Feedback of coastal marshes to climate change: Long‐term phenological shifts

Abstract Coastal marshes are important carbon sinks facing serious threats from climatic stressors. Current research reveals that the growth of individual marsh plants is susceptible to a changing climate, but the responses of different marsh systems at a landscape scale are less clear. Here, we document the multi‐decadal changes in the phenology and the area of the extensive coastal marshes in Louisiana, USA, a representative of coastal ecosystems around the world that currently experiencing sea‐level rise, temperature warming, and atmospheric CO 2 increase. The phenological records are constructed using the longest continuous satellite‐based record of the Earth's ecosystems, the Landsat data, and an advanced modeling technique, the nonlinear mixed model. We find that the length of the growing seasons of the intermediate and brackish marshes increased concomitantly with the atmospheric CO 2 concentration over the last 30 years, and predict that such changes will continue and accelerate in the future. These phenological changes suggest a potential increase in CO 2 uptake and thus a negative feedback mechanism to climate change. The areas of the freshwater and intermediate marshes were stable over the period studied, but the areas of the brackish and saline marshes decreased substantially, suggesting ecosystem instability and carbon storage loss under the anticipated sea‐level rise. The marshes' phenological shifts portend their potentially critical role in climate mitigation, and the different responses among systems shed light on the underlying mechanisms of such changes.

2016). Depending on the marshes' vegetation composition, the current ambient conditions, and other environmental factors such as nutrients, the marsh plant growth, and the resulting primary production can be stimulated or hindered by temperature warming, sea-level rise, and elevated atmospheric CO 2 concentration (Charles & Dukes, 2009;Erickson, Megonigal, Peresta, & Drake, 2007;Langley & Megonigal, 2010;Langley, Mozdzer, Shepard, Hagerty, & Megonigal, 2013;Morris, Sundareshwar, Nietch, Kjerfve, & Cahoon, 2002). The decomposition of marsh substrates and hence the carbon sequestration rate can be promoted or inhibited by saltwater intrusion and global warming, varying among different marsh systems and being controlled by concurrent stressors (Craft, 2007;Wu, Huang, Biber, & Bethel, 2017). Given the complex and dynamic outcomes of the various factors, it is hard to predict how coastal marshes will respond to the simultaneous climatic stresses at a landscape scale. Yet, this is a key piece of information needed in order to assess ecosystem sustainability and to predict future changes.
Here, we study the multi-decadal phenology of four distinct coastal marsh systems to gain insights into their ability to uptake carbon in relation to climate change at a broad scale (Walther et al., 2002). We use the longest continuous satellite-based record of the Earth's ecosystems, the Landsat Climatic Data Records (CDRs) and an advanced modeling technique, the nonlinear mixed model, to reconstruct the phenology between 1984 and 2014 for the different coastal marsh systems in Louisiana, USA, one of world's largest coastal marsh habitats that share the same climatic stressors as many other coastal ecosystems around the world. We speculate that climate change (i.e., temperature, sea-level, and atmospheric CO 2 concentrations) has influenced the phenology of the marshes at the ecosystem level and that the influence is a function of whether the marshes are tidal freshwater, intermediate, brackish, or saline systems. We also hypothesize that the climatic influences will continue, if not increase, under future climate scenarios.

| Study area
The study area is in four major basins in Louisiana, USA, at the north-  (Gosselink, 1984). Freshwater marshes occupy habitats with a salinity <0.5 ppt and are dominated by Panicum hemitomon, Sagittaria falcata, and Eleocharis sp.

| Marsh phenology modeling
We use the longest continuous satellite record of the Earth's ecosystems, the Landsat CDRs from 1984 to 2014, to create the phenological records of the marshes (Table A1) Mo, Kearney, Riter, Zhao, & Tilley, 2018), and the NDVIbased phenological records of the marshes are modeled using an advanced modeling technique, the nonlinear mixed model (Mo, Momen, & Kearney, 2015). This method is developed by Mo et al. (2015) to provide a rigorous statistical analysis for phenological curves of different vegetation that are represented by nonlinear functions with repeated-measure variables. Phenological measurements (i.e., the NDVI) made on the same observational units (i.e., marsh systems) over time are treated as repeated measurements. The phenological records of the different marsh systems are fitted into three nonlinear models, the Gaussian, the stepwise Gaussian, and the stepwise logistic functions. The goodness-of-fit of the models is assessed via the Efron's pseudo R 2 and the Akaike Information Criterion (AIC), the Akaike Information Criterion Correction (AICC), and the Bayesian Information Criterion (BIC).
The pseudo R 2 , a statistic similar to R 2 in the linear regression, indicates the percent variance explained by the nonlinear models (Hardin, Hilbe, & Hilbe, 2007). The pseudo R 2 ranges from −∞ to 1. A pseudo R 2 closer to 1 indicates more variability in the data is explained. The AIC, AICC, and BIC indices evaluate models based on the principle of parsimony, that is, a model explains more variation in the data with fewer variables is considered a better fit (Boyce, Vernier, Nielsen, & Schmiegelow, 2002;Richards, 2005).
Key phenological parameters, that is, peak NDVI, peak NDVI day, and growing season length (bracketing days that had NDVI >90% of peak NDVI) for each marsh system in each year are estimated from the best-fit model. The phenology modeling only considered existing marshes, that is, it is corrected for the marsh area changes over the 30 years.

| Marsh area estimation
The areas of the freshwater, intermediate, brackish, and saline marshes are estimated using cloud-free Landsat 5 TM and Landsat 8 OLI data.
The marshes type boundaries are determined using the USGS vegetative survey for Louisiana coastal marshes done in 1988, 1997, 2001, 2007, and 2013 (Chabreck & Linscombe, 1988Linscombe & Chabreck, 2001;Sasser et al., 2008Sasser et al., , 2014. The marshland area within the boundaries is estimated using the C version of the Function Mask (CFMask) that comes with the Landsat CDRs (Zhu & Woodcock, 2012).

| Climatic and environmental data
We acquire records of the atmospheric CO 2 , air temperature, Oceanic Niño Index (ONI), sea-level, and salinity of the study area

| Phenology prediction
Linear models describing the correlations between the marsh phenology and the climatic variables are built on the historical data and used to predict the marsh phenology in the future (until 2050). The future sea-level and air temperature in the study area are estimated using linear models based on the sea-level data from the NOAA Grand Isle station, and the air temperature data from the NWS New Orleans Airport station, both dating back to the 1940s ( Figure A1a, b). The slope of the sea-level increase is 9 mm/year (p < 0.01), which is in consistence with the literature (González & Törnqvist, 2011;Jankowski, Törnqvist, & Fernandes, 2017). The temperature increases at a speed of 0.016°C/ year (p < 0.01), falling within the lower ranges of projections from the  Figure A1c).

| Marsh phenological changes
The NDVI-derived phenological records of the freshwater, intermediate, brackish, and saline marsh systems from 1984 to 2014 are well-described by our models (pseudo-R 2 0.86 ± 0.11; Table 1).
Exceptions are years when not enough relatively cloudless images were collected, and thus the phenological parameters of the respective marsh units in those years cannot be estimated: that is, all marshes in 1990, 1991, 1997, 2001, 2002, and 2012
The areas of the freshwater and intermediate marshes were quite stable, and their increased percentage was mostly a result of the decrease of the total marsh area. The marsh loss mainly occurred in the brackish and saline marshes, contributing to most of the 16% change of marsh-to-water from 1988 to 2013. The decrease of the brackish and saline marshes' areas were significant over time (p < 0.05 in both cases; Figure 2d). The area of the brackish marshes decreased for 52.9% (from 2,326 km 2 in 1988 to 1,541 km 2 in 2013), and the area of the saline marshes decreased for 17.9% (from 2,080 to 1,631 km 2 ). The area of the brackish marshes was negatively correlated with sea-level, CO 2 concentration, and temperature, and positively with precipitation (r = −0.8, −0.7, −0.9, and 0.7, respectively; p < 0.05 in all cases; Table A4); the area of the saline marshes was negatively correlated with sea-level rise rates and CO 2 concentration (r = −0.5 and −0.6, respectively; p < 0.05 in both cases; Table A4).

| Future marsh phenology
In Section 3.1, we find that the peak NDVI day of the saline marshes was significantly correlated with air temperature and that the length of the growing seasons of the intermediate and brackish marshes was significantly correlated with atmospheric CO 2 (p < 0.05 in all cases). Based on these correlations, we use the phenological and environmental data from 1984 to 2014 to build models to predict the marshes' future phenology. The model for predicting the brackish marsh growing season length (day)  The changes in the peak NDVI day for saline marshes will continue at the same rate ( Figure 4a). The peak NDVI day of the saline marshes moved from July to August during the past 30 years and is projected to be in September in 2050. It should be noted that based on our analysis, the changes of the peak NDVI day of saline marshes were not directly correlated to time, but were correlated to air temperate in the study area which increases over with the BERN, the length of the growing seasons of the intermediate and brackish marshes will increase at the lowest rates-but still faster than the past 30 years-for around 80 and 90 days in the next 30 years, equals to 2.2 and 2.7 days/year, respectively.
It should be noted that these predictions rely solely on air temperature and CO 2 emissions scenarios and do not take nutrient limitations or other limiting factors into account.

| The marshes' phenological changes in the last 30 years
The correlations between the growing season length and the atmospheric CO 2 concentration may be the result of the stimulation of elevated CO 2 concentration on photosynthesis in marsh plants (Cherry, McKee, & Grace, 2009;Rasse, Peresta, & Drake, 2005). This is the first study we know of that reports a positive correlation between atmospheric CO 2 concentration and coastal marshes' growing season length, which reflects a broad pattern of ecosystem change due to a changing climate (Walther et al., 2002).  (Arp, Drake, Pockman, Curtis, & Whigham, 1993;Cherry et al., 2009;Drake, 2014;Erickson et al., 2007;Rasse et al., 2005). This is because the increased atmospheric CO 2 stimulates photosynthesis of C3 plants, but not the photosynthesis of C4 plants that is nearly saturated under ambient conditions as C4 plants concentrate CO 2 at the site with their primary CO 2 -fixing enzyme. Moreover, the elevated CO 2 may even inhibit the plant growth of C4 species by reducing their stomatal conductance, transpiration, and ion uptake (Ghannoum, 2009;Rozema et al., 1991). Yet, the freshwater marshes, which have the highest percentage of C3 plants among the different marsh systems, demonstrate no response to higher CO 2 levels in our study. One possible explanation is that the high nutrient loading in freshwater marshes-freshwater marshes are closest to the surface runoff-favors species that are unresponsive to elevated atmospheric CO 2 (Langley & Megonigal, 2010;Langley et al., 2013).
Global warming has a general effect of promoting plant growth that is manifested in increasing the growing season NDVI and lengthening the active growth season, especially in the middle and high latitudes (Myneni, Keeling, Tucker, Asrar, & Nemani, 1997;Zhou et al., 2001). Although it was reported that warming increased the annual peak biomass of the marshes TA B L E 1 The pseudo R 2 , the Akaike Information Criterion (AIC), the Akaike Information Criterion Correction (AICC), and the Bayesian Information Criterion (BIC) of the best-fit phenological model (i.e., the Gaussian, G; the stepwise Gaussian, SG; or the stepwise logistic, SL, function) of the freshwater, intermediate, brackish, and saline marshes from 1984 to 2014 (Exceptions are years when not enough cloudless images were collected, including all marshes in 1990, 1991, 1997, 2001, 2002, and 2012 and saline marshes in 1994 and 1998 Indeed, the linear models used in this study are highly simplified. These models do not consider other environmental factors, such as water temperature and nutrient availability, which can vary substantially with in situ conditions, or the nonlinear effects and the interactions among the different factors. For instance, the enhancement of CO 2 uptake by C3 plants may be eventually slowed down by photosynthetic downregulation, nutrient limitations, or increased disturbance from sea-level rise, thus the CO 2 fertilization on the marshes will not continue over long time frames as decades (Erickson et al., 2007;Langley et al., 2013). In addition, the marshes may suffer from other stresses such as pests, herbivores, and pathogens that can also be intensified by a warmer climate (Van der Putten, Macel, & Visser, 2010).
Nevertheless, this study provides a "baseline" scenario for future marsh phenology with unconstrained sole impacts from temperature or CO 2 , as well as observational inputs for more advanced modeling studies. Couvillion et al. reported that 25% of Louisiana's coastal marshes that existed in 1932 had been lost by 2010 (Couvillion et al., 2011), and this study shows that 20% of the marshes in 1988 were lost by 2013. These findings document a highly vulnerable ecosystem.

| The marsh area changes in the last 30 years
Recent studies have found strong correlations between marsh area loss and several climatic variables (Turner, Kearney, & Parkinson, 2017;Turner, Baustian, Swenson, & Spicer, 2006;Kearney & Turner, 2015). In this study, we further the research by separating the area changes of different marsh systems. We find that the areas of the freshwater and intermediate marshes were quite stable; while the area of the brackish and saline marshes decreased significantly and were negatively correlated with sea-level and CO 2 concentration.
F I G U R E 2 Phenology and areas of the coastal freshwater (dark green), intermediate (light green), brackish (light purple), and saline (dark purple) marshes in Louisiana from 1984 to 2014. The three key phenological parameters estimated are peak NDVI (Panel a), peak NDVI day (Panel b), and growing season length (bracketing days that had NDVI greater than 90% of peak NDVI; Panel c). The marsh areas are estimated in 1988. Sea-level rise may decrease the marsh elevation by reducing their organic accretion (Nyman, Delaune, Roberts, & Patrick, 1993), promoting decomposition of the marsh substrates (Craft, 2007;Weston, Vile, Neubauer, & Velinsky, 2011;Stagg, Schoolmaster, Krauss, Cormier, & Conner, 2017), and increasing erosion (Smith, Cialone, Wamsley, & McAlpin, 2010). Hence, the negative correlation between the marsh area and sea-level can be expected. On contrary, increased air temperature and CO 2 concentration can contribute to the marsh stability against sea-level rise. The elevated air temperature and atmospheric CO 2 may promote the marsh plant growth (for both aboveground and belowground) and thus their organic accretion (Langley, McKee, Cahoon, Cherry, & Megonigal, 2009;Ratliff, Braswell, & Marani, 2015). Although the enhancement of plant growth from CO 2 , as discussed earlier, is mainly for C3 plants, the fertilization of CO 2 , based on a modeling study, can increase the marsh elevation for a mixed C3 and C4 plant community (by increasing plant production) in similar magnitude to the effect of increasing inorganic sediment input (Ratliff et al., 2015). The maintenance of the marsh elevation can further benefit from the increased belowground biomass: the increased shoot density will enhance the trapping of tidally driven sediment and provide stronger protection against erosion (Temmerman, Moonen, Schoelynck, Govers, & Bouma, 2012;Mudd, D'Alpaos, & Morris, 2010). It should also be noted that the influence of CO 2 fertilization on marsh elevation is likely to depend on many other factors such as the inorganic sedimentation rate (Langley & Megonigal, 2010;Langley et al., 2009;Ratliff et al., 2015) and may be offset by the enhanced decomposition resulted from a rising air temperature (Charles & Dukes, 2009;Kirwan & Blum, 2011). The fertilization effects of the increased air temperature and atmospheric CO 2 allow the coastal marshes to be more resilient against sea-level rise, but this study documents a sig-  (Jankowski et al., 2017), the relative relationship between sediment supply and sea-level rise of different locations (Mariotti & Fagherazzi, 2010), and the availability of accommodation space (Schuerch et al., 2018) may provide more insights the spatial patterns of the area changes. an increase in photosynthesis and CO 2 uptake, providing a negative feedback mechanism to the elevated atmospheric CO 2 and climate change. On the other hand, the loss of the brackish and saline marshes impairs the ecosystems' potential to capture CO 2 and the stability of the existing carbon storage, which is quite large in the coastal marsh anaerobic substrates (long-term C accumulation rate at 18-1713 g C m −2 yr −1 ; McLeod et al., 2011;Nahlik & Fennessy, 2016). The stored carbon will be released into the ocean or the atmosphere-which can also be in the form of methane, a more potent greenhouse gas, under the marshes' anaerobic conditions (Whiting & Chanton, 1993)-providing a positive feedback mechanism to climate change. The Louisiana coastal marshes are representatives of coastal ecosystems around the world experiencing various climatic and anthropogenic stressors (Bianchi & Allison, 2009;Wang et al., 2007). This study documents the climate-driven long-term phenological shifts of the marshes that, in turn, provide a negative feedback mechanism to the changing climate. A stable coastal marsh system will capture and store more carbon under a changing climate, and compensate, to some extent, for anthropogenic carbon emissions. Such mechanisms highlight the marshes' critical role in climate mitigation and emphasize the importance of the conservation and restoration of coastal ecosystems under the changing climate. Year 2005 42, 106, 122, 138, 154, 170, 202, 234, 250, 282, 298, 314 Landsat 5 24 TA B L E A 4 Significant correlations (r) between marsh area and phenological changes, and the environmental variables, that is, year, sealevel (SL), atmospheric CO 2 , temperature (T), precipitation (P), salinity (S), discharge (D), Oceanic Niño Index (ONI) from 1984 to 2014. All the correlations between the variables are tested, but only significant correlations are shown in the