Ecological risk assessment based on land cover change: A case of Zanzibar-Tanzania, 2003-2027

10 10 ECOLOGICAL RISK ASSESSMENT BASED ON LAND COVER CHANGE: A CASE OF ZANZIBAR-TANZANIA, 2003-2027 Hassan Mohammed Oma i ECOLOGICAL RISK ASSESSMENT BASED ON LAND COVER CHANGE: A CASE OF ZANZIBAR-TANZANIA, 2003-2027 Disse a ion supe ised by Ped o da Cos a B i o Cab al, PHD P o esso , No a In o ma ion Managemen School Uni e si y o No a Lisbon, Po ugal Hanna Meye , PhD P o esso , Ins i u e o Geoin o ma ics Uni e si y o Muns e , Ge many Ca los G anell Canu , PhD P o esso , Ins i u e o New Imaging Technologies Uni e si y o Jaume I Cas ellon, Spain Feb ua y, 2020 ii DECLARATION OF ORIGINALITY I decla e ha he wo k desc ibed in his documen is my own and no om someone else. All he assis ance I ha e ecei ed om o he people is duly acknowledged and all he sou ces (published o no published) a e e e enced. This wo k has no been p e iously e alua ed o submi ed o NOVA In o ma ion Managemen School o elsewhe e. Lisbon, 17 h Feb ua y 2020 Hassan M. Oma iii ACKNOWLEDGEMENTS Fi s ly, I would like o acknowledge ha , i is only by he G ace o Allah he Almigh y ha I ha e made his a . The gi o li e, good heal h and sound mind ha has been g an ed has enable me o accomplish his wo k. I would like o hank all conso ium o E asmus Mundus Mas e 's p og am in Geospa ial Technologies o hei inancial and ma e ial suppo du ing all pe iod o my s udy. Special hanks a e due o my supe iso s P o . Ped o Cab al, P o . Ca los G anell Canu , and P o . Hanna Meye , who p o ided imme se suppo o ensu e my wo k con o ms o s anda ds. I would like o hank P o . D . Ma co Painho o con inuous hesis ollow up and encou agemen and ensu ing ha , his wo k is comple e a he igh ime and in he igh s anda ds. Finally, bu no leas , I wish o hank my daugh e , Haj a Hassan, my lo ely wi e Asha Sei , and my whole amily o hei pa ience, inspi a ion, and unde s anding du ing he en i e pe iod o my s udy. My iends and all ela i es o hei suppo , encou agemen and all hei con ibu ions o make his s udy success. i ECOLOGICAL RISK ASSESSMENT BASED ON LAND COVER CHANGE: A CASE OF ZANZIBAR-TANZANIA, 2003-2027 ABSTRACT Land use unde imp ope land managemen is a majo challenge in sub-Saha an A ica, and his has d as ically a ec ed ecological secu i y. Add essing en i onmen al impac s ela ed o his majo challenge equi es as e and mo e e icien planning s a egies ha a e based on measu ed in o ma ion on land-use pa e ns. This s udy was employed o access he ecological isk index o Zanziba using land co e change. We i s employed Random Fo es classi ie o classi y h ee Landsa images o Zanziba o he yea 2003, 2009 and 2018. And hen he land change modele was employed o simula e he land co e o Zanziba Ci y up o 2027 om land-use maps o 2009 and 2018 unde business-as-usual and o he wo al e na i e scena ios (conse a ion and ex eme scena io). Nex , he ecological isk index o Zanziba o each land co e was assessed based on he heo ies o landscape ecology and ecological isk model. The esul s show ha he buil -up a eas and a mland o Zanziba island ha e been inc eased cons an ly, while he na u al g assland and o es co e we e sh inking. The o es , ag icul u al and g assland ha e been highly agmen ed in o se e al small pa ches ela i e o he dec ease in hei pa ch a eas. On he o he hand, he ecological isk index o Zanziba island has appea ed o inc ease a a cons an a e and i he cu en end con inues his index will inc ease by up o 8.9% in 2027. In compa ing he h ee u u e scena ios he esul s show ha he ERI o he conse a ion scena io will inc ease by only 4.6% which is a leas 1.6% less compa ed o 6.2% o he business as usual, while he ex eme scena io will p o ide a high inc ease o ERI o up o 8.9%. This s udy will help au ho i ies o unde s and ecological p ocesses and land use dynamics o a ious land co e classes, along wi h p e en ing unmanaged g ow h and haphaza d de elopmen o in o mal housing and in as uc u e. KEYWORDS Ecological isk assessmen Ecosys ems se ices Land co e changes modelling Landscape ecological s a is ics (LECOS) Zanziba i ACRONYMS ANN A i icial Neu al Ne wo k BAU Business as Usual COLA Commission o Land CONSV Conse a ion Scena io DEM Digi al Ele a ion Model ERI Ecological Risk Index EXTRM Ex eme Scena io FAO Food and Ag icul u e O ganiza ion GDP G oss Domes ic P oduc LCM Land Change Modele LECOS Landscape ecological s a is ics LUCC Land Use Land Co e Change LULC Land Use Land Co e MCE Mul i C i e ia E alua ion MLP Mul i-laye Pe cep on NEMC Na ional En i onmen Managemen Council NBS Na ional Bu eau o S a is ics RGZ Re olu ion Go e nmen o Zanziba USGS Uni ed S a e Geological Su ey UTM Uni e sal T ans e se Me ca o WB Wo ld Bank WWF Wo ld Wide Fund o Na u e ii INDEX OF THE TEXT ACKNOWLEDGEMENTS ......................................................................................................... iii ABSTRACT ................................................................................................................................. i KEYWORDS ................................................................................................................................ ACRONYMS ............................................................................................................................... i INDEX OF TABLES ................................................................................................................... ix INDEX OF FIGURES ................................................................................................................... x 1. INTRODUCTION ................................................................................................................. 1 2. LITERATURE REVIEW ...................................................................................................... 4 2.1. Ecosys em se ices and hei bene i s ............................................................................ 4 2.2. The Cu en S a us o Ecosys em Se ices ..................................................................... 5 2.3. The Role o Remo e Sensing on Land use and Land Co e Change ............................. 6 2.4. Fu u e P o isions o Ecological Moni o ing h ough Land Change Modele ................ 8 3. STUDY AREA ...................................................................................................................... 8 4. DATA AND METHODS .................................................................................................... 11 4.1. Desc ip ion o he da a .................................................................................................. 11 4.1.1. Ex e nal a iables .................................................................................................. 12 4.2. Me hods ........................................................................................................................ 14 4.2.1. Image classi ica ion ............................................................................................... 14 4.2.2. Accu acy assessmen ............................................................................................. 15 4.3. Desc ip ions o he chosen scena ios ............................................................................ 16 4.3.1. Business as usual (2027BAU) ............................................................................... 16 4.3.2. Conse a ion scena io (2027CONSV) .................................................................. 16 4.3.3. Ex eme scena io (2027EXTM) ............................................................................ 16 4.4. P epa a ion o ex e nal (independen ) a iables ....................................................... 17 4.4.1. Fac o s ............................................................................................................... 17 4.4.2. Cons ain s ......................................................................................................... 19 4.5. Modelling Fu u e Scena ios wi h LCM ........................................................................ 20 4.5.1. Change Analysis .................................................................................................... 20 iii 4.5.2. T ansi ion po en ial ............................................................................................... 21 4.5.3. Change p edic ion ................................................................................................. 23 4.5.4. Model alida ion ................................................................................................... 24 4.6. Calcula ion o Ecological isk indices .......................................................................... 25 Ecological Risk index (ERI) ................................................................................................ 25 5. RESULTS ............................................................................................................................ 27 5.1. LULC changes 2003 - 2018 .............................................................................................. 27 5.2. Accu acy assessmen .................................................................................................... 28 5.3. LULC changes 2003 - 2018.......................................................................................... 30 5.4. Land Co e Change o u u e Scena ios ..................................................................... 31 5.4.1. Model alida ion esul s ........................................................................................ 31 5.4.2. Simula ed Maps o 2027 scena ios ...................................................................... 33 5.5. Ecological Risk Assessmen ......................................................................................... 35 5.5.1. Ecological Risk Indices o he yea 2003-2018 ................................................... 35 5.5.2. Ecological Risk Indices o 2027 scena ios .......................................................... 36 5.5.3. Dis ibu ion o ERI pe dis ic .............................................................................. 38 6. DISCUSSION ...................................................................................................................... 40 7. CONCLUSIONS AND RECOMMENDATIONS .............................................................. 42 Bibliog aphic Re e ences ............................................................................................................ 43 Annexes ................................................................................................................................... 49 Change analysis ....................................................................................................................... 49 P obabili y Ma ix.................................................................................................................... 49 5 Al e na i ely, he coas al ecosys em like mang o e o es , seag ass beds, and co al ee s, p o ide s ong suppo o hund eds o species o bo h ha d and so co als and ish as well as sea u les, c us aceans, and ma ine mammals (A S aeh , 2018). 2.2. The Cu en S a us o Ecosys em Se ices As (MA, 2005) sugges ed, ecosys em se ice bene i “is he bene i ha humans de i e om ecosys ems”. Bu in e e se, human ac i i ies ha e been coined as he main agen o a majo loses o des uc ion o global na u al ecosys ems (Chi, Zhang, Xie, & Wang, 2019). They cause a majo p essu e in ecosys em species, by ei he imp ope use o na u al esou ces, o due o inc easing in demands o ecosys em se ices in a pa icula socie y. In ha sense, he ph ase like “ecosys em se ice loss” is mo e connec ed wi h he cos s ha mus be paid o elie e he impac o ecosys em des uc ion on human p oduc ion and ea nings (Haden e al., 2019). Acco ding o Millennium Ecosys em Assessmen epo o 2005, he e is a massi e decline and h ea ening o ecosys em heal h which a e comp ises wi h he p o ision o ecosys em se ices upon which indi iduals, communi ies and en i e cul u es depend (O e peck e al., 2013; Rowland, 2019). This 2005 MA epo had p o ided an es ima ion o mo e han 60% o he ecosys em se ices which ha e been assessed ac oss he globe a e al eady being deg aded o a e used unsus ainably wi h he po en ial o become mo e deg aded in he i s hal o his cen u y. The d i e s like he apid inc ease in human popula ion, mining ac i i ies, subsequen g ow h in u baniza ion and in o mal se lemen s, oge he wi h o he h ea s, place eno mous s ess on he ecosys em hea h and hus impac s hei u u e sus ainabili y (Engel, 2012). Zanziba and Tanzania o example, ha e been epo ed o expe ience an ex ensi e decline in he s a e o i s en i onmen h ough loss o na u al habi a s and biodi e si y o mo e han h ee decades (A. I. Ali, 2016). Much o hese changes a e caused by he changes in LULC which a e in luenced by inc ease in human popula ion, economic ac i i ies and poo managemen o u ban de elopmen s (NEMC, 2006). Al e na i ely, he 2018 WWF’s li ing Plane epo , p o ide an es ima ion o up o 50% o he global eshwa e and we land habi a s which ha e al eady been los in he pas 30 yea s since 1970 due o a ious human ac i i ies and clima ic a ia ion (Wo ld Wide Fund Fo Na u e (WWF), 2018). These eshwa e and we land habi a s a e mo e essen ial in main aining ecological p ocesses, as well as p o iding an app op ia e wa e lows wi hin an en i e wa e ca chmen (Paul, 2013). Losing o des uc ing hese habi a s, would a ec he en i onmen al lows including educing he olume, iming and e en he quali y o wa e lows which a e e y 6 essen ial o he su i al o downs eam habi a s and species wi hin hose a eas (Paul, 2013). In case o o es ecosys ems, which appea o be he mos undamen al sou ces o se ices and global biodi e si y, hei abili y o main ain hei u u e sus ainabili y is po en ially h ea ened. The ecen epo o Food and Ag icul u e O ganiza ion, p o ide an es ima ion o abou i e million hec a e pe yea o global o es which ha e al eady been los be ween 2005 and 2015 (FAO, 2018). This es ima ion is e y wo se o he u u e de elopmen o biodi e si y and ecosys em sus ainabili y, as well as h ea co esponding o he eedback on clima ic sys em and ood secu i y (Ga zuglia, 2018). Wi hou p ope plan and managemen , he s a us o ecosys em se ices will con inue o ace unp eceden ed s ess and p essu e, ha could esul in hei deg ada ion and con e sion, he eby a ec ing hei sus ainabili y in bo h cu en and u u e gene a ions. 2.3. The Role o Remo e Sensing on Land use and Land Co e Change Remo e sensing is he science and o some ex en , he a o acqui ing in o ma ion abou he Ea h’s su ace wi hou ac ually being in con ac wi h i (Paul, 2013). In combina ion wi h GIS, Remo e Sensing (RS) p o ide a be e oppo uni y in s udying and analyzing he a ious o m o he ea h's in o ma ion (Paul, 2013). I p o ides ooms o iew he ea h phenomenon om he space, which enables he comp ehension o he cumula i e in luence o human ac i i ies on he ea h's su ace's na u al s a e. One o he majo applica ions o emo e sensing echniques is he classi ica ion o emo ely sensed da a o mul i- empo al images. These echniques ha e been conside ed as an ul ima e oo o a ious analysis and applica ions in he ield o emo e sensing (Mülle o á, Pe gl, & Pyšek, 2013). The undamen al objec i e o emo e sensing image classi ica ion is o di ide he emo ely sensed image in o a numbe o classes (pixel g oup) in o de o e ec i ely examine and asses wha changes ha e been occu ed in each o hose g oups (J. Xu, Feng, Zhao, Sun, & Zhu, 2019). Howe e , he classi ica ion p ocess o emo ely sensed images is some ime conside ed o be he mos complex ask in luenced by a ious ac o s such as a ailabili y o high-quali y images, ancilla y da a, p ope classi ica ion p ocedu e, and analy ical abili y o he esea che (Gao & Xu, 2015) . In mos cases, i appea s o be much ha de o iden i y which is he bes classi ie o be used in a pa icula s udy due o ei he he lack o p ope guidelines in selec ing he algo i hm o e en he lack o a ailabili y o app op ia e classi ica ion algo i hms o a pa icula band (Gao & Xu, 2015). In ha sense, many esea che s ha e made a g ea e o in p oposing he mos e ec i e classi ica ion me hods o imp o e classi ica ion accu acy and as well as hei esul s. 7 The mos popula emo e sensing classi ica ion me hods p oposed in di e en s udies a e supe ised, and unsupe ised classi ica ion. In supe ised classi ica ion, he image pixels a e classi ied by selec ing ep esen a i e samples o each land co e class (T aining Si es o A eas) and hen he classi ica ion algo i hm applies hese samples o he en i e image (Bha acha ya, Ca , & Pal, 2016). This p ocess is done in h ee majo s eps including selec ing aining a eas, gene a e he signa u e ile and classi y he image. In ano he case, in unsupe ised classi ica ion he images a e classi ied only by g ouping he pixels in o clus e s based on hei p ope ies i s , and hen he classi ica ion o each clus e wi h land co e classes a e done. In gene al, his classi ica ion me hod is conside ed as he mos basic echnique, since he e is no need o aining samples o classi y he image. I is a simple way o classi y he image wi h only wo majo s eps including gene a ing clus e s and assigning classes (Bha acha ya e al., 2016). All o hese classi ica ion me hods can be u he g ouped in ei he s a is ical me hods, decision ees, a i icial neu al ne wo ks, and o he a i icial in elligence algo i hms (Xie, Li, Xiao, & Peng, 2016). S a is ical me hods such as maximum likelihood classi ie and Bayesian classi ie wo ks in an assump ion ha he membe s o each class in each band ollow a no mal dis ibu ion in he ea u e space and calcula es he p obabili y ha a gi en pixel belongs o a speci ic class. In his pa icula me hod, he classi ica ion accu acy dec eases as he dimension o ea u e dec eases (Xie e al., 2016). The decision ee me hods like andom o es classi ie make no assump ions conce ning he dis ibu ion o he inpu ea u es and he e o e hey a e obus and e ec i e me hods in managing nonlinea ela ionships among he class membe (J. Xu e al., 2019). Howe e , he e ec i eness o he decision ee me hods is highly in luenced by he size o he ea u e space and hus a e no app op ia e me hods o da a wi h a high dimension o ea u e space (J. Xu e al., 2019). In he case o a i icial neu al ne wo ks (ANN) me hods p o e a powe ul pe o mance wi h a highe dimensional ea u e space compa ed o s a is ical classi ica ion me hods because hey a e dis ibu ion- ee (J. Xu e al., 2019). Howe e , he mos di icul ness o using ANN models is i s demands o equi es a signi ican amoun o aining da a and a conside able numbe o i e a i e aining p ocedu es o ensu e ha he models a e ained success ully. Al e na i ely, he compu a ion o ANN algo i hms can be ex ao dina ily complex (J. Xu e al., 2019). 8 2.4. Fu u e P o isions o Ecological Moni o ing h ough Land Change Modele Spa ial and dynamic changes in LULC o he ea h’s su ace ha e a majo in luence on he des uc ion and u u e p o ision o ecosys em sus ainabili y o bo h landscape s uc u e and agmen a ion o g een space (Tian, Jim, Tao, & Shi, 2011). These dynamic changes in LULC can esul in an inc ease in he le el o land abandonmen , s a a ion o e en mig a ion ou o he a ec ed egion (Y. Wang e al., 2018). In ha kind o si ua ion, unde s anding he dynamics and u u e land co e ends has o be conside ed as a majo componen in app o al s a egies o bo h he planning and implemen a ion o app op ia e ools in conse ing he deg aded land. A numbe o compu a ion algo i hms o moni o ing and p edic ing u u e land co e changes ha e been widely inc eased in ecen yea s. The models like Ma ko chain (MC) (Al-sha i & P adhan, 2014), a i icial neu al ne wo k (Pijanowski, B own, Shelli o, & Manik, 2001), cellula au oma a (Cla ke & Gaydos, 1998), cellula Au oma a-Ma ko model (A sanjani, Kainz, & Mousi and, 2011), bina y logis ic eg ession (A sanjani, Helbich, Kainz, & Boloo ani, 2012), and simila i y weigh ed ins ance-based machine lea ning algo i hm (Sange mano, Eas man, & Zhu, 2010) a e among he common used models o he u u e p edic ion and simula ion o changes in land co e . Land Change Modele (LCM) o example, was ound o be one among he e ec i e modeling ool which inco po a es CA-Ma ko chain based on a neu al ne wo k o p edic u u e land-use change (Eas man, 2006). This CA-Ma ko chain models inco po a ed in LCM a e ai ly simple and much powe ul in modeling he complex p ocess and changes in land use o planning pu poses (Eas man, 2006). I models he u u e land co e changes by making an assump ion ha , he p obabili y o sys em being in a ce ain ime can only be de e mined i i s p e ious s a e is known wi h he assump ion ha a es o change obse ed du ing he calib a ion pe iod (T1 o T2), will emain he same du ing he simula ion pe iod (T2 o T3). LCM does also p o ides he ools ha can simula e he u u e land-use changes and pa e ns and allowing es ing o al e na i e planning scena ios. Simula ing u u e land co e change unde wha called scena ios, can p o ide a be e unde s anding o a ious planning al e na i es, s imula ing policy discussions be ween en i onmen al and de elopmen goals, and he e o e helping he esponsible au ho i ies in designing he new policies and p ocedu es ha c ea e incen i es o ecological conse a ion (McKenzie, E., Rosen hal, A., Be nha d , J., Gi e z, E., Ko acs, K., Olwe o, N. and To , 2012). 9 3. STUDY AREA This s udy was conduc ed in Zanziba (Figu e 3.1), he small coas al island o abou 2,461 km² in o al, loca ed a 39° 05’ E o 39° 55’ E and 4° 45’S o 6° 30’S Wes e n Indian Ocean, in which ou ism and ag icul u al ac i i ies a e he majo economic sec o s con ibu ing abou 28% and 40% o i s GDP espec i ely (WB, 2019). Zanziba is a pa o he Uni ed Republic o Tanzania, and consis s o wo main islands which a e Unguja and Pemba and wi h almos 50 o he small isle s (Mye s, 2010; RGZ, 2008). These wo islands a e loca ed 40 km o he mainland coas o Eas A ica in he Indian Ocean (Mye s, 2010). The wo main islands a e 50 km apa and a e sepa a ed by he 700 me e deep Pemba channel. Howe e , he cu en s udy uses he e m Zanziba e e ing o he island o Unguja as mos o he people li e in his island (Unguja), and because o Zanziba ci y loca ed in i (A S aeh , 2018). Acco ding o 2018 Na ional Bu eau o S a is ics (NBS) census epo , he o e all popula ion o Zanziba island (include Pemba) is abou 1,348,776 people and i s annual g ow h inc ease in a e y apid a e. In he ecen yea s, he popula ion g ow h a e o Zanziba has been epo ed o g ow a a a e o 3.4 % annually (NBS, 2018). Na u al popula ion g ow h is conside ed as he main agen o his high apid a e, bu o he ac o s including g owing o ou ism indus y and economic in luence has also a ac ed a signi ican numbe o mig an s om Tanzania mainland and o he pa s o Eas A ica, making his island o be one among he highly popula ed islands in he wo ld. Zanziba has a opical clima e wi h ou dis inc seasons. “Kaskazi” ( he ho season, which is be ween Decembe and Feb ua y and associa ed wi h ei he li le o no ains), “Masika” which is he long ainy season om Ma ch – May, “Kipupwe” ( he cold season, wi h s ong winds be ween June and Sep embe ), and “Vuli” which is e y sho ainy season om Oc obe – Decembe . The annual ain all o island is anging om 1600 mm o Unguja and 1900 mm o Pemba espec i ely, and he ai empe a u es is anging be ween 29 and 32ºC on a e age. Zanziba island was o iginally o es ed, bu p essu es like popula ion inc ease, human habi a ion and clima e change and a iabili y ha e esul ed in widesp ead clea ing o he o es and ege a ion co e (M. Kukkonen & Käyhkö, 2014). 10 Figu e 3. 1: S udy A ea-Zanziba The land co e pa e ns a ound Zanziba island a e gene ally dis ibu ed unde wo di e en soil classes, namely he deep soil and he co al ag a eas. The deep soil a eas a e mainly a ibu ed by pe manen cul i a ion and o es , while he la e co al ag a e cha ac e ized by cul i a ion and consequen sc ublands (RGZ, 2008). Howe e , all o hese wo soil classes a e no s able in e ms o i s land co e and bio ope cha ac e is ics. The hilly deep soil a eas a e epo ed o be mo e ulne able o soil e osion, highly land use demands and simul aneously acing popula ion p essu e, which all o hese in luences he apid and cons an changes in land co e pa e ns and na u al esou ce (M. Kukkonen & Käyhkö, 2014; Mye s, 2010). Al e na i ely, shi ing cul i a ion cha ac e ized in co al ag a eas c ea e a cons an elemen o changes, which unde he p essu e o diminishing a ea due o comme cial and conse a ion land use may leads owa ds de e io a ion o ecosys ems and aluable na u al esou ces (Mye s, 2010). 11 4. DATA AND METHODS 4.1. Desc ip ion o he da a Da a om di e en sou ces we e used o achie e he objec i es o his s udy. A e y i s s ages, he emo e sensed sa elli e images o Zanziba island (Table 4. 1) a di e en ime we e ob ained om USGS websi e (www.ea hexplo e .usgs.go /), and hen compa ed based on hei di e en empo al phenomenon. In ha sense, h ee mul ispec al Landsa images o he yea 2003, 2009 and 2018 wi h spa ial esolu ion o 30 m we e downloaded om USGS websi e (Table 4.1). Ancilla y da a ep esen ing biophysical p ope ies ha we e conside ed o in luence land use changes we e also applied in his analysis. These ancilla y da a we e ob ained by i s downloading hei o iginal shape iles including Zanziba oad, p o ec ed o es , buildings and illages as well as coas al egion om (www.zansdi.en i onmen . i ). By means o digi al ele a ion model (DEM) which was ob ained om ( h p://openda a. cm d.o g ), we calcula e he dis ance o each o hese a iables o c ea e he biophysical p ope ies o dis ance o he oad, dis ance o he buildings, dis ance o he coas , dis ance o he o es and slope (Table 4.2). P o ec ed a eas including p o ec ed o es , go e nmen al ag icul u al ield, as well as open and es ic ed zones we e also used o ep esen he cons ain s o land co e changes in he s udy egion. These p o ec ed a eas da a we e ob ained by eclassi ying he 2012 Zanziba LULC map which was ob ained om Zanziba commission o land (COLA) o c ea e a Boolean map o 0 o p o ec ed a eas and 1 o all o he classes (Figu e 4.3). 12 Da ase Desc ip ion P ojec ion Resolu ion Sou ce Landsa 7, ETM+ Raw = 064, pa h = 166, Da e: 14/01/2003 WGS 1984, UTM Zone 37N 30 m www.ea hexplo e .usgs.go / Landsa 5 TM Raw = 064, pa h = 166, Da e: 1/07/2009 WGS 1984, UTM Zone 37N 30 m www.ea hexplo e .usgs.go / Landsa 7, ETM+ Raw = 064, pa h = 166, Da e: 18/07/2018 WGS 1984, UTM Zone 37N 30 m www.ea hexplo e .usgs.go / Digi al Ele a ion Model (DEM) SRTM Clipped om Tanzania Dem WGS 1984, UTM Zone 37N STRM 30 m h p://openda a. cm d.o g Table 4.1: Sa elli e Da a 4.1.1. Ex e nal a iables In modeling LULC change o a speci ic a ea, he a iables in luencing o con olling he land co e change on ha a ea should be p ope ly used as independen a iables, and he LU maps should be used as dependen a iables. The inclusion o independen a iables in land change modeling is e y essen ial since hese a iables ensu e he u u e land co e maps a e always p edic ed wi h exis ing pa e ns and d i e s o changes in ha egion (Bulley, & Fü s , 2017). In his s udy, he selec ion o independen a iables was based on e iews ela ed o simila models, esea ches conduc ed in o he de eloping coun ies as well as he local ci cums ances o Zanziba island. In ha sense, he a iables slope, dis ance o he oad, dis ance o he o es , dis ance o buildings, dis ance o he coas and p o ec ed a eas we e all selec ed as independen a iables o simula e he u u e land co e change o Zanziba ci y. The a iable “slope “was chosen o ep esen he biophysical condi ions o he s udy a ea. Acco ding o (Liu, 2009; A sanjani e al., 2013), he la opog aphy has much in luence on land co e changes han ugged opog aphy, and his heo y has al eady been used in many simila models wi h posi i e esul s (Zhao e al., 2014). “Dis ance o he oads” and “dis ance o he o es ” we e all connec ed o economic ci cums ances. The chosen o dis ance o he o es in economic ac o s is connec ed wi h he ac ha in Zanziba , he cus oms o con e ing o es ed land in o ei he buil -up a eas o a mland is one o he majo eason o land-use dynamics, and his is highly caused by he economic si ua ion o Zanziba popula ion (M. Kukkonen & Käyhkö, 2014). This a iable is also ela ed o local 13 spa ial policies o bidding any cons uc ions o de elopmen ac i i ies in p o ec ed o es (M. Kukkonen & Käyhkö, 2014). We ega d “dis ance o he coas ” and “dis ance o buildings” o be connec ed in he social ac o s. In his case, he a iable “dis ance o he coas ” con i ms he global end owa ds highe quan i ies o land-use changes in coas al egions (Ebada i, 2018). And o he second a iable “dis ance o he buildings” exp ess he local social and planning condi ions o Zanziba , in which he houses a e buil ela i ely close o each o he due o subdi isions o small landholdings. In o he case, his phenomenon is also ela ed o spa ial in e ac ions, p o ided ha he land-use changes a e highly a ac ed by simila changes in hei neighbo ing a eas (M. O. Kukkonen e al., 2018; Toble sFi s Lawo Geog aphy, n.d.). And in he case o he emaining a iable “p o ec ed a eas”, i has been used o ep esen he local spa ial policies ha ha e ei he obs uc ed o indi ec ly s imula ed land-use dynamics (M. O. Kukkonen e al., 2018). Da ase Desc ip ion P ojec ion Resolu ion Sou ce Dis - - oad Vec o con e ed o as e ; Euclidian dis ance o he oad WGS 1984, UTM Zone 37N Re-sampled o 30m www.zansdi.en i onmen . i (Commission o Land adminis a ion Zanziba ) Dis - - build Vec o con e ed o as e ; Euclidian dis ance o he building WGS 1984, UTM Zone 37N Re-sampled o 30m www.zansdi.en i onmen . i Dis - - o es Vec o con e ed o as e ; Euclidian dis ance o he building WGS 1984, UTM Zone 37N Re-sampled o 30m www.zansdi.en i onmen . i Dis - - coas Vec o con e ed o as e ; Euclidian dis ance o he beaches WGS 1984, UTM Zone 37N Re-sampled o 30m www.zansdi.en i onmen . i slope Slope calcula ed om digi al ele a ion model WGS 1984, UTM Zone 37N Re-sampled o 30m h p://openda a. cm d.o g P o ec ed a eas/ Cons ain Vec o con e ed o as e ; Reclassi y o ha e 0 o p o ec ed a eas and 1 o all o he classes WGS 1984, UTM Zone 37N Re-sampled o 30m www.zansdi.en i onmen . i (Commission o Land adminis a ion Zanziba Table 4. 2: Ex e nal Va iables 14 4.2. Me hods The o e all me hodology o his s udy was ca ied ou in o h ee majo s eps (Figu e 4.1). Fi s , he land use land co e classi ica ion was pe o med using andom o es classi ie o gene a e h ee LULC maps o 2003, 2009 and 2018. Nex he land co e maps o he u u e scena ios we e modeled using land change modele o he Te Se 18.3 so wa e. And inally, he ecological isk indices o each map and each dis ic o Zanziba we e calcula ed by i s compu ing he landscape me ics using LECOS plugin in QGIS 3.8, ollowed by applying he o mulas om a ious li e a u es o ERI assessmen (Xue e al., 2019; F. Zhang e al., 2018; X. Zhang e al., 2013). 4.2.1. Image classi ica ion P e-p ocessing o all h ee sa elli e images (2003, 2009 and 2018), Zanziba adminis a i e bounda y, oge he wi h all ancilla y da a Shape iles, was pe o med by using A cGIS 10.6. All da ase s we e p ojec ed o WGS 1984, UTM Zone 37S coo dina e sys em, ollowed by band combina ion, clipping o he s udy egion, and esampling o each one o hem o appea in he same spa ial esolu ion o 30 m. T aining Samples Shape ile o i e classes including buil -up, g assland, o es , we land and a mland we e gene a ed by digi izing each o he sa elli e images. Wi h hese gene a ed i e land Figu e 4. 1: Me hodological amewo k 21 4.5.2. T ansi ion po en ial A e inishing he i s s ep o change analysis, he nex s ep was o es he explana o y powe o each ex e nal a iable, as well as selec ing ansi ion sub-models ha a e ele an in he p edic ion o he u u e scena ios. The e o e, in his s age, i s , he C ame ’s powe o all ex e nal a iables we e es ed o iden i y all a iables wi h a po en ial explana o y powe . These C ame ’s powe a e anging om 0 – 1, in which he a iables wi h a leas 0.15 C ame ’s alue a e conside ed o be po en ial in land use changes, whe eas hose ha ing alue om 0.4 onwa d a e much po en ial (Eas man, 2006). The esul s o C ame ’s powe o each es ed a iable a e illus a ed in (Table 4.1) below. Explana o y Va iable O e all C ame ’s V alue Dis ance o buildings 0.2852 Dis ance o oad 0.1836 Dis ance o he coas al a ea 0.1653 Dis ance o he o es 0.1354 Slope 0.0725 Res ic ed a eas 0.1932 Table 4. 3: C ame 's powe o ex e nal a iables Due o he lowe C ame ’s alue, nega i e in luence on he modelling esul s, as well as less li e a u es in o ma ion ega ding o Slope issues in Zanziba , his a iable was excluded in analysis. In he same sense, he a iable dis ance o he o es was also emo ed om he analysis due i s lowe C ame ’s powe and i s nega i e in luence on he analysis accu acy. The e o e, only h ee ex e nal ac o s including dis ance o he building, dis ance o he oad and dis ance o he coas , and one cons ain map o each scena io (excep o ex eme scena io, whe e cons ain s we e no employed) we e used in his analysis. The nex s ep was o combine all selec ed a iables ( ac o s and cons ain ) oge he wi h ansi ions sub-models be ween 2003 and 2009 and employ Mul i-Laye Pe cep on (MLP) o gene a e ansi ion po en ial o 2018 land co e map p edic ion. Mul i-Laye pe cep on (MLP) is a eed o wa d a i icial neu al ne wo k (ANN) composed wi h one o mul iple laye s including inpu , ou pu and hidden laye s be ween inpu and ou pu laye . The e m eed- o wa d me hod implying ha he da a lows in one di ec ion om inpu o ou pu . 22 The main algo i hm o his model is compu ing he linea ou pu om nonlinea inpu s acco ding o he weigh s by using a nonlinea ac i a ion unc ion (Ga dne & Do ling, 1998). MLP employs a supe ised lea ning echnique called backp opaga ion o aining. The back- p opaga ion algo i hm composed o wo s eps which a e o wa d pass and backwa d pass. In he o wa d pass, ac i a ion ansmi s om inpu o ou pu laye which means ha he signal low mo es om he inpu laye h ough he hidden laye s o he ou pu laye . While in backwa d pass e o s p opaga ed om ou pu o hidden laye (Ga dne & Do ling, 1998). To model land-use ansi ions in LCM, ei he Logis ic Reg ession, SimWeigh o MLP neu al ne wo k can be applied. Howe e , in his s udy, he MLP neu al ne wo k was applied because o i s capabili y o unning mul iple ansi ions o up o 9 pe sub-model as well as explana o y a iables a once (Eas man, 2006). To achie e he be e accu acy on he esul s o Mul i-Laye Pe cep on only majo land co e ansi ions oge he wi h only he d i e ’s a iables wi h highe C ame ’s powe should be included in he ansi ion sub-model (Eas man, 2006). Based on his p inciple, oge he wi h he in o ma ion om li e a u es ega ding on s udy egion, only six majo ansi ions including g assland o buil -up, o es o buil -up, g assland o a mland, o es o g assland, o es o a mland and a mland o g assland we e selec ed in his s udy. Toge he wi h ou selec ed a iables, a Mul i-laye pe cep on was employed wi h an accu acy o 68.4% o gene a e he ansi ion po en ials (Figu e 4.5) ha we e hen used by Ma ko chain me hod o p edic he Land use map o he yea 2018. The same p ocedu es we e also applied o model he 2027 maps. 23 Figu e 4. 4: T ansi ion Po en ial 2003 - 2009 4.5.3. Change p edic ion The hi d s age in modeling u u e scena ios was o pe o m change p edic ion o a u u e speci ied da e o he alloca ion o land co e changes. In his s age, he de aul p ocedu e in LCM (Ma ko Chain analysis) was applied o de e mine he amoun o changes a a da e o 2018 based on his o ical land co e maps o 2003-2009. He e, he algo i hm de e mines exac ly how much land would be expec ed o ansi ion om he la e da e (2009) o he p edic ion da e (2018) based on a p ojec ion o he ansi ion po en ials maps ha was gene a ed and hen c ea es a ansi ion p obabili ies ile (Eas man, 2006). Wi h his ansi ion p obabili ies ile, wo p edic o s maps which a e so p edic ion map and a ha d p edic ion map we e gene a ed. While a so p edic ion map is a con inuous mapping o ulne abili y o change which p o ides an indica ion o he deg ee o which he a eas ha e he igh condi ions o p ecipi a e change in 2018, a ha d p edic ion map is an ac ual p ojec ed map o 2018, in which each pixel is assigned one land co e class; he class ha i is mos likely o become (Eas man, 2006). 24 4.5.4. Model alida ion To ensu e he p edic i e abili y and pe o mance o he model. Model alida ion and calib a ion mus be pe o med be o e s a ing he p edic ion o LU changes o he speci ied da e. Pe o ming alida ion o he model is a e y impo an s ep because he e iciency and usabili y o he model a e always depending on he ou pu o he alida ion ( an Vlie e al., 2016). Since se e al s udies ha e al eady p oposed he usage o he con usion ma ix (Congal on, 2001) in assessing he accu acy and pe o mance o he model, we also compu e he con usion ma ix by pe o ming c oss- abula ion o compa e he ac ual and simula ed maps o 2018. The con usion ma ix which was gene a ed a e pe o ming c oss- abula ion was hen used o calcula e he o e all accu acy, use accu acies, p oduce accu acies, and kappa coe icien . In addi ion o ha , wo o he kappa coe icien s we e also compu ed o gain mo e in o ma ion on he pe o mance o he model. The o he wo kappa coe icien s a e Khis o and Kloca ion. Acco ding o Se na, (2011), Khis o is used o measu e he simila i ies in he numbe o cells be ween simula ed and e e ence maps. In o he wo ds, his kappa is esponsible o measu ing quan i a i e simila i y be ween wo compa ed maps. In ano he case, Kloca ion is used o measu e he simila i ies in he spa ial dis ibu ion o classes bu does no di e en ia e be ween classes ha a e close o dis an and i is independen o he o al numbe o cells pe class (Se na, 2011). 25 4.6. Calcula ion o Ecological isk indices The inal s ep o ou analysis was o compu e ecological isk indices o each LULC map, as well as o each dis ic o Zanziba island and hen compa ing hei esul s. This s age was done in wo majo s eps. Fi s he landscape me ics (pa ch numbe and pa ch a ea) o each LULC maps was calcula ed using LECOS plugin in QGIS 3.8, and hen he o mulas desc ibed in sec ion 4.6.1 below we e used o calcula e ecological isk indices o each map. Ecological Risk index (ERI) P o ide an indica o s o he nega i e en i onmen al impac s ha may occu o a e occu ing due o one o mo e ex e nal ac o s (USEPA, 1992; Hakanson, 1980; Hunsake e al., 1990, X. Zhang e al., 2013) . Ma hema ically: 𝐸𝑅𝐼=∑𝐴𝑘𝑖 𝐴𝑘 𝑛 𝑖=1 𝑅𝑖 Whe e: ERI: The ecological isk index o he isk a ea Ak: is he o al a ea o k h egion Aki = is he i h landscape a ea/class a ea Ri: is he i h landscape/class loss index which can be calcula ed h ough o mula below. 𝑅𝑖=𝐹𝑖∗𝑆𝑖 Whe e: Fi, is he ecological agili y index which is e e ed as he abili y o a landscape o esis he human dis u bance. Acco ding o Zhang e al (2013), and he knowledge o ou s udy egion, he Fi alue o 5 o buil up, 4 o wa e body, 3 o g ass land, 2 o a m land and 1 o o es we e used in his s udy. Addi ionally, o an e ec i e esul , hese alues we e no malized be o e di ec ly being used in calcula ions o ecological isk index. Si is he landscape dis u bance deg ee index o i h landscape (Jin e al., 2019), which can also be calcula ed ia a o mula. 𝑆𝑖=𝑎𝐶𝑖+𝑏𝑁𝑖+𝑐𝐷𝑖 26 Whe e: Ci: Land scape agmen a ion, i e lec s he changes in landscape s uc u e and ecological p ocess, 𝐶𝑖=𝑛𝑖 𝐴𝑘𝑖 Ni: Land scape isola ion, e e s o he deg ee o sepa a ion o pa ches in a gi en landscape ype, 𝑁𝑖=𝐴𝑘 2𝐴𝑘𝑖√𝑛𝑖 𝐴𝑘 Di: Landscape dominance index, desc ibe he dominance o pa ches in a gi en landscape, 𝐷𝑖=𝑄𝑖+𝑀𝑖+𝐿𝑖 3 ni: Is pa ch numbe o i h land scape, Qi: Ra io o he cell wi h i h pa ch and o al cell, Mi: Ra io o he numbe o i h pa ches o o al pa ches and Li: Ra io o i h pa ch a ea o he o al a ea. And a, b and c: ep esen he weigh o Ci, Ni and Di espec i ely whe e a + b + c = 1 (Zhang e al, 2013). 27 5. RESULTS 5.1. LULC changes 2003 - 2018 Figu e 5.1 below illus a e he LULC classi ied maps o he yea 2003-2009. As i can be seen, he isual in e p e a ion o he classi ied images explo es ex ensi e changes in a ious land co e classes in he s udy egion especially he apid expansion o buil up a ea as well as dec ease in o es co e . The high inc ease in buil -up a ea is highly exposed in eas e n zone o he s udy egion which a e nea o he u ban cen e o Zanziba island. Figu e 5. 1: LULC classi ied maps o he yea 2003-2018 28 5.2. Accu acy assessmen Accu acy assessmen o each land use map was pe o med in o de o ind classi ica ion e o s and make he p oduced land co e maps become eliable and easily in e p e able by use s. Kappa coe icien , o e all accu acy, use accu acy as well as p oduce accu acy we e all calcula ed and he assessmen esul s we e summa ized in ( able 5.1, 5.2 and 5.3) below. The assessmen esul s o each o he p oduced land co e maps p o ide an o e all accu acy anging om 74% - 78%, and o e all kappa coe icien which is anging om 0.68 – 0.71. On he o he hand, he use s and p oduce ’s accu acies compu ed o each o he classi ied maps we e all abo e 65%. These esul s indica ing ha he classi ied images ha e me a good le el o accu acy and hus hey a e sa is ying o he s udy analysis. In all o he h ee classi ied images, he we land has shown a high use ’s accu acy (80% - 89%) compa ed wi h o he land use classes. This implied ha he majo i y o hei pixels we e co ec ly classi ied in hei espec i e classes. Howe e , in each classi ied image ei he a mland, g assland o o es co e p o ides a minimum use s o p oduce s accu acy (63%-80%). The maximum use ’s accu acies p o ided by he we land can be jus i ied by hei dis inc i e cha ac e is ics in compa isons wi h o he LU classes. This cha ac e is ic makes hem easily being disc imina ed by classi ica ion algo i hms agains o he land use classes du ing he classi ica ion p ocess. Howe e , on ano he side, he lowes use s and p oduce ’s accu acies o a mland, g assland and o es co e can be explained by hei spec al p ope y simila i ies among hem (Mel ille, Luciee , & A yal, 2018). o e all accu acy 76% kappa 0.69 LULC classes Buil -up Fo es g assland a mland We land To al Use accu acy Buildup 32 0 5 3 2 42 76% Fo es 2 37 6 0 1 46 80% G assland 1 6 36 6 1 50 72% Fa mland 5 5 3 31 0 44 70% We land 0 2 0 0 16 18 88% To al 40 50 50 40 20 200 P oduce s accu acy 80% 74% 72% 77% 80% Table 5. 1: Con usion ma ix o LULC map o he yea 2003 29 O e all accu acy 78% kappa 0.71 LULC classes Buil -up Fo es g assland a mland We land To al Use accu acy Buildup 34 2 2 3 3 44 77% Fo es 0 41 9 0 1 51 80% G assland 1 3 35 8 0 47 74% a mland 5 3 3 29 0 40 73% We land 0 1 1 0 16 18 89% To al 40 50 50 40 20 200 P oduce s accu acy 85% 82.00% 70% 72% 80% Table 5. 2: Con usion ma ix o LULC map o he yea 2009 O e all accu acy 75% kappa 0.68 LULC classes Buil -up Fo es g assland a mland We land To al Use accu acy Buildup 33 1 1 4 2 41 80% Fo es 1 39 6 3 0 49 79% G assland 1 6 34 6 1 48 70% a mland 5 1 9 26 0 41 63% We land 0 3 0 1 17 21 80% To al 40 50 50 40 20 200 P oduce s accu acy 82% 78.00% 68% 65% 85% Table 5. 3: Con usion ma ix o LULC map o he yea 2018 30 5.3. LULC changes 2003 - 2018 The s a is ical esul s co esponding o LULC change in he s udy a eas (Table 5.4) indica ing ha , he e is an ex ensi e inc ease in he buil -up a ea as equi alen o he dec ease in o es co e . The buil -up a ea o Zanziba island has been expanding om 4132.44 Hec a e in he yea 2003 o 7201.35 Hec a e in 2018. These a eal changes in he buil -up a ea make up o 42.6% inc ease in cons uc ion land o only 15 yea s’ (2003-2018) pe iod (2.84% inc ease annually). The Fa mland/ag icul u al land has also shown a p og essi e expansion wi hin hese 15 yea s o he s udy, in which 31.7% (2.1% annually) o ag icul u al land we e inc eased. Fo he o es co e and we land, he ends on hei land co e changes ha e shown a con inuously dec easing, in which 15.8% o o es co e (1.05% annually) and 24.6% o we land (1.64% annually) we e disappea ed. Howe e , in he case o g assland, he end in i s LULC changes does no show a smoo h a ia ion. This is because, in he i s 6 yea s (2003-2009), he g ass co e o he s udy a ea has been dec eased by 4.5%, ollowed by inc easing o 7.48% o he nex 9 yea s (2009 – 2018). LULC classes A ea (ha) (2003) A ea (ha) (2009) A ea ( (2018) % changes (2003-2009) % changes (2009-2018) % changes (2003-2018) Buil -up 4132.44 5081.94 7201.35 18.68 29.43 42.62 Fo es 73329.39 69460.56 61692.03 -5.28 -11.18 -15.87 Fa mland 22171.77 27669.15 29206.17 24.79 5.26 31.73 G assland 56569.41 54107.73 58484.43 -4.55 7.48 3.27 We land 1925.55 1809.18 1544.58 -6.43 -17.13 -24.66 Table 5. 4: LULC a eal s a is ics om 2003 -2018 This apid LULC changes o se e al land co e classes wi hin he s udy egion could be a ibu ed by he imp ope way o managing he land including he high a e o in o mal se lemen , casual a ming me hods as well as he highe le el o de o es a ion ac i i ies in Zanziba (M. O. Kukkonen e al., 2018). 37 The ex eme scena io shows a e y high inc ease in ERI end compa ed o hose o business as usual and conse a ion scena io (Table 5.10). Howe e , i has o be no ed he e ha , al hough in ex eme scena ios he e was no cons ain s ha was employed, i ’s ERI is mo e likely o be applicable in Zanziba e en mo e han ha o business as usual. This is due o he ac ha in Zanziba he de elopmen and cons uc ion o buildings in a andom way and wi hou ollowing es ic ions and go e nmen egula ions a e one among he common and majo p oblem acing u ban planning and hus hey make no much signi icance in he land change esul s ob ained by employing cons ain in e e y p o ec ed a ea. In summa y, he o e all ERI o he s udy a ea shows an inc easing end om 2003 o 2027. Whe eas, he e is also a signi ican change (mos ly inc easing) in he numbe o pa ches, agmen a ion and sepa a ion indices as well as a con inuous dec ease in he g ea es pa ch a ea o he o es , a mland, and g assland. Yea /Scena io O e all ERI % Changes 2003 0.0714 - 2009 0.0867 17.7 2018 0.0957 9.4 BAU 0.1020 6.2 EXTREME 0.1050 8.9 CONSERVATION 0.1002 4.6 Table 5. 10: O e all Ecological isk index changes 2003 – 2027 Figu e 5. 5: T ends o o e all ERI om 2003 - 2027 38 5.5.3. Dis ibu ion o ERI pe dis ic While compa ing he dis ibu ions o ERI a a dis ic le el (Figu e 5.6 and Table 5.11), i appea s ha he dis ic s U ban-wes , No h A and No h B expe iences he high and p og essi ely inc eases in hei ERI alues, while he ERI o he U ban dis ic s seems o be s able. The ex ensi e inc eases in ERI o hese h ee dis ic s could be highly in luenced by he high a e o casual ag icul u al ac i i ies, ou ism ac i i ies as well as in o mal housing in hese egions. In he case o he h ee u u e scena ios, he esul s show ha he ex eme scena io p o ides he highe ERI inc eases in almos all dis ic s o Zanziba especially in No h A and No h B whe e he inc ease in ERI alues co esponding o he ex eme scena io a e 23.5% and 17.92 espec i ely. The conse a ion scena io in o he case shows he ai ERI esul s in almos i e dis ic s o Zanziba excep o he No h A dis ic which show he highe inc eases in ERI o all o he 2027 scena ios. Dis ic T end o ERI Change (%) 2003-2009 2009-2018 2018-2027 BAU 2018-2027 EXTREM 2018-2027 CONSERVATION No h A 16.63 8.95 18.51 23.45 17.13 No h B 9.89 14.78 13.51 17.92 2.79 Cen al 10.37 4.39 10.82 8.64 1.82 Sou h 10.37 10.63 10.94 8.35 3.67 Wes 14.57 23.41 5.39 11.61 6.93 U ban -1.35 5.27 0.24 2.19 0.49 Table 5. 11: ERI Changes by dis ic 2003-2027 39 Figu e 5. 6: T end o ERI pe dis ic 2003 - 2027 40 6. DISCUSSION This s udy was employed o assess ecological isk condi ions in Zanziba based on land co e change, wi h he ocus o p o iding a holis ic unde s anding o land co e dynamics and hei en i onmen al impac s along wi h p e en ing unmanaged g ow h and andom de elopmen o in o mal housing and in as uc u e. Based on Random Fo es Classi ie , Mul i-Laye Pe cep on (ANN) and Ma ko Chain, we success ully assessed he end o land-use dynamics om 2003- 2018 i s and hen he land use maps o 2027 was p edic ed. LECOS plugin in QGIS 3.8 was e ec i ely applied o calcula e landscape me ics in land use maps and hen employed hem o calcula e ecological isk indices o he landscape. These indices including sepa a ion, agili y, dominance, loose as well as o e all ecological isk index o he landscape. Du ing he i s s ages o analyzing land-use dynamics, we ha e ealized ha Zanziba Ci y has expe ienced a end o apid land-use changes and ex ensi e buil -up expansion mos ly in a eas nea exis ing buildings, and some o he coas al zones. The esponse dis ance o he buildings and dis ance o coas a iable suppo s his obse a ion as hey p o ide he highe C amme ’s alues ensu ing ha hese a iables ha e a g ea in luence on land co e dynamics in Zanziba . I he cu en end con inues, he buil -up expansion o Zanziba ci y will be expanded o mo e han 40% by 2027. The La ge pa o ag icul u al and o es ed land has al eady been agmen ed o housing pa cels. And because o hese, he nega i e en i onmen al e ec s ele an o land esou ces, such as loss o na u al o es s, open c opland, and we lands ha e been al eady being highly des oyed in Zanziba (M. H. Ali & Sulaiman, 2006). This is somewha ala ming ha he go e nmen should en o ce es ic ions and s ong land policies lea ing he emaining land co e classes o s op being u he des oyed. The analysis esul s gi e no signs indica ing ha hese land-use changes and ex ensi e buil -up expansion would a leas decline in he u u e. And his is possibly caused by he poo quali y o housing and minimum go e nmen e o s o en o ce he op imal usage o land esou ces in Zanziba island (Ameyibo e al., 2003). While analyzing he o e all ecological isk indices o he landscape, he esul s sc eened ha , he o e all ERI o Zanziba Ci y has been cons an ly g own om 2003 onwa d. This end is pa allel o he inc ease in he numbe o smalle pa ches o some land co e classes, ela i e o he dec easing in hei la ges pa ch a eas. 41 The obse a ion o he u u e scena ios p o ide he wo s ecological isk indices o he ex eme whe eas conse a ion scena io p o ides a li le ai esul s. The d as ic ecological isk sco es p o ided by ex eme scena io, sugges ed ha , he land use es ic ions in Zanziba should be p ope ly espec ed, a leas by p ope ly u ilizing he al eady exis ing land policies and es ic ions. Finally, i would be e y impo an o highligh ha , in his s udy we ha e assessed ecological isk indices o Zanziba island wi h land co e change using ee sa elli e da a. This p ocess equi ed a consis en and accu a e land co e da a o he pu pose o p o iding eliable and accu a e in o ma ion. Howe e , i was a bi challenge in acqui ing ee sa elli e images wi h desi ed s anda d and a desi ed ime. This is due o he geog aphical egion o Zanziba island making mos o hei sa elli e images being co e ed wi h cloud. In ha sense, i akes a much ime o sea ch and acqui e he needed sa elli e da a, and e en hough he image o 2018 ha was ob ained was no in e y good quali y and con ain some scanned lined which somehow a ec ed he analysis esul s. Howe e , in u u e hese esul s can be imp o ed by using mo e accu a e and upda e da a, as well as inco po a ing ex ensi e land co e change d i e s. 42 7. CONCLUSIONS AND RECOMMENDATIONS The indings in his esea ch suppo he e idence o he ecen s udy which ound ha he apid land-use changes in Zanziba ha e imposed an in ensi e deg ada ion in he ecosys em o Zanziba (A S aeh , 2018). The main d i e s behind his in ensi e deg ada ion a e inc easing in uncon olled and poo ly managed in o mal housing, high a e o casual a ming and imp o ised plan in he ou ism sec o (WB, 2019). E idence has been clea ly exposed by inc easing he agmen a ion index, sepa a ion index, loss index as well as o e all ERI o he en i e landscape. In o de o educe such kind o h ea s, mo e sys ema ic app oaches in he ecological assessmen a e equi ed. Coope a i e esea ch among esea ch g oups, land planne s, and all o he s akeholde s including ou ism sec o s should be highly en o ced as a ool o moni o and ecommend a long- e m plan o sus ainable ecosys ems in Zanziba . In u u e scena ios, he indings ha e p o en ha disobeying land policies and egula ions can in luence a d as ic shock o he ecological condi ions o Zanziba in he u u e. This has been p o en by he esul s o an ex eme scena io which was e y wo s . Conse a ion scena io o somehow had shown li le in luence on di ec ing u u e land co e changes and ecological isk con ol, bu howe e , he esul s we e only mode a e and he e o e i is eally impo an o he go e nmen o ake a special e o on se ing up new land-use app oaches such as p omo ing he densi ica ion o e ical buildings a he han elying in one- loo ho izon al buildings. In gene al, all o he indings in his s udy can p o ide use ul insigh s o land-use planne s and policymake s ega ding he cu en ends and u u e p ojec ion o land co e changes in Zanziba , he ac o s d i ing hese changes as well as he e ec ha hey ha e imposed in hei ecological heal h. This can help he esponsible au ho i ies on se ing up he p ope ules and egula ions on land use con ol. The s udy also assessed he ecological heal h condi ions o he Zanziba islands in h ee di e en scena ios. In his case, he esul s can se e as he basic no ion in explo ing he sus ainabili y o he landscape s uc u e o he Zanziba island as well as i s ecological heal h secu i y unde a ious land-use p ac ices. This can con ibu e o he designa ion o he app op ia e s a egies owa d he land use managemen so ha a sys ema ic balance be ween economic and sus ainable ecosys ems can be main ained. 43 Bibliog aphic Re e ences A S aeh , P. (2018). Managing human p essu es o es o e ecosys em heal h o zanziba coas al wa e s. Jou nal o Aquacul u e & Ma ine Biology, 7(2), 59–70. h ps://doi.o g/10.15406/jamb.2018.07.00185, accessed on 15 Oc obe , 2019. Ali, A. I. (2016). Assessmen o Mang o e Phenology and he Role o Insec Pollina o s in F ui P oduc ion A Nyeke and Micham i Mang o e Fo es s, Zanziba . (May). Ali, M. H., & Sulaiman, M. S. (2006). The Causes and Consequences o he In o mal Se lemen s in Zanziba . Shaping he Change, 8–13. Al-sha i , A. A. A., & P adhan, B. (2014). Moni o ing and p edic ing land use change in T ipoli Me opoli an Ci y using an in eg a ed Ma ko chain and cellula au oma a models in GIS. A abian Jou nal o Geosciences, 7(10), 4291–4301. h ps://doi.o g/10.1007/s12517-013-1119- 7, accessed on 22 No embe , 2019. A aya, Y. H., & Cab al, P. (2010). Analysis and modeling o u ban land co e change in Se úbal and Sesimb a, Po ugal. Remo e Sensing, 2(6), 1549–1563. h ps://doi.o g/10.3390/ s2061549, accessed on 18 No embe , 2019. A sanjani, J. J., Helbich, M., Kainz, W., & Boloo ani, A. D. (2012). In eg a ion o logis ic eg ession, Ma ko chain and cellula au oma a models o simula e u ban expansion. In e na ional Jou nal o Applied Ea h Obse a ion and Geoin o ma ion, 21(1), 265–275. h ps://doi.o g/10.1016/j.jag.2011.12.014, accessed on 28 Oc obe , 2019. Bha acha ya, S., Ca , T. R., & Pal, M. (2016). Compa ison o supe ised and unsupe ised app oaches o muds one li ho acies classi ica ion: Case s udies om he Bakken and Mahan ango-Ma cellus Shale, USA. Jou nal o Na u al Gas Science and Enginee ing, 33, 1119–1133. h ps://doi.o g/10.1016/j.jngse.2016.04.055, accessed on 27 Oc obe , 2019. B own, S. S., & Reine , K. H. (1992). A concep ual amewo k o ecological isk assessmen . En i onmen al Toxicology and Chemis y, 11(2), 143–144. h ps://doi.o g/10.1002/e c.5620110202, accessed on 5 Decembe , 2019. Chi, Y., Zhang, Z., Xie, Z., & Wang, J. (2019). How human ac i i ies in luence he island ecosys em h ough damaging he na u al ecosys em and suppo ing he social ecosys em? Jou nal o Cleane P oduc ion, (xxxx), 119203. h ps://doi.o g/10.1016/J.JCLEPRO.2019.119203, accessed on 18 No embe , 2019. 44 Cla ke, K. C., & Gaydos, L. J. (1998). Loose-coupling a cellula au oma on model and GIS: Long- e m u ban g ow h p edic ion o San F ancisco and Washing on/Bal imo e. In e na ional Jou nal o Geog aphical In o ma ion Science, 12(7), 699–714. h ps://doi.o g/10.1080/136588198241617, accessed on 4 Decembe , 2019. Eas man, J. R. (2006). IDRISI Andes Tu o ial. Cla k Uni e si y, (Ap il), 284. Ebada i, N. (2018). Land Use Changes in Coas al A eas and he Need o P o ec he Coas al En i onmen Limi . 269–270. h ps://doi.o g/10.31031/EAES.2018.03.000566, accessed on 26 Decembe , 2019. Egoh, B. N. (2009). In eg a ing ecosys em se ices in o conse a ion planning in Sou h A ica. (Ma ch), 171. Engel, M. (2012). In o mal u banism. A chi ec s’ Jou nal, 236(15), 59–63. Re ie ed Decembe 18, 2019. om h p://sea ch.p oques .com/doc iew/1346114189?accoun id=8630%5Cnh p://diglib1.bham.ac. uk:3210/s xlcl3?u l_ e =Z39.88- 2004& _ al_ m =in o:o i/ m :ke :m x:jou nal&gen e=a icle&sid=P oQ:P oQ%3Abhi&a i l e=In o mal+u banism.& i le=A chi ec s%27+Jou nal&iss FAO. (2018). Global Fo es Resou ces Assessmen 2020. Guidelines and speci ica ions. 58. Re ie ed Janua y 17, 2020 om www. ao.o g/ o es y%0Ah p://www. ao.o g/3/I8699EN/i8699en.pd Gao, J., & Xu, L. (2015). An e icien me hod o sol e he classi ica ion p oblem o emo e sensing image. AEU - In e na ional Jou nal o Elec onics and Communica ions, 69(1), 198–205. h ps://doi.o g/10.1016/j.aeue.2014.09.001, accessed on 28 Oc obe , 2019. Ga dne , M. W., & Do ling, S. R. (1998). A i icial neu al ne wo ks ( he mul ilaye pe cep on) - a e iew o applica ions in he a mosphe ic sciences. A mosphe ic En i onmen , 32(14–15), 2627–2636. h ps://doi.o g/10.1016/S1352-2310(97)00447-0, accessed on 7 Janua y, 2020. Ga zuglia, M. (FAO). (2018). Se en y yea s o FAO’s Global Fo es Resou ces Assessmen (1948- 2018). In His o ical o e iew and u u e p ospec s. Gislason, P. O., Benedik sson, J. A., & S einsson, J. R. (2006). Random o es s o land co e classi ica ion. Pa e n Recogni ion Le e s, 27(4), 294–300. h ps://doi.o g/10.1016/j.pa ec.2005.08.011, accessed on 28 Sep embe , 2019. 45 Haden, V. R., Lip zin, D., Rosens ock, T. S., Vande slice, J., B od , S., Yeo, B. L., … Tomich, T. P. (2019). 5. Ecosys em Se ices and human well-being. The Cali o nia Ni ogen Assessmen , 113–194. h ps://doi.o g/10.1525/9780520962231-019, accessed on 9 Oc obe , 2019. Hakanson, L. (1980). An ecological isk index o aqua ic pollu ion con ol.a sedimen ological app oach. Wa e Resea ch, 14(8), 975–1001. h ps://doi.o g/10.1016/0043-1354(80)90143-8, accessed on 21 Decembe , 2019. Hi sche , M., Schweize , E., Welle , M., & K onmülle , H. (1996). In e nal ic ion in NiAl single c ys als. Philosophical Magazine Le e s, 74(3), 189–194. h ps://doi.o g/10.1080/095008396180344, accessed on 12 No embe , 2019. Hunsake , C. T., G aham, R. L., Su e , G. W., O’Neill, R. V., Ba n house, L. W., & Ga dne , R. H. (1990). Assessing ecological isk on a egional scale. En i onmen al Managemen , 14(3), 325– 332. h ps://doi.o g/10.1007/BF02394200, accessed on 29 Janua y, 2020. Jin, X., Jin, Y., & Mao, X. (2019). Ecological isk assessmen o ci ies on he Tibe an Pla eau based on land use/land co e changes – Case s udy o Delingha Ci y. Ecological Indica o s, 101(No embe 2018), 185–191. June, M. (2015). Uni e si y o Nai obi School o Public Heal h. (July), 2. Kukkonen, M., & Käyhkö, N. (2014). Spa io- empo al analysis o o es changes in con as ing land use egimes o Zanziba , Tanzania. Applied Geog aphy, 55, 193–202. Kukkonen, M. O., Muhammad, M. J., Käyhkö, N., & Luo o, M. (2018). U ban expansion in Zanziba Ci y, Tanzania: Analyzing quan i y, spa ial pa e ns and e ec s o al e na i e planning app oaches. Land Use Policy, 71(No embe 2017), 554–565. Lakes, T., & Kim, H. O. (2012). The u ban en i onmen al indica o “bio ope A ea Ra io” - An enhanced app oach o assess and manage he u ban ecosys em se ices using high esolu ion emo e-sensing. Ecological Indica o s, 13(1), 93–103. Langan, C., Fa me , J., Ri ing on, M., & Smi h, J. U. (2018). T opical we land ecosys em se ice assessmen s in Eas A ica; A e iew o app oaches and challenges. En i onmen al Modelling and So wa e, 102, 260–273. McKenzie, E., Rosen hal, A., Be nha d , J., Gi e z, E., Ko acs, K., Olwe o, N. and To , J. (2012). De eloping Scena ios o Assess Ecosys em Se ice T adeo s : Guidance and Case S udies o InVEST Use s. Wo ld Wildli e Fund, Washing on, D.C. (Decembe 2014). 46 McPhea son, T., Picke , S. T. A., G imm, N. B., Niemelä, J., Albe i, M., Elmq is , T., … Qu eshi, S. (2016). Ad ancing U ban Ecology owa d a Science o Ci ies. BioScience, 66(3), 198–212. h ps://doi.o g/10.1093/biosci/biw002, accessed on 5 Decembe , 2019. Mülle o á, J., Pe gl, J., & Pyšek, P. (2013). Remo e sensing as a ool o moni o ing plan in asions: Tes ing he e ec s o da a esolu ion and image classi ica ion app oach on he de ec ion o a model plan species He acleum man egazzianum (gian hogweed). In e na ional Jou nal o Applied Ea h Obse a ion and Geoin o ma ion, 25(1), 55–65. h ps://doi.o g/10.1016/j.jag.2013.03.004, accessed on 26 Sep embe , 2019. Mye s, G. A. (2010). Social cons uc ion o pe i-u ban places and al e na i e planning in Zanziba . A ican A ai s, 109(437), 575–595. h ps://doi.o g/10.1093/a a /adq044, accessed on 21 Oc obe , 2019. Na ional Bu eau o S a is ics. (2018). The Uni ed Republic o Tanzania; Na ional Popula ion P ojec ions O ice o he Chie Go e nmen S a is ician Minis y o Finance and Planning Zanziba Na ional Bu eau o S a is ics Minis y o Finance and Planning Da es Salaam. Re ie ed No embe 15, 2019, om h p://www.nbs.go. z/nbs/ akwimu/census2012/P ojec ion- Repo -20132035.pd NEMC. (2006). In eg a ed Ecosys ems Assessmen in Tanzania :A Repo on Expe iences in Ecosys ems Managemen . pg 52. O e peck, J., Ga in, G., Ja dine, A., Busch, D. E., Cayan, D., De inge , M., … Udall, B. (2013). Summa y o decision make s. In Assessmen o Clima e Change in he Sou hwes Uni ed S a es: A Repo P epa ed o he Na ional Clima e Assessmen . h ps://doi.o g/10.5822/978-1-61091- 484-0_1, accessed on 19 No embe , 2019. Paul, S. S. (2013). Analysis o land use and land co e change in kiska inaw i e wa e shed: a emo e sensing, gis & modeling app oach. Thesis. Pijanowski, B. C., B own, D. G., Shelli o, B. A., & Manik, G. A. (2001). Using Neu al Ne wo k and GIS o o ecas .. Compu e s, En i onmen and U ban Sys ems. Sange mano, F., Eas man, J. R., & Zhu, H. (2010). Simila i y Weigh ed Ins ance-based Lea ning o he Gene a ion o T ansi ion Po en ials in Land Use Change Modeling. T ansac ions in GIS, 14(5), 569–580. Rimal, B., Sha ma, R., Kunwa , R., Kesh ka , H., S o k, N. E., Rijal, S., … Ba al, H. (2019). E ec s o land use and land co e change on ecosys em se ices in he Koshi Ri e Basin, Eas e n