	region	of	the	globe	contains	50	countries	(UNESCO,	2014)	from	the	borders	of	China-Mongolia	to	the	north,	and	the	southern	tips	of	Australia	and	New	Zealand	to	the	south.	This	region	covers	approximately	2.8	billion	hectares	of	land,	approximately	22%	of	the	Earth’s	land	surface,	30%	of	the	world’s	natural	forest,	and	54%	of	the	world’s	plantations	(UNESCAP,	2012).	In	addition	to	Asia	and	Oceania,	the	pan	Pacific	region	as	recognised	in	this	dissertation	also	contains	North	and	Central	America,	producing	a	diversity	of	geopolitical,	social,	and	ecological	environments,	making	it	an	ideal	regional	focus	area	for	the	questions	posed	herein.		Throughout	this	dissertation	I	focused	on	25	cities	across	the	pan	Pacific	region	covering	a	range	of	population	sizes	and	economic	development	statuses	(Table	2.1,	Figure	2.1).	Nine	cities	were	selected	in	China,	and	six	in	Southern	Asia	and	South	America,	all	of	which	are	highly	dynamic	urban	areas.	The	rest	of	the	cities	are	located	in	more	developed	areas	with	four	in	the	United	States,	three	in	Canada,	two	in	Australia,	one	in	Japan,	and	one	in	South	Korea.	Among	the	25	urban	environments,	eight	are	mega-cities	(defined	as	having	a	population	over	10	million),	namely,	Tokyo,	Shanghai,	Seoul,	Mexico	City,	Tianjin,	Bangkok,	Shenzhen,	and	Harbin.	Smaller	cities,	particularly	those	in	developing	regions,	are	often	less	studied	(Bell	&	Jayne,	2009)	and	were	thus	included	in	this	dissertation.	As	a	result,	the	thesis	includes	cities	such	as	Changsha	that	are	not	as	economically	developed	as	other	selected	mega-cities.		Urban	environments	were	also	located	across	a	variety	of	landscapes	from	costal	mountainous	regions	(e.g.	Vancouver,	Dalian)	to	plains	and	dry	inland	areas	(e.g.	Las	Vegas,	Denver).	Climatically,	using	Köppen	climate	scheme	(Kottek,	Grieser,	Beck,	Rudolf,	&	Rubel,	2006),	10	cities	are	located	within	a	temperate	climate	scheme	(Class	C)	with	the	temperature	of	the	coolest	month	at	18	°C	or	higher.	The	cold	continental	(Class	D)	climate	scheme	contained	six	cities	dominantly	from	North	America	and	East	Asia.	Cities	located	in	South	East	Asia	were	primarily	within	the	Tropic	(Class	A)	climate	scheme,	and	three	cities	from	North	America	(i.e.	Denver,	Las	Vegas,	and	Phoenix)	were	from	the	Arid	(Class	B)	climate	scheme.		13			Table	2.1	Summary	table	of	studied	cities.	City	(code)	 Country	 Latitude/	Longitude	1	 Location	 Köppen	climate	scheme*	Word	Bank	Economy	Class*	Bangkok	(bak)	 Thailand	 13.75°N/100.49°E	 South	East	Asia	 Tropical	(Aw)	 Upper	middle	Calgary	(cal)	 Canada	 51.05°N/114.08°W	 North	America	 Continental	(Dfb)	 High	Changsha	(csx)	 China	 28.20°N/112.92°E	 East	Asia	 Temperate	(Cfa)	 Upper	middle	Dalian	(dal)	 China	 38.92°N/121.64°E	 East	Asia	 Continental	(Dwa)	 Upper	middle	Denver	(den)	 USA	 39.75°N/111.00°W	 North	America	 Arid	(Bsk)	 High	Edmonton(edm)	 Canada	 53.55°N/113.49°W	 North	America	 Continental	(Dfb)	 High	Fuzhou	(fuz)	 China	 26.08°N/113.31°E	 East	Asia	 Temperate	(Cfa)	 Upper	middle	Haikou	(hak)	 China	 20.03°N/110.33°E	 East	Asia	 Tropical	(Am)	 Upper	middle	Harbin	(har)	 China	 45.77°N/126.63°E	 East	Asia	 Continental	(Dwa)	 Upper	middle	Shenzhen	(hksz)	 China	 22.53°N/114.05°E	 East	Asia	 Temperate	(Cwa)	 High	Kuala	Lumpur	(kul)	 Malaysia	 3.16°N/101.70°E	 South	East	Asia	 Tropical	(Af)	 Upper	middle	Las	Vegas	(lav)	 USA	 36.17°N/115.14°W	 North	America	 Arid	(Bwk)	 High	Manila	(man)	 Philippines	 14.58°N/120.99°E	 South	East	Asia	 Tropical	(Af)	 Lower	middle	Melbourne	(mel)	 Australia	 37.82°S/144.96°E	 Oceania	 Temperate	(Cfb)	 High	Mexico	City	(mex)	 Mexico	 19.42°N/99.13°W	 Central	America	 Temperate	(Cwb)	 Upper	middle	Nanchang	(ncx)	 China	 28.67°N/115.90°E	 East	Asia	 Temperate	(Cfa)	 Upper	middle	Phoenix	(phx)	 USA	 33.44°N/112.07°W	 North	America	 Arid	(Bwh)	 High	Seattle	(sea)	 USA	 47.61°N/122.34°W	 North	America	 Temperate	(Csb)	 High	Seoul	(sel)	 South	Korea	 37.57°N/126.98°E	 East	Asia	 Continental	(Dwa)	 High	Shanghai	(shh)	 China	 31.24°N/121.50°E	 East	Asia	 Temperate	(Cfa)	 Upper	middle	Singapore	City	(sin)	 Singapore	 1.30°N/103.84°E	 South	East	Asia	 Tropical	(Af)	 High	Tianjin	(tjn)	 China	 33.86°S/151.21°E	 East	Asia	 Continental	(Dwa)	 Upper	middle	Tokyo	(tko)	 Japan	 39.13°N/117.20°E	 East	Asia	 Temperate	(Cfa)	 High	Vancouver	(van)	 Canada	 35.70°N/139.70°E	 North	America	 Temperate	(Cfb)	 High	1	Latitude/Longitude	of	each	city	is	the	location	of	the	city	center	used	in	this	dissertation			 			2.2. Landsat	time	series			Landsat	data	have	been	recorded,	organized,	and	distributed	by	the	U.S.	Geological	Survey	(USGS)	since	1972.	One	of	the	most	critical	components	of	Earth	Observation	(EO)	and	land	use	monitoring	is	a	continuous	archive	of	images	(Wulder	et	al.,	2008).	Landsat	has	continually	imaged	the	Earth	surface	every	16	days	for	almost	40	years	and	become	freely	available	since	2008.	Data	acquired	by	the	Landsat	program	represent	a	unique	combination	of	spatial,	spectral,	and	temporal	resolutions	that	are	desirable	to	chronicle	both	anthropogenic	and	natural	impacts	of	the	land	status	and	dynamics	for	the	past	three	decades	(Woodcock	et	al.,	2008).		Launched	in	1972,	Landsat-1-3,	the	Multi-Spectral	Scanner	(MSS),	pioneered	some	of	the	earliest	planetary	observation	programmes.	The	onboard	MSS	sensor,	capable	of	capturing	multi-spectral	Figure	2.1	Geographic	locations	of	studied	cities.			15	information	at	a	80-meter	pixel	size,	covered	four	electromagnetic	bands,	namely,	green	(500-600nm),	red	(600-700nm),	and	near	infrared	(700-800nm,	800-1100nm).			The	era	of	Thematic	Mapper	(TM)	started	with	the	launch	of	Landsat-4	in	1982.	Comparing	to	MSS,	the	TM	sensor	was	spatially	and	spectrally	more	capable,	adding	two	critical	spectra	(i.e.	shortwave	infrared,	and	thermal	infrared)	at	a	30-meter	resolution.	Landsat-5,	launched	in	1984,	carried	both	MSS	and	TM,	revolutionized	Earth	observation.	Landsat-5	remained	operational	until	2011,	ensuring	a	continuous	time	series	despite	the	failure	of	Landsat-6	in	1993.		Landsat-7,	carrying	an	Enhanced	Thematic	Mapper	(ETM+),	was	launched	in	1999	and	continued	imaging	the	Earth	with	the	addition	of	a	panchromatic	band	at	a	15-meter	spatial	resolution	that	is	finer	than	any	other	previous	Landsat.	Unfortunately	from	May	31	2003	onward,	Landsat-7	suffered	from	a	technical	failure,	causing	approximately	22%	per	image	area	with	no	data	collected.	This	issue	was	also	known	as	the	scanline	off	effect	(SLC-off).		The	Operational	Land	Imager	(OLI)	was	introduced	in	2013	with	the	launch	of	Landsat-8.	The	OLS	added	two	more	spectral	bands,	ultra	blue	for	coastal	application	and	another	infrared	band	for	cloud	detection.	At	the	same	spatial	and	spectral	resolution	as	the	EMT+	sensor	on	Landsat-7,	the	addition	of	Landsat-8	continued	the	Landsat	tradition,	seamlessly	collecting	information	all	around	the	global.		Another	turning	point	during	the	history	of	Landsat	was	the	open-access	policy	implemented	in	2008	when	all	existing	and	future	Landsat	data	were	made	freely	accessible	to	the	general	public	(Wulder	et	al.,	2008).	The	de-commercialization	of	Landsat	data	skyrocketed	the	number	of	Landsat	images	used	in	research	worldwide,	marking	the	transition	between	conventional	scene-based	processing	to	more	pixel-based	compositing	approaches	possible	(Wulder	et	al.,	2008).		This	dissertation	utilized	data	acquired	from	Landsat-5	TM	and	Landsat-7	TM/EMT+	from	1984	to	2013.	Images	captured	using	Landsat-1-3	were	not	considered	due	to	limited	spatial	details	and	spectral	bands.	There	was	a	distinct	difference	between	algorithms	used	to	process	Landsat-8	images	and	Landsat-4–5	TM,	and	Landsat-7	ETM+	Surface	Reflectance,	known	as	the	Landsat	Ecosystem	Disturbance	Adaptive	Processing	System	(LEADPS).	However,	LEADPS	was	not	available	at	the	time	of	commencing	this	project.	Therefore,	Landsat-8	data	were	not	included	in	this	dissertation.					16	2.3. DMSP-OLS	nighttime	lights	time	series			Originally	designed	as	a	meteorology	sensor,	the	Operational	Linescan	System	(OLS)	initiate	the	acquisition	of	Nighttime	lights	(NTL),	a	collection	of	satellite	images	taken	during	the	night.	The	earliest	OLS	flown	by	U.S.	Air	Force	Meteorological	Satellite	Program	(DMSP)	started	collecting	data	in	the	early	1970s.	The	DMSP	satellites	are	orbiting	in	a	near	polar	sun	synchronous	orbit	with	an	altitude	of	approximately	830	km.	Each	satellite	passes	over	any	location	on	Earth	twice	a	day,	providing	a	complete	global	coverage	in	about	six	hours.	The	Operational	Linescan	System	(OLS)	records	images	along	a	3000	km	scan,	corresponding	to	a	temperature	range	from	190	to	310	Kelvins	in	256	equal	intervals.	Onboard	calibration	is	performed	in	each	scan.	Final	pixel	values	are	shown	as	Digital	Numbers	(DN)	rather	than	absolute	values	in	Watts	per	m2.	A	telescope	pixel	is	approximately	500m	at	high	resolution	mode.	The	final	product	was	distributed	on	a	global	latitude-longitude	grid	at	a	spatial	resolution	of	30	arcs	second	grid,	approximately	1-km	at	the	equator.	Two	composites	were	produced	for	years	when	data	were	acquired	by	two	sensor	spontaneously	(Highlighted	in	Table	2.2).		Since	1992,	the	OLS	data	were	archived	and	distributed	digitally	by	the	National	Oceanic	and	Atmospheric	Administration	(NOAA)	National	Geophysical	Data	Center	(NGDC).	Compared	to	other	remote	sensing	products	such	as	Landsat,	NTL	is	an	exceptional	geographic	data	that	highlights	human	activities	and	reveals	the	“cultural	footprint”	of	each	individual	settlement	(Kyba	et	al.,	2014).	Artificial	light	during	nighttime	is	undoubtable	one	of	the	most	direct	measurements	of	human	activity	available	through	remote	sensing	in	a	way	that	daylight	data	is	not	capable	of	(Kyba	et	al.,	2014).		Shortly	after	its	initial	release	in	1992,	the	DMSP-OLS	has	been	collecting	continuous	images	of	the	Earth	at	night	for	over	21	years,	pushing	remote	sensing	data	into	urbanization	and	socio-economic	studies,	a	domain	that	was	before	dominated	mostly	by	census	data	(Bennett	&	Smith,	2017).	With	its	annual	stable	cloud-free	composite	of	average	brightness	the	flagship	product,	DMSP-OLS	has	been	widely	used	proxy	for	variables	that	are	difficult	to	assess	and	reproduce	at	a	global	scale,	for	example,	economic	activity,	carbon	emissions,	poverty,	impervious	surface	density,	and	energy	and	water	use.				This	dissertation	utilized	the	DMSP	version	4.0	stable	lights	time	series	which	was	pre-processed	on	an	annual	increment	using	methods	described	by	Baugh,	Elvidge,	Ghosh,	&	Ziskin	(2010),	following	a	set	of	criteria	(C.	Elvidge,	Hsu,	Baugh,	&	Ghosh,	2014):				17	1. Center	half	of	orbital	swath	for	optimal	image	quality	and	reduced	noise	2. No	sunlight	and	moonlight	present		3. No	solar	glare	contamination		4. No	cloud	coverage		5. No	auroral	activities/emissions		6. Normal	gain	settings		7. No	gas	flaring	contamination			Table	2.2	NTL	DMSP-OLS	Version	4	average	visible	stable	cloud-free	lights	(Years	highlighted	contain	data	acquired	by	two	sensors.				 Satellites	Year	 F10	 F12	 F14	 F15	 F16	 F18	1992	 F101992	 	 	 	 	 	1993	 F101993	 	 	 	 	 	1994	 F101994	 F121994	 	 	 	 	1995	 	 F121995	 	 	 	 	1996	 	 F121996	 	 	 	 	1997	 	 F121997	 F141997	 	 	 	1998	 	 F121998	 F141998	 	 	 	1999	 	 F121999	 F141999	 	 	 	2000	 	 	 F142000	 F152000	 	 	2001	 	 	 F142001	 F152001	 	 	2002	 	 	 F142002	 F152002	 	 	2003	 	 	 F142003	 F152003	 	 	2004	 	 	 	 F152004	 F162004	 	2005	 	 	 	 F152005	 F162005	 	2006	 	 	 	 F152006	 F162006	 	2007	 	 	 	 F152007	 F162007	 	2008	 	 	 	 	 F162008	 	2009	 	 	 	 	 F162009	 	2010	 	 	 	 	 	 F182010	2011	 	 	 	 	 	 F182011	2012	 	 	 	 	 	 F182012							18	2.4. Data	overview		Landsat	surface	reflectance	images	were	used	throughout	this	dissertation.	Particularly	in	Chapter	3	and	4,	I	utilized	all	surface	reflectance	data	from	1984	to	2012	with	less	than	70%	cloud	cover.	Chapter	5	used	all	images	from	DMSP-OLS	version	4	images	from	1992	to	2012.	Images	acquired	by	two	sensors	(Table	2.2)	were	all	downloaded	for	calibration	purposes.	Chapter	6	used	data	from	both	Landsat	and	NTL	time	series.	Thus,	only	images	from	1992	and	2012	were	used	in	Chapter	6	(Figure	2.2).															Figure	2.2	Time	frame	and	sensors	used	for	each	research	chapter.			19	Chapter	3		3. How	can	metrics	derived	from	remotely	sensed	data	inform	environmental	and	socio-economic	dynamics	within	and	across	cities	in	the	pan	Pacific	region?	Introduction			In	the	pan	Pacific	region,	urbanization	has	become	a	major	driver	of	local	and	regional	development	and	is	responsible	for	major	vegetation	loss	from	land	clearing	(Lo	&	Marcotullio,	2000b).	Increasing	numbers	of	people	have	migrated	into	urban	environments	over	the	last	decade	and	by	2026	the	urban	growth	rate	is	projected	to	exceed	50%,	making	the	pan	Pacific	region	the	fastest	urbanizing	area	in	the	world	(UNESCAP,	2012).	Large	differences	in	socio-economic	factors,	such	as	population	size	and	density,	political	system,	and	economic	development	cause	the	trend	in	mass	migration	to	vary	significantly	among	cities	(Bagan	&	Yamagata,	2014).	As	a	result,	the	urban	environment	is	highly	spatially	complex,	yet	some	common	patterns	exist	(Marshall,	2013).	Many	cities	are	proximal	to	productive	agricultural	land	and	other	natural	assets	such	as	forests	and	fresh	water	resources.	Further	outward	from	urban	cores,	cities	often	become	less	urbanized	and	more	vegetated	(Chen	et	al	2014),	resulting	in	an	inverse	relationship	between	urban	areas	and	vegetative	conditions.	As	an	urban	area	develops,	outer	rural	areas	often	undergo	more	intense	development	and	restructuring	compared	to	urban	core	areas	(Champion,	2001;	Lo	&	Marcotullio,	2000a).	The	concentric	ring	model	has	been	a	successful	tool	for	investigating	urban	structure,	even	for	cities	with	less	regular	concentric	growth	patterns	(Dietzel,	Herold,	Hemphill,	&	Clarke,	2005;	Handayani	&	Rudiarto,	2014).	For	cities	located	along	coastlines	or	in	mountainous	areas,	a	concentric	ring	model	can	clearly	differentiate	the	urban	core	from	its	more	peripheral	structures	(Guérois	&	Pumain,	2008).		Detecting	and	analyzing	the	spatio-temporal	dynamics	of	urban	environments	has,	therefore,	become	an	increasingly	critical	research	topic	with	real	management	applications	(Masek,	Lindsay,	&	Goward,	2000;	K.	C.	Seto	&	Fragkias,	2005;	Sexton	et	al.,	2013;	X.	Yang	&	Lo,	2002).	A	key	element	to	developing	an	understanding	of	urbanization	processes	globally	is	the	consistent	monitoring	of	cities	over	space	and	time	(Sexton	et	al.,	2013).	This	complexity	makes	urban	land	monitoring	a	challenging	task	(Sexton	et	al.,	2013).	Remote	sensing	technology	offers	an	exceptional	resource	that	allows	assessment	of	urban	environments	over	time	and	space	by	collecting	reflectance	of	urban	land-cover	characteristics	(Masek	et	al.,	2000;	Woodcock	et	al.,	2008).	One	common	method	of	using	remote			20	sensing	in	the	urban	environments	is	through	land	use	classification	(e.g.,	urban,	vegetation,	agriculture,	etc.).		However,	the	limitation	of	static	categorical	classifications	is	that	they	do	not	represent	dynamics	among	different	land-cover	and	land	use	types	and	therefore	is	insufficient	for	further	statistical	analysis	given	its	categorical	nature	of	measurement.	Furthermore,	categorical	classification	requires	a	set	of	pre-defined	classes	which	might	be	incomparable	for	all	selected	cities.	An	alternative	method	of	monitoring	urban	environments	using	remote	sensing	is	through	spectral	indices	–	numerical	indicators	primarily	derived	from	spectral	band	combinations	(Wentz	et	al.,	2014).		Spectral	indices	produce	a	continuous	measurement	of	urban	land-cover	rather	than	classifying	pixels	into	categorical	classes	(e.g.	urban	vs.	non-urban;	vegetation	vs.	non-vegetation).	This	minimizes	the	problems	that	arise	from	spectral	mixing	and	inferential	land-cover	interpretation	that	result	from	the	heterogeneity	of	urban	environments	(Barnsley	&	Barr,	1997;	Wentz	et	al.,	2014).		Two	common	indices	based	on	vegetation	and	urbanized	area	detection	are,	respectively,	the	Enhanced	Vegetation	Index	(EVI)	and	the	Normalized	Built-up	Index	(NDBI),	both	of	which	have	been	used	to	assist	in	characterizing	and	quantifying	vegetation	condition	and	urbanization	(Lyon,	Yuan,	Lunetta,	&	Elvidge,	1998;	Varshney,	2013).	(Huete,	Jackson,	&	Post	(1985)	and	Huete	&	Tucker	(1991)	suggested	that	EVI	can	be	used	as	an	alternative	to	other	vegetation	indices	(e.g.,	Normalized	Difference	Vegetation	Index)	for	minimizing	the	negative	effects	caused	by	canopy	background	and	atmosphere	interference,	and	thus	enhance	vegetation	signals.	Conversely,	NDBI	has	been	demonstrated	to	offer	accurate	and	objective	delineations	of	urban	settlements	(Zha	et	al.,	2003).	Combined,	these	two	spectral	indices	provide	a	means	to	track	the	trends	of	urban	and	vegetation	covers	across	cities	over	long	time	periods.	In	general,	for	any	land-cover	observation	and	monitoring	program,	one	of	the	most	crucial	components	is	a	continuous	archive	of	imagery	(Wulder	et	al.,	2008).	However,	the	availability	of	cloud-	and	haze-free	images	relies	heavily	on	variable	local	weather	conditions	and	as	a	result	regional	investigations	over	long	time	periods	at	moderate	spatial	resolutions	(i.e.,	30m)	are	still	rare.	Data	gaps	are	common	for	a	given	pixel	in	any	particular	year,	which	may	limit	the	utility	of	images	given	most	existing	image	processing	algorithms	(Hermosilla,	Wulder,	White,	Coops,	&	Hobart,	2015).		Thus,	data	gaps	contribute	to	lost	information	for	long	term	urban	land-cover	monitoring	projects.	Pixel-based	compositing	provides	a	solution	to	long	term,	seamless	land	monitoring	challenges	and	enables	a	new	paradigm	of	Earth	observation	programs	due	to	the	freely	accessible	Landsat	image	archive	(White	et	al.,	2014).	These	new	compositing	methods	take	full	advantage	of	multiple	decades	of	Landsat	imagery			21	and	are	able	to	detect	and	in-fill	missing	pixel	reflectance	values	based	on	the	entire	temporal	trajectory	of	any	given	pixel	(Hermosilla	et	al.,	2015).	In	this	chapter	I	utilized	a	relatively	simple	urban	expansion	morphological	model	to	compare	and	contrast	a	dense	time	series	of	moderate	spatial	resolution	imagery	and	assess	how	urban	and	vegetation	vary	at	regional	scales	in	time	and	space.	To	do	so	annual	gap-free	Landsat	imagery	were	created	from	1984	-	2012	of	25	cities	across	12	countries	in	the	pan	Pacific	region.	Two	spectral	indices—Enhanced	Vegetation	Index	and	Normalized	Difference	Built-up	Index—were	calculated	at	an	annual	basis	and	resampled	using	a	hexagon-based	ring	model	for	all	urban	environments.	Trajectories	of	the	two	spectral	indices	were	then	analyzed	based	on	dynamic	time	warping	and	K-means	clustering	analysis	to	investigate	intra-	and	inter-urban	variations	over	time	and	space.		3.1. Materials	and	methods		This	chapter	used	imagery	from	the	entire	Landsat	Thematic	Mapper	(TM)	Landsat	Enhanced	Thematic	Mapper	Plus	(ETM+)	archive	from	1984	-	2012	(section	2.2)	using	a	pixel-based	compositing	method	(section	3.2.1)	to	generate	annual	gap-free	image	composites	for	25	selected	cities.		I	used	previously	defined	Global	Administrative	Areas	(GADM	version	2.0,	Areas,	2012)	with	a	hexagon	based	ring	model	to	assess	urban	vegetation	and	urban	built-up	conditions	within	each	city	(section	2.4)	using	a	mask	image	to	minimize	seasonal	variations	and	exclude	water	and	snow	cover	from	any	subsequent	analysis	(section	2.5).	I	then	clustered	the	performance	of	each	urban	environment	based	on	its	temporal	and	spatial	characteristics	(section	2.6)	to	allow	inter-	and	intra	city	comparisons.			3.1.1. Landsat	surface	reflectance	product			Surface	reflectance	images	(L1T),	including	Landsat	Thematic	Mapper	(TM)	and	Landsat	Enhanced	Thematic	Mapper	Plus	(ETM+),	were	acquired	for	each	urban	environment	through	the	United	States	Geological	Survey	(http://espa.cr.usgs.gov).	I	downloaded	images	with	less	than	70%	cloud	cover	that	were	acquired	within	a	specific	temporal	window,	from	May	1st	to	September	30th	of	the	years	1985	to	2012.	As	I	focused	on	differentiating	urban	land-cover	from	vegetation,	winter	imagery	in	Australia	was	more	suitable	given	the	high	curing	rate	of	vegetation	and	grasses	in	Australian	urban	areas	in	summer			22	(Keast,	2013).	With	mild	winter	temperature	(Keast,	2013),	and	with	much	less	deciduous	vegetation	in	winter	with	no	snow,	vegetation	is	often	very	green	in	winter	and	in	fact	produces	a	much	greater	contrast	with	urban	features	than	in	summer.	As	a	result,	I	applied	the	same	uniform	temporal	window	to	Melbourne	and	Sydney	as	all	other	cities.	Candidate	images	were	all	pre-processed	using	a	mask	function	(Fmask)	(Zhu	&	Woodcock,	2012)	and	the	Landsat	Ecosystem	Disturbance	Adaptive	System	(LEDAPS;	Schmidt,	Jenkerson,	Masek,	Vermote,	&	Gao,	2013).		Given	the	wide	temporal	window	used	for	image	selection,	seasonal	spectral	variation	due	to	agriculture	practices	can	confuse	the	spectral	indices	(Hill	&	Donald,	2003).	To	minimize	the	impact	caused	by	climate	and	phenological	variations	at	each	urban	location,	an	agricultural	mask	was	generated	using	a	simple	random	forest	classification.	To	do	so	the	variables	including	the	average,	maximum,	minimum,	and	standard	deviation	of	the	two	spectral	indices	from	2008	–	2012	were	used	on	the	assumption	that	if	a	pixel	was	identified	as	agriculture	in	the	last	five-year	period,	it	would	likely	to	have	been	agriculture	or	non-urban	land-cover	in	the	earlier	years.	Training	sites	of	agriculture,	forest,	and	impervious	surface	were	identified	from	the	available	Google	Earth	imagery.		3.1.2. Pixel-based	image	compositing		The	best	available	pixel	(BAP)	approach	was	used	to	produce	image	composites	to	mitigate	data	gaps	in	the	time	series	caused	by	cloud,	cloud	shadow,	or	haze.	BAP	scored	every	pixel	using	multiple	Landsat	images	based	on	the	following	criteria:	sensor,	day-of-year,	atmospheric	opacity,	and	proximity	to	cloud	or	cloud	shadow	using	the	same	criteria	as	White	et	al.,	(2014).	Pixels	identified	by	the	Fmask	algorithm	(Zhu	&	Woodcock,	2012)	as	cloud	or	cloud	shadows	were	masked	from	the	composites.	Additionally,	a	50	pixel	buffer	was	applied	around	clouds	and	cloud	shadows	to	reduce	misclassification	errors	(Griffiths	et	al.,	2010).	The	Thematic	Mapper	(TM)	sensor	was	scored	higher	than	the	Enhanced	Thematic	Mapper	Plus	(ETM+)	sensor	after	2003	to	reduce	the	influence	of	the	scan-line	corrector	malfunction.	Day-of-year	was	scored	according	to	a	Gaussian	function	with	a	maximum	score	equivalent	to	the	middle	day-of-year	of	the	temporal	window.	Pixels	with	atmospheric	opacity	higher	than	10%	were	scored	lower	to	avoid	the	selection	of	hazy	observations.	The	pixels	with	the	highest	scores	based	on	these	criteria	were	then	composited	into	annual	images	that	were	used	in	the	disturbance	mapping	process	(White	et	al.,	2014).	Finally,	any	remaining	data	gaps	were	infilled	using	linear	interpolation	of	the	values	on	the	spectral	trends.	Two	spectral	indices,	the	EVI	and	NDBI,	were	calculated	to	assess	urbanization	and			23	vegetation	changes	(Equations	1	and	2).	Urbanization	and	vegetation	dynamics	were	then	assessed	by	analyzing	the	temporal	trajectories	from	1985	–	2012	of	a	given	spectral	index	(i.e.	EVI	or	NDBI).		EVI = 2.5	× NIR − RedNIR + 6×Red − 7.5×Blue + 1 																			6789:;<=	1	NDBI = SNIR − NIRSNIR + NIR 																																																																	6789:;<=	2		3.1.3. Concentric	ring	model			I	used	the	Global	Administrative	Areas	(GADM	version	2.0,	Areas,	2012)	to	define	administrative	borders	(GADM	version	2.0,	Areas,	2012),	which	has	also	been	used	by	a	number	of	regional	and	global	geospatial	urban	studies	(Gaston,	Duffy,	&	Bennie,	2015;	Hawelka	et	al.,	2014).	Conventional	urban	ring	models	developed	to	monitor	urban	development	often	utilize	absolute	distances	to	define	ring	widths,	and	therefore	cities	with	more	variable	forms	and	sizes	can	confound	effective	comparisons	and	cross-analysis.	Given	I	am	considering	multiple	cities,	the	distance	between	concentric	rings	was	normalized	to	produce	a	normalized	distance	index	(NDI)	which	represents	how	close	a	particular	pixel	is	to	the	urban	center	(Equation	3).	To	produce	general	trends,	and	reduce	high	frequency	noise,	the	indices	were	then	summarised	within	a	hexagon-based	(~1km2)	concentric	ring	model	which	divides	each	urban	environment	into	a	lattice	of	rings	which	were	defined	using	a	normalized	distance	index	(Figure	3.1).	The	distance	of	each	hexagon	to	the	urban	centre	is	normalized	to	values	between	0	(urban	core)	and	1	(urban	edge).	The	central	hexagon	representing	the	urban	core	for	each	individual	urban	environment	was	located	at	the	central	business	districts	(CBD).		Temporally,	to	simplify	the	subsequent	statistical	analysis	and	minimize	the	impact	of	any	proxy	outliers,	I	produced	epoch	images,	which	were	five-year	averaged	image	composite.	Epoch	images	were	then	used	in	the	clustering	analysis.							24	Normalized	Distance	Index	(NDI) = MNMOPQ#	ST	UVQPWSXYN 																																	6789:;<=	3	where	d = x] − x^ _ + y] − y^ _		X]/^	is	the	x	coordinate	of	hexagon	m/n	Y]/^	is	the	y	coordinate	of	hexagon	m/n	df	is	the	distance	between	hexagon	i	to	the	central	hexagon	d]hi	is	the	greatest	distance	among	all	hexagons							Figure	3.1	Examples	of	a	hexagon	based	ring	model;	(a)	Vancouver,	Canada;	(b)	Shanghai,	China;	(c)	Bangkok,	Thailand;	(d)	Melbourne,	Australia.			25		3.1.4. Dynamic	time	warping	and	cluster	analysis			Once	spectral	trajectories	of	EVI	and	NDBI	from	1985	–	2012	were	developed,	cities	were	clustered	based	on	either	the	temporal	changes	or	the	spatial	distribution	of	urban	vegetation	and	urban	built-ups	using	Dynamic	time	warping	and	K-means	cluster	analysis.		Dynamic	time	warping	(DTW)	is	an	approach	originally	developed	to	deal	with	sequential	data	was	used	to	compare	each	epoch.	DTW	has	been	widely	used	in	voice	recognition	(Müller	2007),	pattern	matching	(Rath	&	Manmatha,	2003),	and	time	series	analysis	(Bemdt	&	Clifford,	1994;	Keogh	&	Ratanamahatana,	2005).	Specifically,	DTW	identified	the	optimal	alignment	between	two	temporal	sequences	that	may	vary	in	time	or	rate	(Bemdt	&	Clifford,	1994;	Salvador	&	Chan,	2007).	Although	the	approach	was	developed	for	time-dependent	sequences,	it	could	also	be	used	to	examine	trends	in	any	data	stream	presented	as	a	linear	sequence	(Giorgino,	2009),	such	as	in	this	case	where	trajectories	along	a	distance	index	were	compared.	The	approach	computed	the	similarity	of	spectral	indices	through	time	by	calculating	the	difference	in	the	index	between	each	epoch	pair	for	each	hexagon-based	concentric	ring	from	the	centre	to	the	edge	of	the	urban	environment.			The	mean	of	the	differences	was	then	compared	to	contrast	cities.	Cities	with	similar	trajectories	through	time	were	expected	to	have	minimal	differences	and	thus	a	low	overall	mean	across	the	epochs.	In	contrast,	if	a	city	has	been	changing	markedly	the	mean	differences	would	appear	greater.	The	magnitude	of	EVI	and	NDBI	were	also	used	to	assess	and	compare	the	vegetation	and	urban	dynamics,	with	an	urban	environment	with	more	vegetation	cover	having	a	greater	EVI	and	lower	NDBI.	In	general,	cities	that	were	undergoing	more	rapid	development	were	expected	to	have	a	greater	NDBI	and	lower	EVI.		Once	temporal	and	spatial	change	were	captured,	a	K-means	cluster	methodology	was	used	to	group	the	mean	and	standard	deviations	of	the	difference	curves	for	both	EVI	and	NDBI	as	well	as	the	absolute	spectral	values	across	cities.		I	selected	the	k	value	based	on	the	highest	silhouette	widths	by	iterating	a	K-means	clustering	analysis	25	times	using	k	from	1	to	25.		I	then	applied	K-means	clustering	four	times	using	the	four	sets	of	variables	representing	the	spatial	and	temporal	changes	of	vegetation	and	urban	conditions.	The	silhouette	width	provided	an	indication	of	the	accuracy	of	the	classification	for	each	individual	urban	environment	and	was	used	to	assess	the			26	clustering	performance	(Bolshakova	&	Azuaje,	2003).		As	silhouette	widths	approached	1,	it	was	indicative	of	a	strong	likelihood	a	given	urban	environment	was	assigned	to	the	correct	cluster	(Bolshakova	&	Azuaje,	2003;	Rousseeuw,	1987).	I	also	computed	the	ratio	of	between	(BSS)	to	total	sum	of	square	(TSS)	to	evaluate	the	K-means	classifications,	where	BSS	was	the	sum	of	squared	distances	per	group	means	while	TSS	was	the	sum	of	squared	distances	to	the	global	means.		All	image	processing	in	this	paper	was	done	in	IDL	8.3,	ENVI	5.2,	and	ArcMap	10.2.2	and	statistical	analysis	in	R	3.1.1	using	the	Cluster	package	(Maechler,	Rousseeuw,	Struyf,	Hubert,	&	Hornik,	2012),	and	the	DTW	package	(Giorgino,	2009).		3.2. Results	3.2.1. Spectral	indices	Figure	3.2	demonstrated	the	hexagon	based	rings	for	four	example	cities’	spatial	distribution	of	vegetation	(colored	in	green)	and	urban	conditions	(coloured	in	red).	A	high	EVI	value	was	indicative	of	marked	vegetated	cover	whereas	NDBI	corresponded	to	greater	urban	expansion	and/or	urban	densification.	In	general,	EVI	and	NDBI	showed	contrasting	patterns	with	urban	cores	having	a	much	higher	NDBI	and	lower	EVI	values	than	the	outer	areas.				27	As	expected,	regardless	of	the	magnitude	of	the	EVI	and	NDBI,	all	cities	had	undergone	expansions	and/or	densifications	over	the	epochs	but	at	varying	rates.	For	example,	cities	such	as	Vancouver	and	Melbourne	experienced	less	urban	expansion	than	Shanghai	and	Manila	where	EVI	experienced	a	sustained	reduction.	Likewise,	cities	where	substantial	vegetation	loss	were	observed	to	have	a	marked	increase	in	the	NDBI.	Small	satellite	settlements	around	urban	centers	could	also	be	detected	in	the	epoch	images,	such	as	in	Shanghai	at	epochs	3	and	4,	which	showed	a	number	of	small	“cities”	developing	around	the	urban	core.	Figure	3.2	Temporal	changes	of	EVI	and	NDBI	of	four	studied	urban	environments.			28	Figure	3.3	showed	that	spatially	consistent	across	all	urban	environments	outer	areas	generally	experienced	the	highest	EVI	value	and	the	lowest	NDBI	value.	Despite	the	varying	differences	among	each	trajectory,	EVI	values	tended	to	decrease	as	the	distance	index	increases	from	the	urban	center.	A	similar	pattern	can	also	be	observed	as	NDBI	decreases	when	moving	towards	outer	area	of	the	urban	environment.	However,	exceptions	can	be	found	in	cities	such	as	Melbourne	where	the	highest	EVI	value	was	located	in	the	transition	area	between	inner	and	outer	urban	areas.	Figure	3.3	Example	annual	spectral	trajectories	of	EVI	(left	column)	and	NDBI	(right	column).	See	Appendix	1	for	all	25	cities.				29	Temporally,	Figure	3.3	also	demonstrated	the	overall	change	rates	by	comparing	the	differences	between	each	temporal	trajectory.	For	a	given	urban	environment,	a	greater	distance	between	trajectories	was	expected	when	there	was	a	marked	change	in	EVI	and/or	NDBI	values.	For	example,	Manila	had	much	more	notable	EVI	decrease	and	NDBI	increase	comparing	to	cities	like	Tokyo	and	Vancouver.	Combined	with	the	distance	index,	Figure	3.3	illustrated	where	and	how	rapidly	different	urban	environments	changed	over	time.	For	example,	Melbourne	and	Shanghai	experienced	more	changes	in	the	transition	areas	than	the	centers	and	edges	of	the	urban	environments.			3.2.2. Dynamic	time	warping	and	cluster	analysis	Figure	3.4	showed	the	spatial	distribution	of	vegetation	and	urban	built-up	represented	by	their	separability	as	derived	from	DTW	analysis	across	all	urban	environments.	Greater	separability	values	were	indicative	of	more	different	trajectory	shapes	hence	dissimilar	vegetation	or	urban	spatial	distribution.	Overall,	the	results	suggested	that	the	separability	was	more	apparent	for	the	EVI	trajectory	than	for	NDBI	as	indicated	by	a	slightly	higher	average	separability	of	EVI	(0.36).	As	DTW	estimated	the	separability	using	only	the	spatial	information	which	incorporated	the	absolute	EVI	and	NDBI	values	for	each	of	the	concentric	rings,	Figure	3.4	highlighted	urban	environments	that	experienced	a	relatively	unusual	spatial	distribution	of	vegetation	and/or	urban	patterns.	For	example,	Phoenix	(phx)	had	a	large	separability	in	EVI	compared	to	the	other	urban	environments	while	Vancouver	had	an	overall	higher	separability	in	terms	of	NDBI.												30			A	K-means	clustering	analysis	was	then	applied	to	group	spatially	similar	urban	environments.	A	K	value	(i.e.	number	of	clusters)	will	be	favored	if	it	generated	greater	silhouette	width	(i.e.	closer	to	1)	and	a	meaningful	clustering	analysis	(i.e.	k	<	15).	The	results	indicated	5	optimum	clusters	resulted	in	the	highest	silhouette	width.	The	averaged	trajectory	of	each	cluster	is	shown	in	Figure	3.5	for	both	EVI	and	NDBI.	Results	show	variations	in	EVI	and	NDBI	values	for	each	of	the	clusters.	Cluster	5	showed	the	highest	EVI	and	the	lowest	NDBI	in	the	urban	centre,	suggesting	a	relatively	greener	and	less	urbanized	environment	core	than	other	groups.	Cluster	2	and	4	showed	a	much	lower	EVI	and	higher	NDBI,	indicating	more	developed	and	less	vegetative	urban	environments.	Interestingly,	although	cluster	3	had	an	overall	higher	EVI	than	cluster	1,	2,	and	4,	it	also	had	a	greater	NDBI	index,	particularly	in	the	sub-urban	areas.		Figure	3.4	Separability	metrics	estimated	by	DTW	for	a.	EVI	and	b.	NDBI.				31	Figure	3.5	summarized	the	temporal	changes	for	each	epoch	periods.	Temporal	changes	were	represented	by	DTW	derived	distances	between	pairs	of	epoch	trajectories.	Urban	environments	were	ordered	based	on	their	averaged	temporal	changes	(e.g.	distances)	between	trajectories	as	shown	in	the	last	columns	in	Figure	3.5.	Although	the	absolute	order	varied	when	comparing	EVI	or	NDBI	changes,	more	developed	urban	environments	such	as	Vancouver	(van),	Seattle	(sea),	and	Phoenix	(phx)	tended	to	have	less	temporal	changes	of	both	EVI	and	NDBI	as	indicated	by	relatively	small	distances	between	epoch	trajectories.	Urban	environments	located	in	less	developed	regions	such	South	and	South-east	Asia	tended	to	experience	more	temporal	changes	with	varying	change	magnitudes	over	different	epoch	periods.	For	example,	Dalian	(dal)	showed	marked	temporal	changes	of	both	EVI	and	NDBI.	Nanchang	(ncx)	showed	more	notable	changes	of	EVI	between	epoch	1	and	2	and	less	change	in	NDBI.	Interestingly,	urban	environments	such	Las	Vegas	experienced	a	relatively	great	temporal	change	of	EVI	yet	remained	constantly	low	in	NDBI	changes	suggesting	a	marked	change	of	vegetation	and	a	constant	stable	trend	in	urban.		Figure	3.5	Average	trajectories	of	a.	EVI	and	b.	NDBI	for	each	cluster.			32	A	K-means	clustering	analysis	was	also	applied	to	group	temporally	similar	urban	environments.	Combined	with	the	cluster	results	from	Figure	3.5,	the	final	results	of	K-means	clustering	were	summarized	in	Figure	3.7	using	the	silhouette	width	of	each	individual	urban	environment.	The	average	silhouette	width	ranged	from	0.37	to	0.60,	and	the	ratio	of	the	between	to	the	total	sum	of	square	(BSS/TSS)	ranged	from	69%	to	92%	(Table	3.1).	Overall,	using	temporal	changes	in	NDBI	and	EVI	offered	higher	silhouette	width	and	BSS/TSS	ratio	than	spatial	separability.				Table	3.1	Cluster	analysis	summary		 BSS/TSS	ratio	 Silhouette	width	EVI	 NDBI	 EVI	 NDBI	Spatial	distribution	 68.8%	 89.2%	 0.37	 0.48	Temporal	changes	 91%	 92.3%	 0.60	 0.50		Silhouette	width	of	each	individual	urban	environment	indicated	how	well	an	urban	environment	fits	its	cluster	(Figure	3.7).	Urban	environments	with	low	silhouette	width	were	likely	to	be	assigned	to	the	Figure	3.6	Temporal	change	metrics	of	a.	EVI	and	b.	NDBI.			33	incorrect	cluster.	Low	silhouette	widths	were	therefore	useful	as	a	measure	of	uniqueness	of	a	given	urban	environment.	For	example	Melbourne	and	Calgary	had	low	silhouette	width	(<	0.1)	and	therefore	had	limited	similarity	in	terms	of	spatial	distribution	of	vegetation	(Figure	3.7a).	Dalian	and	Phoenix	were	also	relatively	unique	in	terms	of	their	spatial	distribution	of	urban	vegetation	and	built-ups	(Figure	3.7b).	Temporal	changes	yielded	slightly	better	overall	clustering	results	with	a	number	of	urban	environments	such	as	Kuala	Lumpur	and	Phoenix	having	a	low	silhouette	width	(Figure	3.7c-d).			Figure	3.8	summarized	the	degree	of	similarity	both	spatially	and	temporally	across	urban	environments	in	the	pan	Pacific	region.	A	likely	strong	connection	(e.g.,	thicker	lines	in	Figure	3.8)	between	two	urban	environments	represented	high	spatio-temporal	similarity.	In	total,	there	were	three	pairs	of	urban	environments	that	were	consistently	being	grouped	together—Melbourne	and	Sydney;	Tianjin	with	Manila;	and	Singapore	City	with	Kuala	Lumpur.	In	contract,	cities	such	as	Las	Vegas	and	Vancouver	demonstrated	less	similar	features	both	spatially	and	temporally	with	any	of	the	other	urban	environments.		Figure	3.7	Silhouette	width	based	on	a.	temporal	vegetation	changes,	b.	temporal	built-up	changes,	c.	spatial	vegetation	patterns,	and	d.	spatial	built-up	patterns	(groups	are	identified	by	colors).			34		Figure	3.8	Number	of	times	cities	were	grouped	together	during	K-means	classification	based	upon	spatial	and	temporal	variables.	Number	of	groupings	ranged	from	2	to	4.			3.3. Discussion			3.3.1. Urbanization	patterns	in	pan	Pacific	cities				This	chapter	integrated	a	hexagon	based	ring	model	with	the	pixel-based	image	compositing	to	investigate	spatio-temporal	changes	of	vegetation	and	urban	built-up	among	25	cities	across	the	pan	Pacific	region.	As	anticipated,	the	majority	of	the	cities	in	developing	countries	observed	marked	development	rates	compared	to	the	urban	environments	in	more	developed	regions.	Still,	exceptions	existed	where	urban	environments	that	were	not	geographically	proximal	could	also	be	similar	in	terms	of	the	vegetation	and	urban	dynamics	over	time	and	space.	This	was	likely	due	to	the	highly	concentrated	early	urbanization	of	large	cities	in	some	of	the	developing	nations,	such	as	Hong	Kong,	Bangkok,	and	Shanghai,	which	may	resemble	the	urbanization	patterns	of	cities	in	a	more	developed	regions	(Henderson,	Yeh,	Gong,	Elvidge,	&	Baugh,	2003).	I	concluded	with	a	simple	model	which			35	generalizes	the	urbanization	process	as	starting	as	a	centralized	landscape,	then	expanding	radially	(Figure	3.9).	More	mature	urban	environments,	such	as	Vancouver,	Melbourne,	and	Tokyo	tended	to	decentralize	over	time,	characterizing	an	evenly	spread	spatial	pattern	with	less	visible	urban	cores	(e.g.,	Figure	3.8,	Melbourne	and	Vancouver).	A	similar	trend	occurred	on	less	developed	urban	environments	where	small	and	medium	sized	settlements	started	blending	into	each	other,	forming	a	more	decentralized	urban	environment,	such	as	Shanghai.	Interestingly,	although	the	trend	seemed	to	be	highly	similar	across	majority	of	the	urban	environments,	the	time	required	to	reach	a	similar	level	of	urbanization	magnitude	was	much	shorter	for	developing	than	developed	urban.					36	Remote	sensing	based	assessment	of	urban	environments	also	reflected	some	of	the	socio-economic	pressures,	particularly	for	cities	that	have	undergone	substantial	development	over	the	past	few	decades.	For	example,	the	population	of	Shanghai	has	a	95%	growth	rate	with	a	population	increase	from	11.8	million	to	over	23.0	million	since	the	1980s	(World	Population	Statistics,	2013).	To	accommodate	such	a	massive	population	growth,	as	shown	in	Figure	3.9,	Shanghai	has	massively	expanded	and	already	started	decentralizing	and	in-filling	gaps	between	the	pre-existing	urban	core	and	the	surrounded	settlements.	During	the	studied	period	of	this	work	(i.e.	1984-2012),	the	population	of	greater	Vancouver	has	only	increased	from	1.4	million	to	2.3	million,	which	is	about	48%	slower	than	Shanghai	(BC	Statistics,	2012).	Figure	3.9	A	conceptual	representation	of	urbanization.			37	The	uneven	development	rates	were	likely	due	to	a	number	of	geographical	factors	including	the	current	urban	areal	extent,	political	system,	and	urban	population	distribution	(Birch	&	Wachter,	2011).	In	addition,	the	historical	development	of	each	urban	environment	was	also	a	critical	variable.	Regions	such	as	North	America	and	Australia	were	generally	developed	much	earlier	than	urban	environments	in	less	developed	countries	and	therefore	were	likely	to	be	saturated	and	more	stable	in	terms	of	urban	expansion	and	vegetation	conditions.	Since	all	images	were	acquired	after	the	mid-1980s,	it	was	difficult	to	show	patterns	of	urbanization	in	these	developed	regions.	An	uneven	developing	rate	could	also	be	found	among	different	sections	(i.e.	concentric	rings)	within	the	urban	environments,	suggesting	an	intra-urban	developing	variation.	Understanding	intra-	urban	variation	is	critical	for	urban	planners	to	accomplish	a	balanced	and	sustainable	urban	development.	By	2050,	if	all	cities	doubled	their	size,	roughly	10	to	15%	of	the	productive	agricultural	land	would	be	converted	into	concrete	to	house	the	rapidly	growing	population	(Birch	&	Wachter,	2011).	Such	a	trend	poses	a	major	threat	to	sub-urban	environments	around	agricultural	and	farm	land	as	well	as	water	pollution.	Therefore	quantitative	information	is	needed	for	sustainable	urban	planning,	crafting	local	developing	polices,	and	setting	land	use	priority	to	optimize	both	urban	development	and	resources	allocation.		3.3.2. Urban	densification	and	expansion			However,	urbanization	is	not	only	occurring	in	the	form	of	physical	expansion,	but	also	densification.	In	most	cases,	expansion	is	occurring	in	rural	or	sub-urban	areas,	while	intensification	can	often	be	seen	in	existing	urban	areas	as	well	as	the	transition	zones	between	urban	core	and	urban	edge.	Maps	are	needed	that	are	capable	of	representing	urban	land	extent	(i.e.	expansion)	as	well	as	measuring	the	intensity	of	changes	in	both	pre-existing	and	newly	developed	urban	environments.	Although	mapping	of	the	urban	environments	has	been	undertaken	through	a	variety	of	perspectives	and	models,	such	as	urban	land-cover	classification,	urban	population,	urban	heat	island	effect,	etc.,	I	chose	an	alternative	method	using	urban	and	vegetation	metrics,	using	two	distinct	features	of	the	urban	environment.	Spectral	indices	(i.e.,	EVI	and	NDBI)	avoid	using	hard	land-cover	classifications,	and	offer	quantitative	insights	into	the	physical	changes	of	the	urban	environments.	Another	key	benefit	of	spectral	indices	over	hard	classifications	is	that	a	continuous	numeric	measurement	provides	more	options	for	further	statistical	analysis	which	can	help	better	reveal	the	relationships	and	patterns	in	a	dataset.			38	Relatively	small	urban	areas,	in	particular	those	in	less	developed	countries,	are	often	the	subject	of	less	research	interest	than	larger	urban	environments	and	therefore	have	been	often	overlooked	in	many	studies.	Previous	studies	have	indicated	that	about	2/3	of	urban	residents	live	in	cities	of	less	than	1	million	people	(Clancey,	2004).	Even	in	United	States,	where	almost	45	million	people	live	in	cities	with	a	population	of	over	250,000,	and	another	40	million	live	in	places	of	between	50,000	and	250,000	(Clancey,	2004).	Assess	changes	in	these	urban	environments	using	remote	sensing	offers	a	way	to	provide	data	to	urban	planners	for	closer	examination	of	the	dynamics	in	these	urban	areas.	By	including	these	smaller	urban	environments	in	this	paper	enables	us	to	provide	a	more	comprehensive	picture	of	urbanization	in	the	pan	Pacific	region.	This	chapter	used	the	traditional	concentric	ring	model	to	quantify	and	compare	vegetation	and	urban	dynamics	within	and	among	different	urban	environments.	Each	urban	environment	was	assigned	with	a	single	urban	core,	which	is	a	monocentric	model.	Although	polycentric	or	multi-nuclei	models	suggest	that	modern	urban	environments	are	likely	to	have	more	than	one	functional	urban	core,	in	this	research	I	believe	that	using	a	single	urban	core	was	sufficient	for	meeting	our	key	objective	of	analyzing	changes	to	urban	and	vegetation	cover	over	long	time	periods.	First,	although	the	concentric	ring	model	assigns	a	single	core	to	each	urban	environment,	I	are	still	able	to	quantify	intra-urban	growth	by	using	the	Normalized	Distance	Index	(NDI).	The	NDI	not	only	allows	us	to	compare	urban	environment	with	varying	physical	sizes,	it	also	provided	a	sense	of	urban	development	in	the	surrounding	satellite	settlements.	For	example,	a	decreasing	EVI	and	increasing	NDBI	will	likely	to	occur	near	NDI=0.5	which	indicate	some	degree	of	development	in	the	outer	areas.	Secondly,	the	major	CBD	is	more	likely	to	be	the	most	developed	area	of	an	urban	environment	even	in	a	multi-nuclei	spatial	organization	(Wheaton	2004).	Lastly,	given	the	wide	spectrum	of	urban	environments	I	studied	in	this	work,	some	less	developed	urban	environments	(e.g.	Fuzhou	and	Nanchang)	are	still	highly	concentric	and	yet	have	not	clearly	shown	evidences	of	forming	additional	urban	cores.	Thus,	I	decided	to	apply	the	concentric	ring	model	with	one	major	urban	core.	In	this	chapter,	I	focused	on	examining	the	overall	trends	in	vegetation	and	urban	built	up	than	the	actual	timing	of	the	change.	Using	a	combination	of	epoch	images	and	linear	interpolation	was	sufficient	to	capture	overall	EVI	and	NDBI	trends	over	the	study	period.	Secondly,	image	quality	and	availability	varied	dramatically	across	the	globe	(Wulder	et	al.,	2015).	The	purpose	of	using	epoch	images	and	linear	interpolation	was	to	essentially	further	reduce	unnecessary	noises	and	outliers	while	still	retain	the	trend	of	land-cover	changes.				39	Yet,	there	were	two	key	caveats	to	be	addressed	in	the	next	chapter.	Firstly,	spectral	saturation	is	known	to	cause	inaccurate	descriptions	of	urban	vegetation	(Van	Der	Meer	&	De	Jong,	2000).	Studies	have	proven	the	superiority	of	spectral	unmixing	compared	to	spectral	indices	for	quantitatively	estimating	vegetation	(Elmore,	Mustard,	Manning,	&	Lobell,	2000;	Hostert,	Röder,	&	Hill,	2003).		Vegetation	estimated	from	spectral	indices	is	not	compatible	particularly	across	a	heterogeneous	region	such	as	an	urbanized	environment	(C.	D.	Elvidge	&	Lyon,	1985;	Huete	et	al.,	1985;	Huete	&	Tucker,	1991;	Major,	Baret,	&	Guyot,	1990;	Todd	&	Hoffer,	1998).	Additionally,	many	spectral	indices	derived	from	multispectral	images	(e.g.	NDVI,	EVI,	and	NDBI)	are	typically	constrained	to	two	or	three	spectral	bands,	leaving	other	spectral	information	under	utilized	(Lyon	et	al.,	1998).	In	this	chapter,	certain	land-cover	types	such	as	snow	and	ice	also	showed	similar	EVI	and	NDBI	values	to	urban	dominated	pixels.	Thus,	by	spectrally	unmixing	vegetation	the	influence	of	these	other	land-cover	types	is	minimised.	Since	EVI	and	NDBI	both	use	the	Near	Infrared	part	of	the	spectra,	information	overlapping	and	redundancy	might	be	expected	when	comparing	EVI	directly	against	NDBI.		In	Chapter	4,	an	alternative	approach,	Spectral	Mixture	Analysis	(SMA)	was	used	to	spectrally	unmix	each	pixel	using	a	linear	unmixing	algorithm.	A	vegetation	fraction	value	(VF)	was	then	derived	to	represent	how	vegetated	each	pixel	is	relative	to	the	most	spectrally	pure	pixel	(i.e.	endmember)	within	each	city	over	the	study	period.													40	Chapter	4		4. How	to	can	urban	vegetation	be	mapped	at	the	sub-pixel	scale?		4.1. Introduction			In	the	previous	chapter,	I	discussed	the	benefit	of	using	remote	sensing	to	chronologically	monitor	urbanization	in	terms	of	urban	expansion	and	urban	densification	(Chapter	3,	Section	3.4.2).	In	this	chapter,	I	focus	on	investigating	an	alternative	approach	to	measure	temporal	change	in	vegetation	using	spectral	unmixing	analysis	and	a	Theil-Sen	slope	estimator.	The	combination	of	these	two	processing	techniques	aims	to	minimize	the	possible	saturation	issues	from	spectral	indices	(e.g.	EVI	in	Chapter	3)	and	data	outliers.			Urbanization	can	be	defined	as	a	gradual	land-cover	change	in	the	form	of	urban	sprawl	and	densification	(Sexton	et	al.,	2013).	Urban	sprawl,	or	urban	expansion	is	the	physical	growth	of	a	city,	primarily	through	conversion	from	non-urban	land-cover	(e.g.	vegetation)	to	the	presence	of	urban	land-cover	(e.g.	impervious	surfaces).	Urban	densification	most	often	occurs	adjacent	to	existing	urban	areas	where	the	natural	land-cover	has	already	been	disturbed	to	some	extent.	The	interplay	between	urban	sprawl	and	densification	is	consistently	re-shaping	the	local	geometric	and	ecological	properties	of	urban	environments,	increasing	the	density	of	anthropogenic	infrastructure	while	replacing	local	vegetation,	interrupting	micro	climate,	habitat	loss,	energy	fluxes,	and	modifying	the	water	and	carbon	cycles	(Groffman	et	al.,	2014;	Kahn,	2000;	Ziter,	2016).	Previous	studies	have	suggested	that	urbanization	has	a	direct	association	with	a	series	of	environmental	issues,	such	as	urban	heat	island	(UHI,	Oke,	1982),	habitat	loss	(McKinney,	2002,	2006),	and	water	shortage	(Gober,	2010;	Kummu,	Ward,	De	Moel,	&	Varis,	2010;	Wu	&	Tan,	2012).		An	effective	way	of	mitigating	the	negative	impacts	brought	about	by	urbanization	is	through	urban	vegetation	(Escobedo,	Kroeger,	&	Wagner,	2011;	Nowak	&	Dwyer,	2007).	Urban	vegetation	is	a	term	that	collectively	describes	urban	greenspaces,	including	parks,	wetland,	grassland,	and	patches	of	urban	gardens	(Kumagai,	n.d.;	Ridd,	1995;	Tooke	et	al.,	2009).	The	presence	of	urban	vegetation	is	known	to	be	beneficial	to	modifying	the	local	climate,	and	thus	the	social,	and	physical	environments	through	temperature	control	(Oke,	1982),	air	pollution	reduction	(Nowak	et	al.,	2006),	noise	and	storm	water	control	(Glass	&	Singer,	1972),	and	habitat	preservation	(Nowak	&	Dwyer,	2007).	Studies	have	also			41	indicated	the	significant	social	(Grahn	&	Stigsdotter,	2003),	economic	(Tyrväinen	et	al.,	2005),	and	aesthetic	values	(Jim	&	Chen,	2006;	Tyrväinen	et	al.,	2005)	associated	with	urban	vegetation.	As	a	result,	urban	vegetation	has	been	utilized	as	an	effective	tool	to	achieve	sustainable	and	functional	urban	environments.	Efforts	towards	preserving	healthy	urban	vegetation	can	therefore	be	found	worldwide,	particularly	in	developed	regions,	such	as	in	North	America	and	Europe	(Nowak,	2002).		In	less	developed	areas,	despite	the	benefits	and	services	offered	by	urban	vegetation,	economic	growth	and	urbanization	often	receive	higher	prioritization	than	preserving	and	maintaining	urban	vegetation	(Grimm	et	al.,	2008).	Vegetation	in	these	urban	environments	often	grows	in	more	isolated	and	fragmented	patches	compared	to	vegetation	grown	in	more	novel	and	well-managed	urban	environments,	making	it	more	challenging	to	manage	for	local	urban	planners.	Another	concern	associated	with	poorly	managed	and	fragmented	urban	vegetation	is	ecological	inequity	(N.	Heynen	et	al.,	2006),	causing	uneven	access	for	local	residents	to	quality	urban	green	space.	With	the	majority	of	new	urban	residents	located	in	less	developed	regions	(Grimm	et	al.,	2008),	there	is	a	strong	need	for	data	and	methodological	approaches	that	are	capable	of	quantifying	and	tracking	vegetation	changes	over	space	and	time	(Clancey,	2004).		Remote	sensing	is	an	exceptional	data	source	to	urban	planners	and	researchers	(Jensen	&	Cowen,	1999).		Although	studies	using	fine	spatial	resolution	imagery	(Benz,	Hofmann,	Willhauck,	Lingenfelder,	&	Heynen,	2004),	hyperspectral	data	(Heiden,	Segl,	Roessner,	&	Kaufmann,	2007;	Roberts	et	al.,	1998),	and	aerial	photography	(Hodgson,	Jensen,	Tullis,	Riordan,	&	Archer,	2003)	have	shown	promising	results	all	have	limited	spatial	and	temporal	coverage	limiting	their	global	application.	A	recent	review	by	(Schneider,	2012)highlighted	the	potential	value	of	multi-temporal	dense	image	stacks	generated	from	moderate	spatial	resolution	(e.g.	30meters)	remote	sensing	platforms	such	as	the	Landsat	series	of	satellites	to	urban	remote	sensing	research.	Recently,	increased	accessibility	of	the	Landsat	data	archives	(Schneider	&	Woodcock,	2008;	Wulder	&	Coops,	2013)	with	more	sophisticated	image	processing	procedures	(Griffiths	et	al.,	2010;	White	et	al.,	2014)	and	compositing	techniques	(Hermosilla	et	al.,	2015)	have	allowed	mapping	of	urban	land-cover	as	well	as	quantitatively	describing	urban	physical	features	and	patterns	at	regional	scales	over	30	years.			However,	with	a	30-meter	pixel	size,	spectral	information	collected	by	the	Landsat	Thematic	Mapper	(TM),	and	Enhanced	Thematic	Mapper	(ETM+)	are	more	likely	to	contain	a	mixture	of	surface	materials	(Small,	Elvidge,	Balk,	&	Montgomery,	2011).	Spectrally	mixed	pixels	are	commonly	seen	in	Landsat	images	where	multiple	surface	materials	jointly	occupy	one	pixel	(Keshava	&	Mustard,	2002).	Sub-pixel			42	analysis	or	spectral	unmixing	has	been	developed	to	determine	the	areal	amount	of	pure,	distinct,	surface	materials	within	a	single	pixel.	Spectral	unmixing	is	well	established	in	the	remote	sensing	literature	and	has	applied	in	a	large	number	of	studies	across	a	broad	range	of	spatial	resolutions	(Asner	&	Heidebrecht,	2002;	Van	Der	Meer	&	De	Jong,	2000;	Vikhamar	&	Solberg,	2003).		While	urban	environments	are	highly	heterogeneous,	there	are	some	common	land-cover	properties	consistent	across	all	cities,	such	as	vegetation,	impervious	surfaces,	and	soil	(e.g.	the	V-I-S	model)	(Ridd,	1995)	and	as	a	result	systematic	unmixing	models	can	be	generated.	Spectral	unmixing	involves	two	critical	steps,	identifying	pure	surface	materials	(i.e.	endmember)	followed	by	decomposing	mixed	pixels	(Shi	and	Wang	2014).	Theoretically,	the	selected	endmembers	should	represent	all	spectral	variations	in	the	image.	Based	on	previous	work	(Van	Der	Meer	&	De	Jong,	2000),	although	endmembers	derived	directly	from	the	image	are	relatively	less	divergent	compared	to		laboratory	measured	spectra,	they	have	the	advantage	of	sharing	a	more	similar	atmospheric	conditions	which	is	essential	for	unmixing	Landsat	time	series.	The	majority	of	previous	research	has	focused	on	unmixing	single	images	with	less	research	on	unmixing	multi-temporal	time	series	data.	Building	upon	previous	urban	spectral	unmixing	research	(Ridd,	1995;	Tooke	et	al.,	2009),	this	chapter	aims	to	further	contribute	to	this	field	by	i)	incorporating	pixel	based	compositing	(PBC)	techniques	to	produce	seamless	annual	image	composites,	ii)	examine	the	capacity	of	spectral	unmixing	approaches	to	be	applied	to	dense	annual	Landsat	composites	from	1984	to	2012	to	determine	vegetation	cover	of	25	urban	environments	in	the	pan	Pacific	region;	and	iii)	generate	temporal	and	spatial	information	on	urban	vegetation	features	based	on	the	distance	and	orientation	from	urban	centers	and	boundaries.	Such	information	is	valuable	for	local	urban	planners	as	it	offers	insights	into	the	within-urban	spatial	and	temporal	dynamics	of	urban	vegetation	change,	and	importantly,	it	has	potential	to	assist	regional	cross-urban	study	in	areas	such	as	the	pan	Pacific	region,	one	of	the	most	diverse	and	fastest	growing	areas	in	terms	of	urbanization	(Lo	&	Marcotullio,	2000b).		4.2. Materials	and	methods		The	opening	of	Landsat	archive	has	allowed	the	chronicling	of	land-cover	changes	over	a	large	spatial	area	with	longer	and	denser	temporal	dimensions	than	previously	available.	The	amount	of	vegetation	in	each	pixel	and	year	was	determined	by	applying	spectral	unmixing	analysis	to	each	multi-temporal	urban	image	stack	(section	4.2.1).	The	temporal	trends	in	urban	vegetation	were	then	estimated	using			43	the	Theil–Sen	estimator	(Sen,	1968;	Theil,	1992)	on	the	estimated	vegetation	fraction	(section	4.2.1).	Image	processing	in	this	paper	was	done	in	IDL	8.3,	ENVI	5.2,	and	ArcMap	10.2.2	and	statistical	analysis	in	R	3.1.1.		4.2.1. Spectral	unmixing			Conventionally,	spectral	indices	have	been	widely	used	to	extract	and	monitor	vegetation	dynamics.	However,	spectral	indices	can	saturate	in	area	s	of	high	canopy	cover	and	leaf	area	index	(Jackson	et	al.,	2004).	Alternatively,	spectral	unmixing	approaches	allows	users	to	pre-define	endmembers	and	compute	a	fraction	score	representing	the	abundance	of	a	given	endmember	(i.e.	vegetation).	The	classic	unmixing	process	can	be	summarized	in	three	major	steps:	(1)	identify	endmembers	(i.e.	spectrally	pure	pixels);	(2)	build	spectral	library;	and	(3)	apply	unmixing	algorithm	(Singer	and	McCord	1979;	Ridd	1995).	I	used	the	pixel	based	compositing	(PBC)	image	stack	for	each	city	for	endmember	selection	and	followed	the	same	procedure	as	Small	and	Lu	(2006)	and	Tooke	et	al.	(2009)	who	incorporated	a	three	endmember	model,	including	a	vegetation,	a	high	albedo	(i.e.	reflected	brightness)	Figure	4.1	Image	mosaicking	and	transformation	process	using	a).	Pixel	based	compositing	(PBC)	multispectral	images	(6	spectral	bands);	b).	Mosaicked	PBC	image	(6	spectral	bands)	and	c).	Minimal	Noise	Fraction	transformed	PBC	mega	(3	components	bands).			44	and	dark	endmember.	As	suggested	by	others	(Wu	and	Murray	2003),	I	also	included	a	water	mask	and	undertook	a	Minimal	Fraction	Noise	transformation	(Green,	Berman,	Switzer,	&	Craig,	1988)	prior	to	the	unmixing	(Figure	4.1).	I	selected	the	first	3	components	for	collecting	endmembers	as	they	explained	over	98%	of	the	variance	in	all	sample	urban	environments.		A	spectral	library	of	the	selected	endmembers	was	then	developed	for	each	MNF	transformed	proxy	image	based	on	the	following	two	assumptions	(Equation	1).	First	was	linearity	where	the	spectra	of	each	mixed	pixel	is	a	linear	combination	of	the	endmembers	as	recorded	in	the	spectral	library.	Second	was	unity	which	implies	that	for	each	mixed	pixel,	the	sum	of	fractions	(f),	should	equal	to	one.	Fraction	images	with	values	indicating	fractions	of	each	endmember	for	mixed	pixels	were	generated	for	each	urban	environment	and	used	for	the	subsequent	trend	analysis.																																																			j = kNlN + m																	6789:;<=	1				XNno 																																												Where	R	is	the	unmixed	surface	reflectance;	fi	is	the	fraction	or	the	fraction	of	the	surface	reflectance	value	of	endmember	ei;	ε	is	the	root	mean	square	error;	n	is	the	total	number	of	endmembers.			4.2.2. Validation			Unmixed	vegetation	fractions	were	validated	using	high	spatial	resolution	Google	Earth	images	for	the	corresponding	year.	As	quality	and	availability	of	cloud	free	Google	Earth	images	varied	greatly	from	region	to	region,	I	chose	the	best	available	images	for	the	following	city,	Vancouver	(year	2009),	Tokyo	(year	2006),	Las	Vegas	(year	2011)	and	Shenzhen-Hong	Kong	area	(year	2009).	The	reference	vegetation	fraction	was	determined	by	defining	a	random	sample	of	100	pixels	stratified	across	10	vegetation	fraction	classes.	Each	pixel	was	then	divided	into	a	6x6	m	grid	(i.e.	25	grids	per	pixel).	Each	grid	was	interpreted	as	either	having	a	presence	or	absence	of	vegetation.	The	reference	vegetation	fraction	value	was	calculated	by	counting	the	number	of	vegetated	grids	within	each	pixel	(e.g.	20	vegetated	grids	equal	to	80%	vegetation	fraction).	Spectrally	unmixed	vegetation	fraction	was	then	compared	against	the	interpreted	reference	vegetation	fraction	using	correlation	analysis.					45	4.2.3. Vegetation	trend	analysis			When	using	annual	vegetation	fraction	images,	cities	located	in	high	latitudes	(e.g.	Harbin)	and	particularly	mountainous	regions	(e.g.	Vancouver)	may	require	extra	caution.	In	those	cases,	pixels	with	low	vegetation	fraction	may	not	necessarily	indicate	intense	urbanization,	but	rather	low	vegetated	land-cover	types,	such	as	bare	rock	and	snow.	One	major	benefit	of	utilizing	the	entire	time	series	is	that	it	allows	changes	in	vegetation	fraction	over	time	to	be	easily	assessed	rather	than	classifying	images	into	discrete	land-cover	types	based	on	limited	temporal	snapshots	of	the	city.		I	applied	a	non-parametric	Mann-Kendall	test	for	each	pixel	to	determine	whether	or	not	the	trend	was	significantly	monotonic	(Mann,	1945).	I	then	applied	the	Theil-Sen	estimator	(TS;	Sen,	1968;	Theil,	1992)	to	fit	only	significant	vegetation	change	trends	identified	by	the	Mann-Kendall	test.	Given	its	robustness	and	computational	efficiency	with	distribution-free,	non-normal	input	data	requirements	(Wilcox,	2010),	the	TS	estimator	has	been	widely	used	in	remote	sensing	trend	analysis	(Fernandes	&	G.	Leblanc,	2005;	Hansen,	M.	C.,	Roy,	D.	P.,	Lindquist,	E.,	Adusei,	B.,	Justice,	C.	O.,	Alstatt,	2008).	The	TS	estimator	provided	a	slope	value	by	taking	the	median	slope	of	all	pairwise	time	series	data	points.	This	procedure	was	repeated	for	every	non-water	pixel	of	all	25	cities.	An	alpha	level	of	0.05	was	used	for	all	statistical	tests.		To	derive	the	spatial	and	temporal	patterns	of	urban	vegetation	changes	I	analyzed	the	trend	results	using	concentric	rings	which	have	been	widely	applied	in	previous	urban	studies	(Handayani	&	Rudiarto,	2014)	and	chapter	3	to	demonstrate	the	spatial	distribution	of	urban	vegetation	changes.	I	divided	the	60-km	radius	circular	study	area	into	100	concentric	rings	(i.e.	600-meter	per	ring)	to	summarize	vegetation	trends	from	urban	core	to	urban	edge.	The	60-km	radius	circle	was	also	divided	into	360	slices	(i.e.	1-degree	per	slice)	to	generate	a	directional	circular	histogram	to	examine	how	vegetation	changes	based	on	its	direction	to	urban	core.	I	summarized	the	median	of	the	direction	and	slope	of	the	change	trends	for	each	concentric	ring	and	slice.								46	4.3. Results		4.3.1. Vegetation	fraction			Vegetation	fraction	was	estimated	for	each	urban	environment	(examples	in	Figure	4.2).	Pixels	with	high	vegetation	fraction	were	largely	located	outside	urban	centers.	Low	vegetation	fraction	pixels	also	successfully	delineated	small	satellite	settlements	as	well	as	urban	corridors	that	connect	existing	urban	centers	and	developing	sub-urban	settlements	(Figure	4.2	a-b).	Exceptions	can	be	found	in	cities	located	in	dry	and	deserted	regions	such	as	Las	Vegas	(Figure	4.2	e)	where	opposite	vegetation	patterns	occur	as	urbanized	area	was	relatively	much	greener	than	rural	and	undeveloped	areas.				Validation	using	high	spatial	resolution	Google	Earth	images	showed	a	correlation	coefficient	of	0.66,	0.72,	0.72,	and	0.77	for	Vancouver,	Las	Vegas,	Tokyo,	and	Shenzhen-Hong	Kong	area,	respectively	(Table	Figure	4.2	Landsat	proxy	image	(left	panel)	and	unmixed	vegetation	fraction	results	(right	panel)	in	year	2000	of	a).	Tokyo,	b).	Shanghai,	c).	Melbourne,	d).	Mexico	City	e).	Las	Vegas,	and	f).	Vancouver	(scale	1:	800,000).			47	4.1).	As	shown	by	the	median	values	in	Table	4.1,	I	found	that	overall	in	Shenzhen-Hong	Kong	and	Vancouver,	the	estimated	vegetation	fraction	were	lower	than	the	reference	vegetation	fraction	values	while	in	Las	Vegas	and	Tokyo,	the	estimated	vegetation	fraction	was	slightly	higher	than	the	interpreted	one.	Table	4.1	.	Correlations	of	interpreted	versus	estimated	vegetation	fractions	with	median	values.		 Correlation	coefficient	 Median	(Interpreted)	Median	(Estimated)	Shenzhen-Hong	Kong	 0.77	 0.62	 0.50	Las	Vegas	 0.72	 0.36	 0.49	Tokyo	 0.72	 0.48	 0.50	Vancouver	 0.66	 0.88	 0.51		Figure	4.3	shows	the	median	value	between	interpreted	and	estimated	vegetation	fraction	for	every	20%	increment.	The	interpreted	vegetation	fraction	from	Google	Earth	images	showed	greater	variations	than	Landsat	estimated	vegetation	fraction	as	indicated	by	a	wider	standard	deviation	error	bar	(Figure	4.3).	The	estimated	vegetation	fractions	were	higher	for	0-20%	stratum	while	the	interpreted	Figure	4.3	Comparison	between	estimated	and	interpreted	vegetation	fraction	of	a)	Shenzhen-Hong	Kong,	b)	Las	Vegas,	c)	Tokyo,	and	d)	Vancouver.	Error	bars	represented	the	standard	deviation	of	all	sample	points	within	each	20%	increment.			48	vegetation	fractions	exceeded	the	estimated	values	for	the	higher	vegetation	fraction	samples	(Figure	4.3a,	4.3c,	and	4.3d).	Las	Vegas,	however,	showed	an	opposite	pattern	where	the	interpreted	values	were	lower	than	the	estimated	vegetation	fraction	for	all	strata	(Figure	4.3b).		Figure	4.4	illustrates	an	annual	vegetation	fraction	time	series	in	Las	Vegas.	Unlike	other	urban	environments,	as	urbanization	develops	in	Las	Vegas,	vegetation	fraction	gradually	increases	particularly	near	the	sub-urban	and	the	small	satellite	urban	areas.	Urban	densification	can	also	be	seen	through	time	as	vegetation	fraction	decreased	in	urban	center.					49		4.3.2. Temporal	characteristics	of	vegetation	fraction			As	shown	in	Figure	4.5,	for	a	given	pixel,	a	negative	trend	slope	indicates	a	decreasing	vegetation	fraction	while	a	positive	slope	indicates	an	increasing	vegetation	fraction	during	the	study	period	(i.e.	1984-2012).	By	comparing	all	25	urban	environments,	it	was	apparent	that	each	urban	environment	was	highly	variable	in	terms	of	temporal	vegetation	changes.	To	capture	only	the	statistically	significant	trend,	slope	values	with	a	Man-Kendall	p-value	greater	than	0.05	were	not	included	in	the	subsequent	analysis	hence	masked	out	in	Figure	4.5.	The	absolute	value	of	the	trend	slope	(refer	to	as	vegetation	slope)	represented	the	magnitude	of	vegetation	loss	and/or	gains.	In	general,	most	vegetation	loss	Figure	4.4	Annual	vegetation	fraction	(0	–	100%)	results	of	Las	Vegas.			50	occurred	outside	the	existing	urban	core,	except	in	Las	Vegas	where	urbanization	caused	vegetation	gains	in	sub-urban	areas	and	the	urban	core	only	experienced	a	slight	vegetation	fraction	decrease.			4.3.3. Spatial	dynamics	of	vegetation	fraction			Median	vegetation	trend	slope	value	per	ring	was	extracted	and	plotted	for	all	25	urban	environments	(Figure	4.6).	In	general,	in	terms	of	spatial	distribution	of	the	trends,	there	were	four	main	types	of	urban	environment	trends.	The	first	includes	cities	such	as	Shenzhen-Hong	Kong	area	(Figure	4.6a)	which	exhibited	a	gradual	decline	in	vegetation	from	the	urban	center	through	to	the	outer	areas.	Second	set	of	cities	included	cities	such	as	Las	Vegas	that	had	consistently	increasing	vegetation	slope	as	the	distance	from	urban	center	increases	(out	to	20-km	in	the	case	of	Las	Vegas)	(Figure	4.6b).	The	third	type	contains	cities	such	as	Shanghai	(Figure	4.6c)	where	vegetation	slope	was	mostly	negative	across	the	60km	radius	circle.	These	types	of	cities	were	mostly	located	in	China,	including	Dalian,	and	Nanchang.	The	last	type	of	cities	such	as	Vancouver	(Figure	4.6d),	Tokyo,	Sydney,	Edmonton,	and	Calgary,	where	vegetation	changes	were	relatively	minimal	as	indicated	by	a	near	zero	vegetation	change	trend.	These	types	of	cities	were	mostly	located	in	developed	regions.		Figure	4.5	Theil-Sen	estimated	vegetation	trend	slope	(p	<	0.05)	of	a).	Calgary,	b).	Dalian,	c).	Seattle,	d).	Las	Vegas,	e).	Manila,	f).	Shenzhen-Hong	Kong,	g).	Seoul,	and	h).	Vancouver	(scale:	1:700,000).	Water	is	colored	as	grey.			51		Figure	4.7	exhibits	the	temporal	vegetation	trend	in	terms	of	vegetation	loss	or	gain	with	respect	to	its	direction	from	the	urban	center	between	1984	and	2012.	The	results	reflected	the	variability	across	cities	with	each	city	having	varying	magnitudes	and	spatial	distribution	of	vegetation.	Dalian,	for	example,	had	a	great	amount	of	vegetation	decrease	in	the	west	of	the	city	(Figure	4.7a)	while	Shenzhen-Hong	Kong	area	experienced	more	intense	vegetation	loss	in	the	north	(Figure	4.7b).	Most	inland	cities	such	as	Nanchang	and	Changsha	have	a	greater	spread	of	vegetation	loss	(Figure	4.7c	and	4.7e).	More	developed	cities	such	as	Vancouver	(Figure	4.7d),	although	not	showing	as	much	vegetation	loss	as	other	less	developed	urban	environments,	it	still	showed	some	degree	of	vegetation	decrease	(e.g.	the	northern	and	south-eastern	part	of	the	Vancouver).	Figure	4.6	Vegetation	trend	slope	median	per	ring	(600-meter	per	ring).		Green	represents	median	slope	value.	Grey	represents	confidence	interval	of	smoothed	slope	value	(black).	Appendix	2	shows	circular	histogram	of	all	25	cities.			52			Figure	4.7	Circular	histogram	of	vegetation	trend	with	slope	median	value	per	bin	(1-degree	per	bin).	The	bar	length	indicated	the	percentage	of	pixels	with	a	decreasing	vegetation	trend.	The	median	slope	value	was	colored	with	a	red-yellow-green	scheme	with	red	representing	the	most	negative	slope,	green	representing	the	most	positive	slope,	and	yellow	represent	stable	vegetation	fraction.	Water	is	colored	as	black.	Appendix	3	shows	circular	histogram	of	all	25	cities.				53	4.1. Discussion			4.1.1. Urban	boundaries		This	chapter	demonstrated	advances	in	previous	chapter.	These	include	more	appropriate	definitions	of	city	limits,	using	advanced	image	processing	approaches	to	extract	the	vegetation	of	pixels	and	updated	validation	approaches.	Compared	to	chapter	1,	I	defined	urban	boundaries	using	a	60-km	radius	buffer	from	the	city	center,	rather	than	using	an	inconsistent	global	database	of	administrative	borders.	Although	radius	buffers	are	an	unconventional	and	less	intuitive	way	to	define	urban	boundaries,	it	minimizes	issues	associated	with	finding	temporally	consistent	administrative	boundaries	across	multiple	different	jurisdictions	(e.g.	China	vs	United	States).	Global	Administrative	Areas	(GADM)	is	an	exceptional	data	source	that	provides	urban	boundaries	of	individual	countries	and	is	often	used	in	defining	urban	boundaries	however	it	is	apparent	that	the	boundaries	are	at	a	spatial	scale	not	compatible	with	the	30-meter	spatial	resolution	of	Landsat	imagery.	For	example,	rapidly	expanding	cities	such	as	Changsha,	which	comprises	a	number	of	urban	centers	was	not	well	captured	as	they	fall	outside	Changsha’s	administrative	border	(Figure	4.7e).	The	dimension	of	the	radius	buffer	is	however	dependent	on	the	location	(i.e.	coastal	vs	inland)	and	the	physical	footprint	of	the	city.	The	larger	the	radius	buffer,	the	more	image	data	that	is	required	to	be	downloaded	and	processed.	For	very	large	megacities	such	Shanghai	and	Tokyo,	a	60-km	radius	buffer	may	not	be	sufficient	cover	all	of	the	vegetation	changes	associated	with	urban	development	as	opposed	to	cities	such	as	Las	Vegas	and	Calgary	where	the	majority	of	the	urban	environments	easily	reside	within	the	60-km	radius	buffer.	I	acknowledge	that	a	fixed	radius	buffer	is	likely	less	optimal	for	coastal	cities	such	Dalian	(Figure	4.5b)	as	most	of	the	area	within	the	60-km	buffer	is	water	with	limited	land	mass	for	investigating	vegetation	changes	at	the	further	distances.			4.1.2. Spectral	unmixing	vs	spectral	index		Secondly,	compared	to	the	traditional	spectral	indices,	the	integration	of	spectral	unmixing	and	Theil-Sen	(TS)	estimated	trend	slopes	offers	increased	ability	to	compare	and	contrast	vegetation	across	urban	environments.	A	spectral	index	for	a	given	pixel	can	be	difficult	to	compare	across	different	urban	environment	over	time.	A	number	of	studies	have	discussed	the	ecological	meaning	of	vegetation	index	values	compared	to	vegetation	fraction	estimates	(Pettorelli	et	al.,	2005).	In	this	chapter	I	observed			54	similar	vegetation	fraction	values	from	natural	forest,	wetland,	and	agricultural	vegetation	across	the	25	cities	(e.g.	Figure	4.2).	The	absolute	score	of	an	unmixed	vegetation	fraction	pixel	can	be	interpreted	as	the	quantity	and/or	the	quality	of	different	types	of	vegetation.	Using	the	vegetation	fraction	of	a	single	pixel	at	a	single	snapshot	can	be	prone	to	a	range	of	errors	and	distortions	as	discussed	as	discussed	through-out	this	paper	resulting	in	potential	confusion	especially	in	heterogonous	urban	environments.	Therefore,	the	slope	of	change	in	vegetation	fractions	derived	from	a	time	series	of	spectrally	unmixed	vegetation	fractions	is	likely	to	be	a	more	robust	representation	of	the	relative	vegetation	change	at	a	sub-pixel	level	which	is	less	sensitive	to	individual	data	outliers	caused	by	factors	such	as	phenological	variations	and	vegetation	types.	Another	well-known	challenge	in	urban	remote	sensing	is	separating	bare	ground	and	soil	due	to	their	spectral	similarity	(Weng,	2012).	Therefore,	I	reduced	reliance	on	the	urban	built	up	index	(e.g.	Normalized	Difference	Built-up	Index	in	chapter	3)	and	assumed	vegetation	loss	was	primarily	caused	by	urban	growth	within	the	60-km	radius	buffer.			4.1.3. Google	Earth	as	a	validation	tool			Lastly,	I	included	a	validation	procedure	to	examine	the	vegetation	fractions	using	Google	Earth	Desktop.	Google	released	Google	Earth	in	2005	allowing	users	to	seamlessly	examine	any	location	on	the	globe	by	streaming	the	best	available	imagery.	With	up	to	10	petabytes	of	data	and	sophisticated	built-in	images	pre-processing	such	as	geo-referencing	and	image	mosaicking,	Google	Earth	is	a	reliable	data	source	for	visual	assessment	and	interpretation	has	previously	been	unavailable	to	remote	sensing	researchers	(Bey	et	al.,	2016).	I	found	that	for	two	of	the	Asian	megacities,	namely,	Bangkok	and	Manila,	the	results	confirm	those	of	Murakami,	Medrial	Zain,	Takeuchi,	Tsunekawa,	&	Yokota	(2005)	who	detected	a	decreasing	population	density	from	the	urban	center	to	urban	boundaries	indicating	a	less	active	urban	development	in	the	rural	and	suburban	area.	Although,	the	administrative	boundaries	were	used	in	Murakami,	Medrial	Zain,	Takeuchi,	Tsunekawa,	&	Yokota	(2005)	they	concluded	that	the	urban	growth	had	continued	beyond	the	city	limits.	Similarly,	I	detected	that	vegetation	condition	was	less	disturbed	as	the	distance	from	the	urban	center	increases,	likely	causing	by	less	urban	activity	in	such	areas.	Xian	&	Crane	(2006)	also	reported	the	reverse	trend	between	vegetation	and	urban	impervious	area	in	Las	Vegas,	indicating	urbanization	can	potentially	increase	the	vegetation	cover.	Practically,	Google	Earth	is	an	ideal	alternative	for	validation	purposes.	Compared	to	conventional	high	spatial	resolution	aerial	photography	and	satellite	imagery,	Google	Earth	imagery	has	two	main	advantages.			55	First,	Google	Earth	imagery	is	freely	accessible.	To	date,	most	of	high	spatial	resolution	remote	sensing	data	is	expensive	with	limited	access.	Second,	Google	Earth	images	offers	a	rich	temporal	record	of	locations	worldwide.	The	timeline	option	in	Google	Earth	enables	easily	navigate	through	time	allowing	users	to	select	the	highest	possible	quality,	cloud	free	imagery,	for	validation	purpose.	However,	the	validation	process	also	exposed	uncertainties	and	new	challenges.	Compared	to	previous	studies	(Small	et	al.,	2011;	Small	&	Lu,	2006),	our	correlation	coefficients	between	0.66	and	0.77	were	relatively	low.	This	is	partially	due	to	challenges	associated	with	human	interpretation	of	vegetation	fraction	scores	in	highly	vegetated	locations.	In	spectral	mixture	analysis,	a	common	validation	process	often	involves	applying	spectral	unmixing	on	high	resolution	aerial	photography	or	satellite	imagery.	Given	digital	high	spatial	resolution	imagery	was	not	readily	available	for	all	of	the	cities	of	interest	I	believe	visual	interpretation	approaches	were	appropriate	recognizing	the	benefit	of	Google	Earth	(Dorais	&	Cardille,	2011)	as	a	reliable	source	of	validation	images	that	match	the	scale	of	this	work.		4.1.4. Potential	drivers	of	urban	vegetation	change			Previous	studies	(N.	C.	Heynen	&	Lindsey,	2003;	J.	Liu,	Zhan,	&	Deng,	2005;	Luck,	Smallbone,	&	O’Brien,	2009)	have	summarized	a	list	of	potential	physical	drivers	behind	urban	vegetation	change,	including	topography,	proximity	to	water,	climate,	as	well	as	vegetation	type.	As	a	result	common	patterns	across	a	number	of	cities	are	evident.	Cities	such	as	Mexico	City	and	Edmonton	developed	in	flat	terrains	are	more	likely	to	expand	evenly	around	urban	center	hence	vegetation	loss	in	all	directions.		Cities	such	as	Dalian	and	Shenzhen-Hong	Kong	exhibited	a	typical	linear	urbanization	pattern,	which	is	often	seen	in	coastal	cities.	In	such	cases,	urbanization	is	largely	restricted	by	water	bodies.	Although,	due	to	the	poor	accessibility,	mountainous	topography	can	also	be	a	limiting	factor	in	spatial	urbanization,	it	does	not	restrict	urbanization	as	much	as	large	water	bodies.	Climate	and	vegetation	type,	have	more	absolute	effects	on	the	amount	and	condition	of	vegetation	within	an	urban	area.	Cities	located	in	dry	and	desert	regions	such	as	Las	Vegas	and	Phoenix	shown	opposite	vegetation	change	patterns	compared	to	more	temperate	or	tropical	urban	environments.	In	such	cities,	urbanization	often	brings	with	it	increases	in	vegetation,	replacing	inhabitable	landscapes	with	vegetated	human	settlements	such	as	manicured	gardens,	parks,	golf	courses	and	public	spaces.	Such	vegetation	gains	are	not	always	viewed	positively	and	be	harmful	for	local	environments	by	disturbing	underground	water	tables	(Gober,	2010)	and	introducing	invasive	species	(McKinney,	2002).				56	Besides	the	aforementioned	physical	factors,	there	are	a	number	of	other	social	and	cultural	factors	that	are	likely	to	contribute	to	the	observed	vegetation	changes	and	their	spatial	distribution.	Drivers	such	as	the	increasing	demand	of	single-family	housing		(Kestens,	Thériault,	&	Des	Rosiers,	2004)	and	advanced	fast	transportation	systems	(Janelle	&	Beuthe,	1997)	not	only	boost	people’s	desire	to	live	further	away	from	the	urban	center,	but	also	increase	the	inequality	within	urban	environments	in	terms	of	accessing	quality	urban	vegetation	(Lu	&	Chen,	2004).		Recent	efforts	by	urban	researchers	and	local	managers	have	focused	on	the	relationships	between	social	and	cultural	drivers	with	urban	related	land-cover	changes	(Swetnam	et	al.,	2011).	Next	chapter,	another	remote	sensing	derived	metric,	nighttime	lights	(NTL)	was	introduced	and	compared	against	conventional	census	date	to	examine	the	casual	relationship	between	population	and	GDP	on	urban	brightness.																		57	Chapter	5		5. Are	bright	cities	big	cities	?		5.1. Introduction					In	this	chapter,	I	first	intercalibrated	Nighttime	lights	(NTL)	data	using	a	localized	modelling	approach	to	ensure	temporal	consistency	and	minimize	effect	caused	by	saturated	NTL	pixels	(Small	&	Elvidge,	2013).	Second,	I	spatially	delineated	and	track	urbanization	patterns	using	the	Theil-Sen	estimator	for	25	urban	environments	across	the	pan	Pacific	region.	Then,	I	examined	the	causality	of	two	common	socio-economic	variables	(population	and	GDP)	on	NTL	using	panel	Granger	causality	procedures.	This	approach	allows	a	statistical	verification	of	the	possible	drivers	of	urban	development,	by	examining	the	effect	of	exogenous	macro-level	socio-economic	factors	on	physical	city	growth.	This	chapter	demonstrates	new	ways	of	investigating	relationships	between	NTL	data	and	socio-economic	development.			Cities	have	multi-faceted	definitions,	including	the	permanent	areas	of	heavily	human-induced	infrastructure	and	the	socio-economic	entities	that	facilitate	industrial	development	and	population	growth	(Lo	&	Marcotullio,	2000a;	Montgomery,	2008).	City	growth,	commonly	known	as	urbanization,	is	thus	the	interplay	between	its	physical	and	socio-economic	environments.	Reliable	assessment	and	quantification	of	urbanization	is	critical	to	better	allocate	resources	and	optimize	developing	efficiency.		Sustainable	city	growths	rely	primarily	on	reliable	and	consistent	measurements	of	urbanization.	The	key	metrics	that	have	been	utilized	to	examine	such	activity	and	its	associated	variations	fall	into	two	main	categories.		First	are	demographic	metrics	such	as	births	and	deaths,	immigration	and	emigration,	leading	to	estimates	of	population	size	and	density.	The	second	set	of	metrics	are	associated	with	the	wealth	of	a	city	such	as	regional	gross	domestic	product	(GDP).	These	data	can	be	acquired	in	a	number	of	ways.	Population	data	are	often	recorded	through	censuses	where	the	resident	population	is	polled	locally	using	forms	and	interviews.	Alternatively,	economic	data	are	most	often	complied	directly	by	state	government	or	local	administrative	units.		Census	and	economic	data	are	often	in	tabular	format	with	limited	value	for	monitoring	spatially	explicit	changes	that	are	much	needed	in	urban	studies	(Jensen	&	Cowen,	1999).	It	was	not	until	the	1970s	that	remote	sensing	satellite	imagery	become	an	alternative	data	source	for	monitoring	city			58	growth	in	a	more	repeatable	and	comprehensive	manner	and	offers	a	much	richer	source	of	information	than	conventional	survey	data	(Schneider,	Friedl,	&	Potere,	2009).	However,	most	urban	remote	sensing	applications	mainly	focused	on	extracting	physical	features	such	as	delineating	city	boundaries	(Henderson	et	al.,	2003)	or	mapping	and	quantifying	land-cover	changes	(Venter	et	al.,	2016).	Characterizing	the	socio-economic	nature	of	cities	has	still	primarily	remained	the	domain	of	census	data.		Early	studies	such	as	(Welch,	1980)	uses	lights	to	model	urban	population	and	energy,	while	Croft	(1973)	uses	the	nighttime	space	photographs	to	map	burning	waste	in	oil	fields.	More	recently,	digital	NTL	data	have	been	increasingly	used	on	mapping	urban	and	urbanization	related	human	activities	such	as	delineating	urban	expansion	(Small	et	al.,	2005),	modelling	economic	activities	(Ebener	et	al.,	2005),	and	CO2	emissions	(Ghosh	et	al.,	2010).	Yet,	the	interpretation	of	NTL	brightness,	also	known	as	digital	number	values	(DN),	is	highly	subjective	and	varies	from	study	to	study	(Donaldson	&	Storeygard,	2016).	In	this	chapter,	I	interpreted	NTL	values	in	a	more	general	fashion	to	represent	overall	human	activities.	Thus,	I	assume	that	an	increasing	NTL	value	is	indicative	of	growing	human	activities	rather	than	one	specific	variable	in	previous	studies.		Much	of	the	existing	research	has	shown	encouraging	results	correlating	NTL	with	other	ancillary	variables	such	as	GDP	and	population	size.	However,	few	have	investigated	the	causal	interaction	between	NTL	and	socio-economic	development.	Analyzing	the	causal	relationships	between	NTL	and	socio-economic	variables	can	be	more	valuable	than	traditional	correlation	approaches	for	understanding	the	drivers	of	city	growth,	and	prioritizing	long-term	policy	drafting	and	practical	urban	planning.			5.2. Materials	and	methods		In	1992,	the	first	digital	NTL	acquired	by	the	Defense	Meteorological	Satellite	Program's	Operational	Linescan	System	(DMDP/OLS)	was	released	by	NOAA’s	National	Geographical	Data	Center	(NGDC).	NTL	has	been	used	extensively	to	track	urban	activity	and	its	associated	temporal	characteristics,	enabling	researchers	and	urban	planners	to	quantitatively	compare	and	contrast	spatio-temporal	patterns.	The	full	NTL	temporal	record	enables	us	to	chronicle	the	development	of	urban	patterns	and	produce	spatially	explicit	estimates	that	reflect	a	city’s	growth	or	decline	(Ma	et	al.,	2012).	However,	in	most	years,	DMSP-OLS	operates	a	dual-sensor	system,	meaning	that	there	are	two	sensors	recording			59	spontaneously.	Thus,	an	intercalibration	was	needed	(section	5.2.1)	to	build	a	robust	time	series	of	NTL	(section	5.2.2).	Using	a	panel	version	of	Granger	causality	test	(section	5.2.3),	census	data	were	then	compared	against	NTL	time	series.			5.2.1. Intercalibrate	nighttime	lights	time	series		Annual	average	visible	cloud-free	nighttime	lights	composites	(Version	4)	were	acquired	from	NOAA	(http://ngdc.noaa.gov/eog/dmsp.html)	covering	1992	-	2013.	Images	were	formatted	as	Digital	Numbers	(DNs)	ranging	from	0	to	63	with	a	higher	DN	representing	higher	illumination	or	brightness	of	lights.		Due	to	the	lack	of	an	onboard	calibration	mechanism,	robust	intercalibration	is	a	critical	step	to	allow	images	from	different	years	or	sensors	to	be	directly	comparable.	Recently,	Pandey,	Zhang,	&	Seto	(2017)	quantitatively	evaluated	nine	most	commonly	used	intercalibration	techniques	using	a	Summed	Normalized	Difference	Index	(SNDI,	Equation	1).		Similar	to	Zhang	et	al.	(2016)	and	Elvidge	et	al.	(2014)	I	built	a	3rd	degree	polynomial	model	to	calibrate	each	image	to	a	reference	year.	A	reference	year	was	selected	based	on	maximal	DN	values	across	the	selected	cites,	an	approach	which	has	been	used	previously	(Bennie	et	al.,	2015;	C.	Elvidge	et	al.,	2014;	Li	et	al.,	2013;	Z.	Liu,	He,	Zhang,	Huang,	&	Yang,	2012;	Pandey,	Joshi,	&	Seto,	2013).	Rather	than	using	one	single	model	for	all	cities,	I	fitted	a	polynomial	model	for	each	individual	city	to	account	for	local	NTL	variations.		I	evaluated	the	calibration	results	for	each	city	using	SNDI,	which	quantified	the	level	of	convergence	in	NTL	temporal	series	of	a	given	city.	SNDI	is	the	total	of	Normalized	Difference	Index	(NDI,	Equation	2)	which	assessed	the	absolute	difference	of	total	DN	values	(TDN,	Equation	3)	between	two	sensors	in	the	same	year	between	two	different	sensors.	As	suggested	by	Zhang	et	al.	(2016)	and	Pandey	et	al.	(2017)	an	effective	intercalibration	should	yield	a	much	lower	SNDI	than	the	raw	images.	My	intercalibration	SNDI	was	then	compared	against	raw	data,	Zhang	et	al.	(2016),	and	Elvidge	et	al.	(2014).							60	pqrs = qrst																6789:;<=	1ooNno 	qrst = urqot − urq_turqot + urq_t 							6789:;<=	2		urq = rq	N																					6789:;<=	3					XNno 	:	v	(1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007)			A	challenge	associated	with	NTL	data	was	a	pixel	saturation	issue	which	can	occur	due	to	the	limited	radiometric	range	of	NTL	sensors.	Recently	Zhang	et	al.	(2013)	incorporated	a	series	of	vegetation	images	to	desaturate	NTL	data	on	the	assumption	that	there	is	an	inverse	relationship	between	vegetation	abundance	and	NTL	brightness.	However,	since	our	goal	of	this	chapter	was	to	investigate	the	casual	relationship	between	NTL,	GDP	and	population,	the	inclusion	of	another	input	variable	(e.g.	vegetation)	would	complicate	the	process	of	interpreting	statistic	analysis.	In	addition,	Zhang	et	al.	(2013)	suggested	a	limited	improvement	of	NTL	variability	for	fast	growing	cities	compared	to	more	established	legacy	cities.	The	inverse	relationship	between	NTL	and	vegetation	may	not	hold	for	developing	cities	in	this	chapter.		As	a	result,	NTL	images	used	in	this	chapter	were	calibrated	but	not	alerted	to	accommodate	potential	saturation	issues.			5.2.2. Generate	NTL	temporal	trend		As	indicated	by	previous	studies,	a	“lit”	pixel	did	not	necessarily	coincide	with	human	activities	due	to	the	potential	“blooming”	effect	caused	by	diffused	or	scattered	light	from	neighboring	pixels	(Small	&	Elvidge,	2013).	I	therefore	used	a	threshold	of	DN	=	12	as	a	threshold	between	lit	and	non-lit	or	dimed	pixels	(Small	et	al.	2011).	I	then	generated	NTL	trends	for	all	25	urban	environments	over	the	21	years.		In	order	to	capture	any	development	in	initially	low-lit	area,	NTL	trends	were	generated	for	all	non-water	pixels,	including	the	ones	with	a	DN	value	below	12.	A	Mann-Kendall	non-parametric	test	(Mann,	1945)	was	used	to	determine	the	significance	of	the	monotonic	trend	in	NTL.	The	TS	estimator,	which	has	been	widely	used	with	time	series	data	(Hansen,	M.	C.,	Roy,	D.	P.,	Lindquist,	E.,	Adusei,	B.,	Justice,	C.	O.,	Alstatt,	2008)	to	describe	temporal	change	in	intensity,	was	applied	to	pixels	identified	by	Mann-		61	Kendall	as	statistically	significant	(p	<	0.05).	Those	pixels	were	then	used	to	calculate	the	trend	slope	values	based	on	the	median	of	pairwise	data	points	from	1992	to	2013.		Based	on	the	slope	values,	I	then	grouped	the	25	cities	into	two	classes.	The	first	class	contained	cities	which	have	experienced	rapid	NTL	growth	over	the	21-year	period.	The	second	class	represented	cities	with	much	lower	or	no	slope	in	trend	in	NTL,	indicative	of	little	urban	growth	over	the	time.	Specifically,	cities	in	rapid	NTL	growth	group	needed	to	have	at	least	20%	of	pixels	experience	significant	changes.	In	addition	to	the	slope	values	I	also	examined	the	NTL	with	a	predefined	threshold	(e.g.	DN=12)	to	determine	for	which	year	a	given	pixel	exceeded	the	threshold	value,	indicating	the	year	urban	establishment	in	that	pixel	passed	the	brightness	threshold.		5.2.3. Granger	causality	test		Although	NTL	has	been	extensively	used	as	a	proxy	to	anthropogenic	activities,	many	econometric	theories	and	tools	have	rarely	been	applied	with	the	NTL	time	series.	The	most	notable	challenge	is	that	econometric	tools	often	require	decadal	or	even	centurial	time	series	as	input	in	order	to	capture	the	often	weak	relationship	between	two	given	economic	variables	(Levin,	Lin,	&	Chu,	2002).	NTL	imagery	collected	by	DMSP/OSL	has	a	relatively	short	time	span	(i.e.	1992-2013)	and	therefore	are	often	not	well	suited	to	econometric	tools.	Recent	studies	(Hsiao,	2007)	however	have	shown	encouraging	results	for	utilizing	relatively	short	time	series	for	causation	testing	through	panel	data	that	are	a	collection	of	entities	(e.g.	cities)	where	the	variables	are	observed	across	time.		Statistically,	the	Granger	causality	test	(Granger,	1969)	describes	the	strength	of	association	between	two	time	series	by	testing	whether	or	not	the	inclusion	of	one	time	series	(xt)	can	improve	the	forecasting	of	future	values	in	another	time	series	(yt).	If	the	addition	of	xt	significantly	improves	a	model’s	explanatory	power	in	predicting	yt,	I	could	conclude	that	xt	“granger	causes”	yt.	In	this	chapter,	the	panel	version	of	the	Granger	causality	test	combined	individual	short-time	series	data	in	a	form	of	cross-sectional	structures	that	increased	the	test	efficiency	and	power	by	raising	the	number	of	observations	and	degrees	of	freedom	(Hoffmann,	Lee,	Ramasamy,	&	Yeung,	2005).	In	this	chapter,	a	total	of	three	panel	data	sets	were	generated	for	causality	test,	namely,	total	DN	(TDN),	total	population	(TPOP_Total),	and	total	GDP	(TGDP_Total)	for	each	of	the	25	cities.	As	a	result,	each	panel	data	set	had	a	total	of	25	cross-sections	(N=25	cities)	and	22	temporal	units	(T=22	years).	Total	DN	was			62	calculated	as	the	sum	of	DN	values	of	all	lit	pixels	for	each	year	to	represent	both	the	area	and	intensity	of	the	NTL.	I	also	examined	the	differences	between	cities	which	are	rapidly	developing	(N=13)	versus	those	which	are	more	established	(N=12).	Granger	causality	tests	required	all	panel	data	to	be	stationary	and	co-integrated	based	on	two	panel	unit	root	tests;	the	Levin	&	Lin	(Levin	et	al.,	2002),	known	hereafter	as	the	LLC	test	and	Im	&	Pesaran	(2003),	known	hereafter	as	IPS.	The	panel	co-integration	test	of	Johansen	(Johansen,	1988)	was	applied	to	examine	co-integration	among	all	pairs	of	temporal	variables	(test	results	in	Appendix	4).					The	rejection	of	Granger	causality	tests	H0	(p	<	0.01)	indicated	a	unidirectional	causal	relationship	from	one	input	variable	to	the	other.	I	employed	the	panel	Granger	causality	test	proposed	by	Dumitrescu	&	Hurlin	(2012),	thereafter	DH,	which	respected	the	heterogeneity	within	relatively	small	panel	data	sets.		5.3. Results			5.3.1. Intercalibrating	NTL	time	series		Overall,	all	calibration	methods	successfully	reduced	the	systematic	biases	in	the	NTL	images	with	a	lower	SNDI	than	for	the	raw	data	across	most	of	the	cities	(Figure	5.1).		Although	Zhang	et	al.	(2016)	and	Elvidge	et	al.	(2014)	yield	lower	SNDI	at	the	global	scale,	our	city	level	calibration	shows	a	marginally	better	calibration	result	in	terms	of	minimizing	systematic	biases.	Haikou	(HAK),	Nanchang	(NCX),	and	Vancouver	(VAN)	all	have	a	relatively	higher	SNDI	value	compared	to	the	other	cities	tested.			63			5.3.2. Quantifying	spatio-temporal	changes		Large	inter-	and	intra-city	variations	were	apparent.	For	example,	in	Denver	(DEN),	steeper	slopes	were	clustered	in	the	north	and	east	side	of	the	city	while	in	Kuala	Lumpur	(KUL),	intensive	NTL	changes	were	located	in	the	south	(Figure	5.2).	The	majority	of	pixels	with	rapid	change	were	found	in	less	developed	cities	(e.g.	HAR)	while	more	developed	cities	exhibited	more	stable	NTL	trends	(e.g.	CAL).	Variations	occurring	within	the	same	city	also	clearly	exposed	NTL	change	hotspots	and	the	growth	of	surrounding	satellite	cities	during	the	study	period	(e.g.	BAK,	SHH).				00.20.40.60.811.21.41.61.8bakcalcsxdaldenedm fuzhakharhksz kullavman melmex ncxphxseaselshh sin syd tjn tkovanSNDIRaw Zhang Elvidge LuFigure	5.1	Sum	of	normalized	difference	index	(SNDI)	derived	from	raw	image,	Zhang	et	al.,	(2016),	Elvidge	et	al.,	(2014),	and	Lu	(this	chapter).			64		Spatially,	the	recent	urban	development	generally	occurred	on	the	outer	rings	of	each	urban	area	(Figure	5.3;	e.g.	FUZ	and	CSX).	Timing	of	urban	development	was	also	variable	with	cities	such	as	Seoul	and	Kalua	Lumpur	dominated	by	land-cover	changes	in	the	early	stage	of	the	time	series	while	changes	in	Changsha	and	Dalian	were	relatively	more	recent.		Figure	5.2	NTL	change	rate	represented	by	Theil-Sen	slope	values	showing	the	rate	of	change	from	1992	to	2013.	Water	is	colored	as	white.	Cities	were	grouped	based	on	its	growth	intensity.	Left	panel	contains	cities	with	fast	and	more	dynamic	urban	growth	while	the	right	panel	include	cities	with	more	stable	and	less	development.			65		I	observed	a	wide	range	of	variation	within	and	across	all	25	cities	in	urban	development	(Figure	5.4).	For	example,	Tokyo	(TKO)	and	Hong	Kong	(HKSZ)	have	over	75%	of	land	urbanized	prior	to	1992	while	most	cities	in	China	had	less	than	10%.	Cities	such	as	Shanghai	(SHH)	and	Tianjin	(TJN)	experienced	substantial	growth	over	the	period	studied	with	nearly	50%	of	the	land	crossing	the	pre-defined	threshold	value.	A	few	cities,	however,	had	less	growth	with	approximately	75%	of	land	remaining	undeveloped.			Figure	5.3	.	The	year	when	a	given	pixel	within	each	urban	environment	exceeded	the	pre-defined	DN	value.	Dark	grey	pixels	represent	existing	urban	areas	prior	to	1992	while	light	grey	indicating	areas	with	no	sufficient	light	sources	in	2013.			66		Figure	5.4	Urban	land	breakdown	of	changed,	undeveloped,	and	existing	urban	areas,	ranking	the	25	cities	from	the	highest	proportion	of	lit	pixels	(i.e.	TKO)	to	the	least	(i.e.	HAK).		5.3.3. Granger	causality	test		The	Causality	test	results	differ	depending	on	the	cities	analyzed	(Figure	5.5).		Expectedly,	across	all	cities,	both	population	and	GDP	played	a	major	role	in	directing	changes	of	NTL.	Additionally	GDP	and	NTL	also	“granger	caused”	changes	in	population	(p	<	0.01,	Figure	5.5a).	This	implied	that	the	brightness	of	cities	follows	increases	in	both	population	and	GDP	equally	and	that	neither	population	nor	GDP	alone	is	responsible	for	increasing	the	NTL.	Unexpectedly,	the	test	also	suggested	GDP	and	NTL	“granger	cause”	population	growth	suggesting	that	population	change	was	the	outcome	rather	than	the	cause	of	urban	development	(Figure	5a).		Stratifying	the	cities	by	development	stage	I	found	contrasting	and	unexpected	results.	In	the	case	of	more	established	cities	with	few	NTL	changes	over	the	analysis	period,	the	causal	relationship	from	NTL	to	population	was	no	longer	significant	yet	changes	in	population	“granger	caused”	both	GDP	and	NTL	(Figure	5.5b).	This	suggested	that	in	cities	with	relatively	stable	NTL,	population	and	GDP	were	likely	the	key	driver	of	local	economic	and	urban	development	but	not	the	other	way	around.			67	For	fast	changing	and	more	dynamic	cities,	there	were	only	two	significant	casual	relationships	–	growth	in	NTL	and	GDP	leading	to	an	increase	in	population.		Unexpectedly,	there	was	no	significant	causal	association	between	GDP	and	NTL	(dash	lines	in	Figure	5.5c).	This	suggested	that	in	rapidly	changing	cities	population	increases	were	driven	by	brighter	and	more	economically	active	urbanization.		5.4. Discussion			5.4.1. NTL	calibration			The	calibration	method	in	this	chapter	successfully	minimized	the	systematic	biases	at	the	city	scale,	enabling	direct	comparisons	among	images	taken	by	different	sensors	(Figure	5.1).	Pandey	et	al.	(2017)	suggested	that	a	global	calibration	(i.e.	national	level)	could	outperform	regional	models	which	does	not	appear	to	be	the	case	in	this	study.	I	found	that	intercalibrated	models	at	city	scale	achieved	relatively	lower	SNDI	across	all	25	cities	than	using	calibration	parameters	from	previous	studies	(Figure	5.1).	One	rationale	was	that	the	majority	of	the	pixels	used	in	the	calibration	process	were	brightly	lit	(i.e.	pixels	Figure	5.5	Causal	interactions	among	NTL	(nighttime	lights),	POP	(population	size),	and	GDP	(Gross	Domestic	Product)	of	a)	all	cities,	b)	established	cities,	and	c)	dynamic	cities.	A	solid	line	represents	a	statistically	significant	causal	relationship	while	a	dotted	line	indicates	no	significant	causality.		Arrow	head	indicates	the	direction	of	causal	relationship	and	a	double-headed	arrow	represents	a	bi-directional	causal	relationship.			68	located	in	urban	area)	which	had	a	much	higher	contribution	to	the	overall	SNDI	statistics	than	dimly	lit	pixels	(Pandey	et	al.,	2017).	It	was	also	noticeable	that	the	calibration	performance	varies	across	cites.	Images	with	large	portion	of	dimly	lit	pixels	were	more	likely	to	suffer	from	less	optimal	calibration	due	to	existence	of	random	noises	and	the	skewed	radiometric	DN	values.	Island	cities	or	cities	surrounded	by	large	green	spaces	may	have	a	relatively	less	even	distribution	of	DN	values,	which	may	explain	our	inconsistent	calibration	performance	in	Figure	5.1.	Cities	with	higher	SNDI	values	were	either	located	in	developing	regions	(e.g.	Changsha	and	Haikou)	or	cities	with	higher	cover	of	vegetation	cover	(e.g.	Vancouver).	Generally,	those	cities	had	fewer	brightly	lit	pixels	than	cities	such	as	Tokyo.	Therefore,	I	concluded	that	a	locally	fitted	intercalibration	model	will	likely	work	better	in	areas	dominated	by	high	DN	pixels.			5.4.2. Is	60-km	buffer	ideal		It	was	unsurprising	to	see	that	cities	with	more	dynamic	and	fast	changing	rates	were	located	in	Asia.	According	to	United	Nation’s	review	in	2001,	on	average,	Asian	cities	were	at	least	50	years	behind	Europe	and	North	America	in	terms	of	urbanization	level	(United	Nations,	2002).	Mega-cities	in	Asia	on	the	other	hand	showed	highly	dominating	and	disproportional	impact	on	regional	and	national	economic	development.	Studies	(Jones,	2002)	have	suggested	that	urban	dwellers	have	an	overall	better	living	standard	such	as	education	and	consumption	level,	hence	attracting	substantial	amount	of	in-migration	from	rural	areas.	Understanding	the	spatial	pattern	and	the	timing	of	urban	development	in	those	fast	changing	cities	could	offer	valuable	information	on	efficient	land	resources	allocation	which	can	further	reduce	the	per	capita	cost	of	infrastructure	and	basic	services	(Cohen,	2006).	In	contrast,	urbanization	tended	to	be	less	concentrated	in	more	developed	cities	due	to	their	advanced	urban	network	(Cohen,	2006).	As	a	result	I	noticed	that	while	a	60-km	radius	buffer	was	sufficient	for	fast	changing	and	more	dynamic	cities,	it	was	clearly	not	large	enough	to	capture	recent	urbanization	activities	in	more	developed	cities	(Figure	5.2	&	5.3).	Other	alternatives	such	as	algorithmically	derived	urban	extent	have	been	used	in	previous	literatures,	focusing	on	primarily	tracking	urban	land-cover	and	land	use	over	time.	This	approach	however,	still	required	a	fixed	boundary	to	define	where	the	city	ends.		Although	this	chapter	used	a	ground	distance	of	60	km	to	delineate	urban	boundary,	other	distance	measuring	approaches	such	as	travel	time	ratio	(Dijst	&	Vidakovic,	2000)	or	Manhattan	distance	(Apparicio,	Abdelmajid,	Riva,	&	Shearmur,	2008)	may	affect	the	casualty	tests.	One	limitation	regarding			69	to	the	boundaries	of	selected	cities	was	the	spatial	scale	difference	between	remote	sensing	data	and	census	record	which	was	often	collected	using	administrative	units.	The	scale	inconsistency	between	these	two	data	sets	may	alter	the	final	results.	Yet,	since	the	census	data	used	in	this	chapter	represented	metropolitan	area	which	in	general	covers	a	relative	larger	area	than	normal	administrative	units,	the	results	were	still	able	to	offer	valuable	insights	on	decoupling	the	relationship	between	NTL	and	socio-economic	development.		5.4.3. Chicken	or	the	egg:	causal	relationship	between	NTL	and	socio-economic	factors		A	large	number	of	econometric	studies	have	reported	inconsistent	results	when	examining	interactions	between	socio-economic	and	environmental	variables.	Mozumder	and	Marathe	(2007)	summarized	a	number	of	studies	and	found	mixed	causal	relationship	results	depending	on	the	study	location,	types	of	variables,	and	duration	of	time	series	used.	Knapp	&	Mookerjee	(1996)	tested	the	underlying	interaction	between	population	growth	and	global	CO2	and	concluded	a	weak	long-term	equilibrium	but	strong	short-term	relationships	between	population	and	CO2.	Seto	&	Kaufmann	(2003)	also	employed	panel	causality	procedures	with	remotely	sensed	images	to	estimate	the	economic	drivers	of	land	conversion	in	urban	areas	and	concluded	that	investment	in	capital	construction	is	driving	urban	land	conversion.		It	has	long	been	thought	that	population	was	the	primary	driver	of	urban	brightness	while	economic	development	was	rather	a	form	of	outcome	of	urbanization.	In	this	chapter,	I	found	that	population	and	GDP	revealed	contrasting	effects	on	NTL	trends	between	stable	and	more	dynamic	cities.	Statistically,	changes	in	NTL	were	significantly	driven	by	both	population	and	GDP	growth	in	more	established,	slow	changing	cities.	Previous	work	(Dietz,	Rosa,	&	York,	2007;	Satterthwaite,	2009)	has	indicated	that	rather	than	growing	population	alone,	it	was	the	high	consumption	lifestyle,	economic	and	political	decisions	that	lead	to	urban	growth.	My	results	showed	that	in	more	developed	cities,	it	was	in	fact	both	the	population	and	economic	development	that	drives	urban	NTL	changes.		However,	in	fast	changing,	yet	often	less	developed	cities,	the	growth	of	NTL	brightness	and	GDP	were	driving	population	changes	rather	than	the	other	way	around.	In	those	cities,	a	major	source	of	population	increase	was	through	large	in-migration	from	rural	and	neighboring	areas	and	involves	densification	and	conversion	of	existing	farm,	forest	or	barren	land	to	urban	land-cover	types	(Jones,			70	2002).	My	results	implied	that	migration	was	more	attracted	to	cities	with	promising	economic	conditions	and	undergoing	fast	urbanization	paces.	In	the	following	chapter,	vegetation	fraction	(chapter	3)	and	NTL	time	series	(chapter	4-5)	were	compared	against	each	other	using	three	candidate	relationships,	namely,	linear,	quadratic,	and	cubic	models,	in	order	to	confirm	the	Environmental	Kuznets	Curve	(EKC)	theory.																						71	Chapter	6		6. Testing	EKC	theory:	What	is	the	relationship	between	urban	vegetation	and	nighttime	brightness	across	pan	Pacific	cites?			6.1. Introduction			My	goal	in	this	chapter	is	to	evaluate	the	relationship	between	human	development	and	environmental	quality	within,	and	across,	cities	at	a	pixel	level	using	results	from	previous	chapters.	I	used	vegetation	fraction	value	(Chapter	3)	to	represent	urban	vegetation	cover	while	NTL	time	series	(Chapter	4)	was	used	as	a	proxy	to	urban	economic	development.	Cities	worldwide	strive	to	grow	not	only	economically	strong	but	also	environmentally	sustainable.	Understanding	relationships	between	urban	activity	and	the	environment	has	critical	implications	for	achieving	long-term	sustainability	and	evaluating	policy	decisions	(Panayotou,	1997).	Declines	in	environment	conditions	such	as	the	loss	of	vegetation	(Chapter	3),	highlight	the	need	for	stricter	regulations	and	government	efforts	that	prioritize	environmental	recovery	over	economic	development.	Conversely,	improvements	in	the	environmental	conditions	within	cities	suggest	sustainable	and	planned	development.		Despite	the	fact	that	the	relationship	between	urban	economic	development	and	the	impact	on	environment	has	been	widely	debated	(Schaltegger	&	Synnestvedt,	2002),	there	is	no	consistent	framework	from	both	theoretical	and	empirical	studies	that	provide	suggestions	as	to	how	cities	can	develop	in	a	sustainable	and	environmentally	friendly	manner.	Many	studies	conclude	an	irreversible	and	monotonic	impact	on	the	environment	brought	about	by	human	development	primarily	as	a	consequence	of	economic	activities	(Akbostanci,	Türüt-Aşik,	&	Tunç,	2009;	Stern,	2004).	High	rates	of	economic	activity	such	as	extraction	of	natural	resources	typically	produce	large	quantities	of	waste	and	pollution	(Hoornweg,	Bhada-Tata,	&	Kennedy,	2013).	For	example,	Fodha	&	Zaghdoud	(2010)	found	a	linear	trend	between	per	capita	CO2	emissions	versus	per	capita	GDP	in	Tunisia.	Similarly,	Akbostanci,	Türüt-Aşik,	&	Tunç	(2009)	showed	that	per	capita	CO2	emissions	increased	monotonically	with	per	capita	income	in	Turkey.				72	Others	argue	that	the	negative	impact	of	human	development	in	cities	decreases	over	time	in	response	to	more	efficient	production	(Bartlett,	1994;	Beckerman,	1992).	These	studies	suggested	that	urban	environments	ultimately	reduce	the	impact	of	environmental	degradation	through	increased	wealth	and	education.	For	example,	Panayotou	(1997)	found	across	30	countries	the	level	of	ambient	SO2	(a	key	indicator	of	pollution)	decreased	as	GDP	per	capita	increased.	A	more	recent	study	(Aldy,	2005)	investigated	the	relationship	between	the	consumption	of	CO2	per	capita	versus	income	per	capita	in	the	United	States,	and	suggested	a	similar	trend	where	CO2	levels	reduced	with	increasing	income.		Schaltegger	&	Synnestvedt	(2002)	suggested	one	hypothesis	for	the	often	conflicting	conclusions	of	these	previous	studies.	First,	a	lack	of	consistent	and	compatible	data,	with	which	to	assess	both	environmental	degradation	and	economic	activity,	largely	complicates	the	process	of	comparing	cities	across	time	and	space.	Common	datasets	on	economic	development	include	statistics	on	local	Gross	Domestic	Production	(GDP)	and	income	level.	However,	these	were	commonly	compiled	by	local	census	departments,	which	often	do	not	share	similar	scales	and	resolutions	in	terms	of	temporal	updates	and	spatial	details.	Various	measurements	of	airborne	and	water	pollutions	were	frequently	used	as	the	basis	of	environmental	performance	(Grossman	&	Krueger,	1991;	Selden	&	Song,	1994;	Shafik,	1994).	However,	using	pollution	emissions	as	an	indictor	of	environmental	performance	received	criticism	(Stern,	1998).	For	example,	the	Hecksher-Ohlin	hypothesis	(Heckscher,	1919;	Ohlin,	1952)	suggested	that	developing	countries	are	more	likely	to	produce	goods	that	are	labour-	and	resources-intensive	given	their	inexpensive	and	abundant	resources.	More	developed	countries	tend	to	focus	on	tertiary	goods	and	services	and	human	capital	development	with	a	relatively	small	environmental	footprint.	Additionally,	it	is	difficult	to	compare	spatially	and	temporally	inconsistent	measurements	over	time	that	are	exclusively	recorded	for	a	specific	production	or	industry	type	and	are	not	representative	of	other	environmental	issues	over	the	entire	cityscape	(Cadenasso,	Pickett,	&	Schwarz,	2007).		Environmental	Kuznets	Curve	(EKC)	theory	hypothesizes	a	U-shaped	relationship	where	environmental	performance	decreases	at	the	early	stage	of	economic	development	and	recovers	as	the	economy	reaches	a	certain	turning	point	(Kuznets,	1955).	Indicators	such	as	income	levels	and	measurements	of	certain	pollutants	are	often	used	to	represent	economic	development	and	environmental	performance,	respectively.	Appling	EKC	based	approaches	at	both	inter-	and	intra-	city	scales	provides	insights	into	the	balance	between	human	activity	and	environmental	sustainability.	For	example,	at	the	urban	core,	downtown	or	city	centre	there	may	be	low	vegetation	cover	and	high	levels			73	of	human	development	and	density,	whereas	outer	suburban	areas	may	be	more	likely	to	exhibit	an	opposing	pattern	with	increased	vegetation	cover	and	lower	human	activities.	As	a	result,	the	relationship	between	human	development	and	environment	performance	is	not	a	simple	linear	decline	and	will	likely	vary	within	and	across	cities.	Indicators	of	both	the	economic	and	environmental	conditions	within	city	environments	range	from	simple	tabular	data	to	more	complex	spatially	explicit	predictions.	In	comparison	to	statistics	on	airborne	and	water	pollution,	or	industrial	output	that	have	commonly	been	used	for	EKC	analysis,	changes	in	urban	vegetation	cover	is	a	relatively	direct	measurement	of	the	environmental	conditions	within	a	city	with	urban	greenspace	being	shown	to	be	indicative	of	effective	environmental	management	and	government	regulation	(Zhao	et	al.,	2016).	Conversely	declining	urban	vegetation	cover	has	been	linked	to	intensification	of	human	development	and	increases	in	impervious	surfaces	(Quigley,	2002,	2004).	Conventionally,	information	on	urban	vegetation	cover	has	been	generated	from	land	use	or	land	use	maps,	both	of	which	represent	vegetation	cover	as	a	categorical	variable	(e.g.	vegetation	vs.	non-vegetation).	Vegetation	cover	can	also	be	effectively	observed	from	satellite	imagery	due	to	the	reflective	properties	of	foliage	in	the	visible	and	near	infrared	regions	of	electromagnetic	spectrum.	Changes	in	satellite-derived	nighttime	light	(NTL)	intensity	has	also	been	shown	to	be	a	good	indicator	of	economic	activity	particularly	in	cities	(Bennett	&	Smith,	2017;	Small	&	Elvidge,	2013;	Small	et	al.,	2005).	The	most	common	and	perhaps	the	earliest	interpretation	of	NTL	is	as	a	proxy	to	economic	development,	on	the	assumption	that	a	brighter	city	is	a	wealthier	city	(Doll,	Muller,	&	Morley,	2006).	The	main	benefits	of	NTL	compared	to	conventional	socio-economic	data	is	its	compatibility	and	consistency	over	time	and	its	capability	of	providing	reliable	estimates	especially	for	undocumented	regions	or	areas	with	poor	census	data.	Although	the	use	of	remote	sensing	data	has	been	demonstrated	in	urban	studies	(e.g.	Seto	&	Kaufmann	2003;	Zhang,	Pandey,	&	Seto	2016),	uncoupling	the	relationship	between	human	development	and	the	environment	has	not	been	fully	articulated.	There	is	also	a	need	to	examine	whether	trends	in	these	metrics	are	generalizable	across	and	within	different	cities.		In	this	chapter,	I	statistically	tested	the	relationship	between	economic	and	environmental	conditions	on	fine	spatial	detail	(30-meter	spatial	resolution)	of	25	cities	across	the	pan	Pacific	region	from	1992	to	2012.				74	6.2. Materials	and	methods		Two	sets	of	variables	were	needed	for	testing	the	EKC	theory.	Firstly,	intercalibrated	NTL	time	series	(section	6.2.1)	were	used	as	a	proxy	for	urban	economic	development	(x-axis	on	EKC).	Secondly,	vegetation	fraction	values	(section	6.2.2)	were	used	as	an	indicator	for	environmental	conditions	(y-axis	on	EKC).	A	pixel-based	model	fitting	procedure	was	used	to	find	the	optimal	relationship	based	on	Akaike	Information	Criteria	(section	6.2.3).	Parameters	of	the	optimal	model	were	then	used	to	determine	the	directionality	and	magnitude	of	the	relationship	between	VF	and	NTL	(section	6.2.4).	Lastly,	a	join-count	statistics	was	used	to	examine	the	spatial	autocorrelation	of	each	individual	models	(section	6.2.5).			6.2.1. Human	development	and	economic	indicator—Nighttime	lights			I	used	NTL	images	acquired	by	the	Defense	Meteorological	Satellite	Program	(DMSP)	from	1992	to	2012	(https://ngdc.noaa.gov/eog/viirs/download_dnb_composites.html).	Cloud-free	NTL	composites	measured	the	brightness	level	of	the	target	and	have	been	widely	used	to	represent	human	activities.	In	order	to	capture	changes	in	the	relationship	between	NTL	and	vegetation	cover,	I	excluded	pixels	that	I	re	already	bright	at	the	beginning	of	the	time	series	(i.e.	1992)	on	the	assumption	that	they	would	not	undergo	any	more	significant	brightness	changes	for	the	analysis	period.	To	be	defined	as	a	bright	pixel,	it	had	to	exceed	the	95-brightness	percentile	of	the	entire	time	series	of	a	given	pixel.			6.2.2. Environment	condition	indicator	–	Vegetation	fraction		I	used	the	annual	vegetation	fraction	(VF)	images	derived	from	spectrally	unmixed	Landsat	composites	as	proxies	to	environmental	condition	for	the	25	cities	from	1992	and	2012.	VF	images	I	re	developed	using	spectral	unmixing	analysis	(SMA)	which	assumes	each	pixel	contains	multiple	pure	land-cover	materials,	known	as	endmembers,	and	decomposes	each	pixel	into	fractions	of	those	endmembers.	Chapter	4	developed	an	annual	VF	layer	at	a	30m	resolution	for	all	25	selected	cities	using	annual	Landsat	composites.					75	6.2.3. Model	fitting	Common	approaches	of	examining	EKC	theory	generally	utilise	two	sets	of	time	series	variables,	one	representing	economic	development	and	the	other	indicating	the	environmental	performance.	Various	forms	of	statistical	fits	have	assessed	the	validity	of	the	EKC	framework	(Figure	6.1)	and	three	common	relationships	from	previous	literature,	namely,	linear,	quadratic,	and	cubic	models.	I	then	attempted	to	identify	the	most	appropriate	model	for	each	qualified	pixel	within	each	city	by	comparing	those	three	relationships	(Figure	6.2).	In	theory,	a	linear	model	represents	monotonic	and	irreversible	relationship	while	a	quadric	model	is	likely	indicative	of	the	EKC	theory.	A	cubic	model	is	more	complicated	than	linear	and	quadratic	relationships,	suggesting	a	rather	more	modulating	relationship	between	vegetation	fraction	and	nighttime	lights.				Specifically,	I	applied	model-fittings	for	each	of	the	pixels	within	the	city	limits	using	the	three	models	and	selected	the	best	model	based	on	two	criteria.	First,	the	best	model	must	have	an	AIC	score	(Akaike	Information	Criteria)	that	is	at	least	two	units	lower	than	other	two	competing	models	(i.e.	∆AIC	≥	2)	Figure	6.1	An	Illustration	of	previous	studies	attempting	to	quantify	the	relationship	between	economic	indicators	versus	an	environmental	variable.	(Note	that	all	figures	have	been	simplified	to	only	highlight	the	relationship	between	the	corresponding	variables.			76	(Burnham	&	Anderson,	2003;	Gergel	et	al.,	2004).	The	second	criteria	relied	on	F	tests	(p	<	0.05)	to	eliminate	statistically	insignificant	models	that	may	score	a	low	AIC	value	(Gergel	et	al.,	2004).	A	water	mask	(Chapter	3)	was	applied	to	exclude	water	bodies	from	model	fitting	processing.					6.2.4. Directionality	and	magnitude		Once	the	best	model	form	was	identified,	I	extract	the	leading	coefficient	(i.e.	the	parameter	of	the	variable	with	the	highest	exponent)	and	its	associated	t	test.	For	each	pixel	with	a	statistical	(p	<	0.05)	trend,	I	then	determined	both	the	direction	and	magnitude	of	the	trend	(i.e.	a	positive	vs	negative	leading	coefficient).	A	positive	leading	coefficient	can	be	interpreted	as	a	growing	vegetation	with	a	booming	economic	development.			6.2.5. Join-count	statistics		In	order	to	examine	the	spatial	patterns	of	the	trends	within	individual	cities	I	computed	the	global	spatial	autocorrelation	index	using	join-count	statistics	(Cliff	&	Ord,	1970).	Join-count	statistics	are	widely	used	to	measure	spatial	association	for	categorical	data	(Getis	&	Ord,	1992).	This	chapter	used	Figure	6.2	Vegetation	fraction	(VF)	time	series	regresses	against	Nighttime	time	light	(NTL)	time	series	pixel	by	pixel.	Three	candidate	models	(linear,	quadratic,	and	cubic)	I	re	used	to	determine	the	most	appropriate	model	based	on	AIC	score	and	statistics	significance.	The	red	pixel	symbolizes	where	the	city	center	is	while	the	blue	pixel	represents	water	which	was	excluded	from	all	subsequent	model	fitting.					77	the	Queen	case	contiguity	to	define	neighbouring	focal	cells	and	three	possible	combinations	of	neighbouring	cell	components,	typically	known	as	BB	(black-black),	WW	(white-black),	and	BW	(black-white)	joins.	A	BB	join	indicated	that	the	neighbouring	cells	have	been	assigned	to	the	same	model	while	a	WW	join	indicated	that	none	of	the	adjacent	cells	had	a	statistically	significant	relationship.	The	BW	join	indicated	that	neighbouring	cells	have	been	assigned	different	trends.		Once	assigned	join-counts	I	re	examined	in	comparison	to	a	random	distribution	to	establish	if	any	trends	within	cities	were	clustered.	I	calculated	the	ratio	between	total	join	counts	for	each	individual	model	and	values	from	a	random	spatial	pattern	(refer	to	as	Rj/r	thereafter).	Theoretically,	with	a	higher	ratio,	I	would	expect	a	stronger	spatial	association.		Lastly	I	divided	the	classic	EKC	figure	into	4	quadrants	to	compare	and	contrast	differences	across	cities	through	time	representing	(i)	initial	development	with	rapid	environment	degradation,	(ii)	slowing	environmental	degradation	with	a	more	mature	economy,	(iii)	economic	prosperity	with	a	recovering	environment	and	finally	(iv)	a	prosperous	economy	and	sustainable	environment.		6.3. Results			6.3.1. Model	selection		Only	pixels	with	statistically	significant	trend	were	qualified	for	model	fitting.	Eight	cities	had	more	than	70%	of	their	area	consisting	of	statistically	significant	(p	<	0.05)	relationships.	More	than	half	of	the	cities	(i.e.	19	out	of	25)	had	over	50%	of	pixels	successfully	fitted	to	one	of	the	three	relationships	(i.e.	linear,	quadratic,	or	cubic).	Overall,	coastal	cities	(e.g.	Shenzhen-Hong	Kong,	Vancouver	etc.)	had	a	relatively	fewer	qualified	pixels	compared	to	inland	cities.		6.3.2. Goodness	of	fit	A	linear	relationship	indicates	a	monotonic	trend	between	VF	and	NTL	while	quadratic	or	cubic	models	were	indicative	of	at	least	one	directional	change	between	urban	VF	and	NTL	with	the	correlation	coefficient	providing	an	indication	of	goodness	of	fit	(Figure	6.3).	Overall,	despite	the	fact	that	linear	models	were	the	most	popular	candidate	relationship	(Figure	6.4),	they	explained	the	least	of	the	variation	between	NTL	and	VF	compared	to	quadratic	and	cubic	models.	Four	cities	from	China	(i.e.			78	Changsha,	Shenzhen,	Shanghai,	and	Tianjin)	exhibited	cubic	patterns	rather	than	quadratic.	Phoenix	and	Tokyo	had	almost	an	equal	proportion	of	linear	and	quadratic	models	comparing	to	the	rest	of	the	cities.	The	majority	of	Asian	cities	showed	a	dominating	pattern	of	cubic	relationships	when	compared	to	North	American	cities	(Figure	6.4).	Cities	from	high	income	countries	were	more	dominated	by	linear	models.	It	was	also	apparent	that	cities	located	in	tropical	and	temperate	climate	schemes	were	more	likely	to	have	more	quadratic	and	cubic	models	than	cities	from	continental	and	arid	climate	schemes.			6.3.3. Spatial	distribution	of	fitted	models		Pixel-based	model	fitting	allowed	visualization	and	investigation	of	the	spatial	distribution	of	all	fitted	models.	Cities	showed	varying	patterns	of	model	distribution.	For	example,	linear	relationships	dominated	the	urban	centers	of	Tokyo	and	Shenzhen	(Figure	6.5a-b)	while	in	Manila	and	Shanghai,	city	centers	were	characterised	by	more	quadratic	models	(Figure	6.5d-e).	Another	noticeable	difference	among	cities	were	the	varying	range	of	leading	coefficients	for	each	model.	Cities	such	as	Calgary	(Figure	Figure	6.3	Histograms	of	r2	values	for	each	selected	model.				79	6.5c)	had	a	narrower	range	of	coefficient	values	(i.e.	from	-0.31	to	0.27)	compared	to	cities	such	as	Shanghai	(i.e.	range	from	-3.23	to	2.79).		Although	clusters	of	pixels	within	each	city	were	apparent,	the	spatial	distribution	of	these	clusters	varied	from	city	to	city.	Figure	6.6	shows	how	strong	the	spatial	association	was	for	each	model	within	a	city.	Quadratic	relationships	tended	to	have	a	similar	or	slightly	lower	spatial	association	than	cubic	models	except	for	Vancouver	and	Tokyo.	Interestingly,	I	also	noticed	that	coastal	cities	had	an	overall	higher	spatial	association	than	inland	cities.			6.3.4. Locations	of	cities	on	an	EKC				I	found	no	cities	located	in	quadrant	i,	confirming	that	all	25	cities	have	likely	passed	the	stage	where	environmental	degradation	drastically	outpaces	economic	growth	(Figure	6.7).	Overall,	four	out	Figure	6.4	Percentage	of	pixels	for	each	selected	best	model	type	at	a	city	level.	Only	pixels	with	significant	changes	were	used	for	calculating	the	percentage	of	each	model.	Pixels	that	did	not	fit	any	of	the	three	functions	were	not	concluded.				80	of	six	cities	positioned	in	quadrant	ii	were	located	in	Asia,	with	the	exception	of	Las	Vegas	(lav)	and	Phoenix	(phx).	Interestingly,	other	cities	from	Asia	such	as	Shanghai,	Seoul,	and	Bangkok	have	been	grouped	with	cities	from	high	income	North	American	cities.	The	majority	of	quadrant	iv	cities	were	located	in	relatively	more	developed	regions	with	a	minimal	increment	on	brightness	but	noticeable	increases	in	vegetation.	Cities	that	experienced	noticeable	vegetation	fraction	decrease	were	mostly	from	Arid	or	Continental	climate	schemes.							Figure	6.5	The	top	panel	represents	the	best	model	selected	based	on	AIC	score	and	F-test	at	pixel	level.	Coefficients	of	leading	variables	for	each	selected	model	were	shown	in	the	bottom	panel.	The	sign	of	each	coefficient	determines	the	directionality	of	the	relationship	while	the	absolute	value	of	the	coefficient	indicates	the	magnitude	of	impact	of	NTL	on	VF.			81		Figure	6.7	Spatial	autocorrelation	represented	by	the	ratio	between	total	counts	of	joins	and	a	random	spatial	pattern	calculated	by	Join-count	statistics	for	each	model.	Theoretically,	with	a	higher	Rj/r,	I	would	expect	a	stronger	spatial	association.	Figure	6.6	Positions	of	the	examined	cities	on	the	EKC	curve.	It	classifies	the	25	cities	using	the	summed	value	of	NTL	and	VF	pixels	excluding	water.			82	6.4. Discussion			Questions	as	to	whether	EKC	trends	exist	were	examined	using	various	socio-economic	and	environmental	indicators	yet	with	a	variety	of	contradictory	conclusions.		I	examined	the	validity	of	the	EKC	hypothesis	at	both	inter	and	intra	urban	scales.	To	do	so	I	examined	the	relationship	between	human	development	and	the	environment	represented	by	nighttime	lights	(NTL)	and	vegetation	fraction	(VF)	by	fitting	a	variety	of	model	forms.			I	found	that	linear	models	were	the	most	dominating	relationship	across	the	wide	range	of	examined	cities,	particularly	in	high-income	cities	from	North	America	and	Oceania.	On	the	other	hand,	the	pattern	was	less	obvious	for	middle	and	low-income	cities.	For	example,	the	city	of	Haikou	in	China	is	a	middle	high-income	city	but	it	was	also	dominated	by	linear	models.	In	terms	of	spatial-temporal	pattern,	I	found	that	in	general,	sub-urban	in	Asian	cities	experienced	more	temporal	changes	of	VF	and	NTL	comparing	to	high	income	cities.	I	also	noticed	dramatic	differences	of	spatial	autocorrelations	(Figure	6.6).	Across	the	pan	Pacific	cities,	there	was	little	evidence	of	quadratic	or	cubic	relationships	(Figure	6.4).	Although	quadratic	models	were	not	the	most	dominating	among	all	three	tested	models,	I	still	found	certain	hotspots	or	clusters	of	areas	displaying	a	typical	EKC	theory.	These	hotspots	of	quadratic	models	also	demonstrated	a	stronger	spatial	association	comparing	to	other	two	candidate	models	(i.e.	Tokyo,	Vancouver).	Together,	these	two	results	indicated	that	the	hypothesized	EKC	likely	exists	within	city	confines	and	is	likely	highly	clustered.	For	example,	in	spite	of	a	general	vegetation	decrease	in	Shanghai’s	metropolitan	area,	a	substantial	vegetation	increase	was	observed	on	Chongming	Island	located	to	the	north	of	Shanghai	city	centre.	The	integration	of	both	within,	and	across,	city	comparisons	furthers	the	discussion	of	the	existence	of	EKC	trends	and	avoids	generalizing	entire	cities	using	measurements	from	a	single	dataset.	Both	positive	and	negative	parameters	were	present	in	all	three	tested	model	by	examining	the	parameter	with	the	highest	order.	Overall,	our	findings,	using	vegetation	fraction	and	artificial	light	brightness	at	a	pixel	level,	suggest	that	at	least	in	certain	part	of	the	cityscape,	the	environment	recovers	with	a	growing	economy.		Clean	energy	supplies,	urban	greenspaces,	and	efficient	public	transportation	systems	are	being	aggressively	built	in	many	cities	worldwide.	Evidences	suggest	that	wealthier	and	more	developed	cities	place	a	higher	value	on	environmentally	conscious	policies.	For	example,	despite	its	strikingly	high	urban	density,	the	city	of	Hong	Kong	reserves	much	of	its	landmass	for	parks	and	nature	reserves	(Corlett,	1999;	Ng,	2010).	Since	1976,	the	government	of	Hong	Kong	enacted	the	Country	Park	Ordinance	to			83	initialize,	monitor,	and	manage	Hong	Kong’s	urban	parks	and	reserves.	In	2011,	approximately	40%	of	Hong	Kong	was	forested	(Ng,	2010).	There	were	a	few	potential	caveats	in	interpreting	the	patterns	that	I	found	in	this	chapter.	First	of	all,	even	with	correction	and	calibration,	the	well-known	saturation	issue	of	NTL	data	could	cause	total	brightness	values	to	plateau,	particularly	in	more	developed	high	income	cities	(Zhang,	Schaaf,	&	Seto,	2013).	Secondly,	the	prevalence	of	more	efficient	lighting	technology	could	potentially	lead	to	a	decrease	in	total	brightness	value	with	a	growing	economy,	explaining	the	minimal	increase	or	even	a	decrease	in	total	brightness.	However,	I	found	limited	literature	on	how	exactly	the	shift	in	lighting	source	affects	the	brightness	values	observed	by	VIIRS	sensors.	Thirdly,	according	to	previous	research,	cities	that	were	historically	green	were	more	likely	to	recover	from	environmental	degradation.	The	results	suggested	that	the	majority	of	vegetation	decreases	occurred	in	Arid	and	continental	climate	schemes	where	vegetation	grows	expectedly	slower	than	temperate	and	tropic	areas.	Much	of	the	current	research	uses	data	with	a	relatively	short	time	span	that	is	unlikely	to	capture	the	full	time	frame	over	which	the	Kuznets	curve	is	based	upon.	As	a	result,	cities	are	undergoing	transitions	across	different	stages	of	the	curve	in	terms	of	economic	development	and	environment	recovery.	Although	in	this	chapter,	I	found	that	every	city	exhibited	some	degree	of	vegetation	recovery,	the	presence	of	EKC	does	not	guarantee	that	economic	development	is	able	to	automatically	resolve	environmental	degradation	issues.	One	of	the	key	assumptions	of	EKC	is	that	the	economy	does	not	suffer	from	a	declining	environmental	quality.	EKC	assumes	that	economy	and	production	will	remain	growing	regardless	of	the	environmental	performance	(Stern,	Common,	&	Barbier,	1996).	There	is	a	chance	that,	even	with	the	existence	of	statistically	significant	EKC,	vegetation	loss	occurred	at	the	take-off	stage	of	the	economy	has	crossed	the	ecological	capacity	threshold	irreversibly.	Many	vegetation-related	ecological	properties	such	as	biodiversity	were	not	as	easy	as	pure	“vegetation”	to	capture	using	satellites.	As	a	result,	it	is	challenging	to	quantitatively	test	the	robustness	of	this	assumption.	Ultimately,	human	development	is	believed	to	be	highly	relevant	to	environmental	performance	(Arrow	et	al.,	1995).	Extending	from	current	chapter,	future	research	could	incorporate	additional	quantitative	and	spatially	classified	climate	and	economic	schemes	to	further	verify	the	existence	of	EKC	theory	and	even	potentially	identify	list	of	drivers	causing	the	varying	developing	patterns	observed	in	this	chapter.					84	Chapter	7		7.	Conclusion	Contemporary	cities	are	collectively	more	dynamic,	multi-dimensional,	and	complex	than	ever	before.	Cities	worldwide	strive	to	grow	not	only	economically	strong	but	also	environmentally	sustainable.	The	balance	between	the	economy	and	environment	has	been	challenging	particularly	for	cities	in	the	pan	Pacific	region,	which	is	seeing	some	of	the	most	rapid	urban	growth	rates	globally.	Urbanization	and	its	associated	physical	and	socio-economic	characteristics	are	interacting	at	a	much	faster	pace	and	occurring	at	a	range	of	spatial	and	temporal	scales.		This	work	applied	time	series	of	satellite	images,	chronically	monitoring	the	relationship	between	urban	environment	and	economics.	From	pixel	to	cityscape	scale,	this	approach	has	the	advantage	of	being	intuitively	appealing,	simple	to	reproduce	and	implementable	in	practical	urban	planning	and	management.	The	derived	information	is	useful	for	planners	to	compare	and	visualize	land	use	patterns	as	well	as	for	policy-makers	to	better	understand	inter-and	intra-	city	development	in	polycentric	and	highly	connected	global	urban	systems.		7.1. Research	innovation			This	dissertation	provides	key	innovations	for	characterizing	regional	urbanization	dynamics	using	open	access	remotely	sensed	images:		• The	integration	of	the	classic	concentric	ring	model	and	gap-free	time	series	of	satellite	data	unlocked	new	ways	to	informing	urban	environment	and	socio-economics	dynamics	(Chapter	3).		• Urban	vegetation	was	derived	using	an	innovative	spectral	unmixing	approach	on	an	entire	time	series	image	stack,	allowing	an	improved	understandings	of	vegetation	in	urban	environments	across	the	pan	Pacific	region	(Chapter	4).		• A	classic	econometric	model	was	used	to	examine	the	casual	relationship	between	conventionally	collected	census	information	and	advanced	remote	sensing	nighttime	lights	data	(Chapter	5).				85	• Relationships	between	urban	environment	and	economic	development	were	tested	using	remote	sensing	derived	proxies.	The	results	advanced	our	understanding	of	changing	environment	and	socio-economic	characteristics	from	pixel	to	regional	scales	(Chapter	6).			7.2. Answers	to	proposed	research	questions			7.2.1. How	can	remotely	sensed	derived	metrics	inform	urban	environment	and	socio-economics	dynamics	within	and	across	cities	in	pan	Pacific	region?			Remote	sensing	data,	particularly	Landsat	and	NTL	time	series	demonstrated	advantages	over	conventional	tabular	formatted	census	data.	Varying	political	and	cultural	backgrounds	within	the	pan	Pacific	region	limited	and	complicated	inter-	and	intra	city	comparisons.	Remote	sensing	derived	metrics	offered	measurements	that	were	not	only	spatio-temporally	consistent	but	also	comparable	across	different	cities,	unlocking	new	ways	for	regional	and	global	scale	urban	studies.			Two	spectral	indices	(i.e.	EVI	and	NDBI)	group	25	pan	Pacific	cities	into	5	classes.	The	average	silhouette	width	ranged	from	0.37	to	0.60,	while	the	ratio	of	the	between	to	the	total	sum	of	square	(BSS/TSS)	ranged	from	69%	to	92%	for	EVI	and	NDBI	respectively.	Dynamic	Time	Warping	(DTW)	derived	separability	metrics	showed	an	averaged	distance	of	0.36	(EVI)	and	0.28	(NDBI)	among	all	cities.	Spectral	unmixing	analysis	successfully	estimated	vegetation	fraction	at	a	sub-pixel	level.	Validation	using	high	spatial	resolution	Google	Earth	images	showed	a	correlation	coefficient	ranging	from	0.66	to	0.77.		Comparing	to	categorically	dividing	the	cities	into	land	use	or	land-cover	classes,	spectral	indices	and	vegetation	fraction	images	offered	continuous	measurements	of	the	urban	environments	and	socio-economic	dynamics	that	were	valuable	for	further	modelling	processes.			7.2.2. What	models	exist	to	examine	the	relationship	between	urban	environment	and	socio-economic	develop	over	time	and	space?		Linear	trends	between	urban	environment	and	economic	development	are	assessed	to	be	the	most	dominating	among	all	three	tested	relationship	(i.e.	linear,	quadratic,	and	cubic).	Remote	sensing	derived	metrics,	namely	vegetation	fraction	(VF)	and	nighttime	lights	(NTL)	time	series	were	used	as	a	proxy	to	urban	environment	and	economic	respectively.				86	The	result	implied	that	the	vegetation	changes	within	cities	in	pan	Pacific	regions	were	monotonic	and	irreversible.	Despite	the	dominance	of	linear	models,	EKC-like	quadratic	models	also	existed	within	all	examine	cities.	In	majority	of	the	cities,	quadratic	models	tended	to	be	more	spatially	clustered	compared	to	linear	and	cubic	models.	The	results	statistically	quantified	the	behaviour	of	how	vegetation	responses	to	city	brightness	changes,	furthering	the	discussion	of	EKC	theory	by	integrating	remote	sensing	observations	as	means	of	measuring	environmental	performance	and	human	development,	bridging	the	gap	between	conventional	econometric	theories	with	Earth	observation	platform.			7.2.3. What	similarities	and	differences	exist	across	cities	in	the	pan	Pacific	region	both	spatially	and	temporally?		Cities	showed	both	inter-	and	intra	variations	in	terms	of	spatial	and	temporal	changes	of	vegetation	and	nighttime	lights	brightness.	Regional	similarities	were	found	(Chapter	3)	particularly	in	Asian	countries	where	the	relationship	between	vegetation	fraction	values	and	NTL	brightness	was	less	linear	than	North	American	cities.	Climatically,	vegetation	dynamics	tended	to	be	more	alike	within	the	same	climate	scheme	(Chapter	6)	among	different	cities.		In	terms	of	spatio-temporal	patterns	of	urban	environment	and	economic	development,	I	found	three	pairs	of	urban	environments	that	were	strongly	similar	to	each	other	namely,	Melbourne	with	Sydney;	Tianjin	with	Manila;	and	Singapore	City	with	Kuala	Lumpur.		Characterizing	the	relationships	between	VF	and	NTL	revealed	new	patterns.	Mostly,	Asian	cities	showed	a	dominating	pattern	of	cubic	relationships	when	compared	to	North	American	cities.	While	cities	with	higher	economic	activity	level	were	more	dominated	by	linear	models	yet	patterns	were	less	obvious	for	middle	and	low-income	cities.			7.2.4. Characterizing	urban	built-up	and	greenspace	over	time	and	space			A	key	element	of	developing	an	understanding	of	urbanization	processes	globally	is	the	consistent	monitoring	of	cities	over	space	and	time	(Sexton	et	al.,	2013).	Detecting	and	analyzing	the	spatio-temporal	patterns	of	urban	environments	have	become	an	increasingly	critical	research	topic	with	practical	management	applications.	Early	studies	indicated	that	a	city	can	be	divided	into	a	series	of			87	expanding	rings,	also	known	as	the	concentric	ring	model	(Burgess,	1925).	This	model	has	been	used	widely	even	for	cities	with	less	regular	concentric	growth	patterns.				In	Chapter	2,	built	upon	the	classic	concentric	ring	model,	I	used	remote	sensing	derived	spectral	indices,	namely	the	Enhanced	Vegetation	Index	(EVI)	and	the	Normalized	Difference	Built-up	Index	(NDBI).	As	expected,	patterns	from	these	two	indices	illustrated	opposing	trends.	Spatially,	urban	cores	had	a	much	higher	NDBI	and	lower	EVI	than	outer	areas.	I	also	observed	the	existence	of	a	developing	multi-core	pattern	in	cities	where	both	EVI	and	NDBI	did	not	follow	a	simple	linear	trend.	Temporally,	urban	vegetation	showed	much	greater	variation	than	built-up	marked	by	results	from	Dynamic	Time	Warping	(DTW)	where	the	separability	among	each	trajectory	was	much	greater	for	the	EVI	trajectories	than	for	NDBI’s.		Commonality	can	still	be	found	among	the	25	cities	however.	Two	sets	of	K-means	cluster	analyses	were	applied	to	group	spatially	and	temporally	similar	cities.	K-means	clustering	analysis	indicated	an	optimal	number	of	five	classes.	As	anticipated,	the	majority	of	the	urban	environments	in	developing	countries	experienced	noticeable	development	rates	compared	to	the	urban	environments	in	more	developed	regions.	I	summarize	the	degree	of	similarity	both	spatially	and	temporally	across	urban	environments	in	the	pan	Pacific	region.	In	total,	there	are	three	pairs	of	urban	environments	consistently	being	grouped	together,	namely,	Melbourne	with	Sydney;	Tianjin	with	Manila;	and	Singapore	City	with	Kuala	Lumpur.	In	contrast,	cities	such	as	Las	Vegas	and	Vancouver	had	less	similar	features	both	spatially	and	temporally	with	any	of	the	other	urban	environments.		Compared	to	the	traditional	spectral	indices	the	integration	of	spectral	unmixing	and	Theil-Sen	(TS)	estimated	trend	slopes	offers	increased	ability	to	compare	and	contrast	vegetation	across	urban	environments	(Chapter	3).	Spectral	indices	for	given	pixels	can	be	difficult	to	compare	across	different	urban	environments	over	time.	Spectral	unmixing	analysis	(SMA)	demonstrated	new	ways	to	study	urban	greenspace	dynamics.	A	spectral	library	of	spatially	and	temporally	pure	pixels	successfully	estimated	vegetation	fraction	value	at	a	sub-pixel	scale	for	all	cities.	When	using	annual	vegetation	fraction	images,	cities	located	in	high	latitudes	(e.g.	Harbin)	and	particularly	mountainous	regions	(e.g.	Vancouver)	may	require	extra	caution.	In	those	cases,	pixels	with	low	vegetation	fraction	may	not	necessarily	indicate	intense	urbanization,	but	rather	low	vegetated	land-cover	types,	such	as	bare	rock,	ice,	and	snow.			88	Temporally,	there	were	four	types	of	urban	vegetation	changing	patterns.	The	first	included	cities	such	as	Shenzhen-Hong	Kong	area	which	exhibited	a	gradual	decline	in	vegetation	from	the	urban	center	through	to	the	outer	areas.	The	second	set	of	cities	includes	cities	such	as	Las	Vegas	that	had	consistently	increasing	vegetation	slope	as	the	distance	from	urban	center	increases.	The	third	type	contains	cities	including	Shanghai	where	vegetation	mostly	decreased	over	time.	And	lastly	cities	such	as	Vancouver,	Tokyo,	Sydney,	Edmonton,	and	Calgary,	where	vegetation	changes	were	relatively	minimal	as	indicated	by	a	near	zero	vegetation	change	trend.	These	types	of	cities	were	mostly	located	in	developed	regions.			7.2.5. Characterizing	social-economic	dynamics	over	time	and	space			Contemporary	cities	are	collectively	dynamic,	multi-dimensional,	and	complex.	Urbanization	and	its	associated	physical	and	socio-economic	characteristics	are	interacting	at	a	much	faster	pace	and	occurring	much	beyond	local	level.	In	this	work,	such	characteristics	were	reflected	by	inter-	and	intra-	city	variations	derived	from	the	nighttime	lights	imagery	for	25	cities	in	pan	Pacific	region.		7.2.6. Contrasting	effects	of	GDP	and	population	on	NTL			Large	inter-	and	intra-city	variations	of	urban	economic	activities	were	apparent	as	indicated	by	nighttime	lights	(NTL)	time	series	(Chapter	4).	Tokyo	and	Shen	Zhen-Hong	Kong	had	over	75%	of	land	with	active	brightness	prior	to	1992	while	most	cities	in	China	had	less	than	10%.	Cities	such	as	Shanghai	and	Tianjin	experienced	substantial	growth	in	their	economies	over	the	study	period	with	over	50%	growth.	Other	cities,	however,	experienced	less	economic	growth	with	approximately	75%	of	land	remaining	undeveloped.			Across	all	cities,	both	population	and	GDP	played	a	major	role	in	directing	changes	of	NTL.	According	to	the	Granger	causality	test,	the	brightness	of	cities	followed	increases	in	both	population	and	GDP	equally	and	neither	population	nor	GDP	alone	is	responsible	for	increasing	the	NTL.	Unexpectedly,	I	found	that	GDP	and	NTL	“granger	caused”	population	growth	suggesting	that	population	change	was	the	outcome	rather	than	the	cause	of	GDP	and	NTL	growth.	A	wealthier	and	more	economically	active	city	would	likely	to	attract	more	population.				89	Population	and	GDP	revealed	contrasting	effects	on	NTL	trends	between	stable	and	more	dynamic	cities.	It	has	long	been	thought	that	population	was	the	primary	driver	of	urban	growth	while	economic	development	was	an	outcome	of	booming	populations.	For	cities	with	minimal	NTL	changes	over	the	analysis	period,	the	causal	relationship	from	NTL	to	population	was	not	statistically	significant	yet	changes	in	population	“granger	caused”	both	GDP	and	NTL.	This	implied	that	in	cities	with	relatively	stable	NTL,	population	and	GDP	were	likely	to	the	key	drivers	of	NTL	changes	but	not	the	other	way	around.	Previous	work	(Dietzel	et	al.,	2005;	Satterthwaite,	2009)	found	that	rather	than	growing	population	alone,	it	was	the	high	consumption	lifestyle,	economic	and	political	decisions	that	led	to	urbanization.		In	fast	changing	cities	however,	growth	in	NTL	and	GDP	unexpectedly	led	to	an	increase	in	population.	There	was	no	significant	causal	association	between	GDP	and	NTL,	suggesting	that	in	rapidly	changing	populations	the	increases	were	driven	by	the	economic	development.	In	those	cities,	the	main	source	of	population	increase	was	through	immigration	from	rural	and	neighboring	areas,	involving	densification	and	conversion	of	existing	farm,	forest	or	barren	land	to	urban	land-cover	types.	Chapter	4	demonstrated	that	migration	was	more	likely	to	be	attracted	to	cities	with	promising	economic	conditions	and	undergoing	fast	urbanization	paces.				7.3. Testing	the	EKC	theory			The	relationship	between	urban	development	and	environment	appeared	to	be	linear	in	the	majority	of	the	cities	(Chapter	5).	There	was	little	evidence	of	quadratic	or	cubic	relationships.	A	linear	relationship	indicated	a	monotonic	irreversible	relationship	between	urban	vegetation	fraction	(VF)	and	NTL.		In	general,	I	demonstrated	that	within	the	study	period	(i.e.	1992-2012),	the	hypothesized	Environmental	Kuznets	Curve	(EKC)	was	not	the	dominating	relationship	among	all	three	tested	models	(i.e.	linear,	quadratic,	and	cubic).	Linear	models	were	the	most	dominating	model	yet	explained	the	least	amount	of	variation	of	all	marked	by	an	overall	relatively	low	r2	value.		The	hypothesized	EKC	likely	existed	within	city	limits	and	was	highly	clustered.	Our	findings,	using	vegetation	fraction	and	artificial	light	brightness	at	a	pixel	level,	suggested	that	at	least	in	certain	part	of	the	cityscape,	the	environment	recovered	with	a	growing	economy.	Cities	that	showed	a	dominating			90	vegetation	decreasing	trend	could	also	contained	substantial	vegetation	recovery	particularly	in	wealthy	cities	where	clean	energy	supplies,	urban	greenspaces,	and	efficient	public	transportation	systems	have	been	aggressively	built.	The	majority	of	Asian	cities	showed	a	dominating	pattern	of	cubic	relationships	when	compared	to	North	American	cities.	While	higher	income	level	cities	were	relatively	more	dominated	by	linear	models,	such	patterns	were	less	obvious	for	middle	and	low-income	cities.	Chapter	5	also	showed	that	cities	located	in	tropical	and	temperate	climate	schemes	had	more	quadratic	and	cubic	models	than	cities	from	continental	and	arid	climate	schemes	likely	due	to	varying	phenological	responses	to	local	climate	conditions.			7.4. Research	challenges			7.4.1. Urban	boundary		Like	other	ecosystems,	urban	environments	rarely	have	definitive	boundaries.	Attempts	to	use	two	sets	of	urban	boundaries,	namely,	conventional	administrative	boundary	(Chapter	3)	and	a	60-km	circular	buffer	(Chapter	4,	5,	and	6)	posted	one	of	the	main	challenges	for	urban	research	in	this	thesis.	The	integration	and	connectivity	among	today’s	urban	environments	had	outpaced	the	conventionally	defined	city	administrative	boundaries.	The	trade-off	however	was	the	ease	of	integrating	census-based	data	with	remotely	sensed	data	when	using	administratively	defined	boundaries.	The	key	factors	determining	how	to	define	urban	boundaries	in	remote	sensing	studies	include	the	consistency,	availability,	and	the	quality	of	local	census	data.	Administrative	boundaries	can	be	considered	when	high	quality	census	data	were	available.	While	an	artificially	drawn	circular	buffer	boundary	was	free	from	the	local	jurisdiction	systems,	offering	a	unique	alternative	for	urban	studies	that	focused	less	on	social	and	economic	aspects	and	more	towards	ecological	and	environmental	dynamics.			7.4.2. Urban	heterogeneity		Another	major	challenge	was	a	function	of	urban	environmental	heterogeneity	where	each	medium	resolution	pixel	(i.e.	30m)	can	contain	more	than	one	land-cover	or	land	use	types.	Although	spectral	unmixing	analysis	(Chapter	4)	showed	potential	in	time	series	urban	vegetation	studies,	the	process	of			91	selecting	spectrally	pure	endmembers	could	be	improved	using	a	more	systematic	procedure.	Regional	and	global	scale	studies	would	likely	to	involve	cities	located	in	varying	climate	and	ecologically	complex	locations,	causing	difficulties	in	collecting	spectrally	consistent	yet	global	comparable	endmembers.	Given	that,	this	dissertation	does	not	directly	compare	the	absolute	vegetation	fraction	values.	Rather	I	computed	the	rate	of	vegetation	for	each	individual	city	at	a	pixel	level	using	Theil-Sen	slope,	a	metric	that	was	more	comparable	both	within	and	across	cities.	Such	a	comparison	however	limited	my	ability	to	compare	and	contrast	cities	on	an	annual	basis.		Additionally,	spectral	responses	from	non-forested	vegetation	(e.g.	agriculture,	golf	courses)	were	likely	to	be	mixed	with	more	ecologically	functional	vegetation	(e.g.	parks,	grassland).	Highly	responsive	pixels	such	as	snow	and	ice	on	the	other	hand	are	spectrally	similar	to	pixels	within	cities,	both	of	which	had	a	high	reflectance	throughout	the	Landsat	spectral	bands.			7.4.3. NTL	Saturation			Due	to	the	limited	radiometric	range	of	NTL	sensors,	NTL	data	were	often	saturated	in	bright	areas	such	as	at	the	city	centre.	Zhang	et	al.	(2013)	incorporates	a	series	of	vegetation	images	to	de-saturate	NTL	data	on	the	assumption	that	there	is	an	inverse	relationship	between	vegetation	fraction	values	and	NTL	brightness.	However,	the	inclusion	of	another	input	variable	(e.g.	vegetation)	complicated	the	process	of	interpreting	statistic	analysis	particularly	in	Chapter	6	where	vegetation	was	compared	directly	against	NTL.	The	use	of	saturated	NTL	data	potentially	could	hinder	our	ability	to	capture	to	full	temporal	profile	in	brightly	lit	areas.			7.4.4. High	quality	census	data			Characterizing	the	socio-economic	nature	of	cities	was	still	primarily	remained	the	domain	of	census	data.	Local	census	and	economic	data	used	in	Chapter	4	had	varying	collecting	intervals	and	qualities.	In	many	developing	regions,	census	data	were	collected	infrequently	with	questionable	qualities.	The	scarcity	of	well	maintained	census	data	was	one	of	the	main	obstacles	when	comparing	against	remotely	sensed	derived	information	that	was	derived	using	an	institutionally	different	collecting	approach	and	scale.				92	However,	local	census	data	still	play	an	important	role	in	urbanization	researches.	One	of	the	main	challenges	of	this	dissertation,	or	remote	sensing	derived	model	in	general,	was	the	lack	of	reliable	validating	data.	High	quality	census	data	can	not	only	be	used	in	the	model	fitting	process	but	also	provide	an	alternative	sources	for	model	validation	and	ground-truthing	purposes.	7.4.5. Future	research	opportunities		Real-time	accurate	information	derived	from	remote	sensing	devices	offers	wall-to-wall	evaluations	of	new	policies	and	regulations.	Previous	literature	highlighted	the	gap	between	on-ground	initiatives	and	the	associated	spectral	responses	from	remote	sensing	sensors	(Liu	et	al.,	2005).	With	new	remote	sensing	sensors	being	designed	to	continuously	image	the	Earth,	it	is	possible	for	future	research	that	quantifies	such	delays	as	a	measurement	of	local	policy	effectiveness.		Beyond	the	local	scale,	many	global	remote	sensing	products	focused	on	urbanization	are	now	available	(e.g.	Román	et	al.,	2018).	Regional	to	global	scale,	temporally	informed,	urban	research	is	starting	to	emerge.	However,	a	more	comparable	metric	is	needed	for	comparing	and	contrasting	cities	on	an	annual	basis,	particularly	for	characterising	vegetation	and	other	ecological	dynamics.	Given	the	diverse	climate	and	ecological	conditions,	future	research	should	focus	on	developing	metrics	integrated	with	climate	variables	to	account	for	local	climate	variations.		Landscape	metrics	(e.g.	measurements	of	area,	density,	and	edge)	have	been	used	widely	with	remote	sensing	data.	The	spatial	configuration	of	each	city	offers	a	unique	perspective	of	the	current	land	use	patterns,	effectively	informing	future	plans.	However,	the	relationship	between	the	landscape	configuration	and	urban	temporal	changes	is	less	explored	(K.	C.	Seto	&	Fragkias,	2005).	Relating	time	series	data	with	landscape	metrics	will	be	critical	to	better	understand	the	rate	of	landuse	changes	in	a	language	that	is	more	familiar	to	urban	planners.		An	even	less	charted	topic	is	the	marriage	between	advanced	remote	sensing	data	and	conventional	econometric	theories	and	models	at	regional	and	global	scales.	Many	classic	econometric	theories	such	as	Environmental	Kuznets	Curve	were	developed	and	tested	using	spotted	and	discontinuous	data,	often	before	the	era	of	advanced	Earth	observation	data.	The	integration	of	remote	sensing	derived	variables	will	greatly	minimize	the	uncertainties,	allowing	a	more	systematic	and	repeatable	test.			 			93	Reference		Akbostanci,	Elif,	Türüt-Aşik,	Serap,	&	Tunç,	G.Ipek.	(2009).	The	relationship	between	income	and	environment	in	Turkey:	Is	there	an	environmental	Kuznets	curve?	Energy	Policy,	37(3),	861–867.	https://doi.org/10.1016/j.enpol.2008.09.088	Aldy,	J.	E.	(2005).	An	Environmental	Kuznets	Curve	Analysis	of	U.S.	State-Level	Carbon	Dioxide	Emissions.	The	Journal	of	Environment	&	Development,	14(1),	48–72.	https://doi.org/10.1177/1070496504273514	Apparicio,	Philippe,	Abdelmajid,	Mohamed,	Riva,	Mylène,	&	Shearmur,	Richard.	(2008).	Comparing	alternative	approaches	to	measuring	the	geographical	accessibility	of	urban	health	services:	Distance	types	and	aggregation-error	issues.	International	Journal	of	Health	Geographics,	7(1),	7.	https://doi.org/10.1186/1476-072X-7-7	Arrow,	Kenneth,	Bolin,	Bert,	Costanza,	Robert,	Dasgupta,	Partha,	Folke,	Carl,	Holling,	C.	S.,	Jansson,	Bengt-Owe,	Levin,	Simon,	M~iler,	Karl-Gsran,	Perrings,	Charles,	&	Pimentel,	David.	(1995).	Economic	growth,	carrying	capacity,	and	the	environment	1.	Ecological	Economics,	268(2),	520–521.	https://doi.org/10.1016/0921-8009(95)00059-3	Ash,	Caroline,	Jasny,	Barbara	R.,	Roberts,	Leslie,	Stone,	Richard,	&	Sugden,	Andrew	M.	(2008).	Reimagining	cities.	Science,	319(5864),	739.	https://doi.org/10.1126/science.319.5864.739	Asner,	G.	P.,	&	Heidebrecht,	K.	B.	(2002).	Spectral	unmixing	of	vegetation,	soil	and	dry	carbon	cover	in	arid	regions:	Comparing	multispectral	and	hyperspectral	observations.	International	Journal	of	Remote	Sensing,	23(19),	3939–3958.	https://doi.org/10.1080/01431160110115960	Bagan,	Hasi,	&	Yamagata,	Yoshiki.	(2014).	Land-cover		change	analysis	in	50	global	cities	by	using	a	combination	of	Landsat	data	and	analysis	of	grid	cells.	Environmental	Research	Letters,	9(6),	64015.	https://doi.org/10.1088/1748-9326/9/6/064015	Balbo,	Marcello,	&	Navez-Bouchanine,	Françoise.	(1995).	Urban	fragmentation	as	a	research	hypothesis:	Rabat-Salé	case	study.	Habitat	International,	19(4),	571–582.	https://doi.org/10.1016/0197-3975(95)00008-4	Barnsley,	Michael	J.,	&	Barr,	Stuart	L.	(1997).	Distinguishing	urban	land-use	categories	in	fine	spatial	resolution	land-cover		data	using	a	graph-based,	structural	pattern	recognition	system.	Computers,	Environment	and	Urban	Systems,	21(3–4),	209–225.	https://doi.org/10.1016/S0198-9715(97)10001-1	Bartlett,	Bruce.	(1994).	The	high	cost	of	turning	green.	The	Wall	Street	Journal,	14(0).	Baugh,	Kimberly,	Elvidge,	Christopher	D.,	Ghosh,	Tilottama,	&	Ziskin,	Daniel.	(2010).	Development	of	a	2009	Stable	Lights	Product	using	DMSP-OLS	data.	Proceedings	of	the	Asia-Pacific	Advanced	Network,	30,	114–130.	https://doi.org/10.7125/APAN.30.17	BC	Statistics.	(2012).	Population	estimates.	Retrieved	from	https://www2.gov.bc.ca/gov/content/data/statistics/people-population-community/population	Beckerman,	Wilfred.	(1992).	Economic	growth	and	the	environment:	Whose	growth?	whose	environment?	World	Development,	20(4),	481–496.	https://doi.org/https://doi.org/10.1016/0305-		94	750X(92)90038-W	Bell,	David,	&	Jayne,	Mark.	(2009).	Small	cities?	Towards	a	research	agenda.	International	Journal	of	Urban	and	Regional	Research,	33(3),	683–699.	https://doi.org/10.1111/j.1468-2427.2009.00886.x	Bemdt,	Donald	J.,	&	Clifford,	James.	(1994).	Using	Dynamic	Time	Warping	to	FindPatterns	in	Time	Series.	Retrieved	from	https://www.aaai.org/Papers/Workshops/1994/WS-94-03/WS94-03-031.pdf	Bennett,	Mia	M.,	&	Smith,	Laurence	C.	(2017).	Advances	in	using	multitemporal	night-time	lights	satellite	imagery	to	detect,	estimate,	and	monitor	socioeconomic	dynamics.	Remote	Sensing	of	Environment,	192,	176–197.	https://doi.org/10.1016/j.rse.2017.01.005	Bennie,	Jonathan,	Davies,	Thomas	W.,	Duffy,	James	P.,	Inger,	Richard,	&	Gaston,	Kevin	J.	(2015).	Contrasting	trends	in	light	pollution	across	Europe	based	on	satellite	observed	night	time	lights.	Scientific	Reports,	4(1),	3789.	https://doi.org/10.1038/srep03789	Benz,	Ursula	C.,	Hofmann,	Peter,	Willhauck,	Gregor,	Lingenfelder,	Iris,	&	Heynen,	Markus.	(2004).	Multi-resolution,	object-oriented	fuzzy	analysis	of	remote	sensing	data	for	GIS-ready	information.	ISPRS	Journal	of	Photogrammetry	and	Remote	Sensing,	58(3–4),	239–258.	https://doi.org/10.1016/J.ISPRSJPRS.2003.10.002	Bey,	Adia,	Sánchez-Paus	Díaz,	Alfonso,	Maniatis,	Danae,	Marchi,	Giulio,	Mollicone,	Danilo,	Ricci,	Stefano,	Bastin,	Jean-François,	Moore,	Rebecca,	Federici,	Sandro,	Rezende,	Marcelo,	Patriarca,	Chiara,	Turia,	Ruth,	Gamoga,	Gewa,	Abe,	Hitofumi,	Kaidong,	Elizabeth,	&	Miceli,	Gino.	(2016).	Collect	Earth:	Land	Use	and	Land-cover	Assessment	through	Augmented	Visual	Interpretation.	Remote	Sensing,	8(10),	807.	https://doi.org/10.3390/rs8100807	Birch,	Eugenie	L.,	&	Wachter,	Susan	M.	(2011).	Global	urbanization.	University	of	Pennsylvania	Press.	Bolshakova,	N.,	&	Azuaje,	F.	(2003).	Cluster	validation	techniques	for	genome	expression	data.	Signal	Processing,	83(4),	825–833.	https://doi.org/10.1016/S0165-1684(02)00475-9	Boone,	Christopher	G.,	&	Fragkias,	Michail.	(2012).	Urbanization	and	Sustainability:	Linking	Urban	Ecology,	Environmental	Justice	and	Global	Environmental	Change.	https://doi.org/10.1007/978-94-007-5666-3	Burgess,	E.	W.	(1925).	The	growth	of	the	city.	In	The	City	(pp.	47–62).	Burnham,	Kenneth	P.,	&	Anderson,	David	R.	(2003).	Model	selection	and	multimodel	inference:	a	practical	information-theoretic	approach.	Springer	Science	&	Business	Media.	Cadenasso,	Mary	L.,	Pickett,	Steward	T.	A.,	&	Schwarz,	Kirsten.	(2007).	Spatial	heterogeneity	in	urban	ecosystems:	reconceptualizing	land-cover	and	a	framework	for	classification.	Frontiers	in	Ecology	and	the	Environment,	5(2),	80–88.	Castrence,	Miguel,	Nong,	Duong,	Tran,	Chinh,	Young,	Luisa,	&	Fox,	Jefferson.	(2014).	Mapping	Urban	Transitions	Using	Multi-Temporal	Landsat	and	DMSP-OLS	Night-Time	Lights	Imagery	of	the	Red	River	Delta	in	Vietnam.	Land,	3(1),	148–166.	https://doi.org/10.3390/land3010148	Champion,	A.	G.	A.	(2001).	Changing	Demographic	Regime	and	Evolving	Polycentric	Urban	Regions:	Consequences	for	the	Size,	Composition	and	Distribution	of	City	Population.	Urban	Studies,	4(38),	657–677.	Chen,	S.	S.,	Yan,	Yue,	Gao,	Qun,	&	Liu,	Dene.	(2014).	Quantifying	circular	urban	expansion	patterns	of			95	compact	Chinese	cities:	the	case	of	Yangtze	River	Delta,	China.	Environment	and	Planning	B:	Planning	and	Design,	41(1963).	https://doi.org/10.1068/b120004p	Cheng,	Xiaoyun,	Wei,	Bensheng,	Chen,	Guojian,	Li,	Junxiang,	&	Song,	Conghe.	(2007).	Influence	of	Park	Size	and	Its	Surrounding	Urban	Landscape	Patterns	on	the	Park	Cooling	Effect,	1–10.	https://doi.org/10.1061/(ASCE)UP.1943-5444.0000256.	Clancey,	Gregory.	(2004).	Local	memory	and	worldly	narrative:	The	remote	city	in	America	and	Japan.	Urban	Studies,	41(12),	2335–2355.	https://doi.org/10.1080/00420980412331297564	Cliff,	Andrew	D.,	&	Ord,	Keith.	(1970).	Spatial	autocorrelation:	a	review	of	existing	and	new	measures	with	applications.	Economic	Geography,	46(sup1),	269–292.	Cohen,	Barney.	(2006).	Urbanization	in	developing	countries:	Current	trends,	future	projections,	and	key	challenges	for	sustainability.	Technology	in	Society.	https://doi.org/10.1016/j.techsoc.2005.10.005	Corlett,	Richard	T.	(1999).	Environmental	forestry	in	Hong	Kong:	1871--1997.	Forest	Ecology	and	Management,	116(1–3),	93–105.	Croft,	Thomas.	(1973).	Burning	Waste	Gas	in	Oil	Fields.	Nature,	245,	375–376.	Dietz,	Thomas,	Rosa,	EA,	&	York,	Richard.	(2007).	Driving	the	human	ecological	footprint.	Frontiers	in	Ecology	and	the	Environment	in	Ecology	and	the	Environment,	5(1),	13–18.	Dietzel,	Charles,	Herold,	Martin,	Hemphill,	Jeffrey	J.,	&	Clarke,	Keith	C.	(2005).	Spatio-temporal	dynamics	in	California’s	Central	Valley:	Empirical	links	to	urban	theory.	International	Journal	of	Geographical	Information	Science,	19(2),	175–195.	https://doi.org/10.1080/13658810410001713407	Dijst,	Martin,	&	Vidakovic,	Velibor.	(2000).	Travel	time	ratio:	the	key	factor	of	spatial	reach.	Transportation,	27(2),	179–199.	https://doi.org/10.1023/A:1005293330869	Doll,	Christopher	N.	H.,	Muller,	Jan-Peter,	&	Morley,	Jeremy	G.	(2006).	Mapping	regional	economic	activity	from	night-time	light	satellite	imagery.	Ecological	Economics,	57(1),	75–92.	https://doi.org/10.1016/j.ecolecon.2005.03.007	Donaldson,	Dave,	&	Storeygard,	Adam.	(2016).	The	View	from	Above:	Applications	of	Satellite	Data	in	Economics.	Journal	of	Economic	Perspectives,	30(4—Fall),	171–198.	https://doi.org/10.1257/jep.30.4.171	Dorais,	Alexis,	&	Cardille,	Jeffrey.	(2011).	Strategies	for	Incorporating	High-Resolution	Google	Earth	Databases	to	Guide	and	Validate	Classifications:	Understanding	Deforestation	in	Borneo.	Remote	Sensing,	3(6),	1157–1176.	https://doi.org/10.3390/rs3061157	Dumitrescu,	Elena	Ivona,	&	Hurlin,	Christophe.	(2012).	Testing	for	Granger	non-causality	in	heterogeneous	panels.	Economic	Modelling,	29(4),	1450–1460.	https://doi.org/10.1016/j.econmod.2012.02.014	Ebener,	Steeve,	Murray,	Christopher,	Tandon,	Ajay,	&	Elvidge,	Christopher	C.	(2005).	From	wealth	to	health:	modelling	the	distribution	of	income	per	capita	at	the	sub-national	level	using	night-time	light	imagery.	International	Journal	of	Health	Geographics,	4(1),	5.	https://doi.org/10.1186/1476-072X-4-5	Egli,	Hannes,	&	Steger,	Thomas	M.	(2007).	A	dynamic	model	of	the	environmental	Kuznets	curve:	Turning	point	and	public	policy.	Environmental	and	Resource	Economics,	36(1),	15–34.			96	https://doi.org/10.1007/s10640-006-9044-9	Elmore,	Andrew	J.,	Mustard,	John	F.,	Manning,	Sara	J.,	&	Lobell,	David	B.	(2000).	Quantifying	vegetation	change	in	semiarid	environments:	precision	and	accuracy	of	spectral	mixture	analysis	and	the	normalized	difference	vegetation	index.	Remote	Sensing	of	Environment,	73(1),	87–102.	Elvidge,	Christopher	D.,	&	Lyon,	Ronald	J.	P.	(1985).	Influence	of	rock-soil	spectral	variation	on	the	assessment	of	green	biomass.	Remote	Sensing	of	Environment,	17(3),	265–279.	Elvidge,	Christopher,	Hsu,	Feng-Chi,	Baugh,	KimberlyE,	&	Ghosh,	Tilottama.	(2014).	National	Trends	in	Satellite-Observed	Lighting:	1992-2012.	In	Global	Urban	Monitoring	and	Assessment	through	Earth	Observation	(pp.	97–120).	CRC	Press.	https://doi.org/doi:10.1201/b17012-9	Escobedo,	Francisco	J.,	Kroeger,	Timm,	&	Wagner,	John	E.	(2011).	Urban	forests	and	pollution	mitigation:	Analyzing	ecosystem	services	and	disservices.	Environmental	Pollution,	159(8–9),	2078–2087.	https://doi.org/10.1016/J.ENVPOL.2011.01.010	Fan,	Fenglei,	&	Fan,	Wei.	(2014).	Understanding	spatial-temporal	urban	expansion	pattern	(1990–2009)	using	impervious	surface	data	and	landscape	indexes:	a	case	study	in	Guangzhou	(China).	Journal	of	Applied	Remote	Sensing,	8(1),	83609.	https://doi.org/10.1117/1.JRS.8.083609	Fauvel,	M.,	&	Benediktsson,	Ja.	(2008).	Spectral	and	Spatial	Classification	of	Hyperspectral	Data	Using	SVMs	and	Morphological	Profile.	…	and	Remote	Sensing	…,	46(11),	3804–3814.	https://doi.org/10.1109/TGRS.2008.922034	Fernandes,	Richard,	&	G.	Leblanc,	Sylvain.	(2005).	Parametric	(modified	least	squares)	and	non-parametric	(Theil–Sen)	linear	regressions	for	predicting	biophysical	parameters	in	the	presence	of	measurement	errors.	Remote	Sensing	of	Environment,	95(3),	303–316.	https://doi.org/10.1016/J.RSE.2005.01.005	Fodha,	Mouez,	&	Zaghdoud,	Oussama.	(2010).	Economic	growth	and	pollutant	emissions	in	Tunisia:	An	empirical	analysis	of	the	environmental	Kuznets	curve.	Energy	Policy,	38(2),	1150–1156.	https://doi.org/10.1016/j.enpol.2009.11.002	Gaston,	Kevin	J.,	Duffy,	James	P.,	&	Bennie,	Jonathan.	(2015).	Quantifying	the	erosion	of	natural	darkness	in	the	global	protected	area	system.	Conservation	Biology,	29(4),	1132–1141.	https://doi.org/10.1111/cobi.12462	Gergel,	Sarah	E.,	Bennett,	Elena	M.,	Greenfield,	Ben	K.,	King,	Susan,	Overdevest,	Christine	A.,	&	Stumborg,	Basil.	(2004).	A	TEST	OF	THE	ENVIRONMENTAL	KUZNETS	CURVE	USING	LONG-TERM	WATERSHED	INPUTS.	Ecological	Applications,	14(2),	555–570.	Getis,	Arthur,	&	Ord,	J.Keith.	(1992).	The	analysis	of	spatial	association	by	use	of	distance	statistics.	Geographical	Analysis,	24(3),	189–206.	Ghosh,	Tilottama,	Elvidge,	Christopher	D.,	Sutton,	Paul	C.,	Baugh,	Kimberly	E.,	Ziskin,	Daniel,	&	Tuttle,	Benjamin	T.	(2010).	Creating	a	global	grid	of	distributed	fossil	fuel	CO2	emissions	from	nighttime	satellite	imagery.	Energies,	3(12),	1895–1913.	https://doi.org/10.3390/en3121895	Giorgino,	Toni.	(2009).	Computing	and	Visualizing	Dynamic	Time	Warping	Alignments	in	R :	The	dtw	Package.	Journal	of	Statistical	Software,	31(7).	https://doi.org/10.18637/jss.v031.i07	Glaeser,	Edward	L.	(2011).	Cities,	productivity,	and	quality	of	life.	Science	(New	York,	N.Y.),	333(6042),			97	592–594.	https://doi.org/10.1126/science.1209264	Glass,	David	C.,	&	Singer,	Jerome	E.	(1972).	Urban	stress:	Experiments	on	noise	and	social	stressors.	Gober,	Patricia.	(2010,	August	1).	Desert	urbanization	and	the	challenges	of	water	sustainability.	Current	Opinion	in	Environmental	Sustainability.	Elsevier.	https://doi.org/10.1016/j.cosust.2010.06.006	Gonçalves,	R.	R.	V.,	Zullo,	J.,	Amaral,	B.	F.,	Coltri,	P.	P.,	M.Sousa,	E.	P.,	&	Romani,	L.	A.	S.	(2014).	LAND	USE	TEMPORAL	ANALYSIS	THROUGH	CLUSTERING	TECHNIQUES	ON	SATELLITE	IMAGE	TIME	SERIES,	2173–2176.	Gordon,	P.,	Richardsonll,	H.	W.,	&	Wong,	H.	L.	(1986).	The	distribution	of	population	and	employment	in	a	polycentric	city :	the	case	of	Los	Angeles	f,	18(November	1984),	161–173.	Grahn,	Patrik,	&	Stigsdotter,	Ulrika	A.	(2003).	Landscape	planning	and	stress.	Urban	Forestry	and	Urban	Greening,	2(1),	1–18.	https://doi.org/10.1078/1618-8667-00019	Granger,	C.	W.	J.	(1969).	Investigating	Causal	Relations	by	Econometric	Models	and	Cross-spectral	Methods	Author	(	s	):	C	.	W	.	J	.	Granger	Published	by :	The	Econometric	Society	Stable	URL :	http://www.jstor.org/stable/1912791	Accessed :	08-07-2016	18 :	00	UTC	Your	use	of	the	JST.	Econometrica:	Journal	of	the	Econometric	Society,	37(3),	424–438.	Green,	A.	A.,	Berman,	M.,	Switzer,	P.,	&	Craig,	M.	D.	(1988).	A	transformation	for	ordering	multispectral	data	in	terms	of	image	quality	with	implications	for	noise	removal.	IEEE	Transactions	on	Geoscience	and	Remote	Sensing,	26(1),	65–74.	https://doi.org/10.1109/36.3001	Griffiths,	Patrick,	Hostert,	Patrick,	Gruebner,	Oliver,	&	der	Linden,	Sebastian	Van.	(2010).	Mapping	megacity	growth	with	multi-sensor	data.	Remote	Sensing	of	Environment,	114(2),	426–439.	https://doi.org/10.1016/j.rse.2009.09.012	Grimm,	N.	B.,	Faeth,	S.	H.,	Golubiewski,	N.	E.,	Redman,	C.	L.,	Wu,	J.,	Bai,	X.,	&	Briggs,	J.	M.	(2008).	Global	Change	and	the	Ecology	of	Cities.	Science,	319(5864),	756–760.	https://doi.org/10.1126/science.1150195	Groffman,	Peter	M.,	Cavender-Bares,	Jeannine,	Bettez,	Neil	D.,	Grove,	J.Morgan,	Hall,	Sharon	J.,	Heffernan,	James	B.,	Hobbie,	Sarah	E.,	Larson,	Kelli	L.,	Morse,	Jennifer	L.,	Neill,	Christopher,	Nelson,	Kristen,	O’Neil-Dunne,	Jarlath,	Ogden,	Laura,	Pataki,	Diane	E.,	Polsky,	Colin,	Chowdhury,	Rinku	Roy,	&	Steele,	Meredith	K.	(2014).	Ecological	homogenization	of	urban	USA.	Frontiers	in	Ecology	and	the	Environment,	12(1),	74–81.	https://doi.org/10.1890/120374	Grossman,	Gene	M.,	&	Krueger,	Alan	B.	(1991).	Environmental	Impacts	of	a	North	American	Free	Trade	Agreement.	National	Bureau	of	Economic	Research	Working	Paper	Series,	No.	3914(3914),	1–57.	https://doi.org/10.3386/w3914	Guérois,	Marianne,	&	Pumain,	Denise.	(2008).	Built-up	encroachment	and	the	urban	field:	A	comparison	of	forty	European	cities.	Environment	and	Planning	A,	40(9),	2186–2203.	https://doi.org/10.1068/a39382	Handayani,	Wiwandari,	&	Rudiarto,	Iwan.	(2014).	Dynamics	of	Urban	Growth	in	Semarang	Metropolitan	–	Central	Java:	An	Examination	Based	on	Built-Up	Area	and	Population	Change.	Journal	of	Geography	and	Geology,	6(4),	80–87.	https://doi.org/10.5539/jgg.v6n4p80	Hansen,	M.	C.,	Roy,	D.	P.,	Lindquist,	E.,	Adusei,	B.,	Justice,	C.	O.,	Alstatt,	A.	(2008).	A	method	for			98	integrating	MODIS	and	Landsat	data	for	systematic	monitoring	of	forest	cover	and	change	in	the	Congo	Basin.	Remote	Sensing	of	Environment,	112(5),	2495–2513.	Hawelka,	Bartosz,	Sitko,	Izabela,	Beinat,	Euro,	Sobolevsky,	Stanislav,	Kazakopoulos,	Pavlos,	&	Ratti,	Carlo.	(2014).	Geo-located	Twitter	as	proxy	for	global	mobility	patterns.	Cartography	and	Geographic	Information	Science,	41(3),	260–271.	https://doi.org/10.1080/15230406.2014.890072	Heckscher,	Eli	Filip.	(1919).	The	effect	of	foreign	trade	on	the	distribution	of	income.	Heiden,	Uta,	Segl,	Karl,	Roessner,	Sigrid,	&	Kaufmann,	Hermann.	(2007).	Determination	of	robust	spectral	features	for	identification	of	urban	surface	materials	in	hyperspectral	remote	sensing	data.	Remote	Sensing	of	Environment,	111(4),	537–552.	https://doi.org/10.1016/J.RSE.2007.04.008	Henderson,	M.,	Yeh,	E.	T.,	Gong,	P.,	Elvidge,	C.,	&	Baugh,	K.	(2003).	Validation	of	urban	boundaries	derived	from	global	night-time	satellite	imagery.	International	Journal	of	Remote	Sensing,	24(3),	595–610.	https://doi.org/10.1080/01431160304982	Hermosilla,	Txomin,	Wulder,	Michael	A.,	White,	Joanne	C.,	Coops,	Nicholas	C.,	&	Hobart,	Geordie	W.	(2015).	An	integrated	Landsat	time	series	protocol	for	change	detection	and	generation	of	annual	gap-free	surface	reflectance	composites.	Remote	Sensing	of	Environment,	158,	220–234.	Heynen,	Nik,	Perkins,	Harold	A.,	&	Roy,	Parama.	(2006).	The	Political	Ecology	of	Uneven	Urban	Green	Space:	The	Impact	of	Political	Economy	on	Race	and	Ethnicity	in	Producing	Environmental	Inequality	in	Milwaukee.	Urban	Affairs	Review,	42(1),	3–25.	https://doi.org/10.1177/1078087406290729	Heynen,	Nikolas	C.,	&	Lindsey,	Greg.	(2003).	Correlates	of	Urban	Forest	Canopy	Cover.	Public	Works	Management	&	Policy,	8(1),	33–47.	https://doi.org/10.1177/1087724X03008001004	Hill,	Michael	J.,	&	Donald,	Graham	E.	(2003).	Estimating	spatio-temporal	patterns	of	agricultural	productivity	in	fragmented	landscapes	using	AVHRR	NDVI	time	series.	Remote	Sensing	of	Environment,	84(3),	367–384.	https://doi.org/10.1016/S0034-4257(02)00128-1	Hodgson,	Michael	E.,	Jensen,	John	R.,	Tullis,	Jason	A.,	Riordan,	Kevin	D.,	&	Archer,	Clark	M.	(2003).	Synergistic	Use	of	Lidar	and	Color	Aerial	Photography	for	Mapping	Urban	Parcel	Imperviousness.	Photogrammetric	Engineering	&	Remote	Sensing,	69(9),	973–980.	https://doi.org/10.14358/PERS.69.9.973	Hoffmann,	Robert,	Lee,	Chew-Ging,	Ramasamy,	Bala,	&	Yeung,	Matthew.	(2005).	FDI	and	pollution:	a	granger	causality	test	using	panel	data.	Journal	of	International	Development,	17(3),	311–317.	https://doi.org/10.1002/jid.1196	Hoornweg,	Daniel,	Bhada-Tata,	Perinaz,	&	Kennedy,	Chris.	(2013).	Waste	production	must	peak	this	century.	Nature,	502(7473),	615–617.	https://doi.org/10.1038/502615a	Hostert,	Patrick,	Röder,	Achim,	&	Hill,	Joachim.	(2003).	Coupling	spectral	unmixing	and	trend	analysis	for	monitoring	of	long-term	vegetation	dynamics	in	Mediterranean	rangelands.	Remote	Sensing	of	Environment,	87(2–3),	183–197.	https://doi.org/10.1016/80034-4257(03)00145-7	Hsiao,	Cheng.	(2007).	Panel	data	analysis-advantages	and	challenges.	Test,	16(1),	1–22.	https://doi.org/10.1007/s11749-007-0046-x	Huang,	Qingxu,	He,	Chunyang,	Gao,	Bin,	Yang,	Yang,	Liu,	Zhifeng,	Zhao,	Yuanyuan,	&	Dou,	Yue.	(2015).			99	Detecting	the	20	year	city-size	dynamics	in	China	with	a	rank	clock	approach	and	DMSP/OLS	nighttime	data.	Landscape	and	Urban	Planning,	137,	138–148.	https://doi.org/10.1016/j.landurbplan.2015.01.004	Huete,	A.	R.,	Jackson,	R.	D.,	&	Post,	D.	F.	(1985).	Spectral	response	of	a	plant	canopy	with	different	soil	backgrounds.	Remote	Sensing	of	Environment,	17(1),	37–53.	Huete,	A.	R.,	&	Tucker,	C.	J.	(1991).	Investigation	of	soil	influences	in	AVHRR	red	and	near-infrared	vegetation	index	imagery.	International	Journal	of	Remote	Sensing,	12(6),	1223–1242.	Hugo,	Graeme.	(2017).	New	forms	of	urbanization:	beyond	the	urban-rural	dichotomy.	Routledge.	Retrieved	from	https://books.google.ca/books?hl=en&lr=&id=aJwuDwAAQBAJ&oi=fnd&pg=PT10&ots=KpH9Fs2b3h&sig=a07XGKAAWOtwo3LY_zSzXabEgw0#v=onepage&q&f=false	Im,	Kyung	So;,	&	Pesaran,	M.	H.	(2003).	on	the	Panel	Unit	Root	Tests	Using	Nonlinear	Instrumental	Variables.	SSRN.	Janelle,	Donald	G.,	&	Beuthe,	Michel.	(1997).	Globalization	and	research	issues	in	transportation.	Journal	of	Transport	Geography,	5(3),	199–206.	https://doi.org/10.1016/S0966-6923(97)00017-3	Jensen,	Jr,	&	Cowen,	Dc.	(1999).	Remote	sensing	of	urban	suburban	infrastructure	and	socio-economic	attributes.	Photogrammetric	Engineering	and	Remote	Sensing,	65(5),	611–622.	Ji,	Wei,	Ma,	Jia,	Twibell,	Rima	Wahab,	&	Underhill,	Karen.	(2006).	Characterizing	urban	sprawl	using	multi-stage	remote	sensing	images	and	landscape	metrics.	Computers,	Environment	and	Urban	Systems,	30(6),	861–879.	https://doi.org/10.1016/j.compenvurbsys.2005.09.002	Jim,	C.	Y.,	&	Chen,	Wendy	Y.	(2006).	Impacts	of	urban	environmental	elements	on	residential	housing	prices	in	Guangzhou	(China).	Landscape	and	Urban	Planning,	78(4),	422–434.	https://doi.org/10.1016/j.landurbplan.2005.12.003	Johansen,	Søren.	(1988).	Statistical	analysis	of	cointegration	vectors.	Journal	of	Economic	Dynamics	and	Control,	12(2–3),	231–254.	https://doi.org/10.1016/0165-1889(88)90041-3	Jones,	Gavin	W.	(2002).	Southeast	Asian	urbanization	and	the	growth	of	mega-urban	regions.	Journal	of	Population	Research,	19(2),	119–136.	https://doi.org/10.1007/BF03031973	Kahn,	Matthew	E.	(2000).	The	environmental	impact	of	suburbanization.	Journal	of	Policy	Analysis	and	Management,	19(4),	569–586.	https://doi.org/10.1002/1520-6688(200023)19:4<569::AID-PAM3>3.0.CO;2-P	Keast,	Allen.	(2013).	Biogeography	and	ecology	in	Australia.	Springer.	Keogh,	Eamonn,	&	Ratanamahatana,	Chotirat	Ann.	(2005).	Exact	indexing	of	dynamic	time	warping.	Knowledge	and	Information	Systems,	7(3),	358–386.	https://doi.org/10.1007/s10115-004-0154-9	Keshava,	Nirmal,	&	Mustard,	John	F.	(2002).	Spectral	unmixing.	IEEE	Signal	Processing	Magazine,	19(1),	44–57.	https://doi.org/10.1109/79.974727	Kestens,	Yan,	Thériault,	Marius,	&	Des	Rosiers,	François.	(2004).	The	Impact	of	Surrounding	Land	Use	and	Vegetation	on	Single-Family	House	Prices.	Environment	and	Planning	B:	Planning	and	Design,	31(4),	539–567.	https://doi.org/10.1068/b3023			100	Klotz,	M.,	Kemper,	T.,	Geiß,	C.,	Esch,	T.,	&	Taubenböck,	H.	(2016).	How	good	is	the	map?	A	multi-scale	cross-comparison	framework	for	global	settlement	layers:	Evidence	from	Central	Europe.	Remote	Sensing	of	Environment,	178,	191–212.	https://doi.org/10.1016/j.rse.2016.03.001	Knapp,	Tom,	&	Mookerjee,	Rajen.	(1996).	Population	growth	and	global	CO2	emissions.	Energy	Policy,	24(1),	31–37.	https://doi.org/10.1016/0301-4215(95)00130-1	Kong,	Fanhua,	Yin,	Haiwei,	&	Nakagoshi,	Nobukazu.	(2007).	Using	GIS	and	landscape	metrics	in	the	hedonic	price	modeling	of	the	amenity	value	of	urban	green	space:	A	case	study	in	Jinan	City,	China.	Landscape	and	Urban	Planning,	79(3–4),	240–252.	https://doi.org/10.1016/j.landurbplan.2006.02.013	Kottek,	Markus,	Grieser,	Jürgen,	Beck,	Christoph,	Rudolf,	Bruno,	&	Rubel,	Franz.	(2006).	World	map	of	the	Köppen-Geiger	climate	classification	updated.	Meteorologische	Zeitschrift,	15(3),	259–263.	Kumagai,	Kiichiro.	(n.d.).	ANALYSIS	OF	VEGETATION	DISTRIBUTION	IN	URBAN	AREAS:	SPATIAL	ANALYSIS	APPROACH	ON	A	REGIONAL	SCALE.	Retrieved	from	https://pdfs.semanticscholar.org/92bf/9658e5c5a739d9b625df8fc7de9b45fa7029.pdf	Kummu,	Matti,	Ward,	Philip	J.,	De	Moel,	Hans,	&	Varis,	Olli.	(2010).	Is	physical	water	scarcity	a	new	phenomenon?	Global	assessment	of	water	shortage	over	the	last	two	millennia.	Environmental	Research	Letters,	5(3),	34006.	https://doi.org/10.1088/1748-9326/5/3/034006	Kuznets,	Simon.	(1955).	Economic	growth	and	income	inequality.	The	American	Economic	Review,	1–28.	Kyba,	Christopher,	Garz,	Stefanie,	Kuechly,	Helga,	de	Miguel,	Alejandro,	Zamorano,	Jaime,	Fischer,	Jürgen,	&	Hölker,	Franz.	(2014).	High-Resolution	Imagery	of	Earth	at	Night:	New	Sources,	Opportunities	and	Challenges.	Remote	Sensing,	7(1),	1–23.	https://doi.org/10.3390/rs70100001	Lee,	Taedong.	(2013).	Global	cities	and	transnational	climate	change	networks.	Global	Environmental	Politics,	13(1),	108–127.	Levin,	Andrew,	Lin,	Chien	Fu,	&	Chu,	Chia	Shang	James.	(2002).	Unit	root	tests	in	panel	data:	Asymptotic	and	finite-sample	properties.	Journal	of	Econometrics,	108(1),	1–24.	https://doi.org/10.1016/S0304-4076(01)00098-7	Li,	Xi,	Chen,	Xiaoling,	Zhao,	Yousong,	Xu,	Jia,	Chen,	Fengrui,	&	Li,	Hui.	(2013).	Automatic	intercalibration	of	night-time	light	imagery	using	robust	regression.	Remote	Sensing	Letters,	4(1),	45–54.	https://doi.org/10.1080/2150704X.2012.687471	Lin,	Jinyao,	Liu,	Xiaoping,	Li,	Kai,	&	Li,	Xia.	(2014).	A	maximum	entropy	method	to	extract	urban	land	by	combining	MODIS	reflectance,	MODIS	NDVI,	and	DMSP-OLS	data.	International	Journal	of	Remote	Sensing,	35(18),	6708–6727.	https://doi.org/10.1080/01431161.2014.960623	Liu,	Jiyuan,	Zhan,	Jinyan,	&	Deng,	Xiangzheng.	(2005).	Spatio-temporal	Patterns	and	Driving	Forces	of	Urban	Land	Expansion	in	China	during	the	Economic	Reform	Era.	AMBIO:	A	Journal	of	the	Human	Environment,	34(6),	450–455.	https://doi.org/10.1579/0044-7447-34.6.450	Liu,	Ting,	&	Yang,	Xiaojun.	(2015).	Monitoring	land	changes	in	an	urban	area	using	satellite	imagery,	GIS	and	landscape	metrics.	Applied	Geography,	56,	42–54.	https://doi.org/10.1016/j.apgeog.2014.10.002	Liu,	Zhifeng,	He,	Chunyang,	Zhang,	Qiaofeng,	Huang,	Qingxu,	&	Yang,	Yang.	(2012).	Extracting	the			101	dynamics	of	urban	expansion	in	China	using	DMSP-OLS	nighttime	light	data	from	1992	to	2008.	Landscape	and	Urban	Planning,	106(1),	62–72.	https://doi.org/10.1016/j.landurbplan.2012.02.013	Lo,	Fu-chen,	&	Marcotullio,	Peter	J.	(2000a).	Globalisation	and	Urban	Transform	ations	in	the	Asia-Pacific	Region :	A	Review.	Urban	Studies,	37(1),	77–111.	Lo,	Fu-chen,	&	Marcotullio,	Peter	J.	(2000b).	Globalisation	and	Urban	Transform	ations	in	the	Asia	±	Paci	®	c	Region :	A	Review,	37(1),	77–111.	Lu,	Ming,	&	Chen,	Zhao.	(2004).	Urbanization,	Urban-Biased	Economic	Policies	and	Urban-Rural	Inequality	[J].	Economic	Research	Journal,	6(5),	50–58.	Luck,	Gary	W.,	Smallbone,	Lisa	T.,	&	O’Brien,	Rachel.	(2009).	Socio-Economics	and	Vegetation	Change	in	Urban	Ecosystems:	Patterns	in	Space	and	Time.	Ecosystems,	12(4),	604–620.	https://doi.org/10.1007/s10021-009-9244-6	Lyon,	John	G.,	Yuan,	Ding,	Lunetta,	Ross	S.,	&	Elvidge,	Chris	D.	(1998).	A	change	detection	experiment	using	vegetation	indices.	Photogrammetric	Engineering	and	Remote	Sensing,	64(2),	143–150.	Ma,	Ting,	Zhou,	Chenghu,	Pei,	Tao,	Haynie,	Susan,	&	Fan,	Junfu.	(2012).	Quantitative	estimation	of	urbanization	dynamics	using	time	series	of	DMSP/OLS	nighttime	light	data:	A	comparative	case	study	from	China’s	cities.	Remote	Sensing	of	Environment,	124,	99–107.	https://doi.org/10.1016/j.rse.2012.04.018	Maechler,	Martin,	Rousseeuw,	Peter,	Struyf,	Anja,	Hubert,	Mia,	&	Hornik,	Kurt.	(2012).	Cluster:	cluster	analysis	basics	and	extensions.	R	Package	Version.	Major,	D.	J.,	Baret,	F.,	&	Guyot,	G.	(1990).	A	ratio	vegetation	index	adjusted	for	soil	brightness.	International	Journal	of	Remote	Sensing,	11(5),	727–740.	Mann,	Henry	B.	..	(1945).	Nonparametric	Tests	Against	Trend.	The	Econometric	Society,	13(3),	245–259.	Marconcini,	Mattia,	Metz,	Annekatrin,	Esch,	Thomas,	&	Zeidler,	Julian.	(2014).	GLOBAL	URBAN	GROWTH	MONITORING	BY	MEANS	OF	SAR	DATA.	German	Aerospace	Center,	(April	2012),	1477–1480.	Marshall,	Richard.	(2013).	Emerging	urbanity:	global	urban	projects	in	the	Asia	Pacific	Rim.	Routledge.	Martinus,	Kirsten,	&	Tonts,	Matthew.	(2015).	Powering	the	world	city	system:	energy	industry	networks	and	interurban	connectivity.	Environment	and	Planning	A,	47(7),	1502–1520.	Masek,	J.	G.,	Lindsay,	F.	E.,	&	Goward,	S.	N.	(2000).	Dynamics	of	urban	growth	in	the	Washington	DC	metropolitan	area,	1973-1996,	from	Landsat	observations.	International	Journal	of	Remote	Sensing,	21(18),	3473–3486.	https://doi.org/10.1080/014311600750037507	McKinney,	Michael	L.	(2002).	Urbanization,	Biodiversity,	and	Conservation:	The	impacts	of	urbanization	on	native	species	are	poorly	studied,	but	educating	a	highly	urbanized	human	population	about	these	impacts	can	greatly	improve	species	conservation	in	all	ecosystems.	BioScience,	52(10),	883–890.	https://doi.org/10.1641/0006-3568(2002)052[0883:UBAC]2.0.CO;2	McKinney,	Michael	L.	(2006).	Urbanization	as	a	major	cause	of	biotic	homogenization.	Biological	Conservation,	127(3),	247–260.	https://doi.org/10.1016/j.biocon.2005.09.005	Michishita,	Ryo,	Jiang,	Zhiben,	&	Xu,	Bing.	(2012).	Monitoring	two	decades	of	urbanization	in	the	Poyang	Lake	area,	China	through	spectral	unmixing.	Remote	Sensing	of	Environment,	117,	3–18.			102	https://doi.org/10.1016/j.rse.2011.06.021	Mieg,	Harald	A.,	&	Töpfer,	Klaus.	(2013).	8	Privatization,	urban	fragmentation,	and	sustainable	development.	In	Institutional	and	Social	Innovation	for	Sustainable	Urban	Development	(p.	130).	Mike	Jenks,	Daniel	Kozak,	Pattaranan	Takkanon.	(2008).	World	Cities	and	Urban	Form:	Fragmented,	Polycentric,	Sustainable?	Montgomery,	Mark	R.	(2008).	The	Urban	Transformation	of	the	Developing	World.	Science,	761(February),	761–765.	https://doi.org/10.1126/science.1153012	Mozumder,	Pallab,	&	Marathe,	Achla.	(2007).	Causality	relationship	between	electricity	consumption	and	GDP	in	Bangladesh.	Energy	Policy,	35(1),	395–402.	https://doi.org/10.1016/j.enpol.2005.11.033	Murakami,	Akinobu,	Medrial	Zain,	Alinda,	Takeuchi,	Kazuhiko,	Tsunekawa,	Atsushi,	&	Yokota,	Shigehiro.	(2005).	Trends	in	urbanization	and	patterns	of	land	use	in	the	Asian	mega	cities	Jakarta,	Bangkok,	and	Metro	Manila.	Landscape	and	Urban	Planning,	70(3–4),	251–259.	https://doi.org/10.1016/J.LANDURBPLAN.2003.10.021	Ng,	Mee	Kam.	(2010).	International	Planning	Studies	Sustainable	Urban	Development	Issues	in	Chinese	Transitional	Cities :	Hong	Kong	and	Shenzhen	Sustainable	Urban	Development	Issues	in	Chinese	Transitional	Cities :	Hong	Kong	and	Shenzhen,	3475(October	2012),	37–41.	https://doi.org/10.1080/1356347022011258	Nicolaou,	N.,	Siddique,	N.,	&	Custovic,	A.	(2005).	Allergic	disease	in	urban	and	rural	populations:	increasing	prevalence	with	increasing	urbanization.	Allergy,	60(11),	1357–1360.	Nowak,	David	J.	(2002).	The	effects	of	urban	trees	on	air	quality.	USDA	Forest	Service,	96–102.	Nowak,	David	J.,	Crane,	Daniel	E.,	&	Stevens,	Jack	C.	(2006).	Air	pollution	removal	by	urban	trees	and	shrubs	in	the	United	States.	Urban	Forestry	and	Urban	Greening,	4(3–4),	115–123.	https://doi.org/10.1016/j.ufug.2006.01.007	Nowak,	David	J.,	&	Dwyer,	John	F.	(2007).	Understanding	the	Benefits	and	Costs	of	Urban	Forest	Ecosystems.	Handbook	of	Urban	and	Community	Forestry	in	the	North	East,	25–46.	https://doi.org/10.1007/978-1-4020-4289-8{_}2	Ohlin,	Berth.	(1952).	Interregional	And	International	Trade.	Vol.	39.	Harvard	University	Press.;	Cambridge.	Oke,	Timothy	R.	(1982).	The	energetic	basis	of	the	urban	heat	island.	Quarterly	Journal	of	the	Royal	Meteorological	Society,	108(455),	1–24.	Panayotou,	Theodore.	(1997).	Demystifying	the	environmental	Kuznets	curve:	turning	a	black	box	into	a	policy	tool.	Environment	and	Development	Economics,	2(4),	S1355770X97000259.	https://doi.org/10.1017/S1355770X97000259	Pandey,	Bhartendu,	Joshi,	P.	K.,	&	Seto,	Karen	C.	(2013).	Monitoring	urbanization	dynamics	in	India	using	DMSP/OLS	night	time	lights	and	SPOT-VGT	data.	International	Journal	of	Applied	Earth	Observation	and	Geoinformation,	23,	49–61.	https://doi.org/https://doi.org/10.1016/j.jag.2012.11.005	Pandey,	Bhartendu,	Zhang,	Qingling,	&	Seto,	Karen	C.	(2017).	Comparative	evaluation	of	relative	calibration	methods	for	DMSP/OLS	nighttime	lights.	Remote	Sensing	of	Environment,	195,	67–78.			103	https://doi.org/10.1016/j.rse.2017.04.011	Pettorelli,	Nathalie,	Vik,	Jon	Olav,	Mysterud,	Atle,	Gaillard,	Jean-Michel,	Tucker,	Compton	J.,	&	Stenseth,	Nils	Chr.	(2005).	Using	the	satellite-derived	NDVI	to	assess	ecological	responses	to	environmental	change.	Trends	in	Ecology	&	Evolution,	20(9),	503–510.	https://doi.org/10.1016/J.TREE.2005.05.011	Phinn,	Stuart,	Stanford,	M.,	Scarth,	P.,	Murray,	A.	T.,	&	Shyy,	P.	T.	(2002).	Monitoring	the	composition	of	urban	environments	based	on	the	vegetation-impervious	surface-soil	(VIS)	model	by	subpixel	analysis	techniques.	International	Journal	of	Remote	Sensing,	23(20),	4131–4153.	Pickett,	Steward	T.	A.,	&	Zhou,	Weiqi.	(2015).	Global	urbanization	as	a	shifting	context	for	applying	ecological	science	toward	the	sustainable	city.	Ecosystem	Health	and	Sustainability,	1(1),	art5-art5.	https://doi.org/10.1890/EHS14-0014.1	Quigley,	Martin	F.	(2002).	Franklin	Park:	150	years	of	changing	design,	disturbance,	and	impact	on	tree	growth.	Urban	Ecosystems,	6(3),	223–235.	Quigley,	Martin	F.	(2004).	Street	trees	and	rural	conspecifics:	will	long-lived	trees	reach	full	size	in	urban	conditions?	Urban	Ecosystems,	7(1),	29–39.	Rath,	T.	M.,	&	Manmatha,	R.	(2003).	Word	image	matching	using	dynamic	time	warping.	2003	IEEE	Computer	Society	Conference	on	Computer	Vision	and	Pattern	Recognition,	2003.	Proceedings.,	2,	II-521-II-527.	https://doi.org/10.1109/CVPR.2003.1211511	Ridd,	Merrill	K.	(1995).	Exploring	a	VIS	(vegetation-impervious	surface-soil)	model	for	urban	ecosystem	analysis	through	remote	sensing:	comparative	anatomy	for	cities.	International	Journal	of	Remote	Sensing,	16(12),	2165–2185.	Roberts,	D.	A.,	Gardner,	M.,	Church,	R.,	Ustin,	S.,	Scheer,	G.,	&	Green,	R.	O.	(1998).	Mapping	Chaparral	in	the	Santa	Monica	Mountains	Using	Multiple	Endmember	Spectral	Mixture	Models.	Remote	Sensing	of	Environment,	65(3),	267–279.	https://doi.org/10.1016/S0034-4257(98)00037-6	Román,	Miguel	O.,	Wang,	Zhuosen,	Sun,	Qingsong,	Kalb,	Virginia,	Miller,	Steven,	Molthan,	Andrew,	Schultz,	Lori,	Bell,	Jordan,	Stokes,	Eleanor,	Pandey,	Bhartendu,	Seto,	Karen,	Hall,	Dorothy,	Oda,	Tomohiro,	Wolfe,	Robert,	Lin,	Gary,	Golpayegani,	Navid,	Devadiga,	Sadashiva,	Davidson,	Carol,	et	al.	(2018).	NASA’s	Black	Marble	Nighttime	Product	Suite.	Remote	Sensing	of	Enviroment,	Revision	S(November	2017),	113–143.	https://doi.org/10.1016/j.rse.2018.03.017	Rottensteiner,	Franz,	&	Briese,	Christian.	(2002).	A	new	method	for	building	extraction	in	urban	areas	from	high-resolution	LIDAR	data.	International	Archives	of	Photogrammetry	Remote	Sensing	and	Spatial	Information	Sciences,	34(3/A),	295–301.	Rousseeuw,	Peter	J.	(1987).	Silhouettes:	A	graphical	aid	to	the	interpretation	and	validation	of	cluster	analysis.	Journal	of	Computational	and	Applied	Mathematics,	20(C),	53–65.	https://doi.org/10.1016/0377-0427(87)90125-7	Salvador,	Stan,	&	Chan,	Philip.	(2007).	Toward	accurate	dynamic	time	warping	in	linear	time	and	space.	Intelligent	Data	Analysis,	11(5),	561–580.	Satterthwaite,	David.	(2009).	The	implications	of	population	growth	and	urbanization	for	climate	change.	Environment	and	Urbanization,	21(2),	545–567.	https://doi.org/10.1177/0956247809344361			104	Schaltegger,	Stefan,	&	Synnestvedt,	Terje.	(2002).	The	link	between	“green”	and	economic	success:	environmental	management	as	the	crucial	trigger	between	environmental	and	economic	performance.	Journal	of	Environmental	Management,	65(4),	339–346.	https://doi.org/https://doi.org/10.1006/jema.2002.0555	Schmidt,	Gail,	Jenkerson,	Calli,	Masek,	Jeffrey,	Vermote,	Eric,	&	Gao,	Feng.	(2013).	Landsat	ecosystem	disturbance	adaptive	processing	system	(LEDAPS)	algorithm	description.	Schneider,	A.	(2012).	Monitoring	land-cover	change	in	urban	and	peri-urban	areas	using	dense	time	stacks	of	Landsat	satellite	data	and	a	data	mining	approach.	Remote	Sensing	of	Environment,	124,	689–704.	https://doi.org/10.1016/j.rse.2012.06.006	Schneider,	A.,	Friedl,	Mark	A.,	&	Potere,	D.	(2009).	A	new	map	of	global	urban	extent	from	MODIS	satellite	data.	Environmental	Research	Letters,	4(4),	44003.	https://doi.org/10.1088/1748-9326/4/4/044003	Schneider,	A.,	&	Woodcock,	C.	E.	(2008).	Compact,	Dispersed,	Fragmented,	Extensive?	A	Comparison	of	Urban	Growth	in	Twenty-five	Global	Cities	using	Remotely	Sensed	Data,	Pattern	Metrics	and	Census	Information.	Urban	Studies,	45(3),	659–692.	https://doi.org/10.1177/0042098007087340	Secord,	John,	&	Zakhor,	Avideh.	(2006).	Tree	detection	in	aerial	lidar	and	image	data.	Proceedings	-	International	Conference	on	Image	Processing,	ICIP,	4(2),	2317–2320.	https://doi.org/10.1109/ICIP.2006.312850	Selden,	Thomas	M.,	&	Song,	Daqing.	(1994).	Environmental	Quality	and	Development:	Is	There	a	Kuznets	Curve	for	Air	Pollution	Emissions?	Journal	of	Environmental	Economics	and	Management,	27(2),	147–162.	https://doi.org/https://doi.org/10.1006/jeem.1994.1031	Sen,	Pranab	Kumar.	(1968).	Estimates	of	the	Regression	Coefficient	Based	on	Kendall’s	Tau.	Journal	of	the	American	Statistical	Association,	63(324),	1379–1389.	https://doi.org/10.1080/01621459.1968.10480934	Seto,	Karen	C.,	&	Fragkias,	Michail.	(2005).	Quantifying	spatiotemporal	patterns	of	urban	land-use	change	in	four	cities	of	China	with	time	series	landscape	metrics.	Landscape	Ecology,	20(7),	871–888.	https://doi.org/10.1007/s10980-005-5238-8	Seto,	Karen	C.,	&	Kaufmann,	Robert	K.	(2003).	Modeling	the	Drivers	of	Urban	Land	Use	Change	in	the	Pearl	River	Delta,	China:	Integrating	Remote	Sensing	with	Socioeconomic	Data.	Land	Economics,	79(1),	106–121.	https://doi.org/10.3368/le.79.1.106	Seto,	Karen	Ching-Yee.	(2014).	Rethinking	Global	Land	Use	in	an	Urban	Era.	Rethinking	global	land	use	in	an	urban	era.	The	MIT	Press.	https://doi.org/10.7551/mitpress/9780262026901.001.0001	Sexton,	Joseph	O.,	Song,	Xiao	Peng,	Huang,	Chengquan,	Channan,	Saurabh,	Baker,	Matthew	E.,	&	Townshend,	John	R.	(2013).	Urban	growth	of	the	Washington,	D.C.-Baltimore,	MD	metropolitan	region	from	1984	to	2010	by	annual,	Landsat-based	estimates	of	impervious	cover.	Remote	Sensing	of	Environment,	129,	42–53.	https://doi.org/10.1016/j.rse.2012.10.025	Shafik,	Nemat.	(1994).	Economic	Development	and	Environmental	Quality :	An	Econometric	Analysis	Source.	Oxford	Economic	Papers,	46(Special	Issue	on	Environmental	Economics),	757–773.	Retrieved	from	http://www.jstor.org/stable/2663498	Shelley,	Louise	I.	(1981).	Crime	and	modernization:	The	impact	of	industrialization	and	urbanization	on			105	crime.	Southern	Illinois	University	Press	Carbondale.	Shen,	Li,	Guo,	Xulin,	Xiao,	Kang,	&	Access,	Open.	(2015).	Spatiotemporally	characterizing	urban	temperatures	based	on	remote	sensing	and	GIS	analysis :	a	case	study	in	the	city	of	Saskatoon,	27–39.	https://doi.org/10.1515/geo-2015-0005	Singh,	Kunwar	K.,	Vogler,	John	B.,	Shoemaker,	Douglas	A.,	&	Meentemeyer,	Ross	K.	(2012).	LiDAR-Landsat	data	fusion	for	large-area	assessment	of	urban	land	cover:	Balancing	spatial	resolution,	data	volume	and	mapping	accuracy.	ISPRS	Journal	of	Photogrammetry	and	Remote	Sensing,	74,	110–121.	https://doi.org/10.1016/j.isprsjprs.2012.09.009	Small,	Christopher.	(2003).	High	spatial	resolution	spectral	mixture	analysis	of	urban	reflectance.	Remote	Sensing	of	Environment,	88(1–2),	170–186.	https://doi.org/10.1016/j.rse.2003.04.008	Small,	Christopher,	&	Elvidge,	Christopher	D.	(2013).	Night	on	Earth:	Mapping	decadal	changes	of	anthropogenic	night	light	in	Asia.	International	Journal	of	Applied	Earth	Observation	and	Geoinformation,	22,	40–52.	https://doi.org/10.1016/j.jag.2012.02.009	Small,	Christopher,	Elvidge,	Christopher	D.,	Balk,	Deborah,	&	Montgomery,	Mark.	(2011).	Spatial	scaling	of	stable	night	lights.	Remote	Sensing	of	Environment,	115(2),	269–280.	https://doi.org/10.1016/j.rse.2010.08.021	Small,	Christopher,	&	Lu,	Jacqueline	W.	T.	(2006).	Estimation	and	vicarious	validation	of	urban	vegetation	abundance	by	spectral	mixture	analysis.	Remote	Sensing	of	Environment,	100(4),	441–456.	https://doi.org/10.1016/j.rse.2005.10.023	Small,	Christopher,	Pozzi,	Francesca,	&	Elvidge,	C.	D.	(2005).	Spatial	analysis	of	global	urban	extent	from	DMSP-OLS	night	lights.	Remote	Sensing	of	Environment,	96(3–4),	277–291.	https://doi.org/10.1016/j.rse.2005.02.002	Stern,	David	I.	(1998).	Progress	on	the	Environmental	Kuznets	Curve?	Environment	and	Development	Economics,	3(2),	173–196.	https://doi.org/doi:null	Stern,	David	I.	(2004).	The	Rise	and	Fall	of	the	Environmental	Kuznets	Curve.	World	Development,	32(8),	1419–1439.	https://doi.org/10.1016/j.worlddev.2004.03.004	Stern,	David	I.,	Common,	Michael	S.,	&	Barbier,	Edward	B.	(1996).	Economic	growth	and	environmental	degradation:	The	environmental	Kuznets	curve	and	sustainable	development.	World	Development,	24(7),	1151–1160.	https://doi.org/10.1016/0305-750X(96)00032-0	Sutton,	Paul	C.,	Elvidge,	Christopher	D.,	Ghosh,	Tilottama,	&	others.	(2007).	Estimation	of	gross	domestic	product	at	sub-national	scales	using	nighttime	satellite	imagery.	International	Journal	of	Ecological	Economics	&	Statistics,	8(S07),	5–21.	Sutton,	Paul,	Roberts,	Dar,	Elvidge,	Christopher,	&	Baugh,	Kimberly.	(2001).	Census	from	Heaven:	An	estimate	of	the	global	human	population	using	night-time	satellite	imagery.	International	Journal	of	Remote	Sensing,	22(16),	3061–3076.	Swetnam,	R.	D.,	Fisher,	B.,	Mbilinyi,	B.	P.,	Munishi,	P.	K.	T.,	Willcock,	S.,	Ricketts,	T.,	Mwakalila,	S.,	Balmford,	A.,	Burgess,	N.	D.,	Marshall,	A.	R.,	&	Lewis,	S.	L.	(2011).	Mapping	socio-economic	scenarios	of	land-cover	change:	A	GIS	method	to	enable	ecosystem	service	modelling.	Journal	of	Environmental	Management,	92(3),	563–574.	https://doi.org/10.1016/J.JENVMAN.2010.09.007			106	Theil,	Henri.	(1992).	A	Rank-Invariant	Method	of	Linear	and	Polynomial	Regression	Analysis	(pp.	345–381).	Springer,	Dordrecht.	https://doi.org/10.1007/978-94-011-2546-8_20	Tian,	Yichen,	Yin,	Kai,	Lu,	Dengsheng,	Hua,	Lizhong,	Zhao,	Qianjun,	&	Wen,	Meiping.	(2014).	Examining	Land	Use	and	Land-cover	Spatiotemporal	Change	and	Driving	Forces	in	Beijing	from	1978	to	2010.	Remote	Sensing,	6(11),	10593–10611.	https://doi.org/10.3390/rs61110593	Todd,	Stella	W.,	&	Hoffer,	Roger	M.	(1998).	Responses	of	spectral	indices	to	variations	in	vegetation	cover	and	soil	background.	Photogrammetric	Engineering	and	Remote	Sensing,	64,	915–922.	Tooke,	Thoreau	Rory,	Coops,	Nicholas	C.,	Goodwin,	Nicholas	R.,	&	Voogt,	James	A.	(2009).	Extracting	urban	vegetation	characteristics	using	spectral	mixture	analysis	and	decision	tree	classifications.	Remote	Sensing	of	Environment,	113(2),	398–407.	https://doi.org/10.1016/j.rse.2008.10.005	Tyrväinen,	Liisa,	Pauleit,	Stephan,	Seeland,	Klaus,	&	De	Vries,	Sjerp.	(2005).	Benefits	and	uses	of	urban	forests	and	trees.	In	Urban	Forests	and	Trees:	A	Reference	Book	(pp.	81–114).	https://doi.org/10.1007/3-540-27684-X_5	UNDESA.	(2014).	P.D.	World	Urbanization	Prospects	the	2014	Revision,	Highlights.	New	York.	UNDESA.	(2017).	P.D.	World	Population	Prospects	the	2017	Revision,	Key	Findings.	New	York.	UNESCAP.	(2012).	Statistical	Yearbook	for	Asia	and	the	Pacific	2012.	UNESCO.	(2014).	Asia	and	the	Pacific.	United	Nations.	(2002).	World	Urbanization	Prospects.	The	2001	Revision	Data	Tables	and	Highlights.	World	Urbanization	Prospects:	The	2003	Revision,	(March),	1–195.	https://doi.org/ESA/P/	WP/173	Van	Der	Meer,	F.,	&	De	Jong,	S.	M.	(2000).	Improving	the	results	of	spectral	unmixing	of	Landsat	Thematic	Mapper	imagery	by	enhancing	the	orthogonality	of	end-members.	International	Journal	of	Remote	Sensing,	21(15),	2781–2797.	https://doi.org/10.1080/01431160050121249	Varshney,	Avnish.	(2013).	Improved	NDBI	differencing	algorithm	for	built-up	regions	change	detection	from	remote-sensing	data:	An	automated	approach.	Remote	Sensing	Letters,	4(5),	504–512.	https://doi.org/10.1080/2150704X.2013.763297	Venter,	Oscar,	Sanderson,	Eric	W.,	Magrach,	Ainhoa,	Allan,	James	R.,	Beher,	Jutta,	Jones,	Kendall	R.,	Possingham,	Hugh	P.,	Laurance,	William	F.,	Wood,	Peter,	Fekete,	Balázs	M.,	Levy,	Marc	A.,	&	Watson,	James	E.	M.	(2016).	Sixteen	years	of	change	in	the	global	terrestrial	human	footprint	and	implications	for	biodiversity	conservation.	Nature	Communications,	7,	1–11.	https://doi.org/10.1038/ncomms12558	Vikhamar,	Dagrun,	&	Solberg,	Rune.	(2003).	Snow-cover	mapping	in	forests	by	constrained	linear	spectral	unmixing	of	MODIS	data.	Remote	Sensing	of	Environment,	88(3),	309–323.	https://doi.org/10.1016/J.RSE.2003.06.004	Wang,	Lanying,	Wang,	Shiqian,	Li,	Wei,	&	Li,	Jonathan.	(2014).	Monitoring	urban	expansion	of	the	Greater	Toronto	area	from	1985	to	2013	using	Landsat	images.	In	International	Geoscience	and	Remote	Sensing	Symposium	(IGARSS)	(pp.	4264–4267).	https://doi.org/10.1109/IGARSS.2014.6947431	Waqar,	Mirza	Muhammad,	Mirza,	Johum	Fatimah,	Mumtaz,	Rafia,	&	Hussain,	Ejaz.	(2012).	Development	of	New	Indices	for	Extraction	of	Built-Up	Area	&	Bare	Soil.	Open	Access,	1(1),	1–4.			107	https://doi.org/10.4172/scientificreports.13	Ward,	Douglas,	Phinn,	Stuart	R.,	&	Murray,	Alan	T.	(2000).	Monitoring	Growth	in	Rapidly	Urbanizing	Areas	Using	Remotely	Sensed	Data.	The	Professional	Geographer,	52(3),	371–386.	https://doi.org/10.1111/0033-0124.00232	Welch,	R.	(1980).	Monitoring	urban	population	and	energy	utilization	patterns	from	satellite	Data.	Remote	Sensing	of	Environment,	9(1),	1–9.	https://doi.org/10.1016/0034-4257(80)90043-7	Weng,	Qihao.	(2012).	Remote	sensing	of	impervious	surfaces	in	the	urban	areas:	Requirements,	methods,	and	trends.	Remote	Sensing	of	Environment,	117,	34–49.	https://doi.org/10.1016/j.rse.2011.02.030	Wentz,	Elizabeth	A.,	Anderson,	Sharolyn,	Fragkias,	Michail,	Netzband,	Maik,	Mesev,	Victor,	Myint,	Soe	W.,	Quattrochi,	Dale,	Rahman,	Atiqur,	&	Seto,	Karen	C.	(2014).	Supporting	global	environmental	change	research:	A	review	of	trends	and	knowledge	gaps	in	urban	remote	sensing.	Remote	Sensing,	6(5),	3879–3905.	https://doi.org/10.3390/rs6053879	Wheaton,	William	C.,	&	Shishido,	Hisanobu.	(1981).	Urban	Concentration,	Agglomeration	Economies,	and	the	Level	of	Economic	Development.	Economic	Development	and	Cultural	Change,	30(1),	17–30.	https://doi.org/10.1086/452537	White,	J.	C.,	Wulder,	M.	a.,	Hobart,	G.	W.,	Luther,	J.	E.,	Hermosilla,	T.,	Griffiths,	P.,	Coops,	N.	C.,	Hall,	R.	J.,	Hostert,	P.,	Dyk,	a.,	&	Guindon,	L.	(2014).	Pixel-based	image	compositing	for	large-area	dense	time	series	applications	and	science.	Canadian	Journal	of	Remote	Sensing,	(August	2014),	00–00.	https://doi.org/10.1080/07038992.2014.945827	Woodcock,	Curtis	E.,	Allen,	Richard,	Anderson,	Martha,	Belward,	Alan,	Bindschadler,	Robert,	Cohen,	Warren,	Gao,	Feng,	GOWARD,	SAMUEL	N.,	HELDER,	DENNIS,	HELMER,	EILEEN,	NEMANI,	RAMA,	OREOPOULOS,	LAZAROS,	SCHOTT,	JOHN,	THENKABAI,	PRASAD	S.,	VERMOTE,	ERIC	F.,	VOGELMANN,	JAMES,	WULDER,	MICHAEL	A.,	&	WYNNE,	RANDOLPH.	(2008).	Free	Access	to	Landsat	Imagery.	Science,	320(May),	1011–1012.	World	Population	Statistics.	(2013).	Shanghai	Population	2013.	Retrieved	from	http://www.worldpopulationstatistics.com/shanghai-population-2013/	Wu,	Peilin,	&	Tan,	Minghong.	(2012).	Challenges	for	sustainable	urbanization:	a	case	study	of	water	shortage	and	water	environment	changes	in	Shandong,	China.	Procedia	Environmental	Sciences,	13,	919–927.	https://doi.org/10.1016/j.proenv.2012.01.085	Wulder,	Michael	A.,	&	Coops,	Nicholas	C.	(2013).	Make	Earth	observations	open	access.	Nature,	2013(February),	2013–2014.	Wulder,	Michael	A.,	Hilker,	Thomas,	White,	Joanne	C.,	Coops,	Nicholas	C.,	Masek,	Jeffrey	G.,	Pflugmacher,	Dirk,	&	Crevier,	Yves.	(2015).	Virtual	constellations	for	global	terrestrial	monitoring.	Remote	Sensing	of	Environment,	170,	62–76.	https://doi.org/10.1016/j.rse.2015.09.001	Wulder,	Michael	A.,	White,	Joanne	C.,	Goward,	Samuel	N.,	Masek,	Jeffrey	G.,	Irons,	James	R.,	Herold,	Martin,	Cohen,	Warren	B.,	Loveland,	Thomas	R.,	&	Woodcock,	Curtis	E.	(2008).	Landsat	continuity:	Issues	and	opportunities	for	land-cover	monitoring.	Remote	Sensing	of	Environment,	112(3),	955–969.	https://doi.org/10.1016/j.rse.2007.07.004	Xian,	George,	&	Crane,	Mike.	(2006).	An	analysis	of	urban	thermal	characteristics	and	associated	land-		108	cover	in	Tampa	Bay	and	Las	Vegas	using	Landsat	satellite	data.	Remote	Sensing	of	Environment,	104(2),	147–156.	https://doi.org/10.1016/J.RSE.2005.09.023	Xu,	Xinliang,	&	Min,	Xibi.	(2013).	Quantifying	spatiotemporal	patterns	of	urban	expansion	in	China	using	remote	sensing	data.	Cities,	35,	104–113.	https://doi.org/10.1016/j.cities.2013.05.002	Yang,	Jun,	Huang,	Conghong,	Zhang,	Zhiyong,	&	Wang,	Le.	(2014).	The	temporal	trend	of	urban	green	coverage	in	major	Chinese	cities	between	1990	and	2010.	Urban	Forestry	&	Urban	Greening,	13(1),	19–27.	https://doi.org/10.1016/j.ufug.2013.10.002	Yang,	X.,	&	Lo,	C.	P.	(2002).	Using	a	time	series	of	satellite	imagery	to	detect	land	use	and	land-cover	changes	in	the	Atlanta,	Georgia	metropolitan	area.	International	Journal	of	Remote	Sensing,	23(9),	1775–1798.	https://doi.org/10.1080/01431160110075802	Yang,	Yetao,	Wong,	Louis	Ngai	Yuen,	Chen,	Chao,	&	Chen,	Tao.	(2014).	Using	multitemporal	Landsat	imagery	to	monitor	and	model	the	influences	of	landscape	pattern	on	urban	expansion	in	a	metropolitan	region.	Journal	of	Applied	Remote	Sensing,	8(1),	83639.	https://doi.org/10.1117/1.JRS.8.083639	Zha,	Y.,	Gao,	J.,	&	Ni,	S.	(2003).	Use	of	normalized	difference	built-up	index	in	automatically	mapping	urban	areas	from	TM	imagery.	International	Journal	of	Remote	Sensing,	24(3),	583–594.	https://doi.org/10.1080/01431160304987	Zhang,	Q.,	Pandey,	B.,	&	Seto,	K.	C.	(2016).	A	Robust	Method	to	Generate	a	Consistent	Time	Series	From	DMSP	/	OLS	Nighttime	Light	Data.	IEEE	Transactions	on	Geoscience	and	Remote	Sensing,	54(10),	5821–5831.	https://doi.org/10.1109/TGRS.2016.2572724	Zhang,	Qingling,	Schaaf,	Crystal,	&	Seto,	Karen	C.	(2013).	The	Vegetation	adjusted	NTL	Urban	Index:	A	new	approach	to	reduce	saturation	and	increase	variation	in	nighttime	luminosity.	Remote	Sensing	of	Environment,	129,	32–41.	https://doi.org/10.1016/j.rse.2012.10.022	Zhang,	Xiaoyang,	Friedl,	Mark	A.,	Schaaf,	Crystal	B.,	Strahler,	Alan	H.,	Hodges,	J.	C.	F.,	Gao,	Feng,	Reed,	Bradley	C.,	&	Huete,	Alfredo.	(2003).	Monitoring	vegetation	phenology	using	MODIS.	Remote	Sensing	of	Environment,	84(3),	471–475.	https://doi.org/10.1016/S0034-4257(02)00135-9	Zhao,	Shuqing,	Liu,	Shuguang,	&	Zhou,	Decheng.	(2016).	Prevalent	vegetation	growth	enhancement	in	urban	environment.	Proceedings	of	the	National	Academy	of	Sciences,	113(22),	6313–6318.	https://doi.org/10.1073/pnas.1602312113	Zheng,	Xinyu,	Yu,	Zhoulu,	Ao,	Weijiu,	Wang,	Youfu,	Tahmassebi,	Amir	Reza,	You,	Shucheng,	Deng,	Jinsong,	&	Wang,	Ke.	(2014).	Rural	impervious	surfaces	extraction	from	Landsat-8	imagery	and	rural	impervious	surface	index.	In	Thomas	J.	Jackson,	Jing	Ming	Chen,	Peng	Gong,	&	Shunlin	Liang	(Eds.),	SPIE	Asia	Pacific	Remote	Sensing	(Vol.	9260,	p.	926030).	https://doi.org/10.1117/12.2068807	Zhu,	Zhe,	&	Woodcock,	Curtis	E.	(2012).	Object-based	cloud	and	cloud	shadow	detection	in	Landsat	imagery.	Remote	Sensing	of	Environment,	118,	83–94.	https://doi.org/10.1016/j.rse.2011.10.028	Ziter,	Carly.	(2016).	The	biodiversity-ecosystem	service	relationship	in	urban	areas:	A	quantitative	review.	Oikos,	125(6),	761–768.	https://doi.org/10.1111/oik.02883				109	Appendix	1		110	Appendix	2				111	Appendix	3				112	Appendix	4		Panel	unit	root	tests	indicated	that	not	all	panel	data	sets	were	stationary	in	level.	TDN	was	stationary	in	level	(p	<	0.01)	as	indicated	by	both	LLC	and	IPS	procedures.	However,	the	test	results	were	mixed	for	other	socio-economic	panel	time	series.	TGDP_Total	contained	a	panel	unit	root	in	level	and	achieved	stationarity	at	1st	difference.		TPOP_Total		was	stationary	in	both	level	and	1st	using	IPS	procedure	but	contained	panel	unit	roots	in	levels	using	LLC	test.	Given	the	mixed	test	results,	I	used	only	the	1st	difference	panel	data	sets	for	subsequent	analysis.				 Level	 1st	difference	 Level	 1st	difference		 LLC	 LLC	 IPS	 IPS	TDN	 -2.94605	***	 -13.8317	***	 -4.26640	***	 -14.7862	***	TPOP_Total	 -0.15784	 -5.36286	***	 -2.65526	***	 -5.97486	***	TGDP_Total	 2.06186	 -8.51984	***	 -0.18361	 -8.16482	***	10%	(*),	5%	(**),	1%	(***)		Johansen	Fisher	co-integration	test	indicated	a	significant	presence	of	long-run	relationship	between	TDN,	TPOP_Total,	and	TGDP_Total	.	I	rejected	the	null	hypothesis	–	no	long-run	relationship,	at	the	1%	significance	level	for	all	pairs	of	panel	data	sets.	The	existence	of	a	long	term	equilibrium	allowed	for	testing	Granger	causality	between	nighttime	lights	and	socio-economic	changes.			Time	Series	Pairs	 H0:	number	of	cointegration	vectors	Fisher	statistic	(trace	test)	Fisher	statistics	(max-eigen	test)	TDN	~	TPOP_Total	 None	 140.6***	 128.3***	At	most	1	 55.27	 55.27	TGDP_Total	~	TPOP_	Total	None	 144.6***	 137.4***	At	most	1	 73.44	 73.44	TDN	~	TGDP_Total	 None	 202.0***	 178.3***	At	most	1	 66.83	 66.83	10%	level	(*),	5%	level	(**),	1%	level	(***)		

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0368676/manifest

Comment

Related Items