Task 2: Fragmentation Analysis

Dr. Patrick Biber and Dr. Saranee Dutta in the Center for Plant Restoration (CPR) at the Gulf Coast Research Laboratory are leading the fragmentation analysis task.

Synopsis:

Landscape fragmentation is the “breaking up of vegetation or other land cover types into smaller patches that impedes flows of organisms, energy, and material across a landscape”. Landscape pattern analysis is the quantification of landscape patterns and their spatio-temporal dynamics. A variety of landscape metrics or indices have been proposed to measure landscape composition and configuration, most can be calculated in landscape analysis packages such as FRAGSTATS. Fragmentation will be assessed to generate both patch and landscape metrics. We applied FRAGSTATS analyses on the classified vegetation maps to determine spatial extent and temporal dynamics of fragmentation as a function of marsh area and riverine sediment supply.

Figure 1: In the Grand Bay area we discovered that the marsh is fragmenting over time due to both edge erosion as well as interior subsidence. This graphic illustrates the loss of marsh area over time with the rate of loss increasing into the future.

Methods:

In a transition from marsh to open water as sea-level rises, there are both losses occurring at the marsh edge from erosion, as well as interior change from marsh to water as subsidence opens up channels and low-lying ponded areas. In the absence of sufficient new sediment input, and/or matching high rates of organic matter production to maintain marsh platform elevation above the mean water level, there will be loss of marsh area through fragmentation, or breaking up, or the formerly contiguous marsh area (Figure 1). At the landscape level (bird's eye view) these processes can be captured in three simple metrics: (1) Total Area = TA, which measures the number of square meters of marsh habitat in a classified map; (2) Number of Patches = NP, which simply counts the number of discrete marsh units separated by water or some other category; and (3) Total Edge = TE, which is the sum of all marsh patch edge lengths in a given area. This last metric is actually very important, as it reflects the available habitat to many estuarine species that use marsh as a nursery habitat. The marsh edge is the most frequently inundated during tides and is the transition area where both marine and terrestrial species can overlap in distribution. A simple conceptual diagram (Figure 2) is proposed that captures the changes in TA, NP, and TE occurring during the course of marsh loss over time. Total Area will tend to decline in a mostly linear fashion, notwithstanding large storm events that may trigger an abrupt change. In contrast, both NP and TE will tend to have a hump-shaped response, where during the initial phase of fragmentation these two metrics both increase, because the marsh is now broken up into more individual discrete sub-units (or patches) that are separated from each other. At some point between about 50% and 30% marsh remaining, there appears to be a transition to a decline in both NP and TE as the remaining small marsh patches are completely lost and the landscape converts to open water.

We then tested these ideas in the Grand Bay, a sediment starved and rapidly eroding marsh area, by selecting 250m square grid cells (same size as a MODIS 250m pixel) for marsh categories from 100% marsh to 0% marsh (or 100% open water). These categories (100%, 90%, 75%, 50%, 25%, 10%, 0%) represent a space-for-time proxy for marsh loss and we calculate the mean TA, NP, and TE from 6 replicate grid cells in each of the seven categories (Figure 3). Marsh edge was manually delineated in each grid cell for the years 1955, 1992, and 2014. The resulting marsh class was then summarized either with FRAGSTATS using the landscape level analysis, or by manually calculating the values for each metric in ESRI ArcGIS. Statistical testing was performed in SPSS to determine either significance among categories (One-Way ANOVA), or significance over time within a category (Repeated Measures ANOVA).

Results:

The data appear to support the conceptual model, with the exception of the 50% marsh cover category. This was lower than expected in all three years (Figure 4). For TA (Fig. 4A), there was a higher rate of area loss from 1955 to 2014 at the low marsh cover classes (10%, 25%) than at the high marsh cover classes (90% and 75%), indicating the increased vulnerability of marsh loss once patches become smaller. For NP (Fig. 4B) there was a bimodal shape to the distribution of mean NP across the seven cover classes, although this appears to be driven primarily by the smaller than expected number of patches in the 50% class. For TE (Fig. 4C) results were closer to expectations, notwithstanding the lower TE at 50%, resulting from the small NP. Further investigation of the selected grid cells in the 50% class suggests that many of these cells were approximately equal amounts of contiguous marsh and open water, with little fragmentation of the marsh habitat. This would result in the lower than expected NP and TE values for this class. Finally, in Figure 4D, we present the relationship between NP and TE, indicating that mean NP in a 250m grid cell is overall a good predictor (75%) of mean TE and that an average marsh patch in the Grand Bay marsh landscape represents approximately 280m of marsh edge habitat. This edge habitat is primarily dominated by a narrow zone of Spartina alterniflora.

Transition Matrix:

Another way to consider the change occurring in a landscape cell is the transition from marsh to water over time. A simple two-state transition diagram shows the four different transition probabilities in a given time step (Fig. 5). Application of this transition matrix approach applied to the seven marsh cover categories in each of the two time steps (1955->1992, 1992->2014) as well as the entire 59 year time domain (1955->2014) was investigated. The results for the 59 year transition (Fig. 6) shows that probability of marsh remaining marsh (M->M) declines as percent cover of marsh decreases, and conversely the probability of water remaining water (W->W) increases when marsh cover declines. The "tipping point" where marsh was increasingly likely to transition to water (M->W) occurred in the 25 to 50% marsh cover categories, suggesting that marsh cover becomes more vulnerable to irreversible loss when the fragmentation of a given marsh area exceeds 50%. Also of interest is that marsh accretion did occur in some pixels (W->M) but this was more likely only when marsh cover was still greater than 50%, suggesting a loss of marsh resilience once fragmentation increases beyond a threshold.

Finally, the question of whether transition probabilities may have changed during the two phases of the project, 1955 ->1992 (Phase 1 - 37 years) vs. 1992-> 2014 (Phase 2 - 22 years) was looked at in more detail (Fig. 7). Both the M->M and W->W transitions reiterate the findings from the entire time domain (Fig. 6) with similar transition probabilities occurring between the two phases. The biggest difference between the two phases occurs at the 25% marsh cover category with lower probability of marsh remaining marsh in Phase 2 compared to Phase 1, and higher probability of water remaining water. This result suggests that in the more recent times marsh loss may have become exacerbated in areas where there is already highly fragmented marsh compared to areas with more intact marsh. The transitions from M->W (marsh loss) and W->M (marsh accretion) indicate different landscape dynamics are occurring as a function of marsh coverage. Loss of marsh was more likely between 75 - 25% marsh coverage in both phases. Once marsh cover was down to 10% there was insufficient marsh remaining to allow for a high loss probability, as most of the area was already water. Finally, naturally occurring marsh accretion (W->M), which is a highly desirable state, was only likely when there was >50% of marsh coverage present, and was highest in the 75% marsh category. This situation likely represents areas where small water bodies can become silted in and allows marsh plants to recolonize into the previously unsuitable habitat.

Conclusions:

Marsh fragmentation processes were investigated in detail in the rapidly retreating Grand Bay study site. A space for time substitution approach was tested using seven marsh cover categories spanning the range of 100% marsh to 0% marsh (100% water). Six replicate grid cells of 250 m square (MODIS 250) for each marsh cover category were chosen to represent this landscape and various fragmentation metrics and transition probabilities calculated. Based on these data, there appear to be critical landscape transitions that occur when marsh cover drops below 50%. These include higher fragmentation (based on larger numbers of patches, and longer edge distances) and higher marsh loss transition probabilities. The result of these processes is a landscape condition where marsh recovery becomes less likely, and the ongoing loss of marsh to open water becomes more likely. These conditions are highly likely to worsen as the rate of sea level rise continues to increase in the future.