Widespread mangrove damage resulting from the 2017 Atlantic mega hurricane season

We aimed to investigate the effects of storms on mangroves across the North Atlantic Basin. As such, we included in our analyses the coast of the Gulf of Mexico, the islands of the Caribbean, as well as the Caribbean coasts of Central and South America and hereafter, we refer to this region as the Caribbean/Mesoamerican Region. To formally delineate this region, we limited the extent of our analyses to the rectangle defined by 6.11° to 31.86° latitude and −99.33° to −55.88° longitude.

We quantified annual mangrove damage and mortality (2009–2017) using the NDVI greenness index derived from two satellite instruments: Landsat 7 Enhanced Thematic Mapper+ and Landsat 8 Operational Land Imager using Google Earth Engine following a modified approach from Lagomasino et al [32]. We calculated the response of mangroves to a given hurricane season by first calculating a pre-hurricane season reference NDVI value for each pixel as a median from the end of the previous hurricane season (November 1) to the beginning of the given hurricane season (June 1). We then calculated the average departure from the reference value (hereafter referred to as ΔNDVI) from all cloud-free images during the 7 month post-hurricane period (November 1 of given year to June 1 of the next year; SI appendix (stacks.iop.org/ERL/15/064010/mmedia)). We used a high-resolution (1 ha) map of global mangrove extent in 2010 to extract the ΔNDVI within mangroves only [33]. According to this map, there were 2.26 million ha of mangroves within our study region. Based on a previous investigation of mangrove response to hurricanes [25], we considered pixels with a ΔNDVI < −0.2 to represent mangrove damage [34]. In 2017 only, we also calculated mangrove damage as described below from imaging instruments aboard two additional satellites, MODIS Terra and Sentinel 2, to investigate bias resulting from differences in spatial resolution and optical qualities of the instruments (SI appendix). We further confirmed these ΔNDVI responses reflected actual vegetation change in 2017 by comparing to changes in canopy height within a subset of the larger study area. The canopy height data were derived from high resolution LiDAR (∼1 m) measured with NASA Goddard's LiDAR, Hyperspectral and Thermal Imager (G-LiHT) both before and after the 2017 hurricane season in mangroves in South Florida (figure S2 and SI appendix). The linear relationship between ΔNDVI and canopy height was significant (p < 0.01) and the largest changes in canopy height (2–5 m decreases) occurred at sites with ΔNDVI < −0.2 (figure S2).

In order to distinguish between minor damage, such as defoliation, and more severe, persistent structural damage, we quantified recovery over the 7 month period between the end of the hurricane season and the beginning of the subsequent hurricane season, for each of 2009–2017. Specifically, we calculated the linear slope of NDVI values at a given pixel from all cloud-free images during the post-hurricane season and considered the pixel to represent persistent damage if this NDVI trend was close to or less than zero (slope < .000001, figure 1).

Zoom In Zoom Out Reset image size

To categorize mangrove damage in each year according to storm intensity, we combined the 'best track wind swath' for all storms in each year from 2009 to 2017 [35]. These data consist of three polygons for each storm delineating the area which experienced a storm intensity at or above each of three wind speed categories: (1) 63 km h⁻¹, (2) 93 km h⁻¹, and (3) 119 km h⁻¹. Within a given year, we overlaid the polygons for all storms to delineate the total area within our study region to experience winds of at least 63 km h⁻¹ (Tropical Depression on Saffir-Simpson Scale) and 119 km h⁻¹ (Category 1 on Saffir-Simpson Scale).

For a more detailed analysis of the relationship between storm intensity and mangrove damage and loss in only the year 2017, we used a higher spatial resolution dataset developed by NASA's Global Modelling and Assimilation Office (see SI appendix for more detailed description of data). These data consisted of 10 m surface wind speeds calculated at every hour for the duration of the 2017 hurricane season at a spatial resolution of 7 km. From this dataset we calculated the maximum wind speed experienced at each cell and the duration, in hours, for which a cell experienced winds at or greater than 119 km h⁻¹ (i.e. hurricane force or stronger according to Saffir-Simpson Scale).

We investigated several additional variables other than storm intensity that might explain the spatial variation in mangrove damage and loss that we observed in 2017. We incorporated canopy height using the global mangrove canopy height dataset developed by Simard et al [28]. Though some variation in canopy height is likely explained by previous storm activity, we also included the number of storms experienced since 2009 (hereafter: 'storm history') at a given pixel to potentially explain any additional variation in mangrove damage related to previous storms. To serve as a metric of storm-related flooding, we included the cumulative rainfall for the month of September, as this was the month during which the most severe storm activity occurred [36].

We used logistic regression to investigate which factors were most strongly related to mangrove damage associated with the 2017 hurricane season. Specifically, we extracted the values of NDVI change and the five predictors that we identified as having potential to influence the degree of mangrove damage (maximum wind speed, duration of hurricane force or stronger winds, rainfall, number of storms experience since 2009, and canopy height) at 8576 random pixels within extant mangroves, distributed evenly between damaged and undamaged mangroves (SI appendix). We used half of these data to fit or model and the other half to compare for validating our model. We used the area under the receiver operating characteristic curve (AUC) to evaluate the predictive performance of our model and compared the strength of covariate effects using the magnitude of the parameter estimates and 95% confidence intervals.

As increasingly severe storms become more frequent, our results show a disproportionate amount of mangrove damage will result soon after. Furthermore, considering simply the number and severity of storms in a season may poorly reflect the resulting mangrove damage. Instead, our results suggest that mangrove damage is at least partially stochastic and depends on the extent and spatial distribution of areas affected by severe storms. For example, mangrove damage in 2017 could have been much lower if Hurricane Irma did not track directly over the extensive mangroves of South Florida. Regardless, if severe storms become more frequent, the likelihood that extensive mangroves are impacted also increases.

The exponentially greater mangrove damage we observed in 2017 compared to other hurricane seasons also may result from the threshold response of mangrove damage to maximum wind speed. Our results confirm previous studies that show wind speed was the most important characteristic of hurricane to explaining mangrove damage. Imbert (2018) showed that decreases in mangrove basal area were greatest in tall basin mangrove stands that experienced storms of Category 1 (110–150 km h⁻¹) severity or greater, whereas the threshold of basal area loss for shorter fringe and scrub mangroves was Category 3 (178–209 km h⁻¹). Though we did observe an effect of canopy height, the variation in the threshold of mangrove damage was not as pronounced as that observed by Imbert [21]. This less pronounced effect could result from the difference in how damage was defined (basal area vs. NDVI), or from the consideration of variation by mangrove species. Nonetheless, our results provide further evidence for nonlinear increases in mangrove damage as Category 1 and stronger storms become more frequent.

The significant effect of cumulative rainfall is of particular importance as storms characterized by moderate wind speed but extreme rainfall become more common [8, 37]. Extreme precipitation rates may lead to more extensive flooding and deposition of water-transported sediments, which can limit soil oxygen availability, induce stomatal closure, and limit photosynthesis in mangroves, ultimately leading to mortality [38, 39]. These prolonged inundation events resulting from extreme precipitation may be further exacerbated as global sea level rises. In addition, storm surge can physically damage trees and exacerbate sediment deposition, causing delayed mortality [40]. However, the hydrological characteristics of tropical storms, including precipitation, storm surge, and changes in salinity, are challenging to quantify on large regional scales. Further study is needed to better understand how these hydrological drivers contribute to the effects on mangroves from tropical storms. Regardless, our results support redefining storm intensity to include characteristics other than just maximum wind speed, namely cumulative precipitation and storm surge.

Building on previous studies that have documented the effects of hurricanes on mangroves, our results reveal scale of these effects during a mega hurricane season; an example of what can be expected in the future as the global climate warms. Previous studies have shown many mangroves have been experiencing severe storms for centuries and are resilient to these disturbances, sometimes recovering in as few as 5 years [41]. However, recovery rates may depend on several factors including the nature and severity of the disturbance, mangrove stand characteristics, as well as interaction with other environmental factors [15, 42]. Given the vast majority of the mangrove damage we observed in 2017 was severe structural damage that persisted through the start of the 2018 hurricane season, recovery may be slow. The increased frequency of mega hurricane seasons has potential to compromise mangrove resilience and the associated ecosystem services mangroves provide. Specifically, an increasing number of mangrove stands will be in recovery stage, and although our results suggest these younger, shorter mangroves are more resistant to future damage, they may also provide different ecosystem services than mature mangroves [15, 29]. Furthermore, these changing disturbance dynamics could then interact with other global change processes, such as rising sea level, increased erosion, and altered nutrient cycling, to further degrade mangroves across large spatial scales [31, 43]. Future research should aim to quantify how ecosystems services are altered in mangrove stands regenerating in the wake of damaging hurricanes and to better understand factors affecting mangrove recovery trajectory. Decreased coastal protection and increased vulnerability of coastal communities in this tourism-dependent region will have yet unforeseen consequences [44].

Our analysis of satellite imagery using cloud-based computing proved essential to investigating temporal trends in hurricane-mangrove dynamics where large-scale field study is often precluded by remote and challenging field conditions [45]. The spatial distribution of mangroves are such that coarse resolution imagery may fail to capture vegetation dynamics in these systems [39] and multi-temporal analyses at fine scales (10 and 30 m) would have been computationally impractical without cloud computing resources, such as Google Earth Engine. Granted, spectral metrics such as NDVI may be biased by environmental changes independent of vegetation, such as surface water; however, our ex-ante and ex-post calculations of NDVI help to address these issues by incorporating 7 months of imagery in each composite image (SI appendix). Alternatively, decreases in NDVI may simply reflect defoliation and thus may not truly reflect changes in vegetation structure [46, 47], but our observation that 73% of damaged mangroves did not recover in the post-hurricane period illustrates that most of this damage was lasting. Moreover, the strong relationship we observed between NDVI and mangrove canopy height change provide further evidence that much of the damaged we observed was indeed structural (figure S2). These results are consistent with another recent study showed that structural damage from Hurricane Maria (2017) in Puerto Rican forests was greater than that observed following previous severe hurricanes [13]. Thus, leveraging contemporary remote sensing data and analysis tools can provide accurate measures of vegetation dynamics at fine scales and across large extents.

Similarly, the spatially explicit GMAO wind data allowed us to rigorously relate mangrove responses to wind across a large region. Though local analyses may incorporate measured wind speeds [48], such data are not available on a regional scale. Instead, regional analyses typically involve modelled wind speeds, for example by assuming wind speed varies as a function of overall storm intensity and the distance from the storm center [21, 49]. Alternatively, wind swaths such as those archived at the National Hurricane Center, provide wind speeds categorized according to coarse intervals, e.g. >119 km h⁻¹. In contrast to these data, the GMAO wind data provided accurate wind speeds that varied across space and time, allowing us to quantify the role of wind speed in predicting mangrove damage across a large region.

Though future projections of tropical cyclone activity are uncertain, studies consistently indicate that anthropogenic climate change will result in a global intensification of storms [4, 50]. Widespread hurricane-induced mangrove damage, such as that observed during the 2017 hurricane season, will likely result from this increased frequency of severe storms. These results are particularly important for the Caribbean/Mesoamerica region which is forecast to incur more damage than any other region from future hurricanes when considered as a proportion of gross domestic product [1]. Thus, the value of mangroves as buffers against future storms will become increasingly important to this region and across the tropics, and our results will aid in prioritizing conservation and restoration efforts.