surface sediment accretion measurements
The RSET measures all processes affecting sediment elevation over the depth of the rod benchmark. To separate below-ground processes (e.g. compaction, decomposition, root production and shrink/swell from water flux) from surface processes of accretion and erosion, artificial soil marker horizon plots were established in the vicinity of each RSET station (Fig. 3) (Cahoon et al. 1995). Three replicate 0.25 m2 markers (local beach sand) were laid at the same time that the baseline RSET readings were taken (4–7 May 2000). Surface sediment accretion was measured from short cores collected from each marker horizon plot. Multiple measurements of depth from the sediment surface to the marker horizon were made within each core. As with the RSET, any material not yet incorporated into the sediment matrix was not considered as the sediment surface. The number of cores in which no marker horizon was recovered was recorded, and generally only one core was measured per horizon per sampling. RSET stations and marker horizons were read in May and August 2000 and January, April and August 2001. Shallow subsidence, or the influence of subsurface processes on sediment elevation, was calculated as surface sediment accretion minus elevation change (Cahoon et al. 1995).
forest structure assessment
Mangrove forest structure characteristics were obtained through a census of live and dead trees, recording mangrove species and trunk diameter, as well as seedlings and propagules within randomly chosen, marked 5 × 10 m quadrats, one within each of the 18 plots, conducted in January 2000 and 2001. Separate diameters were recorded for each main vertical trunk in multiple-trunk individuals (e.g. shoreline red mangrove). Diameters were recorded at the greater of either breast height (1.5 m), or above the highest prop root (in the case of the red mangrove). Tree density per plot was estimated for six age and diameter size classes: seedlings, < 2 cm, 2–4.9 cm, 5–9.9 cm, 10–19.9 cm and 20+ cm. Three additional plots were surveyed within the dead black mangrove zone along the north shore of Roatan in January 2001.
Soil strength, defined as the torque required to shear or to deform the soil, was measured with a Torvane device (H-4212 1, Humbolt Manufacturing Company, Durham Geo-Enterprises, Inc.). Soil cores (12.5 cm diameter × 30 cm deep, 6 per site) were collected in January 2001 and cut in half vertically, creating the flat surface needed for measurement. Measurements were taken at the surface and at 5, 15 and 25 cm and the three subsurface values were averaged per core.
Relative root production rate (g m2 year−1) was determined using the implanted mass technique (Gallagher et al. 1984). A total of six soil cores (7.3 cm diameter × 30 cm deep) were removed from sites within each zone by impact combination with a coring device, and in-growth bags containing root-free, organic material were inserted in January 2000. The in-growth bags were constructed of loose nylon mesh (3 mm2, J & M Industries, Ponchatoula, LA) and filled with milled sphagnum peat that approximated the bulk density and organic matter content of native mangrove peat. Native mangrove peat could not be used because it was primarily composed of fine roots, which would have been indistinguishable from in-grown roots. Bags were positioned with the tops level with the soil surface and tied with monofilament to a nearby prop root and exhumed in January 2001 with a 12.5-cm diameter corer. Soil and roots surrounding the excavated bags were carefully severed at the bag surface before transferring the contents to plastic bags for processing within 48 h of collection.
In-grown root material was washed and separated over a sieve (1 mm2) with freshwater. Large, non-root particles were separated by hand, and roots and root fragments were separated by flotation. Roots were separated into two size categories (coarse, > 2 mm diameter; fine, < 2 mm diameter). Total recovery of root material was about 88% (based on subsamples processed to completion under magnification in the laboratory). Recovery of root fragments from bags deployed at the high impact site (i.e. where total mortality occurred) showed that dead roots could be physically transported into in-growth bags from the surrounding peat. Although these dead roots were not quantified, some of the un-recovered root fraction in other samples is probably attributable to this extraneous source. Root samples were air-dried in the field and later oven-dried in the laboratory at 70 °C and weighed.
Sediment accretion and elevation data resulted from a repeated measures, completely randomized treatment design, with the impact-by-zone treatment measured sequentially over time. The error structure was nested (pins within positions within plots, and cores within marker horizons within plots), and plots were the experimental unit for the treatment factor. Sediment elevation (RSET) and vertical accretion (marker horizon) data were analysed separately, although identical models were used. RSET data were calculated as cumulative change for each pin from the baseline reading. Average cumulative change was then calculated over the nine pins, then over the five positions within each RSET station. Marker horizon data were averaged per core, then per horizon and per plot. One sampling station (RSET with corresponding marker horizons) was lost in the medium impact basin before measurements could be taken, which resulted in an unbalanced design. Durbin-Watson tests (testing for significant first-order serial correlation) were non-significant, and analysis of covariance was therefore used, with time as the covariate (SAS Proc Reg, SAS©Version 8; SAS 1999). The data sets included zeros corresponding to the baseline readings or initial deployment of the marker horizons, and the models forced the intercept through the origin. Different slopes therefore corresponded to the different impact by zone treatment combinations. A stepwise regression procedure was invoked to select for the most efficient model, which grouped similar treatment levels according to their trends over time.
Mangrove mortality recorded in January 2000 was modelled as a logistic function of impact level, zone, and their interactions (Proc Logistic, SAS© Version 8; SAS 1999). Separate models considered all tree diameter size-classes, or only adult trees.
Soil strength and root production data were analysed as a 3 × 2 factorial design, where impact level and spatial position were grouping factors (two-way anova; JMP© Version 4.0.2; SAS 2000). Data were log (ln(x + 1)) or inverse (1/(x + 1)) transformed prior to analysis where necessary to reduce heterogeneity of variance and to reduce deviations from normality. Comparisons after the test among treatments were described with a Tukey HSD test or 1 degree of freedom contrasts for single comparisons of interest between two effects.
relative elevation model
The empirical elevation and accretion data sets described above incorporate not only the processes of sediment deposition and erosion, root growth, compaction and decomposition, but also feedback mechanisms on the processes themselves (e.g. a change in elevation alters flooding patterns that in turn affect rates of sediment deposition, decomposition and autogenic primary production). The short time scale (15 months) of these data sets, however, limits their usefulness in predicting the future relationship between sediment elevation and sea level on a longer time scale (e.g. decades) because compaction, decomposition and the feedback mechanisms are non-linear processes that change with time and elevation. For this reason, we used a previously published wetland sediment development model (Morris & Bowden 1986; Callaway et al. 1996; Rybczyk et al. 1998; Chen & Twilley 1999; Day et al. 1999; Rybczyk & Cahoon 2002) to examine changes in sediment elevation over longer time scales.
The model utilizes a cohort approach (tracking discrete packages of sediments through depth and time) to simulate sediment dynamics (organic and mineral matter accretion, decomposition, compaction and below-ground productivity). These dynamics produce model-generated changes in sediment characteristics including bulk density, organic matter volume and mass, mineral matter volume and mass and pore volume. The model yields total sediment height as an output. Sediment height is then balanced with eustatic sea-level rise and deep subsidence, both forcing functions, to determine wetland elevation relative to sea level. The model was programmed using STELLA iconographic modelling software (Richmond et al. 1987). An Euler numerical method, with Δt = 1 week, was used to solve the finite difference equations generated by the STELLA software. The model consists of three linked submodels or sectors: (i) primary productivity, (ii) sediment dynamics, and (iii) relative elevation. The sediment dynamics submodel was validated previously by Rybczyk et al. (1998) using the dimensionless statistic EF (modelling efficiency) described by Loague & Green (1991). EF parallels the coefficient of determination (R2) except that the lower bound for the EF is negative infinity while the lower bounds for R2 is zero. A perfect fit would be indicated by an EF value of 1, and values less than 0 would be indicative of a poor fit (Mayer & Butler 1993). For the sediment model used in this study, EF values ranged from 0.21 to 0.97.
We applied the model to the basin forest on Guanaja, which suffered virtually complete tree mortality and exhibited little potential for natural forest regeneration because there was no regrowth, propagule colonization or adjacent live forest to serve as a source of propagules. To simulate the effect of Hurricane Mitch in the interior forest on Guanaja, we first ‘turned off’ the primary production sub model and modified leaf, wood and root litter functions to reflect the instantaneous death of all primary production. All leaves were instantaneously pulsed to the forest floor, and all previously live roots were shunted to the soil litter pools (either as labile or refractory material). Wood, however, was fluxed to surface litter at a slower rate to reflect the observations that much remains as standing dead at the site. We ran the impact model for 10 years (simulated years 1998–2008) with no production inputs assuming current sea-level rise of 0.15 cm year−1 (Church et al. 2001).
We used elevation and accretion data collected as part of this study to initialize the model (Table 1). The model was calibrated by comparing observed percentage organic matter over depth (McKee & McGinnis 2003) with the simulated output of the same parameter. A full description of the model is provided in Rybczyk et al. (1998) and Cahoon et al. (2003). Because the primary production submodel was turned off, the sediment dynamics submodel, which simulates sediment collapse and decomposition of sediment organic matter, was the most important part of the model.
Table 1. Initialization parameters for the Guanaja Basin Sediment Elevation Model
|Sea-level rise||0.15 cm year−1||(Church et al. 2001)|
|Deep subsidence rate||0.17 cm/year−1||Emery & Aubrey (1991)|
|Initial wetland elevation||37.5 cm above MLLW||This study|
|Mineral input||385 g/cm2 year−1||McKee & McGinnis (2003)|
|Root standing crop||15 000 g dw m2||Chen & Twilley (1999)|
|Above ground standing crop||20 000 g dw m2||Chen & Twilley (1999)|
|Sediment bulk density at surface||0.16 g dry soil cm−3||McKee & McGinnis (2003)|
|Percent organic matter at surface||57.6%||McKee & McGinnis (2003)|
|Decomposition rate of deep refractory organic matter (kdeep)||0.0009 week−1||By calibration|
|Decomposition rate of labile OM (klab)||0.025 week−1||By calibration|
|Decomposition rate of surface labile OM (klabsurf)||0.30 week−1||By calibration|
|Decomposition rate of refractory OM (kref)||0.0009 week−1||Similar to Chen & Twilley (1999)|
|Labile fraction of above-ground biomass (leaf_lab_frac)||50%||By calibration|
The sediment dynamics submodel has four state variables, each measured once in each of 18 soil cohorts (labile organic matter, refractory organic matter, mineral matter and live root biomass). Maximum mineral inputs (Table 1) are the only forcing functions in this submodel, as other inputs are model generated. This submodel simulates the decomposition of organic matter, the inputs of mineral matter, the distribution of root biomass, sediment compaction and the transfer of material from cohort to cohort. Changes within the cohort caused by decomposition, which is a function of model-generated depth, are calculated on a weekly basis. Sediment compaction, also calculated weekly, is a function of initial pore space (a forcing function) and the mass of material above a particular cohort. Measurements obtained from soil cores (e.g. bulk density, percentage organic matter and mineral matter), along with measurement of accretion rates derived from horizon markers, all collected as part of this study, provide the data which are used to calibrate the submodel.
The model separates all organic matter into labile and refractory pools, each with its own time-dependent decay rate. Additionally, the labile organic matter decomposition rate for the surface cohort is separate from that for the rest of the cohorts (allowing for a distinction from leaf and root labile organic matter). Finally, there is a separate, depth-dependent decomposition rate for deep refractory material. A simple negative exponential (–k) model describes decomposition for each organic matter state variable in each cohort. Required decomposition constants include kdeep, klab, kref, leaf_lab_frac, rlab% and klabsurf (Table 1).
Previous models have simulated mineral inputs as a function of wetland elevation (French 1993; Callaway et al. 1996). A similar approach is used here where mineral inputs are a simple linear function of elevation.
Although root production is simulated in the productivity submodel, root biomass is distributed to the sediment cohorts in the sediment submodel. We used an adaptation of the distribution algorithm originally developed by Morris & Bowden (1986), where root biomass is assumed to be greatest near the surface and decreases exponentially with depth. A complete description of this function is provided in Rybczyk et al. (1998).
Soil compaction is a function of organic matter decomposition and the reduction of sediment pore space (primary consolidation; Penland & Ramsey 1990). Callaway et al. (1996) simulated the compaction of pore space as an asymptotic decrease with depth, bounded by preset minimum and maximum pore space values. We used a modified version of Callaway's algorithm, where the decrease in pore space for a given cohort is a function of the mass of material above it. Again, a complete description of this function is provided in Rybczyk et al. (1998).