Recent Trends in Satellite Vegetation Index Observations Indicate Decreasing Vegetation Biomass in the Southeastern Saline Everglades Wetlands
We analyzed trends in time series of the normalized difference vegetation index (NDVI) from multitemporal satellite imagery for 2001–2010 over the southeastern Everglades where major changes in vegetation structure and type have been associated with sea-level rise and reduced freshwater flow since the 1940s. Non-parametric trend analysis using the Theil-Sen slope revealed that 84.4 % of statistically significant trends in NDVI were negative, mainly concentrated in scrub mangrove, sawgrass (Cladium jamaicense) and spike rush (Eleocharis cellulosa) communities within 5 km of the shoreline. Observed trends were consistent with trends in sawgrass biomass measurements made from 1999 to 2010 in three Long-term Ecological Research (LTER) sites within our study area. A map of significant trends overlaid on a RapidEye high-resolution satellite image showed large patches of negative trends parallel to the shoreline in and around the “white zone,” which corresponds to a low-productivity band that has moved inland over the past 70 years. Significantly positive trends were observed mainly in the halophytic prairie community where highly salt tolerant species are typically found. Taken as a whole, the results suggest that increased saline intrusion associated with sea-level rise continues to reduce the photosynthetic biomass within freshwater and oligohaline marsh communities of the southeastern Everglades.
KeywordsMODIS NDVI Non-parametric trend analysis Sawgrass marsh Mangroves Above-ground biomass production
Coastal wetlands provide a range of essential ecosystem services such as carbon sequestration, protection from erosion, and maintenance of water quality (Webb et al. 2013). Sea-level rise and associated intrusion of salt-water into oligohaline wetland systems can negatively affect primary productivity of wetland plants such that organic accretion rates in marshes may not keep pace with rising water levels and increased salinities (Neubauer 2008, 2013; Barendregt and Swarth 2013). The inundation of coastal wetlands by rising seas may affect as much as 195,000 km2 of tropical and temperate tidal marshes globally (Greenberg et al. 2006; Spalding et al. 2010). A number of large coastal wetland areas are considered especially vulnerable to increased salinity, subsidence, and reduced plant productivity, including the Nile Delta (Hassaan and Abdrabo 2013), major river deltas of China such as the Pearl and Yangtze (Wang et al. 2012), the Sundarbans of Bangladesh and India (Loucks et al. 2010), as well as large areas of eastern North America, Europe and the Gulf of Mexico (Baldwin and Mendelssohn 1998; Neubauer 2008; Couvillion and Beck 2013).
Thus, the productivity of Everglades marsh and mangrove species is strongly influenced by salinity gradients and flushing by fresh and saltwater over different time scales (Childers et al. 2006; Barr et al. 2013). For example, Macek and Rejmánková (2007) found that plant height and shoot/root biomass decreased in C. jamaicense and Eleocharis cellulosa under conditions of elevated salinity. Barr et al. (2009) showed that carbon assimilation in mangrove leaves was limited when salinity exceeded 35 parts per thousand. In the absence of regular flushing by either fresh or brackish waters, Wanless and Vlaswinkel (2005) observed that marsh communities can collapse—a phenomenon has also been noted in mangrove communities that have been migrating inland in recent decades (Davis et al. 2005). Therefore, even relatively salt-tolerant communities may decline as a result of elevated salinity in the southern Everglades.
Systematic estimates of above-ground biomass and culm density have been made in 16 marsh sites in the southern Everglades since 1998 as part of the Florida Coastal Everglades Long-term Ecological Research (FCE LTER) sampling network (Childers et al. 2006; Ewe et al. 2006). These data provide an important baseline for understanding long-term behavior of sawgrass and spike rush communities in relation to salinity, nutrient loading, and hydrological drivers. For example, Childers et al. (2006) showed that above-ground net primary productivity (ANPP) in these sawgrass-dominated communities was negatively related to surface water salinity measured continuously in the Taylor Slough, while more the more hydric spike rush communities possessed higher biomass in sites with long hydroperiods and elevated water levels.
While site-level monitoring of biomass has provided critical understanding of ecosystem processes that control photosynthesis in the southern Everglades, productivity data from satellite observations may be used to assess recent fluctuations and trends in biomass and vegetation cover. In particular, sums of the normalized difference vegetation index (NDVI) obtained from red and near infrared reflectance provide a direct measurement of the fraction of absorbed photosynthetic activity (Goetz et al. 1999) as well as indirect measures of gross and net primary productivity, biomass, and green leaf area in a variety of grassland and forest ecosystems (Green et al. 1997; Paruelo et al. 1997; Myneni et al. 2001; Pineiro et al. 2006; Wessels et al. 2008; An et al. 2013; Barr et al. 2013). Since mid-2000, global NDVI data have been available at 250 m spatial resolution from the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the polar-orbiting Terra Satellite operated by the National Aeronautics and Atmospheric Administration (NASA). As NDVI image archives have grown over the past three decades, various time-series techniques have been applied to these data to identify multi-year trends that may relate to variety of anthropogenic and biophysical factors (Fuller 1998; Herrmann et al. 2005; de Jong et al. 2011, 2013). In this study, we exploit 10 years of MODIS 250 m NDVI imagery covering South Florida to map decadal-scale trends, which we relate to vegetation type and ground-level measurements made in sawgrass sites in the southern Everglades National Park (NP). Our objective, therefore, was to identify statistically significant trends and explain these in terms of current understanding of environmental factors that control ANPP in the southern Everglades ecosystems. Our study area is centered over the Taylor Slough, which is the second-largest flow-way for surface water in the Everglades and stretches approximately 30 km along the eastern boundary of the Everglades NP (Fig. 1).
Data and Methods
Neeti and Eastman (2011) introduced a contextual Mann Kendall (CMK) approach as a way to incorporate local spatial variation of individual pixels with respect to their neighbors. The logic behind contextual analysis is that similar behavior (i.e., spatial autocorrelation) within small neighborhoods of pixels (e.g., 3 x 3) should produce greater confidence in trends. Neeti and Eastman (2011) also showed that this approach increased the number of pixels in satellite time series that have significant slopes.
Implementation of TS slope and CMK was done in Earth Trends Modeler software, which is part of the Idrisi Selva GIS software (Eastman 2012).
To evaluate trends within different vegetation types, we utilized a highly detailed (1:15,000) digital vegetation map produced by Welch et al. (1999). Within our study area, the map contains 57 different dominant vegetation types, so to simplify the analysis we concentrated on seven major vegetation types that cover approximately 95 % of the terrestrial portion of our study area within Everglades National Park (shown in Fig. 1). These major vegetation types include mangrove forest (trees > 5 m), mangrove scrub (trees and shrubs < 5 m), sawgrass prairie, spike rush prairie, other graminoids, bayheads, and halophytic prairie. GIS software was then used to calculate the number of pixels with significant positive and negative TS slopes within each major vegetation type.
Above-ground biomass and culm-density data were obtained from Florida Coastal Everglades LTER site (http://fcelter.fiu.edu/research/working_groups/), which contained sawgrass time series data on biomass for three sites with 10 or more years of temporal overlap with the NDVI time series. The three sites fall along a nutrient and salinity gradient from south to north, with sites TS/Ph-01b and TS/Ph-03 being more typical of marsh-slough habitats with well-developed periphyton mats, and TS/Ph-06b found in the estuarine ecotone (Ewe et al. 2006). Biomass data were derived from bi-monthly estimates of live biomass using a non-destructive phenometric regression model that was calibrated using clipped and dried culms obtained in triplicate 1 m2 permanent plots and cover an area ranging from 0.25 to 1.0 ha. While these sites cover less area than the MODIS sensor field of view (i.e., 6.25 ha), we assumed that the sites were spatially homogeneous with respect to plant diversity and physiognomy as they were selected to be representative of freshwater marshes and mangroves of the larger Everglades landscape (Ewe et al. 2006). Model independent variables included culm diameter, sum of leaf lengths, total culm height, and the height of inflorescences. These four variables in a step-wise regression explained 92 % of the variance in dried, clipped sawgrass samples (Childers et al. 2006). We also used average culm density data collected as part of the same monitoring effort (Childers et al. 2006). These two data sets were used to assess consistency between observed NDVI trends and biomass/cover trends from 2001 to 2010 and thus allow us to evaluate the reliability of estimates of TS slope and p.
Our analysis supports the conceptual model of ecological interactions in the Everglades estuaries advanced by Davis et al. (2005), who postulated that coastal transgression will continue to outpace deposition in coastal marl and mangrove environments in South Florida and that the low-productivity white zone (Fig. 4) will continue to move inland over time. These processes were first observed by Egler (1952) in the 1940s and thus are part of a multi-decadal trend that is driven by reinforcing factors related to global climate change and water management practices within South Florida. Specifically, construction of canals has reduced water levels in the peat mantle and underlying aquifer and withdrawal of groundwater has resulted in lateral saltwater intrusion along the east coast of South Florida. While the process may be reversible through artificial recharge of the aquifer (Barlow and Reichard 2010), continued sea-level rise this century will most likely negate attempts to limit future saline intrusion.
Many of the patterns of significant TS slope in Fig. 2b are difficult to explain based on salinity changes alone. For example, Fig. 4 shows large areas of positive NDVI slope to the north of the white zone and two major areas of significant positive slope along the eastern boundary of Everglades NP and approximately 10 km east of the Park are likely the result of decisions to manage surface water levels for seasonal flood control (Van Lent et al. 1993; Armentano et al. 2006). While increased salinity is a major driver of change in vegetation productivity in sawgrass and spike rush wetlands, shifts in species composition have occurred within 3–4 years in the Taylor Slough as a result of changes in water levels and hydroperiod from 1979 to 2003 (Armentano et al. 2006). For example, increased water levels in portions of the Taylor Slough have been associated with loss of muhly grass (Muhlenbergia capillaris var. filipes) and sawgrass and an increase in spike rush, which is common in long hydroperiod (6–9 month) marshes (Armentano et al. 2006). These observations are consistent with a recent modeling study that suggests major decreases in the area of tall sawgrass vegetation as a result of changes in hydroperiod and land use in the Everglades (Foti et al. 2013). Thus, the changes in sawgrass biomass observed in site TS/Ph-3, which is well outside the oligohaline zone, are likely unrelated to salinity increases. However, it is unclear whether major changes in biomass and ANPP inferred using NDVI time series can be partially ascribed to water level or hydroperiod in areas unaffected by saline intrusion.
Trends in biomass measurements made in sawgrass communities from the FCE LTER support the results obtained from NDVI time series. In particular, the consistency between steep declines in above-ground biomass and culm density at site TS/Ph-03 and NDVI sums for that site provides confidence in our maps of significant trends (Figs. 2b and 4). Lack of significant trends at the other two sites was also consistent with trends in NDVI sums. However, certain limitations in our study should be noted, including lack of concomitant salinity data at the three FCE LTER biomass sites and no biomass time series for major vegetation types other than sawgrass prairies. However, salinity time series from other FCE LTER sites in the Taylor Slough suggest that multi-year trends in this variable are difficult to discern (e.g., Childers et al. 2006). Despite these limitations, the spatial patterns observed in Figs. 2b and 4 reveal changes consistent with general understanding of the long-term (>70 year) trends that have affected vegetation in the southeastern saline Everglades. Therefore, our analysis supports further application of the TS slope and CMK significance to investigate long-term changes in wetland biomass and productivity. While the spatial resolution of MODIS 250 m NDVI may be viewed as somewhat coarse for ecological monitoring, the sensor’s field of view represents a trade-off between temporal and spatial resolution, and was designed to be optimal for detecting land cover changes from space (Justice and Townshend 1988).
The difference in scale between the biomass samples collected in the FCE LTER sites (1 m2 plots) and the MODIS NDVI imagery may partly explain low correlations between biomass and culm density as well. Data on spatial variability at scales comparable to the MODIS sensor resolution were not available to determine if patchiness (i.e., low spatial autocorrelation) may have influenced the results. Moreover, many other factors may influence NDVI including soil background reflectance, atmospheric effects, and view angle effects. Further, the low correlation between biomass, culm density and summed NDVI at site TS/Ph-06b may have been a result of tidal inundation as this location, which may have introduced soil and water background effects on NDVI. Ewe et al. (2006) noted that the low substrate organic content and long hydroperiod may explain the relatively low productivity and low biomass at this particular site. In addition, spatial heterogeneity of plant communities within 250 m MODIS pixels may have resulted in the inclusion of species that were not sampled for biomass as part of the FCE LTER. Thus, as vegetation cover and biomass decline, the soil will play greater role in the radiance received at the satellite sensor. The use of this index is therefore likely to introduce noise in satellite time series and future work using MODIS imagery to isolate significant trends may benefit from use of the Enhanced Vegetation Index (EVI) or EVI-2 indices, which are less sensitive than NDVI to background and atmospheric effects (Jiang et al. 2008).
Many questions remain about how wetland species and communities in the Everglades will respond to the continued sea-level rise and saltwater intrusion in coming decades. The past 70 years of vegetation monitoring and analysis in the southern Everglades suggest that saline communities consisting of salt-tolerant species will continue to expand while oligohaline and salt-intolerant species will decline in productivity, biomass, and density. Significant positive NDVI trends observed in halophytic marshes observed in this study coupled with large areas of negative significant NDVI trends in oligohaline and salt-intolerant vegetation types proximate to the coast such as sawgrass, spike rush, scrub mangrove, and mangrove forest are consistent with long-term (70 year) trends and observed patterns. The white zone, therefore, is a highly sensitive indicator of coastal change and that can be readily monitored using multitemporal vegetation index imagery such as used here. We suggest that studies are needed to remotely monitor change in this sensitive ecotone using similar vegetation indices such as EVI derived from orbital platforms and that further analysis of trends in vegetation indices be evaluated using data on biomass, salinity, water level, and hydroperiod to advance understanding of how saltwater intrusion and inundation will affect the coastal wetlands of the southeastern Everglades.
The authors wish to thank Raymond Turner of Center for Southeast Tropical Advanced Remote Sensing (CSTARS) for providing the RapidEye image. We are grateful to the many scientists of the FCE LTER who have made their field data publicly available. Support for this research was provided by NASA WaterSCAPES (Science of Coupled Aquatic Processes in Ecosystems from Space) Grant NNX08BA43A.