Mangroves facing climate change : landward migration potential in response to projected scenarios of sea level rise

Mangrove forests prominently occupy an intertidal boundary position where the effects of sea level rise will be fast and well visible. This study in East Africa (Gazi Bay, Kenya) addresses the question of whether mangroves can be resilient to a rise in sea level by focusing on their potential to migrate towards landward areas. The combinatory analy­ sis between remote sensing, DGPS-based ground truth and digital terrain models (DTM) unveils how real vegetation assemblages can shift under different projected (minimum (+9 cm), relative (+20 cm), average (+48 cm) and maxi­ mum (+ 8 8 cm)) scenarios of sea level rise (SLR). Under SLR scenarios up to 48 cm by the year 2100, the landward extension remarkably implies an area increase for each of the dominant mangrove assemblages except for Avicennia ma­ rina and Ceriops tagal, both on the landward side. On the one hand, the increase in most species in the first three scenar­ ios, including the socio economically most important species in this area, Rhizophora mucronata and C. tagal on the sea­ ward side, strongly depends on the colonisation rate of these species. On the other hand, a SLR scenario of + 88 cm by the year 2100 indicates that the area flooded only by equinoctial tides strongly decreases due to the topographical settings at the edge of the inhabited area. Consequently, the landward Avicennia-dominated assemblages will further decrease as a formation if they fail to adapt to a more frequent inundation. The topography is site-specific: however non-invadable areas can be typical for many mangrove settings.


Introduction
Inhabiting the interface between land and sea, mangroves are amongst one of the most at-risk ecosystems when sea level rises (McLeod and Salm 2006).Throughout the Quaternary, mangroves have shown high resilience to disruptions from large sea level fluctuations over historic timescales (Woodroffe 1990).However, adaptation probabilities strongly depend on the rates of sea level rise (SLR) and sediment supplies in combination with subsurface processes that affect sediment elevation (Gilman et al., 2007;Gilman et al., 2006;McLeod and Salm 2006;Wolanski and Chappell 1996;Woodroffe 1990).Ellison and Stoddart (1991) suggested that mangroves are stressed by SLRs of between 9 and 12 cm over 100 yr and concluded that faster rates could seriously threaten mangrove ecosystems.This view has been challenged by Snedaker et al. (1994), who cited historical records showing changes nearly twice that high in mangrove expansion under relative sea level; however hard scientific data or SLR simulations are not available.
As mangrove ecosystems are very dynamic, the ability of these forests to migrate to more landward zones is a very important aspect when considering the effect of SLR on mangroves.If the possibility presents itself, mangroves will adjust to a SLR by expanding landward or laterally into areas of higher elevation, or even by growing upward in place (McLeod and Salm 2006).However, the mangrove areas that Published by Copernicus Publications on behalf of the European Geosciences Union.
are most vulnerable are those situated in a physiographic setting that limits landward migration due to obstacles or steep gradients and with a net decrease in sediment elevation or sediment accretion that is insufficient to keep up with SLR (Gilman et al., 2008).Landward obstructions, artificial or natural, have an effect on ecosystems that would normally move landward in response to erosive forces.Where there is a rise in sea level relative to the land, a coastal squeeze takes place (Doody, 2004).On the species level, adaptation can occur through landward migration at different speeds as mangrove species maintain their preferred hydroperiod or by sediment accretion (Gilman et al., 2008).Mangrove species composition can strongly affect a mangrove's resistance and resilience to SLR given that, on the one hand, individual species have varying tolerances of the period, frequency and depth of inundation, and, on the other hand, different vegetation zones have different rates of change in sedimentation elevation (Krauss et al., 2003;McKee et al., 2007;Rogers et al., 2005).Furthermore, several scientists have also investigated how different functional root types of several mangrove species respond to changes in elevation in order to determine the vulnerability to SLR (Ellison and Stoddart 1991;Vincente 1989).
Species-specific competition may allow some species to outcompete others and to become more dominant within the newly formed species composition (Lovelock and Ellison, 2007).Establishment and dispersal play a significant role in these processes.They are, however, different for various species and strongly dependent on many biotic factors such as buoyancy, period of obligate dispersal, longevity and period of establishment (Allen and Krauss, 2006;Clarke et al., 2001;Drexler, 2001;Tomlinson, 1986), whilst wind and hydrodynamics of tides and currents can be equally important abiotic factors (Stieglitz and Ridd, 2001).Additionally, factors like microtopography, the soil type at the top and root structures can also have a significant effect on the fate of propagules once released from their parental tree, as can human-induced degradation, like tree cutting (Di Nitto et al., 2008).
To date, mangroves have been subjected to non-climaterelated anthropogenic stressors which have accounted for most of the global average annual rate of mangrove loss, estimated to be 1-2 %, with losses during the last quarter of a century ranging from 35 to 86 % (Alongi, 2002;Duke et al., 2007;FAO 2003FAO , 2007;;Valiela et al., 2001).So far, relative SLR has been a smaller threat to mangroves.However, it may constitute a substantial proportion of predicted losses (about 10-20 % of total estimated losses) as several studies have already shown that many mangrove areas have not been keeping pace with current rates of relative SLR (Cahoon et al., 2006;Gilman et al., 2007;McKee et al., 2007).We would like to emphasise the importance of understanding mangrove responses to SLR as these ecosystems provide tremendous social, economic and ecological value (Barbier, 2003;Dahdouh-Guebas et al., 2005;Mumby et al., 2004, Nagelkerken et al., 2008;Walters et al., 2008;Wells et al., 2006).
This study focuses on the critical factor "tidal range" in order to investigate the potential for landward migration of mangrove vegetation assemblages in Gazi Bay (Kenya) under different SLR scenarios.As mangrove species have their preferred hydroperiod, the vegetation distribution in the different inundation classes at present is extrapolated towards future SLR scenarios based on a static mangrove surface elevation.Digital terrain modelling is derived from differential GPS field measurements and used to simulate water levels in a GIS environment.In combination with a mangrove species map, preliminary results are generated regarding the effect of SLR in the study site in Gazi Bay (Kenya).The focus resides on individual mangrove species and their possible colonisation of back-mangrove areas that become accessible when sea level rises.We deliberately adopt a reductionistic approach by taking abstraction of alterations in sedimentation and elevation and other consequences of global change such as increases in temperature, CO 2 concentration and storm frequency, as well as possible shifts in seasonal periods (Pernetta, 1993;UNEP, 1994;Woodroffe, 1990;Woodroffe and Grime, 1999).However we feel that, in this context, relevant conclusions can be made.First of all, this study represents the first attempt to simulate the effect of SLR based on a large amount of detailed information on topography and vegetation covering the whole bay.Secondly, many researchers have already gathered valuable information within this study area on diverse subjects like regeneration, vegetation structure dynamics, human impacts and propagule dispersal (e.g.Abuodha and Kairo, 2001;Bosire et al., 2003Bosire et al., , 2008b;;Dahdouh-Guebas Farid and Koedam, 2006;Dahdouh-Guebas et al., 2002a;Di Nitto et al., 2008;Kairo et al., 2001;Kirui et al., 2008;Neukermans et al., 2008).The latter gives us the opportunity to draw preliminary conclusions on the potential for landward migration of mangroves in Gazi Bay and to create some views on the future vegetation structure dynamics, which can contribute to their resilience to SLR.Resilience is understood here as the survival of the formation, even if displaced in space.
Mangrove species distribution within this study area was obtained by Neukermans et al. (2008).A classification of a Standard QuickBird multispectral satellite image was performed in combination with ground truthing based on vegetation transects by the point-centred quarter method (PCQM+) of Dahdouh-Guebas and Koedam (2006).The two socioeconomically most important species within this study area, R. mucronata and C. tagal (Dahdouh-Guebas et al., 2000, 2004a), are mapped with user's accuracies above 85 %, whereas all four dominant mangrove species (A.marina (on the seaward side (Sw) and the landwards side (Lw)), S. alba, R. mucronata and C. tagal (Sw and Lw)) are mapped with an overall accuracy (OA) of 72 %.

Topographical field survey and construction of a DTM
The UTM zone 37S) and assigning their absolute height, a thorough knowledge of the field was used to add extra points and breaklines in order to eventually optimise the constructed DTM.As the height measurements of these points are relative, we followed the high-water line of a chosen spring tide on two consecutive days and collected the xyz data of 116 points using the DGPS.Based on the Kilindini tide tables (Kenya Ports Authority, KPA) the approximated absolute height of the water was calculated, and the relative elevations in the DTM converted to approximate absolute field topography.We recognise a temporal delay in tides between Mombasa and Gazi Bay; however this does not influence our study.
The final coordinates resulting from the topographical measurements were inserted into a geographical information system (GIS) and served as an input to create a triangular irregular network (TIN) of the area.The TIN was based on the (non-constrained) Delauney triangulation of the original set of points by use of Voronoi diagrams, a theory for which we refer to Raper (1990).In this paper it is not the intention to investigate in depth the impact of these elevation errors through principal component analysis (Lopez, 1997), but we give an estimation of the absolute mean error and the standard deviation in densely covered and less densely covered areas.After extracting 30 points from each of the latter areas, the TIN was reconstructed and height values were reassessed for these particular points.

Spatial analyses
IPCC has predicted several SLR scenarios (+ 9 cm (minimum), + 20 cm (relative), + 48 cm (average) and + 88 cm (maximum)) by the year 2100 (IPCC 2001) 1 based on atmosphere-ocean general circulation models and emission scenarios incorporating uncertainties regarding changes in terrestrial ice, permafrost and sediment deposition.The main purpose of the spatial analyses is to predict possible changes in vegetation assemblages under these different scenarios of SLR.
This modelling exercise mainly focuses on the potential of mangroves to migrate towards landward areas, but it is solely based on sea level rise relative to a static mangrove surface elevation.In this stage, data on sediment-related changes are 1 We based our analysis on SLR scenarios of the IPCC Third Assessment Report (TAR) (2001) and not on those of the Fourth Assessment Report (AR4) (2007), which forecast a range from 9 to 88 cm by 2100 and a range from 18 to 59 cm by 2090-2099, respectively.The reason is the following: due to lacking of published literature, AR4 models do not include uncertainties in climate-carboncycle feedback, nor do they include the full effects of changes in ice sheet flow.The AR4 projections do, however, include a contribution due to increased ice flow from Greenland and Antarctica at the rates observed for 1993-2003, but these flow rates could increase or decrease in the future.The AR4 could have similar ranges to those of TAR if uncertainties were to be treated in the same way.
not available; however we do not underestimate the importance of sediment in mangrove vegetation dynamics in view of SLR.
The modelling exercise started with an assessment of the current species-related zonation or spatial structure present in Gazi Bay.First of all, the height boundaries for each inundation class according to Watson (1928) (Table 1) was defined based on the combination of the tide tables (July 2003-July 2004) published by the KPA and the monthly inundation frequencies per class (Table 1).In further analysis, inundation frequencies higher than those of "class 1" will be referred to as "class 0".Using ArcGIS 8.2, these boundaries were classified into inundation classes based on the DTM for the current scenario versus different IPCC scenarios of eustatic SLR.The relative scenario of + 20 cm coincides with the current trend of SLR within the long-term data set (1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003) obtained from gauge measurements by the Kenya Marine and Fisheries Research Institute at the Kilindini Port in Mombasa.This initiative is part of the Global Sea Level Observing System (GLOSS) founded by the Intergovernmental Oceanographic Commission (IOC) of UNESCO.
Secondly, an overlay between the vegetation map and the current inundation classes (Fig. 2b) gives an estimation of the vegetation surface of each species within each inundation class.To review the accuracy of the DTM and/or the classification of the vegetation, it is important to investigate whether the distribution of the species within the inundation classes deviate from a random distribution.To perform the statistical analyses, the complete area was divided into 10 equally sized blocks.Within each block the areal coverage (ha) was calculated for each species in all inundation classes of the current situation.Secondly, a Kolmogorov-Smirnov test was performed to compare the observed cumulative distribution function to a theoretical normal distribution, whereafter Kruskal-Wallis tests were completed to investigate whether the vegetation distribution within the inundation classes is random.Since the species concerned are not randomly distributed, extrapolations of the vegetation structure towards future IPCC scenarios of SLR were performed.The area increase (%) of each inundation class within each scenario was calculated in relation to the current situation, whereafter these percentages were multiplied by the current vegetation area (ha).

Sensitivity analysis
A source of uncertainty in the input data is the DTM's absolute height, which was calibrated using Kilindini port gauge measurements.To address the sensitivity of the model to the absolute height uncertainty of the DTM, we investigated the impact of changes in the height boundaries of the inundation classes.Upper and lower height boundaries are slightly altered at a time and in a systematic manner, more specifically by an increase and decrease of these boundary intervals with 5, 10 and 15 % corresponding to 4, 6 and 8 cm.The Table 1.Inundation classes and monthly inundation frequency according to Watson (1928).Height boundaries (metres above datum) of present and future inundation classes are presented: a minimum (+9 cm), relative (+20 cm), average (+48 cm) and maximum (+88 cm) scenario is based on IPCC eustatic SLR scenarios for the year 2100.In further analysis, inundation frequencies higher than those of "class 1" will be referred to as "class 0".AHT stands for all high tides; MHT, medium high tides; NHT, normal high tides; SHT, spring high tides; and EHT, equinoctial high tides.Inundation frequencies higher than those of "class 1" will be referred to as "class 0".
comparison between the reference map (Fig. 2C1) and the output maps after altering the height boundaries was assessed with an error matrix, giving overall (OA), user's (UA) and producer's accuracies (PA) (for calculations, see Appendix A).

Construction and validation of the digital terrain model
The DTM of the study area is shown in Fig. 2a.After postprocessing in SKI, 4105 points were accepted with an average error on x, y and z of 1.16, 2.08 and 0.89 cm, respectively, whereafter several breaklines and 82 extra points were manually added to optimise the DTM.Breaklines along the creek banks are, however, crucial and had to be added as estimates (based on measurements within the creek) due to high mangrove coverage.Absolute mean error and standard deviation for densely covered and less densely covered areas are 0.013 m ± 0.106 and 0.089 m ± 0.374, respectively.

Simulation of sea level rise scenarios
The current situation covers a total (studied) area of 423.43 ha, of which the regularly flooded area and the nonflooded area encompass 386.53 and 36.90 ha, respectively.When looking at the inundation classes within the different scenarios (Fig. 2C1 to C5), we can conclude that there is an overall trend of transgression into the terrestrial areas.The maximum scenario in particular (+ 88 cm) represents a significant area increase of class 0 and class 1 (AHT) (for abbreviations, see Fig. 2c).More specifically, the percentage area increase of these two classes from the current situation towards the maximum scenario of SLR is 245 and 103 %, respectively.After calculating the extent of each mangrove species within each current inundation class, Kolmogorov-Smirnov tests were completed with results showing significance values < 0.05 for each species.The vegetation distribution is therefore not normal and nonparametric techniques have to be used for further analyses.The following Kruskal-Wallis test proved that the distribution of the vegetation within the inundation classes is not random; all significance values are < 0.05.Each species evaluated within the area has a preference for certain inundation classes, confirming the occurrence of a specific zonation or spatial structure in Gazi Bay and therefore also an adequate accuracy of the field measurements.
Due to the errors on the classification of the vegetation map (see Sect. 2.1 and Neukermans et al., 2008) and the topographical measurements, the total area (ha) occupied by each mangrove species within the whole study area (TMA) does not fully coincide with the total area (ha) occupied by each mangrove species within the inundation classes at present (TMAI).This, however, does not exceed values between 2 and 12 (Table 2), except for Sonneratia alba, which mainly occurs in class 0 (38 %) and class 1 (AHT) (35 %), consequently being the only species with a high difference between TMA and TMAI of 61 %.The high discrepancy between TMA and TMAI for S. alba could be explained by a possible lower accuracy of the DTM at the breaklines marking the creek bank.
All other species appear to have an adequate distribution within the whole study area: Avicennia marina Sw (seaward) mainly resides in class 1 (AHT) (26 %) and class 2 (MHT) (45 %), Rhizophora mucronata mainly appears in class 2 (MHT) (53 %) and class 3 (NHT) (22 %), whilst Ceriops tagal, which is an inner mangrove, occupies the areas in several mid-classes.A. marina Lw (landward) dominates the landward classes with 35 % in class 4 (SHT).An extrapolation of changes in vegetation assemblages towards future scenarios (Fig. 3a) demonstrates that, in comparison to the average scenario of SLR (+ 48 cm), all species will decrease in the maximum scenario (+ 88 cm), resulting in a decline of 13 % in 100 yr.Although most species show a possible area increase throughout the minimum, relative and average scenario, this is not the case for A. marina Lw, as this species will diminish throughout all scenarios, with a highest decrease of 60 % in the maximum scenario.When considering the two socioeconomically most important species R. mucronata and C. tagal in the most probable relative scenario of + 20 cm SLR, an area increase of 15 % occurs in comparison to the current situation.Finally, the area proportions between the total mangrove area, the non-flooded area and class 0 are shown in Fig. 3b as percentage increase or decrease compared to the current situation.The maximum scenario shows a considerable decrease in total mangrove area of 13 %, whereas for the relative scenario this area increases by 4 %.The most marked increase is for the area of class 0, namely 245 % in comparison to the current situation.

Sensitivity analysis and error matrix for map comparison or accuracy assessment
Table B1 (see Appendix B) shows the results of the error matrices for map comparison or accuracy assessment.When comparing the vegetation distribution within adjusted height boundaries for each inundation class, the outcome appears to be relatively sensitive to an increase or decrease of 15 %.The overall accuracy, with a comparable outcome for K hat , fluctuates between 87.34 and 65.88 % when considering an increase or decrease up to 10 %, yet strongly declines towards 53.61 to 48.02 % when height boundaries of each inundation class are adjusted by 15 %.As the applied vegetation classification confirms the occurrence of a specific zonation or spatial structure in Gazi Bay, which is highly related to inundation patterns, we can conclude that sensitivity to alterations in topography can be significant from a certain limit and should therefore be aligned to vegetation distributions when data are available.Furthermore, the wind setup may have affected the high-water-line measurements at spring tide.In addition, the inundation classification according to Watson (1928), which is based on inundation frequency, may not always yield fully satisfactory results, especially in regions with an irregular tidal regime and/or irregular elevation profile, where the duration of inundation seems equally important, as was shown by Van Loon et al. (2007).

Discussion
This study was to investigate whether mangrove assemblages in Gazi Bay have the potential to migrate to more landward areas, which can contribute to their resilience to SLR (Fig. 4), understood as the survival of the formation within the site.Although the focus of this study was mainly on tidal range, we emphasise the importance of sediment supply, especially for scenarios of SLR higher than 20 cm 100 yr −1 (relative scenario).Whether mangroves can be resilient to SLR strongly depends on the physiographic setting in which these ecosystems occur, human activities that are carried out in the wetland and on how species-specific competition and adaptation will unfold.There is no clear-cut answer that can be applied to global mangrove coverage; however, by studying this particular mangrove area with a mesotidal regime and a common vegetation zonation along a gentle slope gra-dient from land to sea, extrapolations can be made to areas with similar characteristics.

Vegetation dynamics of mangrove assemblages under different scenarios of SLR
Bearing in mind the reductionistic approach, the extent of the most common assemblages, apart from Avicennia marina and Ceriops tagal on the landward side (Lw), are forecasted to increase in surface under the different scenarios of SLR (except for the maximum scenario of + 88 cm).This forecast is in line with a few earlier reports that current sea level rise rates do not pose a threat to mangrove ecosystems (e.g.Mc-Kee et al. 2007;Snedaker et al., 1994;Tan and Zhang, 1997), but contradicts many others (e.g.Ellison and Stoddart, 1991;Fujimoto and Miyagi 1990;Parkinson et al., 1994;Pernetta, 1993).However, considering the uncertainties regarding the impact of global change on mangrove growth and development, such contradictions are not unexpected.In addition, our reductionistic approach focuses on tidal range and the possible dispersal range of propagules, but it does not take into account the biogeomorphological capacity to maintain or to protect a mangrove forest.Landward migration of mangroves in Gazi Bay appears to be limited under the maximum scenario as the highest intertidal inundation class strongly decreases due to the topographical settings at the edge of the inhabited area.Consequently, the coastal squeeze will signify a decrease in the Avicennia-dominated assemblages if they fail to adapt to a more frequent inundation or if competition with other species prevails.Dahdouh-Guebas et al. (2004a) made a prediction of future vegetation structure in Gazi Bay based on retrospective remote sensing, social surveys and tree distribution, and results show that the surface extent of A. marina on the landward side has been reducing since 1972.Furthermore, the current situation in Gazi Bay is characterised by large bare and sandy sites on the landward side which have remained in the same state for a substantial time; that is, we have observed no colonisation for at least 16 yr).When landward areas are accessible during SLR, dispersal and early growth become important stages in a plant life that fundamentally determine community structure and population dynamics (Clarke et al., 2001;Sousa et al., 2007).These processes are very complex.A dense mangrove forest can provide an adequate propagule supply for dispersal towards newly colonisable areas, but (1) as Clarke et al. (2001) stated, establishment of young trees is mainly related to the presence of parental trees, while this is not so much the case for juveniles and the hydrochorous dispersal of propagules, and (2) suitability for stranding or self-planting of propagules is strongly dependent on the presence of root structures (which can facilitate the entanglement of propagules) and the compactness of the soil (clay-or siltdominated) (Di Nitto et al., 2008).
As in other transitional systems, plant establishment and community succession is driven by tolerance to physiological  stress and plant-plant interactions (Bertness, 1991;Milbrandt and Tinsley, 2006); hence species-specific competition could signify a natural blockage for landward migration of mangroves.However, in several cases, facilitation is a common mechanism of succession in terrestrial habitats, meaning that an early coloniser changes the abiotic conditions in a way that allows an entry and finally a displacement of a second species to a previous intolerable habitat ( Connell and Slayter, 1977).This was, for instance, the case for (1) saltwort (Batis maritima L.), as it was identified as an abundant initial coloniser of an extensive black mangrove (Avicennia germinans L.) die-off area (Milbrandt and Rinsley, 2006), and (2) salt marsh cordgrass (Spartina alterniflora Loisel.),being a potential initial soil stabiliser creating successional stages firstly for Laguncularia racemosa (L.) C. F. Gaertn, which is secondly outshaded and replaced by Avicennia schaueriana Stapf and Leechm.ex Mold.(Cunha-Lignon et al., 2009).
The reported forecasts can also have an important socioecological implication.Although the forest adjacent to the village has long been overexploited for wood and decreased in area, anthropogenic disturbance has diminished over the last years and some mangrove assemblages have even expanded (Dahdouh-Guebas et al., 2004a).An increase in mangrove area under different scenarios of SLR, provided that it does not go at the expense of qualitative degradation, may imply an increase in anthropogenic threats such as traditional utilisation (McLeod and Salm, 2006).Clear felling of mangroves species can have severe consequences for future vegetation dynamics.Furthermore, most mangrove creeks (like the case in Gazi Bay) are characterised by the occurrence of time-velocity asymmetry in which ebb flow is more dominant than flood flow (Kitheka 1997(Kitheka , 1998;;Kitheka et al., 2002).Sediment trapping occurs during incoming flood tides, and there is no significant export of sediments during ebb tide (Furukawa and Wolanski, 1996;Wattayakorn et al., 1990); however degradation of mangroves can lower trapping efficiency (Kitheka et al., 2002), consequently increasing vulnerability to sea level rise.McLeod and Salm (2006).This three can be applied once candidate sites of high biodiversity have been selected using biological criteria, for instance *, factors that indicate strong recovery potential (see Appendix Table C1).This decision tree was adjusted (**) to implement the possibility of a shift in vegetation structure.MPA = marine protected area.

Vegetation dynamics of individual species under different scenarios of SLR
When landward areas become accessible for the migration and colonisation of mangrove species, we have to ask the same question as Alongi (p4, 2008): "Are trends in mangrove forest replacement in response to catastrophic disturbances the result of somewhat deterministic sequences as in terrestrial forests, or are they the result of a stochastic "first come, first served" opportunistic response or neither?"Empirical data support the idea that recovery is stochastic with distinct succession stages, yet early sequences of species replacement are greatly influenced by species present at initial recovery (Alongi, 2008;Clarke et al., 2001;Sousa et al., 2007).Within this study the extrapolation of the present vegetation distribution towards scenarios under a rising sea level is based on species-specific preference for certain inundation frequencies.The survival of these species in their shift in a more landward direction is strongly dependent on their colonisation rate and interspecific competition.The most seaward mangrove species Sonneratia alba appears in vege-tation zones that are daily inundated and are never submitted to large salinity variations (Tomlinson, 1986).When sea level rises, this species is forecasted to increase in area (except under the maximum scenario), yet as investigated by Dahdouh-Guebas et al. (2004a), the juvenile layer within these S. alba stands is limited and propagule establishment is hampered by currents that are generally known to be strongest along the seaward side (Diop et al., 2001).The distribution of the young individuals of S. alba is more related to the adult trees, whereas juveniles are generally spread over a wider area (Dahdouh-Guebas et al., 2004a).The latter also applies for the species Avicennia marina on the seaward side (Sw).Furthermore, Imai et al. (2006) verified that S. alba seedlings and saplings, which require sunny conditions for their growth, were more abundant in gaps than in the understorey.Competition with a more landward species such as Rhizophora mucronata might demonstrate that an area increase of S. alba could be overestimated by our analyses.However, colonisation by S. alba on seaward sand banks has occurred throughout the years.Additionally, bearing in mind the site-specific rates of sea level rise and sediment input rates, Ellison and Stoddart (1991) claimed that mangrove ecosystems can keep pace with SLR of 8-9 cm per 100 yr, making seaward expansion and colonisation of these daily inundated areas possible.Rates of 9-12 cm per 100 yr will cause stress, yet the ability for mangroves to adjust to even higher rates is unlikely.The minimum scenario of SLR (+ 9 cm) could in fact provide an additional and suitable habitat for S. alba and A. marina (Sw).R. mucronata and Ceriops tagal are two economically valuable pioneer species that will most likely increase as predicted unless anthropogenic impact rises.Multivariate vegetation structure analysis showed that C. tagal is very abundant in the understorey of assemblages dominated by other mangroves, which could mask a dynamic shift (Dahdouh-Guebas et al., 2004a).R. mucronata and C. tagal already occupy the mid-zone within the mangrove area and knowledge on the dispersal of their propagules indicates that prop roots and pencil roots clearly have the ability to entangle propagules and that preference of propagule dispersal goes to flat areas and substrates with a more compact soil structure (clay, silt) (Di Nitto et al., 2008).One disadvantage for R. mucronata could, however, be represented by a further siltation along the seaward sand bank creating a patch of arid conditions and higher light intensity more favourable for A. marina (Dahdouh-Guebas, et al. 2004a).
Avicennia marina (Lw) will have to adapt to greater inundation frequencies.It is known that this species can tolerate high salinity variation, so could the double zonation of this species on the landward side versus the same species on the seaward side support the idea of dynamic adaptation?Genetic analyses based on 48 RAPD (randomly amplified polymorphic DNA) loci have demonstrated that four DNA fragments show a slight differentiation in allelic frequency between the two A. marina stands in spite of their short distance separation (Dahdouh-Guebas et al., 2004b).This indicates that there is less genetic exchange between the disjunctive stands than within one stand, consequently suggesting that an ecological or physical barrier might exist.Tidal range might facilitate the dispersal of propagules in both directions; however obstruction by complex root structures can prevent this exchange.Additionally, interspecific competition with the adjacent species C. tagal could disadvantage A. marina as McCusker (1977) confirms that a salinity increase causes a reduction in water use efficiency for the seedlings of Rhizophora, but not for Ceriops.Furthermore, an elevated CO 2 level will enhance the efficiency of water use (UNEP, 1994); however this advantage is lost when salinity becomes too high, for instance, at low inundation frequency areas on the landward side.Another drawback for A. marina is an increase of temperature, since this species has lowest optimal temperature for leaf development (Hutchings and Saenger, 1987).
There are several well-established physiologic mechanisms influencing mangrove community composition (Duke et al., 1998;McKee, 1995), yet research is needed on interspecies interactions influencing mangrove forest regeneration in post-disturbance mangrove communities.

Recommendation for further research and management strategies
In the light of mangrove ecosystem stresses caused by climate change, managers face the dual challenge of selecting and implementing conservation strategies in order to maintain and restore resilient mangrove forests.
In this study the emphasis resides on tidal range and not on sediment supply; however, we give a preliminary vulnerability assessment of this mangrove area based on a slightly adjusted decision tree (Fig. 5) to aid resilient site selection for mangroves by McLeod and Salm (2006).This decision tree was applied after appointing Gazi Bay as a high biodiversity candidate site based on biological and environmental criteria (Table C1; see Appendix C).Decisions were made based on the available literature involving the mangrove area in Gazi Bay and the relative SLR scenario of + 20 cm, which coincides with the current trend along the coast of Kenyan.
Following this decision tree, the mangrove area in Gazi Bay appears to be adequately resilient for at least 100 yr and can most likely be appointed as a marine protected area (MPA).However, although we do not intend to focus only on MPAs, we want to anticipate a future scenario of sea level rise and indicate gaps in, on the one hand, scientific and, on the other hand, site-specific knowledge that necessitate further research.Given (1) the mesotidal regime and permanent rivers and creeks that provide freshwater and sediment (mainly during wet season), (2) the knowledge that the drainage basin of both Mkurumuji and Kidogoweni rivers, which extend into the coastal ranges of the Shimba Hills National Reserve, has limited anthropogenic pressures with respect to the intactness of the hydrological regime, and given that (3) landward migration in Gazi Bay is possible under the relative scenario of sea level rise, the decision tree leads us towards the question whether recruitment is strong.The answer is definitely "yes"; however we feel that the possibility of a shift in vegetation structure needs to be implemented, rendering Gazi Bay a site that is "Maybe OK for MPA".According to McLeod and Salm (2006) the decision tree would have led towards "Good choice for MPA".
The recommendations for further research and management strategies, which can be applied globally, are the following: (1) identifying an early coloniser to promote early establishment of mangrove seedlings, (2) measuring changes in elevation by means of surface elevation tables (SETs), (3) assuring the possibility of landward migration and (4) investigating propagule dispersal by combined hydrodynamic and ecological behaviour modelling.(2004a), Kitheka (1996Kitheka ( , 1997)), Njambuya (2006)

Fig. 1 .
Fig. 1.Representation of (A) the Kenyan coast (Dahdouh-Guebas et al., 2000) and (B) Gazi Bay.The satellite image (QuickBird) shows the whole bay of Gazi ; however this research focuses on the western part as encompassed by the overlaid vegetation map.S. alba = Sonneratia alba, R. mucronata = Rhizophora mucronata, C. tagal Lw = Ceriops tagal on the landward side, C. tagal Sw = Ceriops tagal on the seaward side A. marina Lw = Avicennia marina on the landward side and A. marina Sw = Avicennia marina on the landward side.Classification of the mangrove species coverage was obtained by Neukermans et al. (2008).

Fig. 3 .
Fig. 3. (A) Graph of the total area (ha) per species within the four SLR scenarios, (B) future prediction of the dynamics of mangroves, non-flooded area and class 0.

Fig. 4 .
Fig. 4. Overview scheme summarising the discussion on resilience of mangroves facing sea level rise, more specifically concerning the case study in Gazi Bay (Kenya).

Fig. 5 .
Fig.5.Decision tree to aid resilient site selection for mangroves according toMcLeod and Salm (2006).This three can be applied once candidate sites of high biodiversity have been selected using biological criteria, for instance *, factors that indicate strong recovery potential (see Appendix TableC1).This decision tree was adjusted (**) to implement the possibility of a shift in vegetation structure.MPA = marine protected area.
,Obade et al. (2004),Ohowa et al. (1997) Sediment-rich macrotidal environments to facilitate sediment redistribution and accretion Yes Actively prograding coast and delta Yes Natural features(bays, barrier islands, beaches, sandbars, reefs)  that reduce wave erosion and storm surge Yes Factors that allow for landward migration Mangroves backed by low-lying retreat areas (for example, salt flats, marshes, coastal plains) which may provide suitable habitat for colonisation and landward movement of mangroves as sea level rises No/Yes in certain places e.g.Di Nitto et al. (2008), Neukermans et al. (2008), Obade et al. (2004) Mangroves in remote areas and distant from human settlements and agriculture, aquaculture, and salt production developments Yes Mangroves in areas where abandoned alternate land use provides opportunities for restoration, for example, flooded villages, tsunami-prone land, unproductive ponds Yes, unmanaged coconut plantations Factors that enhance sediment distribution and propagule dispersal Unencumbered tidal creeks and areas with a large tidal range to improve flushing, reduce ponding and stagnation, and enhance sediment distribution and propagule dispersal Yes e.g.De Ryck (2009), Di Nitto et al. (2008), Kitheka (1996, 1997), Ohowa et al. (1997) Areas with a large tidal range that may be better able to adjust to increases in sea level due to stress toleranceYes Permanent strong currents to redistribute sediment and maintain open channels Yes Factors that indicate survival over time Diverse species assemblage and clear zonation over range of elevation (intertidal to dry land) Yes e.g.Beeckman et al. (1989), Bosire et al. (2008a), Bosire et al. (2006), Dahdouh-Guebas et al. (2002a), Dahdouh-Guebas et al. (2004a), Dahdouh-Guebas et al. (2002b), Kairo (2001), Kairo et al. (2001), Neukermans et al. (2008), Tack et al. (1992), Van Tendeloo (2004) Range in size from new recruits to maximum size class (location and species dependent) Yes Tidal creek and channel banks consolidated by continuous dense mangrove forest (which will keep these channels open) Yes Healthy mangrove systems in areas which have been exposed to large increases in sea level due to climate-induced sea level rise and tectonic subsidence No Factors that indicate strong recovery potential Access to healthy supply of propagules, either internally or from adjacent mangrove areas Yes Strong mangrove recruitment indicated by the presence, variety, and abundance of established mangrove propagules Yes Close proximity and connectivity to neighbouring stands of healthy mangroves Yes Access to sediment and freshwater Yes Limited anthropogenic stressYes, no major residential area in the vicinity, selected as a fairly pristine East African site in the EU PUMPSEA project: http://www.pumpsea.icat.fc.ul.pt/main.phpUnimpeded or easily restorable hydrological regime Yes Effective management regime in place such as the control of usual threats like dredging and filling, conversion to aquaculture ponds, and construction of dams, roads, and dikes that disrupt hydrological regime, etc.YesIntegrated coastal management plan or protected area management plan implemented

Table 2 .
Presentation of the total area (ha) occupied by each mangrove species with the whole studied area (TMA) and the total area occupied by each mangrove species within the inundation classes at present (TMAI).Sw = seaward side and Lw = landward side.

Table C1 .
McLeod and Salm, 2006)ctors that contribute to site selection (according toMcLeod and Salm, 2006).Case study: Gazi Bay, Kenya.Factors that allow for peat building to keep up with sea level rise Applicable to Gazi Bay Yes/No Literature available per factor Association with drainage systems including permanent rivers and creeks that provide freshwater and sediment Yes e.g.Dahdouh-Guebas et al.