土地利用、覆被变化的过程、趋势及其与变化的沿海生态系统的联系-以伊朗克什姆岛为例（外语版）.pdf

Table of Cont ent sTable of ContentsTable of Contents.iList of Tables.ivList of Figures.vAbbreviations.vii摘要.viiiAbstract.xiChapter 1 Introduction.11.1 Research background.11.2 Research objectives.31.3 Analytical framework.4Chapter 2 Literature review.62.1 Land-use/cover change detection.62.2 Land-use/cover modeling.72.3 Response of coastal ecosystems to LUCC.112.3.1 Impact of LUCC on coast al ar eas.112.3.2 Measur ing t he st at e of coast al ecosyst ems under t he changingenvir onment.122.4 Summary.15Chapter 3 Materials,and methods.163.1 Study areas.163.1.1 Qeshm Island.163.1.2 Gabr ik r egion.183.2 Remotely sensed data.193.2.1 Landsat dat a.193.2.2 MODIS dat a.203.3 Methods to measure LUCC.203.3.1 Pr epar ing Land-Use/Cover dat a.203.3.2 Int ensit y analysis t o measur e LUCC.213.4 Methods to simulate LUCC.24Table of Cont ent s3.4.1 Cellular aut omat a.253.4.2 Mar kov chain model.253.4.3 Mult i-Layer per cept r on(MLP)neur al net wor k.；.263.4.4 Logist ic r egr ession(LR).283.4.5 CA-Mar kov.293.4.6 Sensit ivit y analysis t hr ough adding/delet ing appr oach.293.4.7 Validat ion.303.5 Methods to link LUCC with EWS.323.5.1 Google ear t h engine-GEE.333.5.2 R language.333.5.3 Remot ely sensed indices.343.5.4 Used st at ist ical analyses for ear ly war ning signal.36Chapter 4 Measuring LUCC in Qeshm Island.384.1 Land use/cover maps.384.2 Measuring land change at three levels:intervals,category,and transition.424.3 Driving forces of LUCC in Qeshm Island.464.4 Patterns to processes in LUCC.474.5 Summary.50Chapter 5 Simulating LUCC in Qeshm Island.525.1 An adding/deleting approach to improve LUCC modeling.525.1.1 Model configur at ion.525.1.2 Analyzing t r ansit ion pot ent ial maps(TPMs).545.1.3 An adding/delet ing appr oach t o impr ove t he model per for mance 57-5.2 Comparison of four hybrid models to simulate the LUCC.615.3 Prediction of land use/cover change(LUCC)in 2026.655.4 Discussion.665.5 Summary.71Chapter 6 Linking LUCC with shifting regime in coastal ecosystems.736.1 Detecting LUCC.736.2 Time series of three indices derived from MODIS from 2000 to 2018 756.3 Identification of early warning signals(EWS).77iiTable of Cont ent s6.4 Discussion.796.4.1 Effect iveness of t hr ee indices for ident ifying EWS.796.4.2 Linking LUCC wit h shift ing r egime in coast al ecosyst ems.816.5 Summary.82Chapter 7 Conclusion,and recommendation.847.1 Conclusion.847.2 Innovative of this study.867.3 Limitation of this study.867.4 Recommendation.87References 88Acknowledgments.108Projects and publications.109Appendices 1 Google Earth Engine used code to extract RS indices.IllAppendices 2 R used code for statistical analysis related to Early Warning.114111List ofTablesList ofTablesTable 3-1 Mean annual t emper at ur e,and aver age annual pr ecipit at ion.17Table 3-2 List of Landsat sat ellit e images used in t he r esear ch.19Table 3-3 Specificat ions of MODISs bands 1 t o 7(USGS).20Table 3-4 List of LULC classes,and descr ipt ion for each class.21Table 3-5 Mat hemat ical not at ion following Aldwaik,and Pont ius(2012).22Table 4-1 Over all accur acy fbr t he land-use maps.40Table 4-2 Var iat ion mat r ix fbr each land-use class based on t he number of pixels.41Table 5-1 A summar y of t he Mar kov pr obabilit y mat r ix fbr simulat ing t he t r ansit ionbet ween t he pr imar y,and final cellular st at es.54Table 5-2 Result s of Adding/Delet ing appr oach fbr each var iable.59Table 5-3 Validat ion of t he model befor e,and aft er adding/delet ing appr oach.60Table 5-4 Validat ion of t he used modeling appr oach in compar ison wit h ot her modeling met hods include CA-MC-ANN(Cellular Aut omat a-Mar kov Chain-Ar t ificial Neur al Net wor k),MC-ANN(Mar kov Chain-Ar t ificial Neur al Net wor k),and CA-MC-LR(Cellular Aut omat a-Mar kov chain-logist ic r egr ession).62Table 5-5 Kappa st andar d coefficient fbr each class.63Table 5-6 Compar ison of t he ar ea of each land-use class bet ween 2014,and 2026.65Table 6-1 Ar ea of changes r elat ed t o human-made usages include Built-up,and agr icult ur e(Number of pixels).75Table 6-2 Compar at ive t able which shows t he value of t he indices in t he t wo r egions.77Table 6-3 Kendalr s t t r end and it s associat ed significance level.What ever Kendall5s t be closer t o 1,it means mor e upwar d t r end and indicat es an Ear ly War ning at t he ecosyst em.79ivList of Figur esList of FiguresFig.1-1 Analyt ical fr amewor k of t his st udy.5Fig.2-1 Descr ipt ion of shift ing st able st at es,st abilit y,and r esilience t heor y(Clement s&Ozgul 2018).12Fig.3-1 Locat ion of main st udy ar ea(Qeshm Island),and cont r ol ar ea(Gabr ik).16Fig.3-2 Aver age mont hly t emper at ur e,and pr ecipit at ion in t he Qeshm Island.17Fig.3-3 Flowchar t of using int ensit y analysis t o measur e LUCC.24Fig.3-4 Flowchar t of LUCC modeling in t his st udy.31Fig.3-5 Pr oposed appr oach t o link LUCC wit h EWS.32Fig.3-6 The flowchar t of linking LUCC wit h EWS.37Fig.4-1 Land-use/cover maps of Qeshm Island in 6 classes at four t ime int er vals.39Fig.4-2 Rat e of change at t ime level,and int ensit y of changes.42Fig.43 Rat e of change at cat egor y level,and int ensit y of changes associat ed wit h land-use classes.43Fig.4-4 Tr ansit ion int ensit y given cat egor y gains dur ing t hr ee-t ime int er vals.45Fig.5-1 TPMs MLP-ANN a)Agr icult ur e b)Bar e-land c)Built-up d)Dense-veget at ion e)Mangr ove f)Wat er-body.55Fig.5-2 TPMs Logist ic-Regr ession a)Agr icult ur e b)Bar e-land c)Built-up d)Dense-veget at ion e)Mangr ove f)Wat er-body.56Fig.5-3 Value pf ar ea under t he cur ve(AUC).57Fig.5-4 Simulat ed fbr Land-use simulat ed for 2014.This map was used t o compar e wit h t he act ual map of t he same year t o measur e t he accur acy of four modeling met hods t o pr ove t hat t he adding/delet ing appr oach has ult imat ely helped t o incr ease t he accur acy of t he modeling.57Fig.5-5 Gr een bar s show t he cor r ect pr edict ed changes aft er adding/delet ing,and t he r ed bar s show r ight pr edict ed changes befor e Adding/Delet ing(A-left,unit:km2)-The number of var iables used t o model in ever y class(B-r ight).61vList of Figur esFig.5-6 Agr eement,and disagr eement component s in 1)CA-MC-LR 2)MC-ANN3)CA-MC-ANN 4)CA-MC-ANN-SA.62Fig.5-7 Compar ison of t he simulat ed map aft er adding/delet ing(t op)wit h t hesimulat ed map befor e adding/delet ing(down).64Fig.5-8 Pr edict ed map of 2026-Using int egr at ion of CA-Mar kov-ANN-AD.66Fig.5-9 Gains,and Losses bet ween 2014,and 2026(hect ar es)for each land-use class.66Fig.6-1 Land-use maps of Qeshm Island(up),and Gabr ik(down)ext r act ed fr om Landsat image for 1996,and 2014.74Fig.6-2 The t ime ser ies of t hr ee r emot ely sensed indices,Nor malized Differ ence Veget at ion Index(NDVI),Modified Nonnalized Wat er Index(MNDWI),and Modified Veget at ion Wat er Rat io(MVWR)t hat have been used as t he var iables in Qeshm Island(r ight),and Gabr ik(left).The r ed line illust r at es t he t r end obt ained using a moving aver age wit h a window size of 20-t ime st eps.76Fig.6-3 Met r ic-based leading indicat or s based on NDVI,MNDWI,and MVWR for Qeshm Island(3 t op r ows),and Gabr ic(3 bot t om r ows).The fir st,second,and t hir d r ows show aut ocor r elat ion at lag 1(ACF),st andar d deviat ion(SD),and skewness of each spect r al indices,r espect ively.78VIAbbr eviat ionsAbbreviationsANNAr t ificial Neur al Net wor kATFAut o-Cor r elat ion Funct ionAUCAr ea Under t he Cur veCACellular Aut omat aEWSsEar ly War ning SignalsFOMFigur e of Mer itGEEGoogle Ear t h engineGISGeogr aphic Infor mat ion Syst emLCMLand Change ModelerLRLogist ic Regr essionLU/LCLand Use/Land CoverLUCCLand Use/Cover ChangeMCMar kov ChainMLPMult i-Layer Per cept r onMNDWIModified Nor malized Differ ence Wat er IndexMODISModer at e r esolut ion Imaging Spect r or adiomet erMVWRModified Veget at ion Wat er Rat ioNDVINor malized Differ ence Veget at ion IndexNDWINor malized Differ ence Wat er IndexNIRNear-Infr ar edROCRelat ive Oper at ing Char act er ist icRSRemot e SensingSASensit ivit y AnalysisSDSt andar d Deviat ionSKSkewnessSPISt andar dized Pr ecipit at ion IndexTPMsTr ansit ion Pot ent ial Mapsvii摘要摘要土地利用/覆盖变化(Land Use/Cover Change,以下简称LUCC)是全球变化的 重要组成部分，这种变化及其对生态环境的影响也一直是人们关注的焦点。近 年来，地理空间技术的发展和全球覆盖的对地观测能力使研究人员能够以越来 越快的速度处理海量数据。本文就地理空间技术在LUCC调查方面的应用进行 了研究，揭示了干旱地区的滨海生态系统如何应对这些变化，研究结果可为波 斯湾地区人类和自然之间的相互作用机制研究提供科学支撑，并可成为探究生 态环境如何对土地利用变化作出本地响应的参考案例。由于波斯湾地区的Qeshm岛过去二十年发展迅速，因此我们选择以其为例 来开展本项研究。研究过程中我们将地理空间技术、强度分析、土地利用变化 的混合模型和统计分析相结合，以探究Qeshm岛的土地利用变化过程、趋势及 其对生态环境的影响。得出的主要成果如下：第一，本研究从时间间隔、土地利用类型和相互转化等三个方面对Qeshm 岛最近20年的土地利用/覆盖变化进行了系统的调查和分析。我们分别下载了 1996年、2002年、2008年和2014年的Landsat遥感影像，解译时将土地利用 类型分为农业用地、裸地，建设用地，密集植被，红树林和水体等6类，按照 年份的不同分别绘制了土地利用图，并使用强度分析的方法从时间间隔、土地 利用类型和相互转化等三个方面来评价Qeshm岛土地利用的动态变化。研究结 果表明，在第一个六年(1996-2002)和第三个六年Q008-2014)Qeshm岛的土地利 用变化很大，而在第二个六年(2002-2008)土地利用变化呈现缓慢趋势。在建设 高要求和人口不断增长的背景下，Qeshm岛的建设用地面积在三个时间段内都 明显增大。转化强度分析结果表明，1996年至2002年间，其他土地利用类型 范围的扩增主要由裸地转化而来，尤其是建设用地的增加，而建设用地并没有 转化成其他土地利用类型。密集植被的范围持续增加。然而，对于红树林来说,干旱和居民的砍伐对其造成了负面影响，因此建议建立自然保护区来有效地保 护这一类珍贵的生态系统。转化强度分析的结果还表明，与其他土地利用类型 相比，岛上的建设用地和农业用地的需求很高。总之，Qeshm岛的土地利用/覆 盖变化与近年来的快速发展有着十分密切的关系。V111摘要第二，本文提出了一种融合元胞自动机模型（Cellular Aut omat a）、马尔科夫 链（Mar kov Chain）和人工神经网络模型（Ar t ificial Neur al Net wor k）的混合方法，再 结合敏感性分析法（添加/删除）来筛选有效参数，进而模拟土地利用/土地覆盖未 来的变化趋势，并运用此法对2026年Qeshm岛的土地利用变化趋势进行预测。将此法模拟的结果与实际现状（即基于2014年Landsat遥感影像数据制作的地图）对比后发现，使用此方法可将准确预测的土地利用面积增加到7.2km2,而不使 用此方法时，准确预测的面积仅有6.09 km2,可见此法能大大提高土地利用模 型预测的精度。本研究还通过将运用该方法得到的结果与运用CA-MC-ANN阮 胞自动机模型-马尔科夫链-人工神经网络模型）、MC-ANN（马尔科夫链-人工神 经网络模型）和CA-MC-LR（元胞自动机模型-马尔科夫链-罗吉斯回归）得到的结 果进行对比后发现，本文所提出方法的FOM指数（品质因数）值是7.8,而其他 方法得到的FOM指数值分别为6.7、5.1和4.5,进一步证明了此法的优越性。在此准确性评估结果的基础上，我们运用该方法预测了 Qeshm岛2026年的土 地利用变化趋势。模拟结果表明，未来建设用地的面积将保持增加的趋势，而 红树林的面积将会减少，强调了规划建设保护区的重要性。若需充分了解沿海 地区的规划和管理，可将这种方法应用在其他地区（特别是沿海地区）的类似研 究中。第三，本研究进一步将多源遥感影像、数据挖掘技术和三种评价指标相结 合，用来探究滨海生态系统如何对土地利用变化进行响应。由于红树林属于敏 感型生态系统，本研究还指出了这些变化是否会引起红树林生态系统的结构转 换。研究中我们分别采用了归一化植被指数（NDVI）、修正的归一化水指数（MNDWI）和修正的植被水比值（MVWR）等评价指标来检测生态系统动态的预警 信号。本研究将Gabr ik地区作为背景与Qeshm岛进行对比，因为Gabr ik地区 与Qeshm岛属于同一省份，且气候相似，更重要的是该地区具有人为干扰少、生态系统健康的鲜明特点。结果表明，由于人类活动的增加，Qeshm岛出现了 预警信号，但Gabr ic岛却没有出现。由于这两个地区的气候和地理条件极为相 似，Qeshm岛的预警信号不可能是气候变化造成的，因此正如研究假设中所述，预警信号是由于Qeshm岛的土地利用变化和人类活动造成的。然而，将植被评 价指数与遥感影像、实地数据相结合可增加这类研究结果的价值。在这三个评 ix摘要价指标中，MNDWI和NDVI产生的预警信号结果更好。为了检验本研究中所 使用方法的有效性，作者建议在更大范围内进行此项研究。关键词：土地变化;模拟;状态转移;沿海生态系统;伊朗XAbst r actAbstractLand Use/Cover Change(LUCC)is r egar ded as a significant component of global change.Invest igat ing LUCC,and it s subsequent impact s on t he eco-envir onment have been gr eat concer n fr om local t o global scales.The r ecent development of geospat ial t echnology,and t he availabilit y of br oad cover age of t he Ear t h obser vat ions,facilit at es t he r esear cher s t o handle massive dat a wit h incr easing pr ocessing speed.As such,we examined t he abilit y of geospat ial t echnology in t his st udy t o measur e LUCC,and det ect ed how a coast al ecosyst em in t he ar id climat e zone r esponds t o t hese changes.The findings of t his st udy might offer an in-dept h insight int o t he int er act ion mechanism bet ween nat ur e,and humans in t he Per sian Gulf r egion,and it can be demonst r at ed as a r efer ent ial example t o show how local ecological r esponses t o LUCC.Wit h t he aids of geospat ial t echnology,int ensit y analysis,and pr oposed fr amewor ks,LUCC was measur ed,simulat ed,and linked wit h shift ing r egime in coast al ecosyst ems locat ed in t he lar gest island,Per sian Gulf.The major r esult s obt ained ar e as follows:Fir st,t he land use/cover change was syst emat ically invest igat ed in Qeshm Island at t hr ee levels:int er val,cat egor y,and t r ansit ion in t he last t wo decades.Land-use maps for 1996,2002,2008,and 2014 wer e pr epar ed using Landsat sat ellit e imager y in 6 classes including Agr icult ur e,Bar e-land,Built-up,Dense-veget at ion,Mangr ove,and Wat er-body,and t hen t he dynamic of t he classes was evaluat ed using int ensit y analysis by t hr ee levels:int er val,cat egor y,and t r ansit ion.The r esult illust r at ed t hat,while t he land change was fast over t he fir st,and t hir d t ime int er vals(19962002,and 2008-2014),t he change exper ienced a slow t r end in t he second per iod(2002-2008).Dr iven by t he high demand for const r uct ion,and populat ion gr owt h,t he Built-up class was ident ified as an act ive gainer in t he t hr ee t ime int er vals.The r esult s of t r ansit ion int ensit y analysis indicat ed t hat Bar e-land class was r ecognized as t he main supplier of t he land for ot her classes,especially for t he Built-up ar ea,while Built-up did not act as t he act ive supplier of t he land fbr ot her classes.The Dense-veget at ion class was act ive in all t hr ee t ime int er vals.As fbr t he Mangr ove class,dr ought,and cut t ing by t he r esident s had negat ive effect s,while set t ing up pr ot ect ed ar eas can effect ively maint ain t his valuable ecosyst em.The r esult s of t he t r ansit ion int ensit y analysis showed t hat high demands wer e obser ved fbr land change in r elat ion t o Built-up,andXIAbst r actAgr icult ur e classes,among ot her classes.In conclusion,over all land cover changes in Qeshm Island ar e r elat ed t o r apid development dur ing r ecent year s.A fr amewor k t hat int egr at es hybr id models wit h an adding/delet ing appr oach for scr eening t he effect ive par amet er s was pr oposed in t his st udy for a bet t er under st anding of t he fut ur e LUCC t r end in t he Qeshm Island.The r esult s showed t he suggest ed appr oach led t o an incr ease in t he pr edict ive accur acy of t he model for t he major it y of t he land use classes.The accur acy assessment was conduct ed by compar ing t he simulat ion r esult s obt ained fr om t he model,and t he st at us quo(i.e.t he maps pr oduced based on Landsat 2014 dat a).Using suggest ed appr oach,t he ar ea of t he cor r ect pr edict ions incr eased t o 7.2 km2,t hat was 6.09 km2 wit hout using t he pr oposed appr oach.Also,compar ing t he r esult s of pr oposed met hod wit h CA-MC-ANN(Cellular Aut omat a-Mar kov Chain-Ar t ificial Neur al Net wor k),MC-ANN(Mar kov Chain-Ar t ificial Neur al Net wor k),and CA-MC-LR(Cellular Aut omat a-Mar kov chain-logist ic r egr ession)showed t hat t he FOM(Figur e of Mer it)index in t he pr oposed met hod is 7.8 wher eas in ot her met hods wer e 6.7,5.1,4.5 r espect ively.Based on t he r esult of accur acy assessment,suggest ed appr oach is t hus used t o simulat e land-use change in 2026.Result s of simulat ion indicat ed t hat t he Built-up will have an incr easing t r end while t he ar ea of t he Mangr ove for est s will decr ease in t he fut ur e which highlight t he need for conser vat ion planning.To sufficient ly infor m t he planning,and management of coast al zones,t his met hod is also r ecommended for use in similar st udies in ot her r egions,especially coast al ar eas.A fr amewor k in which r emot ely sensed images was acquir ed and pr ocessed t o ext r act t hr ee e