High-Resolution Satellite Imagery to Assess Sargassum Inundation Impacts to Coastal Areas

A change detection analysis utilizing Very High-resolution (VHR) satellite imagery was performed to evaluate the changes in benthic composition and coastal vegetation in La Parguera, southwestern Puerto Rico, attributable to the increased influx of pelagic Sargassum spp and its accumulations in cays, bays, inlets and near-shore environments. Satellite imagery was co-registered, corrected for atmospheric effects, and masked for water and land. A Normalized Difference Vegetation Index (NDVI) and an unsupervised classification scheme were applied to the imagery to evaluate the changes in coastal vegetation and benthic composition. These products were used to calculate the differences from 2010 baseline imagery, to potential hurricane impacts (2018 image), and potential Sargassum impacts (2020 image). Results show a negative trend in Normalized Difference Vegetation Index (NDVI) from 2010 to 2020 for the total pixel area of 24%, or 546,446 m2. These changes were also observed in true color images from 2010 to 2020. Changes in the NDVI negative values from 2018 to 2020 were higher, especially for the Isla Cueva site (97%) and were consistent with the field observations and drone surveys conducted since 2018 in the area. The major changes from 2018 and 2020 occurred mainly in unconsolidated sediments (e.g. sand, mud) and submerged aquatic vegetation (e.g. seagrass, algae), which can have similar spectra limiting the differentiation from multi-spectral imagery. Areas prone to Sargassum accumulation were identified using a combination of 2018 and 2020 true color VHR imagery and drone observations. This approach provides a quantifiable method to evaluate Sargassum impacts to the coastal vegetation and benthic composition using change detection of VHR images, and to separate these effects from other extreme events.


Introduction
The floating brown algae Sargassum sp. provide an essential marine habitat in the open ocean. However, since 2011, there has been a dramatic increase in Sargassum biomass in the tropical Atlantic Ocean and Caribbean Sea and consequently, massive accumulations of Sargassum have been reported along the Eastern and Western Caribbean and Florida coasts [1,2]. From mid-2014 until the end of 2015, the Mexican Caribbean coast experienced a massive influx of drifting Sargassum that accumulated on the shores [3].
This build-up of decaying beach-cast material and nearshore accumulations produced murky brown waters that were called Sargassum-brown-tides (Sbt) with persistent reduction in light, oxygen (hypoxia or anoxia) and pH. In addition to the water quality impacts, they found that seagrass meadows dominated by Thalassia testudinum were replaced by calcareous rhizophytic algae and drifting algae and/or epiphytes, which resulted in a 61.6-99.5% loss of below-ground biomass and reduced value to coastal blue carbon storage. High levels of eutrophication were still present after one year of the Sbt and near-shore corals suffered total or partial mortality.
Remote sensing have been employed by many researchers to map general benthic habitat types (e.g., sand, seagrass, coral reefs, hard substrate) in coral reef environments [4,5,6] and coastal vegetation [7,8] in Puerto Rico. Also, coral reef habitat maps, based on remotely sensed data, are a fundamental tool for management because they summarize ecologically meaningful information across extensive geographic scales in a costeffective manner [9]. In addition, changes to the vegetation coverage from extreme events (e.g. hurricanes) have been documented using remotely sensed data for the estimation of NDVI [7,8].
Initial observations in La Parguera area (SW Puerto Rico) suggest that Sargassum influx and accumulations have increased since 2011, with a significant increase in the peak years of Sargassum influx to the Caribbean [1]. Additionally, original benthic cover and composition, as well as coastal vegetation may have been displaced and/or significantly impacted by persistent Sargassum accumulations in near-shore areas. The purpose of this study was to evaluate the changes in benthic composition and coastal vegetation in La Parguera area due to Sargassum influx and accumulations in cays, bays, inlets and nearshore environments.

Study Area
The study area in La Parguera was selected due to the observed increased frequency of events resulting in the accumulation of Sargassum at some of the coral reefs and mangrove-fringed coasts. The constraints to hydrodynamic activity, provided by reefs and mangrove keys in the study area, coupled to the prevalence of south-easterly winds makes the study area particularly vulnerable to Sargasso accumulation The benthic and coastal vegetation area includes the Boquerón State Forest and La Parguera Natural Reserve, managed by the Puerto Rico Department of Natural and Environmental Resources (DNER) (Figure 2). The area of La Parguera is recognized for the exceptional value of its marine resources, which includes extensive coral reef ecosystems, seagrass beds, coastal mangrove fringe and mangrove islands, and two bioluminescent bays [10].

Materials and Methods
A very high-resolution (VHR) benthic composition and coastal vegetation map, including fringing mangroves, at the study site was developed from satellite imagery to evaluate the changes in benthic composition in La Parguera Marine Reserve potentially attributable to Sargassum influx and accumulations in cays, bays, inlets and near-shore environments. A previous benthic composition map was used as the baseline map for La Parguera [6]. This baseline benthic map was developed from VHR satellite imagery and field observations before the Sargassum influx that started in 2011.
A VHR satellite image was obtained for 2020 for the selected sites to evaluate the effect of Sargassum accumulation on benthic composition and coastal vegetation. To minimize the potential effects of the 2017 hurricane season on the analysis, an additional VHR image (early 2018) was also acquired. The satellite imagery selection was based on the availability of cloud-free data, timeline of the observation (e.g. after Sargassum season, after hurricane impacts), and spatial/spectral resolution of the sensors. The imagery selected and their attributes are described in Table 1. Imagery was pre-processed and ancillary data such as, LiDAR data (bathymetry), drone

Pre-Processing
The images were corrected for radiometric and atmospheric effects and were also pansharpened to improve the spatial resolution of the multispectral bands. Land and water masks were developed and applied to the images to separate the effects on coastal vegetation or benthic cover, respectively and provide and independent analysis of the changes.
The Geoeye-1 imagery was acquired from National Oceanographic and Atmospheric Additional processing included a Dark Object Subtraction atmospheric correction [11] creating a mosaic of the tiles and a subset of the area of interest ( Figure 3).
The Pleiades-1 images were acquired, radiometrically and atmospherically corrected using Dark Object Subtraction atmospheric correction [11] in ENVI 5.3 image processing software. A Gram-Schmidt pan-sharpening was applied to the imagery by including the panchromatic band with ground control points. Processing was completed for the January 06, 2018 and January 19, 2020 imagery to obtain a multi-spectral image at 0.55m spatial resolution ( Figure 3).

Georeferencing
The processed GeoEye-1 image was used as the base image for georeferencing. The base image orthorectification was performed using ENVI's Orthorectify GeoEye-1 with Ground Control module. Due to the difference in spatial resolution, the GeoEye-1 imagery was resampled to the Pleaides-1 spatial resolution of 0.55m. A subset of the Pleaides-1 imagery was developed based on the extent of the GeoEye-1 image.

Change Detection for Vegetation
Various techniques were used to evaluate changes in the imagery over time. For the coastal vegetation, changes in the NDVI index was used to quantify these changes. The NDVI was used here to assess the state of live green vegetation and it also indicates a level of photosynthetic activity [12]. The NDVI is calculated as follows: where RED and NIR are the spectral channels (bands) for the sensors in the red (visible) and near-infrared (NIR) regions, respectively. For both the Pleaides-1 and GeoEye-1 images, these correspond to the third and fourth bands. In general, the range of NDVI values are between -1.0 and +1.0, where negative values can represent water. A water mask was applied to the reduce the effects of water on the NDVI index and classification.
Once the vegetation index has been processed, the new images can be subtracted from the baseline image using map algebra to evaluate the changes [7]. This was done for the 2018 and 2020 images. The NDVI pixel values were then reclassified as positive change or negative change categories if pixel values were higher (positive) or lower (negative) than the baseline image. Once the NDVI changes from the baseline were calculated, a difference image was processed between 2018 and 2020 images to evaluate NDVI changes between these dates. To calculate the net percent change of NDVI that can be attributed to the potential hurricane impacts (2018 image), or the potential Sargassum impacts (2020 image), the difference between 2010-2018-and 2010-2020-pixel count was calculated for each study area.

Change Detection for Benthic Composition
For the benthic composition changes, an unsupervised classification (ISODATA) as an initial evaluation, and an object-based segmentation and further classification was used to quantify the changes using all bands. A land mask was applied to the image to reduce the effects of land features on the classification. The classes and iterations were adjusted until the segmentation of benthic features was achieved. Areas with sunglint and land features still visible were grouped with the mask class. A preliminary map was obtained where major classes and major cover were classified based on the benthic classes and cover from baseline benthic cover map [6]. This initial classification step and draft benthic composition maps were produced for both 2018 and 2020 images. These preliminary benthic maps were then assigned benthic classes based on; 1) baseline survey from previous benthic map, 2) updates from boat surveys and drone images and videos for selected sites. No contextual editing [4] was applied to the segmented benthic composition polygons since the focus of the project was to quantify changes from the baseline image. Final maps classes were compared to evaluate changes in the total area by major classes and coverage from the baseline benthic maps with special interest in the potential changes in benthic cover by hurricanes (2018) and Sargassum (2020).

Results
Coastal vegetation and benthic composition maps were developed to quantify changes in the study area potentially attributable to Sargassum influx and accumulations in cays, bays, inlets and near-shore environments.

Changes in Coastal Vegetation
The processed images were evaluated for change detection using NDVI products. These  The net % change of NDVI was calculated for each study area that can be attributed to the potential hurricane (2018 image) or Sargassum impacts (2020 image) ( Figure 5).

Changes in Benthic Composition
Final benthic composition maps were produced for both 2018 ( Figure 6) and 2020 images ( Figure 7), while 2010 image and NOAA benthic map are included for reference ( Figure   8).

Discussion
The coastal vegetation shows a negative trend in NDVI from 2010 to 2020 and these changes were also observed in the corresponding true color images. The major changes to the benthic composition occurred to the Sand and Mud categories, while the Major Cover category changed mainly between Seagrass and Algae.

Changes in Coastal Vegetation
The NDVI index analysis shows a negative trend from 2010 to 2020. The total pixel area that showed a negative change in values was 546,446 m 2 , or 24% of the total area for the sites from 2010 to 2020. These changes can also be observed in the true color images from the same time period (Figure 3). from the storm center and with lower elevation were the least affected [7]. This information is confirmed by the authors [14] that shows that mangroves were severely impacted from Hurricane Maria in the eastern part of the island when compared to other areas. Also, canopy coverage from mangrove forest recover to 60% pre-hurricane conditions between 3-6 months post-storm for Hurricane Irma impacts in Florida [15].
The image used for our NDVI analysis was from January 2018 which suggest that NDVI values and canopy cover were back to pre-hurricane levels, and that negative NDVI are only showing the areas severely impacted by the hurricane.
Several limitations can be found when using NDVI for change detection analysis that include atmospheric effects, new leaf growth in crops, and seasonality [12]. However, these limitations were minimized in the image pre-processing. A water mask was applied to the images to remove any potential NDVI negative values from water to the pixels. In addition, analysis of changes was completed from the baseline image to ensure a standardization of the differences for the 2018 and 2020 images. In addition, the 2018 and 2020 images were selected from the same period (January) to remove any differences in seasonality that might affect the "greenness values" that are obtained from the NDVI.

Changes in Benthic Composition
Benthic composition maps were developed for 2018 and 2020 imagery to evaluate potential impacts to the benthic community structure from both hurricanes and Sargassum. The main changes to the Major Class category occurred to the Sand and Mud categories, while the Major Cover category changed mainly between Seagrass and Algae.
No major changes were observed to the No Cover class from 2018 to 2020 ( Figure 10).
The major changes from 2018 and 2020 occurred mainly in unconsolidated sediments (e.g. Sand, Mud) and submerged aquatic vegetation (e.g. seagrass, algae), both of which can have similar spectra and can be very difficult to differentiate from multi-spectral imagery [16]. Due to these considerations, the changes from 2018 and 2020 cannot be considered relevant since the No Cover class remained relatively similar for the time period. Benthic structure impacts from storm surge of hurricane Maria and Irma should have been minimum, due to the storm track and distance from study area [7]. Also, the Puerto Rico central mountain chain can create a shade effect, that may have provided additional protection from the storm surge at out sites [17].
Using the 2018 and 2020 true color VHR imagery combined with the drone observations, we could locate the areas where the Sargassum accumulated, decomposed, and deposited ( Figure 11). These selected areas at Isla Cueva and Isla Guayacán show the changes in accumulation of Sargassum from 2018 to 2020, which can also be identified in the drone imagery from September 2019. The accumulation of Sargassum persisted and was visible in the 2020 satellite imagery even after the end of the 2019 Sargassum season.
The decomposed Sargassum in the benthos was not captured by the benthic classification, probably due to the lack of differentiation from unconsolidated sediments like mud.
Quantifying the benthic changes presented some additional challenges due to the water column effect on the signal received by the sensors. Additional factors such as sunglint Our study found major changes to the coastal ecosystems of La Parguera (e.g. fringing mangroves and seagrass) which it's interdependence results in a highly productive and biodiverse marine resource [18]. Hurricane impacts to coastal vegetation and benthic ecosystems have been documented [15,7], but these can still be considered episodic events providing time for the ecosystem to recover. However, the cumulative and chronic impacts of Sargassum [3] may not provide a recovery time to these coastal ecosystems, especially if the system has been severely impacted by a hurricane or storm. The changes to the coastal vegetation and benthic composition we observed in our study occurred in a short time frame, considering that both Sargassum [1] and extreme weather events such as storms and hurricanes [19,20] will continue and potentially increase, natural resource managers need to consider the combination of these scenarios and its potential long-term impacts to the ecosystems.

Conclusions
Significant changes in coastal vegetation from 2010 to 2020 were quantified in the study sites in La Parguera area using the NDVI and it was possible to differentiate between the hurricane and Sargassum impacts. Further studies should focus on how these accumulations may have a detrimental effect in the biochemistry and hydrodynamics of the mangrove root system, which may result in loss of mangrove forest habitat. Although we quantified that the observed percent increase in the NDVI negative values could be due to Sargassum, other factors like hydrological changes to the mangrove forests may have also contributed to these changes. Debris accumulations, sediments redistribution or wind damage to canopy structure [14,15] can have a cumulative effect, which cannot be resolved with the temporal scales used in this study.
Benthic composition maps were completed for the 2018 and 2020 imagery and were used to evaluate the potential changes from 2010 to 2020. However, similarities in the spectral characteristics of the substrate limited the reliability of the retrievals and could not differentiate the accumulated Sargassum in the benthos. Drone surveys using multispectral cameras combined with optical field surveys may provide a more detailed optical information and the high-resolution and temporal scale needed.
This approach provides a quantifiable method to evaluate Sargassum impacts to the coastal vegetation and benthic composition using change detection of VHR images, and to separate these effects from other extreme events.